preface - jherjher.org/archive/vol2/jher vol 2 full volume.pdf · we thank md mahmudul haque for...
TRANSCRIPT
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
Preface
________________________________________
This is the second volume of Journal of Hydrology and Environment Research (JHER). The JHER is a peer reviewed
international journal, which publishes high quality research papers in the fields of hydrology, water and environment. The
JHER is published from Sydney, Australia.
This volume of JHER contains 7 technical papers and one review paper. The 1st paper focuses on Priestley-Taylor
parameter of wet surface evaporation. The 2nd
paper deals with estimation of water surface elevation on inundated area
using satellite data at lower Mekong Basin. The 3rd paper examines the impacts of outliers in flood frequency analysis. The
4th paper presents a new watershed codification system for Indian river basins. The 5
th paper provides an assessment of
heavy metal contamination from municipal solid waste sites in Bangladesh. The 6th paper presents impacts of rating curve
error on flood quantile estimation. The 7th paper presents the challenges in modelling a large river basin with scarce data.
The 8th paper presents a review on uncertainty in design rainfall estimation.
We acknowledge greatly the reviewers who have spent their valuable time in reviewing the papers contained in this
volume. We thank Md Mahmudul Haque for editorial coordination. We also thank Mr Tauqir Ullah and Mr Imran
Rahman for assistance with the cover page design and website maintenance. We welcome papers for 3rd volume which will
be published in mid-2015. Authors are advised to visit website of JHER (www.jher.org) for information on preparation and
submission of manuscripts for possible publication in this journal.
Associate Professor Ataur Rahman, PhD
Editor
Journal of Hydrology and Environment Research
Associate Professor (Water and Environmental Engineering)
School of Computing, Engineering and Mathematics, University of Western Sydney, New South Wales, Australia
The publisher does not take any responsibility of the contents of a paper in this journal. Before using information published in this journal in real design and practice, contents/materials must be verified by the user at his/her own responsibility. The publisher does not take any liability
or loss incurred by a user who has used information from this journal.
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
Editorial Board
Editor: Associate Professor Ataur Rahman, PhD, Fellow (Engineers Australia), Member IWA, Member AGU, Member AWA
Associate Editors: Professor Dr Chi Zhang, Dalian University of Technology, China
Dr Khaled Haddad, University of Western Sydney, Australia
Professor Dr Muthiah Perumal, Indian Institute of Technology, Roorkee, India
Dr Amimul Ahsan, University Putra Malaysia, Malaysia
Dr Fazlul Karim, CSIRO Land and Water, Australia
Editorial Coordinator: Md Mahmudul Haque, MScEng
International Advisory Board: 1. Professor Dr Salaheddine El Adlouni, Université de Moncton, Canada
2. Associate Professor Dr Ramesh Teegavarapu, Florida Atlantic University, USA
3. Dr Tony Ladson, Australia
4. Professor Dr Zulkifli Yusop, University Technology Malaysia, Malaysia
5. Dr Douglas Bertram, University of Strathclyde, UK
6. Professor Dr Abdallah Shanableh, University of Sharjah, United Arab Emirates
7. Professor Dr R B Singh, University of Delhi, India
8. Professor DR Kadri Yurekli, Gaziosmanpasa University, Turkey
9. Associate Professor Dr Arumugam Sathasivan, University of Western Sydney, Australia
10. Associate Professor Dr Mehdi Yasi, Urmia University, Iran
11. Associate Professor Dr Mushtaque Ahmed, Sultan Qaboos University, Oman
12. Professor Dr Fawzi S Awad, King Saud University, Saudi Arabia
13. Professor Dr Md Rezaul Karim, Islamic University of Technology, Bangladesh
14. Dr Namrata Pathak, University of Delhi, India
15. Associate Professor Dr Monzur Imteaz, Swinburne University of Technology, Australia
16. Professor Dr Mahmoud Mohamed Hegazi, Ain Shams University, Egypt
17. Professor Dr Ata-ur-Rehman Tariq, University of Engineering and Technology, Pakistan
18. Dr John OSullivan, National University of Ireland, Dublin, Ireland
The publisher does not take any responsibility of the contents of a paper in this journal. Before using information published in this journal in
real design and practice, contents/materials must be verified by the user at his/her own responsibility. The publisher does not take any liability
or loss incurred by a user who has used information from this journal.
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
List of Reviewers
Professor Dr Rezaul Karim, PhD
Associate Professor Dr Ataur Rahman, PhD
Dr Khaled Haddad, PhD
Dr Wilfredo Caballero, PhD
Dr Fazlul Karim, PhD
Dr Md Al-Amin, PhD
Md Mahmudul Haque, MScEng
Mr Vijay Kumar Eppakayala, MEng
Mrs Orpita Urmi, MEng
Mr Faruk Kader, MEng
Mrs Evan Hajani, MEng
Mr Md Jalal Uddin, MEng
Peer Review Process: Each of the published papers has been reviewed independently. Papers are revised as per
reviewers’ comments before being finally accepted for publication.
We acknowledge the reviewers greatly for their valuable time in reviewing the papers.
The publisher does not take any responsibility of the contents of a paper in this journal. Before using information published in this journal in real design and practice, contents/materials must be verified by the user at his/her own responsibility. The publisher does not take any liability
or loss incurred by a user who has used information from this journal.
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
Table of Contents
Technical papers:
Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data 1
J. Szilagyi, M. B. Parlange, G. G. Katul
Estimation of Water Surface Elevation on Inundated Area Using Satellite Data 8
A. Yorozuya, H. Kamimera, T. Okazumi, Y. Iwami, Y. Kwak
Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia 17
A. S. Rahman, K. Haddad, A. Rahman
New Watershed Codification System for Indian River Basins 31
K. Pareta, U. Pareta
Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in 41
Bangladesh
M. R. Karim, M. Kuraoka, T. Higuchi, M. Sekine, T. Imai
Rating Curve Uncertainty in Flood Frequency Analysis: A Quantitative Assessment 50
M. M. Haque, A. Rahman, K. Haddad
Challenge on Modelling a Large River Basin with Scarce Data: A Case Study of the 59
Indus Upper Catchment
A. Sugiura, S. Fujioka, S. Nabesaka, T. Sayama, Y. Iwami, K. Fukami, S. Tanaka, K. Takeuchi
Review Paper: Uncertainty in Design Rainfall Estimation: A Review 65
A. A. Mamoon, A. Rahman
The publisher does not take any responsibility of the contents of a paper in this journal. Before using information published in this journal in real design and practice, contents/materials must be verified by the user at his/her own responsibility. The publisher does not take any liability
or loss incurred by a user who has used information from this journal.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 1
Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global
Reanalysis Data
Jozsef Szilagyi 1,*
, Marc B. Parlange 2, Gabriel G. Katul
3
1 Department of Hydraulic and Water Resources Engineering, Budapest University of Technology and Economics, Budapest, Hungary; also at
School of Natural Resources, University of Nebraska-Lincoln, 3310 Holdrege St., Lincoln, NE, 68583, USA 2 Department of Civil Engineering, University of British Columbia, Vancouver, Canada
3 Nicholas School of Environment and the Earth Sciences, Duke University, Durham, North Carolina, USA
Peer Review History1
Abstract: The Priestley-Taylor parameter α of wet surface evaporation is investigated using daily, 0.75-degree ERA-Interim reanalysis data for
the winter hemispheres of 2000-2001, 2006-2007, and 2012-2013. Published ERA-Interim sensible and latent heat fluxes over sea and land yield
two distinct best-fit curves for α as a function of air temperature (Ta). When the wet land surface temperature (Tws) was estimated by an
independent method, the two curves largely collapsed. Tropical land areas with low wind formed a subgroup of α distribution, yielding the
lowest overall values of 1.06±0.03. The results for sea corroborate the widely accepted α value of 1.26±0.06 when Ta > 20 °C. At 0 °C the mean
α value over sea is about 1.62±0.23. The α values for wet land surfaces with the independent Tws estimates display about the same mean but
with larger variations. Published values of α scatter around the overlapping α vs Ta curves.
Keywords: Evapotranspiration, land/atmosphere interactions, water/energy interactions, Priestley-Taylor parameter, wet surface temperature
1. Introduction
Following the publication of the Priestley and Taylor (1972) equation describing wet environment evaporation under minimal horizontal
advection of energy, many studies focused on relating the value of its empirical coefficient, α, to different environmental variables (Pereira,
2004). The coefficient α in the Priestley-Taylor equation (PTE):
nLE Q
(1)
is generally accepted to express the evaporation-enhancing effect of large-scale entrainment of drier free-tropospheric air resulting from the
growing daytime convective boundary layer (CBL) (Brutsaert, 1982; deBruin, 1983; Culf, 1994; Lhomme, 1997; Heerwaarden et al., 2009). Here
Qn is the available energy at the wet surface equivalent to the sensible heat (H) and latent heat (LE) fluxes, Δ is the slope of the saturation
vapor pressure curve at the air temperature (Ta), and γ (= cpP / (0.622L)) is the psychrometric constant, where cp is the specific heat of air at
constant pressure (P) and L is latent heat of vaporization for water. From observations over both sea and extended wet land surfaces, Priestley
and Taylor (1972) reported that such large-scale entrainment typically enhanced evaporation by about 20-30% in comparison to what would
result from saturated air, yielding a mean value of 1.26 for α. In subsequent studies, the value of α has been found to vary considerably on a
1 Paper JHER0206, submitted on 19/11/2014, accepted for publication after peer review and subsequent revisions on 15/12/2014
* Corresponding author may be contacted at [email protected]
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 2
sub-daily (Yu, 1977; de Bruin and Keijman, 1979; Viswanadham et al., 1991; Parlange and Katul, 1992), daily (Davies and Allen, 1973), and
seasonal basis (de Bruin and Keijman, 1979).
The limited number of experimental data and the sometimes conflicting results on the value of α motivated the present study of investigating
its possible distribution based on data with a global coverage, such as the daily ERA-Interim reanalysis dataset
(http://www.ecmwf.int/products/) at a 0.75° spatial resolution. The significance of the Priestley-Taylor equation (equation 1) at a temporal scale
of a day (or longer) in hydrology, water resources and their related fields is important. It forms the backbone of numerous classical evaporation
estimation methods such as the soil moisture (Davies and Allen, 1973; Spittlehouse and Black, 1982; Chen and Brutsaert, 1995) or
complementary relationship based techniques (Brutsaert and Stricker, 1979; Morton et al., 1985; Parlange and Katul, 1992; Szilagyi et al., 2009)
as well as remote-sensing enhanced approaches such as the two-source models (Anderson et al., 2008; Kustas and Anderson, 2009). In all these
models PTE is employed with a preset α =1.26. Considering that α has a well-defined lower limit of unity (i.e., when the air is saturated and
equilibrium profiles of Ta and specific humidity, q, exist) for extended wet surfaces, as well as an upper limit of 1 + γΔ-1, a function of Ta (i.e.,
when air stratification is adiabatic, thus Qn = LE), the present study focuses on defining the α-value distribution in relation to Ta as well. The
results below will help future practical evaporation estimations of the above models by enabling prescription of the α value as a function of Ta,
and thus improving their overall performance.
2. Estimation of the α parameter value from daily reanalysis data
Reanalysis data are considered as the best representation of reality because they combine measurements with modeling results by taking into
account the errors in both of them. The European Centre of Medium-Range Weather Forecasts (ECMWF) has been producing reanalysis data
since 1979. The latest such product, the ERA-Interim reanalysis dataset, is available free of charge near real time since 2009 at a spatial
resolution of 0.75°. For the present study, daily and monthly 2 m Ta and dew point (Td) temperature, surface P, 10 m wind velocity (u10), net
radiation (Rn), as well as ECMWF-estimated skin temperature (Ts), H, and LE fluxes were downloaded for the winter periods (i.e., the months
of December, January, February for the northern hemisphere and June, July, August, for the southern one) of 2000-2001, 2006-2007, and 2012-
2013. With the choice of the winter season the strong advection effect of the trade winds were meant to be minimized. The monthly values
served only as a check of the ensuing daily analysis and yielded similar results. ECMWF-published skin temperature, Ts, is not identical to the
radiometric skin temperature. Ts, by virtue of its derivation, should be conceived as an aerodynamic surface temperature.
The α value can be obtained by rearrangement of equation 1 as:
( ) /
Bo 1
(2)
where Bo (= H / LE), is the Bowen ratio. Equation 2 can also be written as (Priestley and Taylor, 1972):
q p z z
n q z p z a z z a
( c / L)Ld q ( )d eLE
Q / ( ) (Ld q c d T ) (d e d T )
(3)
where Δq is the slope of the saturation specific humidity curve, e is vapor pressure and dz denotes the vertical difference in the variable. For
saturated surfaces Eichinger et al. (1996) gave an approximation of equation 3 as:
1*
*
s
(e e)1
( )(e e)
(4)
with saturation vapor pressure values starred and the subscript ‘s’ referring to the surface temperature. Equation 4 requires the same input as
equation 3 and its overall performance is also similar, therefore only results from equation 3 are published below, by assuming saturated
conditions at the surface.
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 3
The value of α was calculated separately for sea and land by equations 2 and 3 employing the ERA-Interim sea-land mask. The spatial extent of
the 0.75° degree cells was considered large enough to be applicable with the PTE. Ts over land was estimated by ECMWF (2007) for obtaining
the H and LE fluxes through a soil-moisture dependent resistance approach. The surface temperature values (Tws) over land, to be used in
equation 3 under assumed surface saturation conditions, have been estimated independently (Szilagyi, 2014) by relating the aerodynamic
resistance of Monteith (1981) to Penman’s (1948) Rome wind function and transforming u10 to the required 2 m value via u2 = u10 0.21/7
(Brutsaert, 1982).
Figure 1 displays the relative histograms of α, as a function of Ta for cells that yielded α values between unity and 1 + γΔ-1. The distribution of
α from H, LE fluxes is much wider over land (1b) than sea (1a). Direct application of the ECMWF-estimated Ts values in equation 3, assuming
surface saturation [in contrast to equation 2], yields a significantly different α distribution (1c), with higher overall α values.
Figure 1 Relative histograms of the Priestley-Taylor α values calculated from mean daily ERA-Interim values of sensible- (H) and latent heat (LE) fluxes
(1a,b), air (Ta), skin (Ts) as well as dew point (Td) temperature values (1c). Wet surface temperature (Tws) has been independently estimated (Szilagyi, 2014)
in 1d. Bin-size is 0.02 °C by 0.02, n is the total number of α values found for the winter hemispheres of 2000-2001, 2006-2007, and 2012-2013 combined,
within the (1, 1 + γΔ-1) limits.
Assuming that the ECMWF-derived fluxes and the Ts values are correct, this discrepancy can only exist if the land surface is not always
saturated when the resulting α value falls between the lower and upper bounds for saturated surfaces. With the independently obtained Tws
values; however, the resulting distribution (1d) is in between the former two, both in spread and location, with a new feature: the emergence of
a distinct sub-group of very low α (< 1.08) values. The highest frequency of days with α < 1.08 is predominantly found in the western part of
the Amazon basin (reaching 79 out of a possible 92 days for the winter of 2000-2001) and in Indonesia (Figure 2), both near mountains,
suggesting that large-scale advection is the weakest in these areas. Viswanadham et al. (1991) also reported a mean daily α value of 1.03 in the
western part of the Amazon basin for the end of the austral winter season. Quite interestingly, these regions correspond to the wettest and
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 4
calmest tropical regions of the world, with mean annual precipitation in excess of 2500 mm and mean wind velocities less than 2.7 ms-1 (source:
climate-charts.com), providing strong support for the derived low α values. Note that over sea such low values cannot occur due to much
stronger winds.
Figure 2 Spatial distribution of the number of winter days (2000-2001) when α over land fell within the interval of (1, 1.08) in Figure 1d. The
highest values correspond to the most humid and least windy areas of the tropics having annual precipitation in excess of 2500 mm and mean
wind velocities lower than 2.7 ms-1 (source: climate-charts.com).
Figure 3 displays the third-order polynomial curves (Table 1) fitted to 15 bin-means of α and to their plus/minus standard deviation (std) values
(dashes). The value of α over sea decreases from a value of 1.62±0.23 at 0°C to a value of 1.22±0.03 at 30 °C. Between 20 and 30 °C, the typical
mean daily summer temperature range for mid-latitudes when LE generally is the highest over land, α stays between 1.29±0.07 and 1.22±0.03,
agreeing well with the original findings of Priestley and Taylor (1972).
Table 1 Parameters (with decreasing power) of the 3rd
-order polynomials of Figure 3, fitted to 15 bin means of α values and to their plus/minus
standard deviations
( ∙ 10-6) (∙ 10
-4) (∙ 10
-2)
LE, H, sea -3.89, -0.89, -6.9 4.78, 5.95, 3.61 -2.54, -3.85, -1.2 1.64, 1.89, 1.39
LE, H, land -4.84, -4.36, -5.31 7.07, 10.1, 3.99 -2.96, -4.85, -1.07 1.51, 1.86, 1.16
Ta, Ts, Td, sea 1.57, -0.85, 4 2.55, 5.64, -0.53 -2.22, -3.66, -0.8 1.61, 1.86, 1.36
Ta, Ts, Td, land -0.7, -14.1, 12.7 7.62, 18.1, -2.91 -4.53, -7.44, -1.62 1.93, 2.25, 1.6
Ta, Td, indep. Tws, land 38.5, 20.8, 56.1 -10.8, -1.39, -20.3 -1.65, -3.98, 0.67 1.69, 2.03, 1.36
Ta, Td, indep. Tws, land, α > 1.08 26.2, 16.9, 35.5 -6.29, 0.28, -12.8 -2.03, -4.15, 0.08 1.71, 2.03, 1.38
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 5
For land, equation 2 with ECMWF fluxes yields an α vs Ta curve distinctly lower than that for sea, while equation 3 with Ts values and under
the assumption of surface saturation, yields another, markedly higher curve, especially when Ta < 15 °C. On the other hand, when the currently-
estimated Tws values are employed, the land and sea curves largely overlap. The overlap improves with the exclusion of the extremely low α
value sub-group in Figure 1d from the fitting. In agreement with Priestley and Taylor (1972) who mixed sea and land measurements, the Bowen
ratio, and thus the α value, regulated by entrainment of free-tropospheric air at the top of the CBL, should not in general differ between sea
and wet land surfaces under largely similar environmental conditions, since at the top of the CBL the influence of the land surface is less
significant. However, when Bo and α are indeed markedly different, then the environmental conditions themselves are significantly different, as
was found for the wind in the α-value subgroup of 1d.
Figure 3 Third-order polynomials fitted to 15 bin means (at Ta = 1, 3, … °C) of ERA-Interim α values (with the corresponding standard deviation values
denoted by dashes) found for the winter hemispheres of 2000-2001, 2006-2007, and 2012-2013 combined, within the (1, 1 + γΔ-1) limits.
The overlap of the sea and land α vs Ta curves is a strong indication of the saturation of the land surface when the α value, resulting from
equation 3 by assuming such saturation, falls between the limits, derivable for wet surfaces. Szilagyi and Schepers (2014) demonstrated that Tws
is invariant to drying of the environment under largely unchanged net radiation and wind conditions. Therefore, the estimated Tws values could
possibly come from drying conditions of the vegetated surface. But then it would be extremely fortuitous to obtain α values from equation 3
that largely coincide with the sea values from such drying land conditions (to be reflected in the measured 2 m Ta and Td values) with the
simultaneous ‘false’ assumption of a saturated surface. It is much more likely that the surface is indeed saturated whenever the derived α value
(by equation 3) falls between unity and 1 + γΔ-1 through the application of the corresponding Tws value.
Figure 4 displays published α values when the corresponding Ta values were available as well. With one single exception, the values fit well
within the corresponding plus/minus std regions, and they are above the ERA-Interim H, LE derived α vs Ta curve for land, in support of the
existence of a single α curve for both sea and land, although with larger variance for the latter.
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 6
Figure 4 Subset of Figure 2, combined with reported values of α as a function of mean daily (or monthly) air temperature. Measurements over
sea are in black, over land in red, and over a shallow lake in blue. The lines between the symbols denote reported intervals of α values as a
function of Ta. See the List of References for the relevant publications.
3. Summary
Using daily ERA-Interim reanalysis data the value of the Priestley-Taylor parameter α with the corresponding interval of standard deviation, has
been obtained as a function of Ta. Via the application of independently obtained wet surface temperature values, Tws, the results support the
existence of a single curve [hypothesized by Priestley and Taylor (1972)] for sea and extensive wet land surfaces, contrary to what is obtainable
from ECMWF-derived LE, H fluxes or directly from Ts values of the same source, by simultaneously assuming saturation at the land surface.
As expected, variability of the derived α is larger for land than for sea (both increasing with decreasing temperatures), due to a higher degree of
inhomogeneity in surface properties determined by topography, soil, and vegetation characteristics. At about 0 °C the average value of α is
around 1.6-1.7, while over 20 °C it is found between 1.2 and 1.3, in agreement with the original, Priestley and Taylor (1972) derived average
value of 1.26. In extreme environmental conditions, such as the most humid and least windy tropical land areas in the western part of the
Amazon basin (as well as in Borneo and Sumatra), the α value may frequently reach as low a value as 1.03, close to its theoretical lower limit of
unity.
Most of the published α values fall above or below the two distinct α vs Ta curves for land obtainable by daily ECMWF LE, H or Ts,
respectively, while they do scatter around the sea α vs Ta curve and the largely overlapping land curve, the latter obtained by independently
derived Tws values.
4. Acknowledgements
This work has been supported by the Agricultural Research Division of the University of Nebraska-Lincoln. The data used for this study are
available from http://old.ecmwf.int/products/.
“Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data” Szilagyi et al.
Journal of Hydrology and Environment Research 7
References Anderson MC, Norman JM, Kustas WP, Houborg R, Starks PJ, Agam N (2008). A thermal-based remote sensing technique for routine
mapping of land-surface carbon, water and energy fluxes from field to regional scales, Remote Sensing of Environment, 112(12), 4227-4241.
Brutsaert W (1982). Evaporation into the atmosphere: Theory, history and applications, D. Reidel, Dordrecht, Holland.
Brutsaert W, Stricker H (1979). An advection-aridity approach to estimate actual regional evapotranspiration, Water Resources Research, 15,
443-449.
Chen D, Brutsaert, W (1995). Diagnostics of land surface spatial variability and water vapor flux, Journal of Geophysical Research, 100(D12),
25595-25606.
Culf AD (1994). Equilibrium evaporation beneath a growing convective boundary layer, Boundary-Layer Meteorology, 70, 37-49.
Davies JA, Allen CD (1973). Equilibrium, potential, and actual evaporation from cropped surfaces in southern Ontario, Journal of Applied
Meteorology, 12, 649-657.
de Bruin HAR (1983). A model for the Priestley-Taylor parameter α, Journal of Climatology and Applied Meteorology, 22, 572-580.
de Bruin HAR, Keijman JQ (1979). The Priestley-Taylor evaporation model applied to a large, shallow lake in the Netherlands, Journal of
Applied Meteorology, 18, 898-903.
Eichinger WE, Parlange MB, Stricker H (1996). On the concept of equilibrium evaporation and the value of the Priestley-Taylor coefficient,
Water Resources Research, 32(1), 161-164.
European Centre for Medium-Range Weather Forecasts (ECMWF) (2007), IFS documentation–Cy31r1, Part IV: Physical Processes, Shinfield
Park, Reading, England.
Heerwaarden CC, Arellano JV, Moene AF, Holtslag AAM (2009). Interactions between dry-air entrainment, surface evaporation and convective
boundary-layer development, Quarterly Journal of the Royal Meteorological Society, 135, 1277-1291.
Kustas W, Anderson M (2009). Advances in thermal infrared remote sensing for land surface modeling, Agricultural and Forest Meteorology,
149(12), 2071-2081.
Lhomme JP (1997). A theoretical basis for the Priestley-Taylor coefficient, Boundary-Layer Meteorology, 82, 179-191.
Monteith JL (1981). Evaporation and surface temperature, Quarterly Journal of the Royal Meteorological Society, 107, 1-27.
Morton FI, Ricard F, Fogarasi F (1985). Operational estimates of areal evapotranspiration and lake evaporation—program WREVAP, National
Hydrologic Research Institute Paper No. 24, Ottawa, Canada.
Parlange MB, Katul GG (1992). Estimation of the diurnal variation of potential evaporation from a wet bare soil surface, Journal of Hydrology,
132, 71-89.
Parlange MB, Katul GG (1992). An advection-aridity evaporation model, Water Resources Research, 28(1), 127-132.
Penman HL (1948). Natural evaporation from open water, bare soil, and grass, Proceedings of the Royal Society, London A193, 120-146.
Pereira AR (2004). The Priestley-Taylor parameter and the decoupling factor for estimating reference evapotranspiration, Agricultural and
Forest Meteorology, 125, 305-313.
Priestley CHB, Taylor RJ (1972). On the assessment of surface heat flux and evaporation using large-scale parameters, Monthly Weather
Review, 100(2), 81-92.
Spittlehouse DL, Black TA (1981). A growing season water balance model applied to two Douglas fir stands, Water Resources Research, 17,
1651-1656.
Szilagyi J (2014). Temperature corrections in the Priestley-Taylor equation of evaporation, Journal of Hydrology, 519, 455-464.
Szilagyi J, Schepers A (2014). Coupled heat and vapor transport: the thermostat effect of a freely evaporating land surface, Geophysical Research
Letters, 41, 435-441, doi:10.1002/2013GL058979.
Szilagyi J, Hobbins MT, Jozsa J (2009). A modified Advection-Aridity model of evapotranspiration, Journal of Hydrologic Engineering, 14(6),
569-574.
Viswanadham Y, Silva Filho VP, Andre RGB (1991). The Priestley-Taylor parameter α for the Amazon forest, Forest Ecology and Management, 38, 211-225.
Yu TW (1977). Parameterization of surface evaporation rate for use in numerical modeling, Journal of Applied Meteorology, 16, 393-400.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 8
Estimation of Water Surface Elevation on Inundated Area Using Satellite Data
A. Yorozuya 1,*
, H. Kamimera 1, T. Okazumi
2, Y. Iwami
1, Y. Kwak
1
1International Centre for Water Hazard and Risk Management under the auspices of UNESCO (ICHARM) Public Works Research Institute,
1-6 Minamihara, Tsukuba, Japan 2Ministry of Land, Infrastructure, Transport and Tourism, Japan
Peer Review History1
Abstract: The present study describes a procedure to estimate the water surface elevation on inundated area applying satellite based
information. As targeted area, Lower Mekong Basin (LMB), which is frequently flooded area, was selected. Firstly, eight days composite data
set of Moderate Resolution Imaging Spectrometer (MODIS) on board Aqua, which is MYD09A1, was applied for obtaining inundated area
with calculating Modified Land Surface Water Index (MLSWI). MLSWI from 2002 to 2012 in the part of LMB was well examined, flood inundated areas were then categorized into four zones with specific hydro-geographical features. Secondary, with combining Digital Surface
Model (DSM) obtained from the Shuttle Radar Topography Mission (SRTM) as well as the inundated area with MLSWI, horizontal
distribution of the water surface elevation (WSE) with the selected types of flooding was estimated. Tuning up of the MLSWI with local gauge information is necessary for better estimation. Finally, an appropriate water surface profile that conforms hydraulic judgment was obtained.
Results have implications for understanding spatio-temporal variations of flood water during past events, and will be useful for validation of
inundation simulations.
Keywords: Inundated area, water surface elevation, satellite information, Lower Mekong Basin.
1. Introduction
Recently, water-related disasters have frequently occurred in the globe, e.g., the large flooding in 2011 at Chao Phraya Basin in Thailand
caused considerable damage and sufferings to the people in the basin (Sayama et al., 2013). Not to repeat such a catastrophe, a risk reduction
strategy needs to be considered with a comprehensive risk assessment as well as appropriate information. Also the climate change is one of
the important factors that need to be included in the risk reduction strategy. Therefore, actual phenomena should be analyzed based on the
past events. Sooner or later, numerical estimation should be conducted involving verification of the past events. Therefore, as the first step of
constructing the risk reduction strategy, the most fundamental information, for this purpose, is the location of water body as well as water
depth at points of interest, especially on inundated areas during flooding.
For the purposes of estimating water depths on floodplains where local Digital Surface Model (DSM) is not well established e.g., developing
countries, a numerical simulation with Shuttle Radar Topography Mission (SRTM) has frequently been applied because of sufficient
availability and easy accessibility as listed here in many different scales. For example, they are a region based scale such as Asia-Pacific (e.g.,
Kwak et al., 2012b), a water shed scale (e.g., Sayama et al., 2013), and a part of river channel (e.g., Yamazaki et al., 2012). On the other hand,
Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) operated by Japan Aerospace Exploration Agency (JAXA) has already
observed most of the part of the globe, and can produce PRISM DSM with a stereo matching processing. Though it has jitter noises with
roughly 5 meter wave height, 10 meter grid size is attractive enough to apply in an inundation simulation. Actually, Yorozuya et al. (2013)
proposed a method of modifying the PRISM DSM to be implemented to the simulation with smoothing the noises. Using those available
1 Paper JHER0204, submitted on 01/11/2014, accepted for publication after peer review and subsequent revisions on 20/12/2014
* Corresponding author may be contacted at [email protected]
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 9
DSM, the objective of this paper is to estimate the water depth without the numerical simulation. In this analysis, some uncertainties cannot
be neglected in the poorly gauged basins since the simulation depends on the initial and boundary conditions.
Another method to estimate water bodies is application of satellite data. For example, Alsdorf et al. (2007) introduced the method with image
sensor, such as Landsat, Moderate Resolution Imaging Spectrometer (MODIS), and SPOT, and also introduced the one with Synthetic-
Aperture Radar (SAR). They mentioned those techniques are good enough, though there are some limitations on usages. On the other
hand, Kwak et al. (2011) developed an index to identify water bodies based on MODIS data, which is Modified Land Surface Water Index
(MLSWI), and estimated water bodies on a floodplain. Later, Kwak et al. (2012a) revised MLSWI and applied it to Chao Phraya river basin.
The final goal of this paper is estimation of water depth based on observed data sets without numerical simulation in Cambodia floodplain of
Lower Mekong Basin (LMB). For this purpose, we estimated water bodies on the floodplain with MLSWI. Thereafter, we obtained water
surface elevation (WSE) by combining SRTM and MLSWI.
2. Study area
The target area of this study is LMB in Cambodia shown in Figure 1. The Mekong River originally starts from Tibet, and goes through
China, a border between Myanmar and Lao PDR, another border between Lao PDR and Thailand, Cambodia, Vietnam, and finally flows into
South China Sea. The Mekong River converges with the Tonle Sap River at Phnom Penh, and diverges from the Bassac River at the same
location. The Tonle Touch River is originated from the Mekong River at the point of diverge, just downstream of Kampong Cham in Figure
1. At this point, the Mekong River has a width of about 1.2 km, when the Tonle Touch River has about 70 m. There are three water-level
gauges in this area, e.g., Kampong Cham, Phnom Penh and Neak Luong along the Mekong River. Other gauges are located at Prek Kdam
along the Tonle Sap River as well as Koh Keol along the Bassac River.
Figure 1 Location of targeted area showing river network, stream gauges and low land area (100 km × 90 km)
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 10
Figure 2 Temporal change of water surface elevation (WSE) with solid curve as well as water surface slope (WSS) with dashed curve
Figure 2 shows temporal changes of water surface elevations as well as water surface slopes at Kampong Cham, Phnom Penh, and Neak
Luong. The water surface slopes were calculated between Kampong Cham and Phnom Penh, and between Phnom Penh and Neak Luong.
The water level at Kampong Cham started from 2.67 m and reached the maximum peak of 16.02 m on 25th of September 2011 (which is 148
days from May 1st). The average peak between 22nd and 30th of September 2011 was 15.74m at Kampong Cham, and 7.84m at Neak
Luong. Both water surface slopes around this period were about 5×10-5.
3. Method
To estimate WSE, the following three steps are required: (1) estimation of water body distribution; (2) smoothing of DSM to obtain modified
DSM; and (3) estimation of WSE based on the water bodies and the modified DSM.
3.1 Estimation of water body distribution
To estimate water body distribution in this target area, 8-day composite information obtained by MODIS and MLSWI were applied. MLSWI
is basically the index to identify muddy water by using band 2, which is near-infrared (NIR), and band 7, which is short-wavelength infrared
(SWIR), with the following equation:
NIRSWIR
NIRSWIR
RA
RAMLSWI
(1)
where Ai and Rj are absorption rate and reflection rate, and i and j are SWIR and NIR, respectively. Since these two rates relate as Ai = 1-Ri,
MLSWI can be determined from MODIS data. Based on the equation and information from MODIS, an MLSWI distribution can be
calculated. MLSWI determines surface types, such as clear water, muddy water, mixed water, soil, vegetation, snow and cloud (e.g., Kwak et
al., 2014). To distinguish water bodies (e.g., clear water, muddy water and mixed waters) from the rest of the surface types, the threshold
value ic should be determined; e.g., if MLSWI are more than ic, they are water bodies, and if less than the ic, they are not.
The ic is usually between 0.5 and 0.8, though it should be defined more specifically based on, firstly, field observation with a spectrometer or,
secondary, by maintaining consistency among observed values of WSE, DSM and an MLSWI distribution. The authors selected the second
approach since there are local gauge stations that can provide necessary data and the former requires field survey during flooding.
The estimation process of ic begins with resampling of the MLSWI distribution to obtain the same grid size with DSM. Secondary, an area of
DSM submerged by the flood water is determined. Here, the submerged area can be estimated at where altitude of the point of interest is less
than the water surface elevation obtained by the local water gauges during the target time period. Another area of MLSWI which is more
than the target value is determined. When both areas are matched, the target MLSWI value is defined as the ic. The area should be an
0.00E+00
2.00E-05
4.00E-05
6.00E-05
8.00E-05
1.00E-04
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
20.0
0 30 60 90 120 150 180 210 240
Wa
ter
Su
rfa
ce S
lop
e
Wa
ter
Su
rfa
ce E
lev
ati
on
, m
Days starting from 1st of May at 2011
Kampong Cham (WSE) Pnom Penh (WSE)
Neak Leurng (WSE) Kampong Cham-Pnom Penh (WSS)
Pnom Penh-Neak Leurng (DSS)
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 11
appropriate size that allows the water surface elevation to be constant within the area with concerning the water surface slope at the targeted
period. In this study, since test case is in between 22nd
and 30th of September in 2011, the water surface slope is about 5×10
-5 as Figure 2
indicates. Therefore, the appropriate size can be the area within 20 km from the water gauge, if 1m error in vertical direction is acceptable.
3.2 Smoothing of DSM to obtain Modified DSM
DSM obtained from SRTM has jitter noises. Such noises should be eliminated for better estimation of water surface elevations. Goldstein
and Werner (1998) applied a high pass filter with FFT to smooth DSM. On the other hand, Yorozuya et al. (2013) modified PRISM-DSM to
eliminate jitter noises. Firstly they selected a window to obtain the least square plain. Then, structures such as levees, embankments, and
houses are determined by subtracting the plain from the original DSM. Thereafter, moving to the next window-overlapping part of the area,
the same process is operated to obtain the plain and the structures. After repeating the process to cover the whole target area, moving
averaged values from the windows are obtained for both the plains and the structures. Finally, by placing the moving averaged structure on
the averaged plain, modified DSM is constructed. If the size of a window is too large, local undulations will disappear. For example, if the
size of a window is the same as that of the whole domain, only flat surface will appear without any undulations. Contrarily, if the size of a
window is too small, the undulation of the plain may be largely affected by noises. Therefore, the size of a window should be carefully
selected by comparing photos or other information. In this study, the data processing to obtain the modified PRISM-DSM was applied to
SRTM DSM to obtain smooth DSMs. Also SRTM DSM was resampled with the nearest neighbour to 100 m grid size.
3.3 Estimation of Water Surface Elevation
Similar to the process used to obtain modified DSM, the moving average process was applied. Every single point except edges in a window
was examined with a neighboring point. If the threshold value, determined in subsection 3.1, lied between two MLSWI values, the point of
interest was recognized as a border between the wet and dry areas. Other seven neighboring points were examined as well. Once borders
were recognized, assuming that MLSWI and DSM had a linear relationship, an elevation from DSM was calculated with inputting the exact
threshold to the linear function. The same data processing is conducted for every point, the elevations are averaged, and the averaged as well
as standard deviation (σ) value is assigned as the WSE in the single window. Finally, an averaged-smooth water surface is obtained after
moving averaging in the whole domain area. Similar to smoothing of DSM, the size of the window is carefully considered. For example,
when the window is small and the target area is wide floodplain, the borders cannot be recognized in the single window. Hence, the elevation
cannot be determined. Contrarily, when the window is large and the target area is sort of valley type geometry, water body from different
sources or that in other valley is recognized as single value. As a consequence, the standard deviation might be larger.
4. Results
4.1 Zoning of inundated area
Figure 3 indicates the MLSWI distribution with averaging of 483 composites, which are collected for about 11 years from 2002 to 2012. The
figure shows the northern half of LMB including Kampong Cham and Phnom Penh compared with Figure 1. As it shows, the areas around
the Mekong River and the Tonle Sap River have most likely higher elevations compared to rest of the floodplain. Also this map is similar to
an inundation map issued by the Mekong River Commission (MRC), though it is not discussed in this paper. Also, Figure 3 indicates four
zones, which were classified based on following discussion with concerning about types of flood based on information, such as temporal
changes of the MLSWI, and the location of colmatage system, which is an irrigation canal system in Cambodia. Again the latter two pieces of
information are not described in detail in this paper. Zone 1 is located in the right side of the Mekong River and surrounded by natural levees
and many colmatage canals. The inundated water body by MLSWI starts to appear even before the water surface in the Mekong River is
lower than the natural levees. Zone 2 is an inundated area along the Tonle Sap River. Hence, the area is affected by the Tonle Sap Lake.
Zone 3 is on the left side of the Mekong River and has similar characteristics with Zone 1. Zone 4 is on the left side of the Mekong River
and the Tonle Touch River flows at the center of the zone. In this area, most of floodwaters come along the Tonle Touch River, or overflows
around the diverge area when the flood flow is large. One of the purposes of this paper is estimation of water surface elevation on the
floodplain. Based on the above discussion, regarding Zones 1 and 3, the water bodies on the floodplain are well connected to that of the
Mekong River, though timing of the elevation changing might be different. Therefore, the elevation can represent that of the Mekong River.
Therefore, as long as the water surface along the Mekong River can be estimated, the elevation on the both zones can also be estimated.
Regarding Zone 2, the hydraulic phenomena might be complicated, since the flood flow is dominated by that of the Tonle Sap Lake.
Regarding Zone 4, the elevation is dependent on the flow capacity of the Tonle Touch River as well as water discharge diverged from Mekong
River.
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 12
Figure 3 MLSWI distribution with averaging of 483 composites from 2002 to 2012 (showing northern half of LMB)
Figure 4 DSM around Kampong Cham (35 km × 39 km) (points above 15.74 m are shown only)
Zone 1
Zone 2
Zone 3
Zone 4
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 13
Figure 5 DSM and MLSWI (points above 0.64 are shown only) (35 km × 39 km)
4.2 Water Surface Elevation
The water surface elevation obtained by the local water-level gauge at Kampong Cham was the highest of 15.74 m as described in Figure 2.
Based on the information, Figure 4 is obtained from DSM with only showing the values, which are more than 15.74 m around the Kampong
Cham area with the size of 35 km × 39 km. Similarly, Figure 5 shows MLSWI which is more than the threshold value of 0.64 as well as same
DSM with Figure 4. Comparison indicates that overall shape is very similar. Also the natural levee around the point of diverge indicates
higher value by the DSM, and actually MLSWI indicates the dry area. However, some natural levees in middle of the Mekong River
indicated wet area, though DSM shows the value higher than 15.74 m. Though this small difference or inconsistency appears, this overall
comparison is good enough to determine the threshold value. It may be noted that Figure 5 is the result of trial and error processes to
determine the threshold value of 0.64. After estimating the whole domain of the targeted area, the distribution of the WSE was superimposed
on Figure 1 including river lines of the Mekong River, the Tonle Sap River, and the Bassac River, as shown in Figure 5. The WSE has no
value when MLSWI is smaller than the threshold value, which indicates the dry area.
As Figure 6 indicates, the WSE shows higher value around the Kampong Cham area. After flood water flows this area, it widely spreads to
Zone 1 and Zone 4. Along the Mekong River from Kampong Cham to Phnom Penh, the altitude of water surface gradually goes down,
though it has about one meter higher value around confluence point of the Mekon and Tone Sap River than that around upstream side.
Because of the complexity of flow around this area or because of difficulty of estimation, the WSE is kind of difficult to understand around
this area. Another difficulty to interpret this figure is longitudinal distribution of WSE along the Tonle Sap River. As it was mentioned
before, the flow around the area, which was classified as Zone 2, might have complex flow pattern because of complicated system involving
different flow system. Therefore, further study related to the flow in the Zone 2 and the Tonle Sap Lake need to be conducted separately. As
flow goes down to downstream along the Mekong River from Phnom Penh, the WSE gradually falls down, this is deemed reasonable. At the
very upstream of the Bassac River, there is area where flood flow does not exist. It might or might not be true, but it should be recalculated
with different threshold value using water level observation at Phnom Penh. Finally, the local maximum showing in the yellow colour in
between the Mekong River and the Tonle Touch River is difficult to understand. Since this place does not have any sources of water, local
maximum of WSE cannot be located in this area. This kind of estimation error occurs because window size was not properly selected.
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 14
Figure 6 Distribution of water surface elevation in LMB as well as the river lines (100 km × 90 km)
To discuss more in detail about the flow in Zone 4, the longitudinal WSE profile is described. In Figure 6, series of black square marks
located along the Tonle Touch River. The WSE values, WSE ± σ and DSM from each single point of the marks were picked up; thereafter,
Figure 7 was drawn as the longitudinal distribution originated from the most upstream mark. Here, the σ is the one explained in subsection
3.3. As it shows, the WSE starts to have slope about 1.4×10-4 between 0 and 22,000 m. Thereafter, it becomes almost flat between 22,000 and
44,000 m, which indicates the M1 curves. In the end, it becomes about 3.8×10-4 between 44,000 and downstream end. As both Figures 6 and
7 indicate, there is the national high way 8 (NHW8) at this location. The NHW8 is the road embankment, where embankment height is
about 4 or 5 m depending on the location. This road has about 20 bridges in this Zone 4; each bridge has about 100 m opening. Thus the
flood water might be not fully, but partially blocked. Therefore, the water surface profile becomes flat at upstream side of NHW8 as Figure 7
indicates. This is one of the verifications of the method described in this paper.
Figure 7 Longitudinal distributions of WSE, DSM and WES±σ along the Tonle Touch River
The adopted method estimated 16.10 m and 7.64 m at upstream and downstream end, respectively. As Figure 2 indicates, the WSE at
Kampong Cham and Neak Luong are 15.74 and 7.89 m respectively. Actually, distance between the downstream end of Figure 6 and Neak
0.0
5.0
10.0
15.0
20.0
25.0
0 10,000 20,000 30,000 40,000 50,000 60,000
Ele
va
tio
n, m
Distance, m
WSE±σ DSM WSE
NHW8
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 15
Luong is about 20 km. Therefore, 1-m height differences can be estimated between both points, since water surface slope in this reach is
about 5.0×10-5 based on observed results as shown in Figure 2. Then, actual WSE at both points are 15.74 and 8.89 m, respectively.
Therefore, differences between estimation and observation at both ends are 0.36 m and -1.25 m, respectively.
Another important aspect is σ, which varied over different locations. As Figure 7 indicates, standard deviations are much higher in the
upstream side, though they are very small in the middle. Actually, geometry around upstream side has larger undulation than that of rest of
area, since natural levee developed along the Mekong River. On the other hand, the geometry in the middle is much flatter since the area is
in the middle of flood plain. As this figure indicates, since σ in the middle is small, the window was adjusted to obtain the suitable results on
the floodplain. From this point of view, the different size of widow has to be selected depending on the undulation, if the WSE needs to be
obtained with similar size of σ.
5. Summary
The inundation area was successfully mapped with averaging of 483 composites of MODIS data applying the MLSWI method, whose area is
similar to the inundated map issued by the MRC. The flood types are not the same in the hole targeted area; therefore, four zones are
classified as the types of the inundation based on the knowledge of temporal change of inundated area as well as the geometrical condition;
e.g., the river system including the colmatage system. The threshold value of MLSWI was determined as 0.64, which is agreeable with
previous study. The value was selected based on the information of the local water gauge, as well as the DSM data, which indicates the
importance of the local information; though most of the discussions are based on the space-based observation. With combing the MLSWI
and the DSM, the plane distribution of water surface elevation were mapped; though there were few areas which are difficult to be
understood. One of the verification was conducted mainly inside of Zone 4 with longitudinal water surface elevation along the Tonle Touch
River. It shows M1 curves, which is agreeable because of the existence of the embankment type hydraulic structure. Also, the water surface
elevations at the upstream/downstream end in the Figure 7 are also agreeable compared with the observed values at the local gauges.
At the same time, several difficulties were pointed out, such as the estimation method is not well enough to understand the complex flow field
like Zone 2, which including the confluence and diffluence of flow as well as Tonle Sap Lake. Also size of the single window to very
beginning of the moving averaging process sensitively affects estimation of WSE. This paper tries to estimate the WSE with single size of the
window in the whole domain, which includes wide floodplain, or valley surrounded by the natural levee. Actually, few difficulties were
pointed out because of the window size. Certainly good estimation was obtained in the Zone 4 as shown in Figures 6 and 7, since
appropriate size was selected, which is suitable to the geometry in Zone 4. One of the important further studies is installing an option to
select the window size depending on the geography.
6. Acknowledgements
The authors acknowledge Mekong River Commission to provide the local gauge data for the study.
References
Alsdorf DE, Rodriguez E, Lettenmaier DP (2007). Measuring surface water from space, Reviews of Geophysics, 45, RG2002,
doi:10.1029/2006RG000197.
Goldstein RM, Werner CL (1998). Radar interferogram filtering for geophysical applications, Geophysical Research Letters, 25, 21, 4035, 1998,
doi:10.1029/1998GL900033.
Kwak Y, Park J, Fukami K (2011). Nation-wide Flood Risk Assessment Using Inundation Level Model and MODIS Time-series Images,
Proceeding IEEE-IGARSS 2011, IEEE, pp.4395-4398.
Kwak Y, Park J, Yorozuya A, Fukami K (2012a). Estimation of flood volume in Chao Phraya river basin, Thailand from MODIS images coupled
with flood Inundation level, the 32nd annual IGARSS symposium 2012, IEEE Geoscience and Remote Sensing Society, pp.887-890.
Kwak Y, Park J, Fukami, K (2014). Near Real-Time Flood Volume Estimation From MODIS Time-Series Imagery in the Indus River Basin,
Selected Topics in Applied Earth Observations and Remote Sensing, IEEE Journal of , 7, 2, 578,586, doi: 10.1109/JSTARS.2013.2284607.
Kwak Y, Takeuchi K, Fukami K, Magome J (2012b). A new approach to flood risk assessment in Asia-Pacific region based on MRI-AGCM
outputs, Hydrological Research Letters, 6, 70-75.
“Estimation of Water Surface Elevation on Inundated Area Using Satellite Data” Yorozuya et al.
Journal of Hydrology and Environment Research 16
Sayama T, Tatebe Y, Fujioka S, Ushiyama T, Yorozuya A, Tanaka S (2013). An Emergency repsonse-type rainfall-raunoff-inundation prediction
for 2011 Thailand flood, Journal of Japan Society of Civil Engineers, Ser. B1 (Hydraulic Engineering), 69, 1, 14-29. (In Japanese).
Yamazaki D, Baugh CA, Bates PD, Kanae S, Alsdorf DE, Oki T (2012). Adjustment of a spaceborne DEM for use in floodplain hydrodynamic
modeling, Journal of Hydrology, 436-437, 81-91, http://dx.doi.org/10.1016/j.jhydrol.2012.02.045.
Yorozuya A, Kwak Y, Shiratori A, Fukami K (2013). Study on PRISM DSM application to inundation analysis and its modification method,
Journal of Civil Engineers, Ser. B1 (Hydraulic Engineering), 69, 4, I_1549-I_1554. (In Japanese).
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 17
Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia
A.S. Rahman 1,*
, K. Haddad 1, A. Rahman
1
1School of Computing, Engineering and Mathematics, University of Western Sydney,
Building XC, Kingswood Campus, Locked Bag1797, Penrith, New South Wales 2751, Australia
Peer Review History1
Abstract: At-site flood frequency analysis is a useful technique to estimate flood quantiles if reasonably long at-site flood record is available.
In Australia, FLIKE software has been proposed for at-site flood frequency analysis. The advantage of FLIKE is that, for a given application, the user can compare a number of most commonly adopted probability distributions and parameter estimation methods relatively quickly
using a windows interface. The new version of FLIKE has been incorporated with the multiple Grubbs and Beck test, which can identify
multiple numbers of potentially influential low flows. This paper presents a case study considering ten catchments in eastern Australia, which compares two outlier identification tests (original Grubbs and Beck and multiple Grubbs and Beck tests) and two commonly applied
probability distributions (Generalized Extreme Value (GEV) and Log Pearson type 3 (LP3)). The results show that the LP3 distribution with
multiple Grubbs and Beck test provides more accurate flood quantile estimates than when LP3 distribution is used with the original Grubbs and Beck test. Between these two methods, the differences in flood quantile estimates have been found to be up to 61% for the ten study
catchments. It has also been found that GEV distribution with L moments and LP3 distribution with the multiple Grubbs and Beck test
provide quite similar results in most of the cases; however, a difference up to 38% has been noted for flood quantiles for annual exceedance
probability (AEP) of 1 in 100 for one catchment. The methodology presented in this paper can be applied to other catchments/countries.
Keywords: Flood, FLIKE, probability distributions, flood frequency, outlier.
1. Introduction
Flooding is a part of the natural cycle of many ecosystems and plays an important role in maintaining ecosystem function and biodiversity
(Poff et al., 1977). The multitude of environmental benefits from flooding also benefits society through continuation of ecosystem services
(Bayley, 1995). However, major modification to natural processes and ecosystems through human activities has created some of the most
destructive hydro-meteorological phenomena in terms of their impacts on human well-being and socioeconomic activities. There had been
significant increases in the total number of natural disaster events in recent years due to climate change. Flood is the most expensive of these
disasters, for example, during the period of 2010-2011, a series of floods affected various parts of Australia, with most significant damage to
the state of Queensland causing damage over $5 billion.
Flood frequency analysis is the most direct method of estimating design floods, which is needed to design bridges, culverts, flood levees and
other drainage infrastructure and in various water resources management tasks such as flood plain management and flood insurance studies.
Griffis and Stedinger (2007) found that estimates of magnitude and frequency of floods using streamflow-gauging stations with shorter
records of annual peak flow data had higher standard errors or uncertainties when compared to estimates using stream gauges with longer
annual peak flow records. Flood estimation should get the maximum information from the available data, be robust with respect to the
distribution model and potentially influential low flows (PILFs). However, streamflow record length at many sites is often insufficient and
identification of PILFs becomes a major issue in fitting a probability distribution to available flood data.
1 Paper JHER0202, submitted on 28/10/2014, accepted for publication after peer review and subsequent revisions on 23/12/2014
* Corresponding author may be contacted at [email protected]
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 18
Comparing various probability distributions and parameter estimation procedure had been done in numerous occasions in the past; however,
due to the limited length of observed flood data as compared to the return period of interest, flood frequency analysis is deemed to be a
difficult task and often associated with controversies (Bobée et al., 1993). The selection of an ‘appropriate’ probability distribution and
associated parameter estimation procedure is an important step in flood frequency analysis. Flood frequency analysis has been widely
researched in the past (e.g. Vogel et al., 1993; Onoz and Bayazit, 1995; Bates et al., 1998; Laio, 2004; Merz et al., 2008; Meshgi and Khalili,
2009a, b; Laio et al., 2009; Ishak et al., 2010, 2011; Haddad et al., 2011, Haddad et al., 2012; Haddad and Rahman, 2012; ; Haddad et al.,
2013; Rahman et al., 2013). In flood frequency analysis, a probability distribution is often selected on the basis of statistical tests or by
graphical methods, and convenience plays an important role in this choice (Bobée et al., 1993). In practical applications, empirical suitability
plays a much larger role in distributional choice than a priori reasoning (Cunnane, 1985; 1989).
One of the earliest studies on the search for the probability distribution of floods was done by Benson (1968). He considered two-parameter
Gamma, Gumbel (EV1), Log Gumbel, Log Normal (LN2), three parameter Log Pearson Type 3 (LP3) distribution for flood data of 10
stations in various parts of the USA. The standardized average deviations were found to be high for the gamma, EV1 and Log Gumbel, but
lower for the LN2 and LP3. Among these, the LP3 distribution was preferred for being in common use, and for being capable of fitting
skewed data. The conclusions and recommendations of this study led to the wide-scale adoption of the LP3 distribution in the USA. In
Australia, an extensive study was done for Queensland data by Kopittke et al., (1976), and another by Conway (1970) for New South Wales
coastal streams. They concluded that the LP3 distribution performed the best among a number of distributions examined. Beard (1974)
estimated the 1000 year flood at 300 stations in the USA with 14200 station-years of data by eight different models (LN2, gamma, EV1 with
two different parameter estimates, Log Gumbel, Pearson type 3 (P3), LP3 and regional LP3). LP3 and LN2 came close to reproducing the
expected 14 exceedances and were concluded to be the preferred ones. The split sample validation confirmed the superiority of these
distributions.
To compare various probability distributions using the data from 172 catchments in Australia, McMahon and Srikanthan (1981) used the
moment ratio diagrams. They also concluded that the LP3 was the most suitable distribution for Australia. Based on the findings of these
studies, it was recommended in Australian Rainfall and Runoff (ARR) (I. E. Aust., 1987) that flood frequency analysis in Australia (I. E.
Aust., 1987) should follow the footsteps of the USA i.e. to use LP3 distribution (IAWCD, 1982).
Since the publication of ARR (I. E. Aust., 1987, 2001), there have been a number of studies to compare various probability distributions
(Rahman et al., 1999). For example, Nathan and Weinmann (1991) examined 53 catchments from Central Victoria, with L-moments-based
goodness-of-fit test, and found that the GEV distribution was the best-fit distribution. Vogel et al. (1993) compared a number of distributions
using data from 61 stations in Australia, using the L-moments ratio diagram; they found that the generalized Pareto distribution (GPA) was
the best-fit distribution followed by the GEV, LN3, and LP3. Haddad and Rahman (2008) compared a number of distributions and parameter
estimation procedures for 18 catchments in southeast Australia and found that the GEV distribution was the best-fit distribution for the
selected catchments. In another study, Haddad and Rahman (2010) found that the two parameter distributions are preferable to Tasmania,
with the lognormal appearing to be the best-fit distribution for Tasmania. As it seems an analyst might choose a different frequency model
and fitting procedure for each catchment, but this could lead to inconsistencies in flood estimates across regions and among governmental
agencies. National consistency in flood frequency estimates is important because these estimates are used in the allocation of resources and
the implementation of the National Flood Insurance Program (Thomas, 1985; Cohn et al., 2013). For this reason, a national methodology
should exhibit the characteristic of robustness, which in this context means that the analysis does not perform poorly when its assumptions
are not fully satisfied.
An important step in flood frequency analysis is the detection of the PILFs in the flood data (Saf, 2010). PILFs are unusually small
observations of flood data which depart significantly from the trend of the rest of the data. Identification and treatment of PILFs are
important issues in flood frequency analysis, because such observations can have a large influence on the estimate of extreme flood quantiles.
In arid regions, even when it rains, channel losses can result in annual flood peaks that are zero or nearly zero, therefore LP3 distribution
cannot fit the entire flood record without censoring zero values. Furthermore, unusually small values can result in relatively poor estimates of
the large flood quantiles. In frequency analyses, one often uses a probability plot to examine if the sample data is consistent with a fitted curve
(Beckman and Cook, 1983; Stedinger et al., 1993), unfortunately such decisions can be relatively subjective. The Bulletin 17B explicitly notes
that not dealing with this issue of PILFs would “significantly affect the [computed] statistical parameters.” Both Barnett and Lewis (1994) and
Beckman and Cook (1983) discussed the notion of using outlier tests to identify unusual (high or low) data points that otherwise might have
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 19
undue influence in flood frequency analysis. Using a good low-outlier identification procedure has the potential for making low-outlier
identification less subjective, by providing “rejection criteria which enable significance to be assessed” (Barnett and Lewis, 1994).
A wide range of test procedures for identifying PILFs has been examined in the past (e.g. Thompson, 1935; Grubbs, 1969; Grubbs and Beck,
1972; Barnett and Lewis, 1994), including methods for dealing with the case of multiple PILFs considered here (Tietjen and Moore, 1972;
Rosner, 1975, 1983; Prescott, 1975, 1978; Gentleman and Wilk, 1975; Marashinghe, 1985; Rousseeuw and Zomeren, 1990; Hadi and
Simonoff, 1993; Spencer and McCuen, 1996; Rousseeuw and Leroy, 2003; Verma and Quiroz-Ruiz, 2006). Thompson (1935) provided an
early criterion for the rejection of an outlier based on the ratio of the sample standard deviation and an observation’s deviation from the
sample mean. An alternative test was proposed by Dixon (1950, 1951), who for high outliers proposed the test statistics for second most
extreme observation in either tail of the distribution. Barnett and Lewis (1994) also noted similar criteria as Dixon (1950, 1951). Grubbs
(1969) and Grubbs and Beck (1972) proposed a one sided 10% significance level criteria to identify PILFs. Rosner (1975, 1983) developed a
sequential two sided outlier test, based on a generalization of the Grubbs (1969) which usually detected outliers either too small or large. This
procedure was found to be less computationally intensive and easy to apply in practice.
Bulletin 17B (IAWCD 1982) was the guideline for flood frequency analysis in the United States for more than 30 years. Recently, there has
been an attempt to revise Bulletin 17B to include recent advances in statistical techniques and computational resources (IAWCD 2013) similar
to the current revision of Australian Rainfall and Runoff. In Bulletin 17C, a new low outlier identification procedure, the multiple Grubbs-
Beck (MGB) test (Lamontagne et al., 2013) has been included. The MGB test is based on significance levels computed using the new
approximations developed by Cohn et al. (2013). The MGB test is a generalization of the old Bulletin 17B original Grubbs-Beck (GB) test
(Grubbs 1969; Grubbs and Beck, 1972).
In this paper the original GB test and the MGB test are compared using data from Australian catchments. The Bulletin 17B’s original GB test
was based only on the distribution of the single smallest observation in a sample. As a result, even though multiple PILFs in flood annual
maximum flood data series may exist, the original GB test rarely identifies more than a single PILF. The MGB test employs the actual
distribution of the kth smallest observation in a sample of n independent normal variates based upon significance levels provided by Cohn et
al. (2013), and is thus suited to test for multiple PILFs.
Kuczera (1999) presented a comprehensive study on flood frequency analysis using Bayesian method and incorporated a number of
probability distributions in his FLIKE software. It has several advantages including the ability to (i) incorporate prior or regional information;
(ii) incorporate stage-discharge uncertainty; (iii) assess parameter uncertainty obtained from regional information; and (iv) allow for threshold
values (censoring). Recently a new version of FLIKE has been released. The older version of FLIKE needed manual identification of PILFs
using the original GB test. The new version of FLIKE is incorporated with the MGB test which attempts to identify multiple PILFs in the
annual maximum flood series data.
The objective of this paper is to compare the performance of two probability distributions in flood frequency analysis (FFA) namely LP3 and
GEV distributions with a special focus on the effects of censoring PILFs using the original GB and MGB tests. To our knowledge, the MGB
test has not been applied before to Australian flood data. The reason for applying the LP3 and GEV distributions with the MGB test is that
these two distributions are most commonly adopted in flood and rainfall frequency analyses in Australia.
2. Study area and data preparation
For this study, ten catchments from eastern Australia are selected from the states of New South Wales (NSW), Queensland (QLD) and
Victoria (VIC) as shown in Table 1 and Figure 1. Catchment area ranges from 87 to 900 km2 with a mean of 345.8 km
2 and median of 160
km2. Record length ranges from 33 to 91 years with a mean of 56 years and median of 54 years. All of the stations have log-space skew values
significantly different from zero. Missing data points in the annual maximum flood series were in-filled where possible by two methods.
Method one involved comparing the monthly instantaneous maximum data (IMD) with monthly maximum mean daily data (MMD) at the
same station. If a missing month of IMD flow corresponded to a month of very low MMD flow, then that was taken to show that the annual
maximum did not occur during that missing month. Method 2 involved a simple linear regression of the annual MMD flow against the
annual IMD series of the same station. It must be mentioned that the regression equations developed were used for filling gaps in the IMD
record, but not to extend the overall period of record.
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 20
Figure 1 Selected ten catchments from eastern Australia
Table 1 Details of the selected study catchments
Station ID Station name River name Catchment
area (km2)
Record length
(years)
Period of record
218005 D/S Wadbilliga R Junct Tuross 900 47 1965-2011 219001 Brown Mountain Rutherford Ck 15 62 1949-2010
219006 Tantawangalo Mountain (Dam) Tantawangalo Ck 87 60 1952-2011
116008 Abergowrie Gowrie Ck 124 58 1954-2011
125002 Sarich's Pioneer 740 51 1961-2011
136202 Litzows Barambah Ck 681 91 1921-2011
136301 Weens Br Stuart 512 76 1936-2011
230204 Riddells Ck Riddells Ck 79 38 1974-2011
232213 U/S of Bungal Dam Lal Lal Ck 157 33 1977-2011
405238 Pyalong Mollison Ck 163 41 1972-2012
3. Methodology
The original GB test (Grubbs, 1969; Grubbs and Beck, 1972) uses the at-site logarithms of the peak-flow data to calculate a one-sided, 10%
significance-level critical value for a normally distributed sample. Although more than one recorded peak flow for a stream gauge may be
smaller than the Grubbs-Beck critical value, usually only one non-zero recorded peak flow is identified from the test as being a PILF. The
original GB test which was recommended in Bulletin 17B (IAWCD, 1982) defines a low outlier (PILF) threshold as:
𝑋𝑐𝑟𝑖𝑡 = µ − 𝑘𝑛𝜎 (1)
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 21
where kn is a one-sided, 10% significance-level critical value for an independent sample of n normal variate, and μ and σ denote the sample
mean and standard deviation of the entire data set, respectively. Any observation less than Xcrit is declared a ‘‘low outlier (PILF)’’ (IAWCD,
1982). As per Bulletin 17B, PILFs are omitted from the sample and the frequency curve is adjusted, using a conditional probability adjustment
(IAWCD, 1982). The kn values are tabulated in section A4 of IACWD (1982) based on Table A1 in Grubbs and Beck (1972).
Stedinger et al (1993) provide an approximation of kn for 5 ≤ n ≤ 150 (where n is sample size):
𝑘𝑛 ≈ −0.9043 + 3.345√log10(𝑛) − 0.4046 log10(𝑛) (2)
The original GB test only identifies one outlier/PILF at a time from a particular data set, but there can be more numbers of PILFs available in
the data. A method for statistically detecting multiple PILFs using a generalized Grubbs-Beck test has been developed (Gotvald et al., 2012).
The MBG test is based on a one-sided, 10% significance-level critical value for a normally distributed sample, but the test is constructed so
that groups of ordered data are examined (for example, the eight smallest values) and excluded from the dataset when the critical value is
calculated. If the critical value is greater than the eighth smallest value in the example, then all eight values are considered to be PILFs
according to this new method.
Here, one considers whether {X[1:n], X[2:n], …. , X[k:n]} are consistent with a normal distribution and the other observations in the sample by
examining the statistic (Cohn et al., 2013):
�̃�[𝑘:𝑛] ≡ (𝑘[𝑘:𝑛] − 𝜇𝑘) 𝜎𝑘⁄ (3)
where X[k:n] denotes the kth smallest observation in the sample, and
𝜇𝑘 ≡1
𝑛−𝑘∑ 𝑋[𝑗:𝑛]
𝑛𝑗=𝑘+1 (4)
𝜎𝑘 ≡1
𝑛−𝑘−1∑ (𝑋[𝑗:𝑛] − 𝜇𝑘)
2 𝑛
𝑗=𝑘+1 (5)
Here the partial mean μk and partial variance σk are computed using only the observations larger than X[k : n] to avoid swamping.
To implement the MGB test, recommended for Bulletin 17C, the following two steps are involved: (i) starting at the median and sweeping
outward towards the smallest observation, each observation is tested with a MGB test significance level αout. If the kth smallest observation is
identified as a low outlier, the outward sweep stops and all observations less than the kth smallest (i.e. j = 1, …, k) are also identified as low
outliers. (ii) An inward sweep always starts at the smallest observation and moves towards the median, with a significance level of αin. If an
observation m ≥ 1 fails to be identified by the inward sweep, the inward sweep stops. The total number of low-outliers/PILFs identified by
the MGB test is then the maximum of k and m − 1. The algorithm has two parameters that need to be specified (Cohn et al., 2013): (i)
outward sweep significance level for each comparison, αout; and (ii) inward sweep significance level for each comparison, αin.
Bulletin 17B used a 10% significance test with a single outlier threshold. The new outlier detection procedure uses two multiple threshold
sweeps. Those thresholds are the Cohn et al. (2013) p(k;n) function which correctly describes if the kth smallest observation in a normal
sample of n variates is unusual. The first outward sweep seeks to determine if there is some break in the lower half of the data that would
suggest the sample is best treated as if it had a number of low outliers. The second sweep using a less severe significance level, say p(k;n) ≤
10%, mimics Bulletin 17B’s willingness to identify one or more of the smallest observations as low outliers so that the frequency analysis is
more robust.
A reasonable concern is that a flood record could contain more than one low outlier and the additional outliers can cause the original GB test
statistic to fail to recognize the smallest observation as an outlier (by inflating the sample mean and variance). This effect is known as
masking (Tietjen and Moore, 1972). Inward sweep tests are particularly susceptible to masking (Branett and Lewis, 1994); therefore an
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 22
outward sweep is desirable to avoid the masking problem (Spencer and McCuen, 1996). Rosner (1983) used a two-sided outward sweep.
McCuen and Ayuub (1996) recommended an outward sweep with their test for multiple outliers when fitting a LP3 distribution.
The GEV distribution is a family of continuous probability distributions developed within extreme value theory to combine the Gumble,
Frѐhet and Weibull families also known as type I, II and III extreme value distributions. Here GEV distribution is used to compare the
results with the LP3 distribution. The LP3 distribution has been used for several decades to model annual maximum flood series. Estimation
of the parameters of the distribution using a method of moment (MOM) estimator in log space was suggested by Beard (1962); this method
was used presumably for computational ease. The only complication was the need for frequency factors to compute quantile estimates given
the sample moments of the logs of the data. The needed frequency factors were tabulated in Benson (1968) and in Bulletin 17B. Kirby (1972)
provided an excellent approximation. Currently, these can be computed directly with built-in functions in many software packages, including
Excel and MATLAB.
To fit the LP3 distribution, it is required to calculate the mean, standard deviation and skew coefficient of the logarithms of the annual
maximum flood data. Estimate of the p percent annual exceedance probability (AEP) flood is computed by inserting the three statistics of the
frequency distribution into the equation:
SKXQ pp log (6)
where QP is the p-percent AEP flood or flood quantile; X the mean of the logarithms of the annual peak flows; kp is the frequency factor
that depends on the skew coefficient and AEP and can be obtained from Bulletin 17B and S is the standard deviation of the logarithms of the
annual peak flows.
The mean, standard deviation and skew coefficient can be estimated from the available sample data (recorded annual-peak flows), but a skew
coefficient calculated from small samples tends to be an unreliable estimator of the population skew coefficient. Accordingly, the guidelines in
Bulletin 17B (IAWCD, 1982) indicates that the skew coefficient calculated from at-site sample data (station skew) needs to be weighted with a
generalized or regional skew determined from an analysis of selected long-term stream gauges in the study region. The value of the skew
coefficient used in equation 6 is the weighted skew that is based on station skew and regional skew. However, Australian Rainfall and Runoff
1987 did not adopt the weighted skew for application in Australia (I. E. Aust., 1987).
4. Results
For five stations (218005, 219001, 219006, 116008, 136301), the original GB test did not find any PILF but the MGB test found 24, 26, 27, 29,
31 PILFs, respectively and for the remaining five stations (125002, 136202, 230204, 232213, 405238), the original GB test found only one PILF
for each of them but the MGB test found 26, 46, 17, 17, 21 PILFs, respectively. These results show a remarkable difference between the results
by the two methods of outlier detection.
Table 3 presents the log space skews of the original annual maximum (AM) flood data set and after removing the PILFs using the original GB
test and the MGB test. This table shows that application of MGB results in a greater reduction in log space skew than the original GB test.
This is likely to affect the quantile estimation by LP3 distribution as skew plays an important role in the fitting of LP3 distribution.
Figures 2 and 3 show how does application of GB and MGB tests affect the fitting of a probability distribution to the AM flood series for
Station 136202. The application of GB test did not identify any PILF for Station 136202; however, the application of MGB test identifies 46
PILFs, the fitting of LP3 distribution is remarkably better in Figure 3 (where MGB test is applied) than in Figure 2 (where GB test is applied).
In another example, Figures 4, 5 and 6 show the effects of PILFs on fitting a probability distribution to the AM flood series for Station 405238
where the application of GB test identified only one PILF; however, the MGB test identified 21 PILFs. Figure 6 shows a better fit of the LP3
distribution to the AM flood data series (where MGB test is applied) than in Figure 5 (where GB test is applied).
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 23
Table 2 Number of PILFs identified by the MGB test and original GB test
Station ID Number (%) of PILFs identified by MGB test Number of PILFs identified by original GB test
218005 24 (51.06%) None
219001 26 (41.94%) None
219006 27 (45.00%) None
116008 29 (50.00%) None
125002 26(50.98%) 1
136202 46 (50.55%) 1
136301 31 (40.79%) None
230204 17 (44.74%) 1
232213 17 (51.52%) 1
405238 21 (51.23%) 1
Table 3 Skew without removing any PILFs, PILFs removed by original GB test and PILFs removed by MGB test, respectively
Station ID Skew (no PILFs removed) Skew (PILFs removed by original GB test) Skew (PILFs removed by MGB test)
218005 -0.375 -0.375 -0.553
219001 -0.525 -0.525 -0.098
219006 -0.514 -0.514 0.367
116008 -0.310 -0.310 -0.079
125002 -0.901 -0.757 0.095
136202 -1.434 -1.059 0.076
136301 -0.738 -0.738 0.477
230204 -0.671 -0.373 0.079
232213 -1.244 -1.197 -0.531
405238 -1.220 -1.151 0.285
Figure 2 Flood frequency curve for Station 136202 using LP3 distribution (there was one PILF as per original GB test)
-2 2
AEP 1 in years
-1.52
-0.55
0.42
1.38
2.35
3.32
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Censored
Expected quantile
90% limit
Expected prob quantile
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 24
Figure 3 Flood frequency curve for Station 136202 using LP3 distribution (46 PILFs censored as per MGB test)
Figure 4 Flood frequency curve for Station 405238 using LP3 distribution (no PILF censored)
-2 2
AEP 1 in years
-2.00
-0.92
0.15
1.23
2.30
3.38
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Censored
Expected quantile
90% limit
Expected prob quantile
-2 2
AEP 1 in years
-3.000
-1.881
-0.763
0.356
1.475
2.593
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Expected quantile
90% limit
Expected prob quantile
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 25
Figure 5 Flood frequency curve for Station 405238 using LP3 distribution (one PILF censored as per original GB test)
Figure 6 Flood frequency curve for Station 405238 using LP3 distribution (21 PILFs censored as per MGB test)
Figure 7 shows the fitting of the GEV distribution to Station 219001 without removing any PILFs and Figure 8 shows the fitting of LP3
distribution after removing 26 PILFs. From these two plots it is evident that LP3 distribution (Figure 8) shows a better fit to the AM flood series
than the GEV distribution.
-2 2
AEP 1 in years
-3.000
-1.884
-0.769
0.347
1.463
2.578
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Censored
Expected quantile
90% limit
Expected prob quantile
-2 2
AEP 1 in years
-1.155
-0.406
0.344
1.093
1.842
2.591
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Censored
Expected quantile
90% limit
Expected prob quantile
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 26
Figure 7 Fitting of the GEV distribution to the AM flood data for Station 219001 (No PILF censored)
Figure 8 Fitting of the LP3 distribution to the AM flood data for Station 219001 (26 PILFs censored by MGB test)
Table 4 shows the flood quantile estimates using LP3 distribution where the PILFs are identified and censored by the original GB test and
MGB test for AEPs of 1 in 10, 1 in 20, 1 in 50 and 1 in 100. It is found that there are notable differences between the two methods where
flood quantiles show a variation in the range of -61% to 28%. Table 5 shows the variation between the flood quantiles estimated by two
methods: LP3 with MGB test and GEV with L moments. It is found that for Station 218005 GEV distribution underestimates 1 in 10 AEP
flood quantile by 6.68%, but for AEPs of 1 in 20, 1 in 50 and 1 in 100, GEV overestimates the flood quantiles by 4.4%, 24.1% and 38%,
respectively. For other 9 stations the variations between the GEV and LP3 estimated quantiles are mixed i.e. a combination of over- and
under-estimation by 0.24% to 26.7%. These results highlight the expected differences in flood quantile estimates between the LP3 and GEV
distributions in eastern Australia.
-2 5
AEP 1 in years
-16
53
122
190
259
328
Peak flow m^3/s
1.5 2 5 10 20 50 100
Gauged
Expected quantile
90% limit
-2 2
AEP 1 in years
-3.000
-1.939
-0.879
0.182
1.243
2.304
log10(Peak flow m^3/s)
1.5 2 5 10 20 50 100
Gauged
Censored
Expected quantile
90% limit
Expected prob quantile
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 27
Table 4 Estimated flood quantiles and percentage difference between two sets of quantiles: LP3 with MGB test and LP3 with original GB test
Table 5 Comparison of flood quantiles by LP3 with MGB test and GEV with L moments
Flood quantiles by GEV-L moments (m3/s) Flood quantiles by LP3 with MGB test (m
3/s)
(% difference between LP3 with MGB test and GEV with L moments)
Station ID AEPs (1 in Y)
10 20 50 100 10 20 50 100
218005 1333.6 1861.3 2719.2 3522.4 1422.74 (-6.68%) 1780.04 (4.37%) 2063.8 (24.10%) 2183.6 (38.01%)
219001 66.3 93.1 137.3 179.5 79.68 (-20.18%) 103.3 (-10.96%) 128.01 (6.77%) 142.16 (20.80%)
219006 151 220.6 344.1 469.8 184.3 (-22.10%) 279.3 (-26.65%) 427.8 (-24.34%) 555.27 (-18.19%)
116008 819.1 1074.6 1457.8 1789.5 920.88 (-12.43%) 1136.2 (-5.73%) 1342.6 (7.90%) 1451.8 (18.87%)
125002 3263.1 4188.7 5488.6 6544.1 3512.61 (-7.65%) 4198.8 (-0.24%) 4887.3 (10.95%) 5276.6 (19.37%)
136202 453.3 662.5 1033.8 1411.9 565.08 (-24.66%) 836.8 (-26.32%) 1221.6 (-18.1%) 1518.39 (-7.54%)
136301 235.1 325.5 474.9 617 267.64 (-13.84%) 371.7 (-14.21%) 520.36 (-9.57%) 638.94 (-3.56%)
230204 37.5 52.4 76.9 100 46.0 (-22.72%) 60.56 (-15.57%) 76.41 (0.64%) 85.86 (14.14%)
232213 24.5 30.3 38 43.9 26.7 (-8.98%) 32.38 (-6.86%) 38.36 (-0.95%) 41.9 (4.56%)
405238 134.9 171.4 221.4 261 143.15 (-6.12%) 180.04 (-5.04%) 224.71 (-1.50%) 255.35 (2.16%)
Estimated quantiles (m3/s) using LP3 distribution
(PILFs removed by MGB test)
Estimated quantiles (m3/s) using LP3 distribution (PILFs removed by original GB
test (% difference between LP3 with MGB test and LP3 with original GB test)
Station ID AEPs ( 1 in Y)
10 20 50 100 10 20 50 100
218005 1422.74 1780.04 2063.82 2183.63 1072.7 (24.60%)
1704.52 (4.24%)
2690.8 (-30.38%)
3519.78 (-61.19%)
219001 79.68 103.3 128.01 142.16 77.55 (2.67%) 104.48 (-1.14%) 135.83 (-6.11%) 155.83 (-9.62%)
219006 184.37 279.38 427.86 555.27 133.51 (27.59%)
228.33 (18.27%)
402.83 (5.85%)
576.06 (-3.74%)
116008 920.88 1136.21 1342.61 1451.81 886.91 (3.69%)
1141.43 (-0.46%)
1454.31 (-8.32%)
1670.81 (-15.08%)
125002 3512.61 4198.86 4887.37 5276.62 2879.16 (18.03%) 3808.77 (9.29%) 4866.33 (0.43%) 5528.19 (-4.77%)
136202 565.08 836.87 1221.68 1518.39 586.6 (-3.81%) 854.9 (-2.15%) 1207.98 (1.12%) 1459.38 (3.89%)
136301 267.64 371.74 520.36 638.94 272.64 (-1.87%) 370.26 (0.40%) 498.4 (4.22%) 592.06 (7.34%)
230204 46.02 60.56 76.41 85.86 34.92 (24.12%) 53.71 (11.31%) 81.32 (-6.43%) 103.21 (-20.21%)
232213 26.7 32.38 38.36 41.9 22.3 (16.48%) 28.38 (12.35%) 34.89 (9.05%) 38.75 (7.52%)
405238 143.15 180.04 224.71 255.35 159.86 (-11.67%) 201.32 (11.82%) 240.6 (-7.07%) 260.91 (-2.18%)
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 28
5. Conclusion
This paper presents a case study using ten catchments from eastern Australia which evaluates two outlier tests being the original Grubbs and
Beck (GB) test and multiple Grubbs and Beck (MGB) test. Two most commonly adopted probability distributions i.e. the GEV and LP3 have
been adopted in the flood frequency analysis. For five stations, the original GB test did not detect any potentially influential low flows
(PILFs); however, for these stations MGB test detected 40% to 50% of the annual maximum flood peaks as PILFs. For the remaining five
stations, the original GB test identified one PILF from each station and the MGB test identified 45% to 50% as PILFs. Between these two
methods, the differences in flood quantile estimates have been found to be up to 61% for the ten study catchments. It has also been found
that GEV distribution (with L moments) and LP3 distribution with the multiple Grubbs and Beck test provide similar results in most of the
cases; however, a difference up to 38% has been found for flood quantiles for AEP of 1 in 100 for one catchment.
6. Acknowledgements
Authors would like to acknowledge Australian Rainfall and Runoff revision Project 5 team for providing the data and FLIKE software for this
study and Engineers Australia and Geosciences Australia for providing financial support for this project and Professor George Kuczera, Mr
Erwin Weinmann, Associate Professor James Ball, Mr Mark Babister and Dr William Weeks for their support and input to this study.
References
Barnett V, Lewis T (1994). Outliers in Statistical Data, John Wiley, New York.
Bates BC, Rahman A, Mein R, Weinmann PE (1998). Climatic and physical factors that influence the homogeneity of regional floods in south-
eastern Australia, Water Resources Research, 34(12), 3369–3381.
Bayley PB (1995). Understanding large river-floodplain ecosystems, Bioscience, 45, 153-162.
Beard LR (1962). Statistical methods in hydrology, Civil works investigation project CW-151, U.S. Army Corps of Engineers, Sacramento,
California.
Beckman RJ, Cook RD (1983). Outlier … … ….s”, Technometrics, 25(2), 119-149.
Benson MA (1968). Uniform flood-frequency estimating methods for federal agencies, Water Resources Research, 4(5), 891–908.
Bobée B, Cavidas G, Ashkar F, Bernier J, Rasmussen P (1993). Towards a systematic approach to comparing distributions used in flood
frequency analysis, Journal of Hydrology, 142, 121–136.
Cohn TA, England JF, Berenbroc CE, Mason RR, Stedinger JR, Lamontagne JR (2013). A generalized Grubbs-Beck test statistic for detecting
multiple potentially influential outliers in flood series, Water Resources Research, 49, 5047–5058.
Conway KM (1970). Flood frequency analysis of some NSW coastal rivers, Thesis (M. Eng. Sc.), University of New South Wales, Australia.
Cunnane C ( 1985). Factors affecting choice of distribution for flood series, Hydrological Sciences Journal, 30, 25–36.
Cunnane C (1989). Statistical distributions for flood frequency analysis, in Proc. Operational hydrological Report No. 5/33, World
Meteorological Organization (WMO), Geneva, Switzerland.
Dixon WJ (1950). Analysis of extreme values, The Annals of Mathematical Statistics, 21(1), 488–506.
Dixon WJ (1951). Ratios involving extreme values, The Annals of Mathematical Statistics, 22(1), 68–78.
Gentleman J, Wilk M (1975). Detecting PILFs. II. Supplementing the direct analysis of residuals, Biometrics, 31(2), 387–410.
Gotvald AJ, Barth NA, Veilleux AG, Parrett C (2012). Methods for determining magnitude and frequency of floods in California, based on data
through water year 2006, U.S. Geological Survey, Reston, Virginia.
Griffis V, Stedinger JR (2007). The LP3 distribution and its application in flood frequency analysis, 2. Parameter estimation methods, Journal of Hydrologic Engineering, 12(5), 492–500.
Grubbs FE (1969). Procedures for Detecting Outlying Observations in Samples, Technometrics, 11(1), 1-21.
Grubbs FE, Beck G (1972). Extension of sample sizes and percentage points for significance tests of outlying observations, Technometrics, 4(14),
847–853.
Haddad K, Rahman A (2008). Investigation on at-site flood frequency analysis in south-east Australia, IEM Journal, The Journal of The
Institution of Engineers, Malaysia, 69(3), 59-64.
Haddad K, Rahman A (2010). Selection of the best fit flood frequency distribution and parameter estimation procedure: a case study for
Tasmania in Australia, Stochastic Environmental Research and Risk Assessment, 25, 415–428.
Haddad K, Rahman A, Kuczera G (2011). Comparison of ordinary and generalised least squares regression models in regional flood frequency
analysis: a case study for New South Wales, Australian Journal of Water Resources, 15(2), 59–70.
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 29
Haddad K, Rahman A, Stedinger JR (2012). Regional Flood Frequency Analysis using Bayesian Generalized Least Squares: a Comparison
between Quantile and Parameter Regression Techniques, Hydrological Processes, 26, 1008-1021.
Haddad K, Rahman A (2012). Regional flood frequency analysis in eastern Australia: Bayesian GLS regression-based methods within fixed
region and ROI framework—quantile regression versus parameter regression technique, Journal of Hydrology, 430–431, 142–161.
Haddad K, Rahman A, Zaman M, Shrestha S (2013). Applicability of Monte Carlo cross validation technique for model development and
validation in hydrologic regression analysis using ordinary and generalised least squares regression, Journal of Hydrology, 482, 119–128.
Hadi A, Simonoff J (1993). Procedures for the identification of multiple outliers in linear models, Journal of the American Statistical
Association, 88(424), 1264– 1272.
Interagency Advisory Committee on Water Data (IAWCD) (1982). Guidelines for Determining Flood Flow Frequency: Bulletin 17-B,
Hydrology Subcommittee, Washington DC.
Institution of Engineers Australia (I. E. Aust.) (1987). Australian Rainfall and Runoff: A Guide to Flood Estimation, Canberra.
Institution of Engineers Australia (I. E. Aust.) (2001). Australian Rainfall and Runoff: A Guide to Flood Estimation, Canberra.
Interagency Advisory Committee on Water Data (IAWCD) (2013). Robust National Flood Frequency Guidelines: What is an Outlier? Bulletin
17C, IAWCD, USA.
Ishak EH, Rahman A, Westra S, Sharma A, Kuczera G (2010). Preliminary analysis of trends in Australian flood data, in Proc.World
Environmental and Water Resources Congress, American Society of Civil Engineers (ASCE), Providence, Rhode Island, USA, 120-124.
Ishak EH, Haddad K, Zaman M, Rahman A (2011). Scaling property of regional floods in New South Wales Australia, Natural Hazards, 58,
1155–1167.
Kirby W (1972). Computer-oriented Wilson–Hilferty transformation that preserves the first three moments and the lower bound of the Pearson
type 3 distribution, Water Resources Research, 8(5), 125l–l254.
Kopittke RA, Stewart BJ, Tickle KS (1976). Frequency analysis of flood data in Queensland, in Proc. Hydrological Symposium, Institution of
Engineers Australia, National Conference, Publication No. 76/2, 2-24.
Kuczera G (1999). Comprehensive at-site flood frequency analysis using Monte Carlo Bayesian inference, Water Resources Research, 35(5),
1551–1557.
Lamontagne JR, Stedinger JR, Cohn TA, Barth NA (2013). Robust national flood frequency guidelines: What is an outlier? In Proc. World
Environmental and Water Resources Congress, ASCE.
Laio F (2004). Cramer-von Mises and Anderson-Darling goodness of fit tests for extreme value distributions with unknown parameters, Water
Resources Research, 40:W09308. doi:10.1029/2004WR003204.
Laio F, Di Baldassarre G, Montanari A (2009). Model selection techniques for the frequency analysis of hydrological extremes, Water Resources
Research, 45:W07416. doi:10.1029/2007/WR006666.
Marasinghe M (1985). A multistage procedure for detecting several PILFs in linear regression, Technometrics, 27(4), 395–399.
McCuen RH, Ayyub BM (1996). Probability, statistics, and reliability for engineers and scientists, A Chapman & Hall book, USA.
McMahon TA, Srikanthan R (1981). Log Pearson III distribution is it applicable to flood frequency analysis of Australian streams? Journal of
Hydrology, 52. 1-2, 139-147.
Merz R, Bloschil G, Humer G (2008). National flood discharge mapping in Austria, Natural Hazards, 46, 53–72.
Meshgi A, Khalili D (2009a). Comprehensive evaluation of regional flood frequency analysis by L-abd LHmoments. 1. A re-visit to regional
homogeneity, Stochastic Environmental Research and Risk Assessment, 23, 119–135.
Meshgi A, Khalili D (2009b). Comprehensive evaluation of regional flood frequency analysis by L-and LHmoments. II. Devewlopment of LH-
moments parameters for the generalized Pareto and generalizedlogistic distributions, Stochastic Environmental Research and Risk
Assessment, 23, 137–152.
Nathan RJ, Weinmann PE (1991). Application of at-site and regional flood frequency analyses, in Proc. International Hydrology Water
Resources Symposium, Perth, 769-774.
Onoz B, Bayazit M (1995). Best-fit distribution of largest available flood samples, Journal of Hydrology, 167(1-4), 195-208.
Poff NL, Allan JD, Bain MB, Karr JR, Prestegaard KL, Richter BD, Sparks RE and Stromberg JC (1997). The natural flow regime: a paradigm
for river conservation an restoration, Bioscience, 47, 769-784.
Prescott P (1975). An approximate test for PILFs in linear models, Technometrics, 17(1), 129–132.
Prescott P (1978). Examination of the behaviour of tests for outliers when more than one outlier is present, Applied Statistics, 27, 10–25.
Rahman A, Weinmann PE, Mein RG (1999). At-site flood frequency analysis: LP3-product moment, GEV-L moment and GEV-LH moment
procedures compared, in Proc. Hydrology and Water Resource Symposium, Brisbane, 715–720.
Rahman AS, Rahman A, Zaman M, Haddad K, Ashan A, Imteaz MA (2013). A Study on Selection of Probability Distributions for At-site Flood
Frequency Analysis in Australia, Natural Hazards, 69, 1803-1813.
“Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia” Rahman et al.
Journal of Hydrology and Environment Research 30
Rosner B (1975). On the detection of many outliers, Technometrics, 17(2), 221–227.
Rosner B (1983). Percentage points for a generalized ESD many outlier procedure, Technometrics, 25(2), 165–172.
Rousseeuw P, Zomeren BV (1990). Unmasking multivariate PILFs and leverage points, Journal of American Statistical Association, 85(411),
633–639.
Rousseeuw P, Leroy A (2003). Robust Regression and Outlier Detection, John Wiley, Hoboken, N. J.
Saf B (2010). Assessment of the effects of discordant sites on regional flood frequency analysis, Journal of Hydrology, 380 (3–4), 362–375.
Spencer C, McCuen R (1996). Detection of PILFs in Pearson type III data, Journal of Hydrologic Engineering, 1, 2–10.
Stedinger JR, Vogel RM, Foufoula-Georgiou E (1993). Frequency Analysis of Extreme Events, Chapter 18, Handbook of Hydrology, McGraw
Hill, New York.
Thomas WO (1985). A uniform technique for flood frequency analysis, Journal of Water Resources Planning and Management, 111(3), 321–
337.
Thompson W (1935). On a criterion for the rejection of observations and the distribution of the ratio of deviation to sample standard deviation,
The Annals of Mathematical Statistics, 6, 214–219.
Tietjen G, Moore R (1972). Some Grubbs-type statistics for the detection of several outliers, Technometrics, 14(3), 583–597.
Verma S, Quiroz-Ruiz A (2006). Critical values for six Dixon tests for outliers in normal samples up to sizes 100, and applications in science
and engineering, Revista Mexicana de Ciencias Geológicas, 23(2), 133–161.
Vogel RM, McMahon TA, Chiew FHS (1993). Flood flow frequency model selection in Australia, Journal of Hydrology, 146, 421-449.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 31
New Watershed Codification System for Indian River Basins
Kuldeep Pareta 1,*
, Upasana Pareta 2
1 Water Resource Division, Spatial Decisions, New Delhi, India
2 Department of Mathematics, PG Collage, District Sagar (M.P.), India
Peer Review History1
Abstract: The international watershed codification system for the Indian River basin is proposed for the better water resource management &
monitoring, river basin planning, innovative research in hydrology, and sustainable water resource development. Based on natural system, the
sub-continent largest transboundary to mini-watershed boundaries have been delineated from SRTM, ASTER, & CARTOSAT DEM data. The nine-digit watershed codification is proposed for the Indian River basins, recognizing each hydrologic unit with unique international code that
provides a single stand to synergize all the development programs related to river basin planning, and natural/water resource management, and
avoiding doubling of interventions of various departments & ministries.
Keywords: Watershed codification, Indian river basin, DEM data, GIS.
1. Introduction
India with 2.4% of the world's total area has 17.31% of the world's population; but has only 4% of the total available fresh water. India has a
3.29 billion hectare geographical area, which is covered by 72 major rivers and its numerous tributaries. A major part of Indian population is
rural and engaged with agricultural activities; for this, the river basin management plays is vital role to source the fresh water. The
Government of India (GoI) has adopted watershed management as a strategy to address the sustainable agricultural productivity in the
rainfed areas as a national policy since 2003. The main objective of this study is to accumulate information on the overall Indian River basins
to smallest hydrological units (mini-WS), and create an international single nine-digit code to the mini-WS that can be applied towards
strategy and policy, primarily for the Ministry of Water Resources as well as general guidelines for the State Water Resources Departments
(GoI, 1999).
2. Method description of existing systems
Various codification systems for river basins have been developed by various organizations in India such as All India Soil and Land Use
Survey, National Commission for Integrated Water Resources Development Plan, Central Water Commission, Central Groundwater Board,
National Remote Sensing Centre of Indian Space Research Organization, and World Resources Institute, which are slightly different to each
other. These codification systems directly address the need for numbering of natural landscape units, which is a focus of river basin
management. Some of the existing codification systems are described below.
2.1 All India Soil and Land Use Survey (AISLUS)
Central Water and Power Corporation (CWPC) initially attempted to defined Indian River basins in 1949 under the intelligent guidance of
A.N. Khosla, and ALS & LUC had developed the “Watershed Atlas of India” on 1:1 million scale following the stream order where the entire
river systems of the country have been divided into 6 water resources regions, which has been further divided into 35 basins and 112
1 Paper JHER0205, submitted on 21/11/2014, accepted for publication after peer review and subsequent revisions on 14/12/2014
* Corresponding author may be contacted at [email protected]
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 32
catchments, 500 sub-catchments and 3,237 watersheds (WAI, 1990). The delineation has been done in seven stages starting with water
resource regions and their subsequent divisions and subdivisions into basins, catchments, sub-catchments, watershed, sub-watershed and
micro-watersheds in decreasing size of the delineated hydrologic unit.
2.2 National Commission for Integrated Water Resources Development Plan (NCIWRDP)
The NCIWRD in 1999 prepared a comprehensive assessment of the water resource availability in India, which estimated the water resource to
be 1969 billion m3 (BCM) including surface and groundwater (NCIWRD, 1999). During the water resource assessment study, NCIWRD
divided the entire country into 23 basins which included 13 major basins, and 10 composite basins.
2.3 Central Water Commission (CWC)
CWC defined river basins of India in 2002 during nation-wide data collection over 878 hydrological observation stations covering both classified
and non-classified river basins. CWC has divided river basins of India into 15 basins, which have been further divided into 3 basins as classified,
and other 12 basins as non-classified (CWC, 1989, 1997).
2.4 Central Ground Water Board (CGWB)
Central Groundwater Board prepared a “Watershed Atlas of India” at 1:250,000 scale using GIS techniques under the guidance of Saleem
Romani in 2006. In this Atlas, the entire river system of the country has been divided into 34 basins, 94 sub-basins, and 3,448 watersheds. They
have also generated various thematic layers significant to groundwater development/assessment on watershed basis and for Hydrological
Information System (HIS).
2.5 Water Resource Information System in India (India-WRIS)
The "River Basin Atlas of India" was launched by Harish Rawat (Minister of Water Resources, India) on 1st November, 2012 (India-WRIS,
2012). It is a joint project “Generation of database and implementation of web enabled Water Resources Information System (India-WRIS) in
the country” of Central Water Commission (CWC) and National Remote Sensing Centre (NRSC) of Indian Space Research Organization
(ISRO). The India-WRIS Project national level watershed atlas has been prepared on 1:50,000 scale by using SRTM DEM data of NASA, having
a spatial resolution of 90 meters, and has been divided into 25 major river basins and 103 sub-basins (STRM, 2006).
2.6 Watersheds of the World
The “Watersheds of the World” has been published by the World Conservation Union (IUCN), the International Water Management Institute
(IWMI) (IWMI, 2001), the Ramsar Convention Bureau and the World Resources Institute (WRI) in July 2003. WRI for the first time presents
and analyzes a wide range of global data at the watershed level, assessing 154 watersheds around the world. River basin boundaries have been
delineated from ETOPO-5, 5 minute gridded elevation data, and USGS’ 30 arc-second digital elevation model of the world (GTOPO30). WRI
revised and checked basin boundaries by overlaying ArcWorld 1:3 million rivers, where rivers (except canals) crossed basin boundaries; the
boundary was edited using a 1-kilometer digital elevation model.
WRI has also published “Primary Watersheds Map” that shows the location of 114 major watersheds of the world with international watershed
code. It includes the world’s largest transboundary watersheds and other small basins that are representative of a particular geographic area.
They have delineated 20 major watersheds from Africa, 29 major watersheds from Europe, 19 major watersheds from North & Central America,
11 major watersheds from South America, 5 major watersheds from Oceania, and 30 major watersheds from Asia. Out of 30’s Asian major
watersheds, 10 major watersheds are situated in the Indian sub-continent area.
3. Data Used
As per Table 1, several types of data have been used for creations of international watershed, water division, water sub-division, basin, sub-basin,
major watershed, and micro-watershed.
4. Methodology
In the present study, the river/drainage network has been generated from ASTER (DEM) data with 30m spatial resolution as well as Cartosat-
1 DEM data with resampled 30m spatial resolution, and the river/stream names have been captured from topographic maps at 1:250,000
scales. Indian sub-continent largest transboundary watersheds, water divisions, and water sub-divisions have been extracted from Shuttle
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 33
Radar Topography Mission (SRTM) DEM Data, their watersheds boundaries have been also redesigned with the help of ASTER (DEM), and
Cartosat-1 DEM data, and also the same data have been used for delineation of basins, sub-basins, watersheds, and sub-watersheds.
Administrative boundaries and major locations of the towns have been obtained from administrative maps of Survey of India/National
Informatics Center (NIC), New Delhi.
Table 1 Data used and sources
S.No. Data Used Sources
1. India and Pakistan
Topographic Map @ 1:250,000
Series U502, U.S. Army Map Service, 1955
http://www.lib.utexas.edu/maps/ams/india
2. Shuttle Radar Topography Mission (SRTM),
DEM Data @ 90m Spatial Resolution
NASA, & USGS EROS Data Center, 2006
http://glcfapp.glcf.umd.edu:8080/esdi
3. ASTER Global Digital Elevation Model (GDEM),
DEM Data @ 30m Spatial Resolution (GDEM,
2009)
Japan Space Systems (J-space systems) Japan, cooperation with US, 2009
http://gdem.ersdac.jspacesystems.or.jp/search.jsp
4. Cartosat-1 Digital Elevation Model (CartoDEM),
DEM Data @ 30m Spatial Resolution
(CartoDEM, 2008)
Indian Earth Observation, National Remote Sensing Centre (ISRO), 2008
http://bhuvan.nrsc.gov.in/data/download/index.php
5. Watersheds of the World http://www.wri.org
5. Proposed Watershed Codification System
The system proposed here for the delineation and codification of the Indian River Basins is established upon concepts, first articulated by the
A.N. Khosla with the Central Water and Power Corporation (CWPC), published by All India Soil and Land Use Survey “AIS&LUS” (WAI,
1949), and Watersheds of the World published by IUCN, IWMI, RCB, & WRI in 2003 (WRI, 2003). It is a common framework based upon
topographic control of regions drained on the earth surface and the topology of the ensuing hydrographic system.
WRI just extracted the 114 international watersheds with international watershed code. As per A.N. Khosla the watershed outline has been
done in seven stages beginning with water resource regions (represent 1 to 6) and their resulting separation and subdivisions into basins
(symbolize A, B, C,…, Z), catchments (represent 1, 2, 3,…, 9), sub-catchments (represent A, B, C,…, Z), watershed (represent 1 to 9), sub
watershed (represented with small English alphabets as a, b, c,…, z) and micro-watersheds (characterized by numerals as 1 to 6) in decreasing
size of the outlined hydrologic unit. For example '2C2C5h1' codes connote ‘2’ - water resource region (Ganga), 'C'- basin (Yamuna), '2' -
Betwa catchment, 'C'- Betwa-Dhasan (Upper Dhasan) sub-catchment, '5'- Karawan watershed, 'h' - Garhpehra sub-watershed, and '1' - micro-
watershed code. The beauty of this code is that it represents a national code to be useful for the easy study of the Indian river system, but due
to lack of international code and detailed information of basin, it does not represent a realistic information of the basin.
After the detailed study of watershed codification, this paper suggests a “New Watershed Codification System” for Indian River Basin, which
will be more useful for the study of systematic river basin planning, watershed management, etc., and this code should represent any micro-
watershed in an international stand.
5.1 Indian Sub-Continent Largest Transboundary Watersheds
The existing international watershed code for 10 watersheds has been taken from the “Watersheds of the World” published by IUCN, IWMI,
RCB, & WRI in 2003, and this paper has suggested some additional international watershed code for the remaining river (12 Watersheds) in
India, which has been used for generation of new watershed codification system for Indian river basins. Table 2 and Figure 1 show the
existing and proposed international watershed code for the Indian River basins.
5.2 Water Divisions
On the basis of drainage flowing into ocean and other basins, three water divisions have been suggested. The water division code has been
used as A for all drainage flowing into the Arabian sea, B for all drainage flowing into the Bay of Bengal, and X for other rivers draining into
other basins. The details of water divisions are (see also Figure 2): (i) All Drainage flowing into Arabian Sea (A); (ii) All Drainage flowing
into Bay of Bengal (B); and (iii) Other River Draining into Other Basin (X).
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 34
5.3 Water Sub-Divisions
On the basis of drainage flowing into ocean from north India, and south India, each water division has been divided into water sub-divisions.
The water sub-division code has been used as 1 for North India, and 2 for South India. The details of six water sub-divisions are shown below
and in Figure 2.
(i) All Drainage flowing into Arabian Sea from North India (A1)
(ii) All Drainage flowing into Arabian Sea from South India (A2)
(iii) All Drainage flowing into Bay of Bengal from North India (B1)
(iv) All Drainage flowing into Bay of Bengal from South India (B2)
(v) Other River Draining into Other Basin from North India (X1)
(vi) Other River Draining into Other Basin from Indian Island (X2)
Table 2 International watershed code, watershed/ basin name for India river basin (WRI, 2003) with modification
Watershed Code Watershed / Basin Name Status Country
AS04 Brahmaputra Existing China, Nepal, India, Bangladesh
AS06 Ganga Existing India, Nepal, Bangladesh
AS07 Godavari Existing India
AS11 Indus Existing Pakistan, India, China, Afghanistan
AS12 Irrawaddy Existing Myanmar, India
AS15 Krishna Existing India
AS18 Mahanadi Existing India
AS20 Narmada Existing India
AS25 Tapti Existing India
AS26 Tarim Existing China, India, Soviet Union
AS31 Drainage in NW India Suggested India
AS32 Drainage in Western Ghat Suggested India
AS33 Kaladan Suggested India, Myanmar
AS34 Damodar Suggested India
AS35 Subernarekha & Other River Suggested India
AS36 Vamsadhara & Nagvati Suggested India
AS37 Penner (Palar) Suggested India
AS38 Ponnaiyar Suggested India
AS39 Cauvery Suggested India
AS40 Pamba &Vaippar Suggested India
AS41 Drainage in Andaman & Nicobar Suggested India
AS42 Drainage in Lakshadweep Suggested India
5.4 Basins
The basin boundaries have been marked as the same boundary of Indian sub-continent largest transboundary. The basin codes are indicated by
suffixing alphabets to the water sub-division code as “Nm” for Narmada, “Gn” for Ganga, etc. The details of 22 basins with code and basin
name are shown in Table 3.
5.5 Sub-Basins
Total 72 sub-basin boundaries have been delineated as the partition of basin boundary. Each basin has been divided into a number of sub-
basins, which connect to main tributaries or individual streams. The sub-basin codes are designated again by suffixing alphabets to basin code as
“(NMD)” for Narmada, “(CMB)” for Chambal, “(LGN)” for Lower Ganga, and “(DSN) for Dhasan, etc. The details of sub-basin code with
name are shown in Table 3. Figure 3 shows the Indian River basin, and sub-basins boundary with respective codes.
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 35
Figure 1 Indian sub-continent largest transboundary watersheds
Figure 2 Water divisions and water sub-divisions
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 36
5.6 Watersheds
Each sub-basin is further divided into watersheds, in which sub-tributaries and streams are taken up for delineation of watersheds. Total 814
watersheds have been delineated from the 72 sub-basin boundaries. Watershed code represented by numerals suffixed to sub-basin code as 1,
2, 3, …, 40. The details of watersheds codes are shown in Table 3. Basin boundary, sub-basin boundary, and major watershed in India with
watershed code in Dhasan sub-basin are shown in Figure 4.
5.7 Sub-Watersheds
Each watershed has been divided into sub-watershed based on main tributaries and streams extracted from Cartosat/ASTER (DEM), and each
sub-watershed code represented by small English alphabets as a, b, c, …, z, which is suffixed to watershed code. Comprehensive code for a sub-
watershed is “AS06B1Gn(DSN)11k” as example of a sub-watershed of Dhasan sub-basin.
Table 3 Basin/sub-basin name with code, & number of watersheds for India river basins
S. N
o.
Inte
rnat
iona
l Wat
ersh
ed
Cod
e
Wat
er D
ivis
ion
Wat
er S
ub-D
ivis
ion
Basin Name (Code) Sub-Basin Name (Code) International Sub-Basin
Code
No.
of W
S
Tota
l No.
of W
S
1
AS11 A 1 Indus (Id)
Gilgit (GGT) AS11A1Id(GGT) 1
65
2 Shyok (SYK) AS11A1Id(SYK) 11
3 Indus (IND) AS11A1Id(IND) 12
4 Jhelum (JHM) AS11A1Id(JHM) 4
5 Chenab (CNB) AS11A1Id(CNB) 12
6 Ravi (RVI) AS11A1Id(RVI) 8
7 Beas (BAS) AS11A1Id(BAS) 1
8 Sutlej (STJ) AS11A1Id(STJ) 9
9 Barmer (BMR) AS11A1Id(BMR) 7
10
AS31 A 1 Drainage in NW India
(Nw)
Ghaghar (GGH) AS31A1Nw(GGH) 10
110
11 Chautang & Other (CTG) AS31A1Nw(CTG) 4
12 Churu (CRU) AS31A1Nw(CRU) 26
13 Luni (LUN) AS31A1Nw(LUN) 35
14 Saraswati (SWT) AS31A1Nw(SWT) 7
15 Sabarmati (SMT) AS31A1Nw(SMT) 7
16 Mahi (MHI) AS31A1Nw(MHI) 10
17 Drainage in Rann (RAN) AS31A1Nw(RAN) 7
18 Bhadra & Other (BDR) AS31A1Nw(BDR) 3
19 Shetranjuli & Other (SJO) AS31A1Nw(SJO) 1
20 AS20 A 1 Narmada (Nm) Narmada (NMD) AS20A1Nm(NMD) 32 32
21
AS32 A 2 Drainage in Western Ghat
(Wg)
Bhatsol and Other (BTS) AS32A2Wg(BTS) 9
37
22 Vasishti and Other (VSO) AS32A2Wg(VSO) 10
23 Nagvati & Other (NGO) AS32A2Wg(NGO) 11
24 Varrae and Other (VRO) AS32A2Wg(VRO) 3
25 Periya and Other (PRO) AS32A2Wg(PRO) 4
26 AS25 A 2 Tapti (Tp) Tapti (TPT) AS25A2Tp(TPT) 16 16
27
AS04 B 1 Brahmaputra (Bp)
Brahmaputra Right Bank (BRB) AS04B1Bp(BRB) 27
70 28 Brahmaputra Left Bank (BLB) AS04B1Bp(BLB) 35
29 Barak & Other (BRK) AS04B1Bp(BRK) 8
30
AS06 B 1 Ganga (Gn)
Upper Ganga (UGN) AS06B1Gn(UGN) 22
204
31 Lower Ganga (LGN) AS06B1Gn(LGN) 30
32 Ramganga (RGN) AS06B1Gn(RGN) 8
33 Ghaghara (GGR) AS06B1Gn(GGR) 14
34 Gandak (GDK) AS06B1Gn(GDK) 2
35 Kosi (KSI) AS06B1Gn(KSI) 4
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 37
36 Mahananda (MHD) AS06B1Gn(MHD) 9
37 Gomti (GMT) AS06B1Gn(GMT) 4
38 Yamuna (YMN) AS06B1Gn(YMN) 24
39 Banas (BNS) AS06B1Gn(BNS) 12
40 Chambal (CMB) AS06B1Gn(CMB) 18
41 Kali Sindh (KSN) AS06B1Gn(KSN) 3
42 Parbati (PBT) AS06B1Gn(PBT) 2
43 Sindh (SND) AS06B1Gn(SND) 2
44 Betwa (BTW) AS06B1Gn(BTW) 7
45 Dhasan (DSN) AS06B1Gn(DSN) 11
46 Ken (KEN) AS06B1Gn(KEN) 8
47 Tons (TNS) AS06B1Gn(TNS) 9
48 Son (SON) AS06B1Gn(SON) 15
49 AS34 B 1 Damodar (Dd) Damodar (DMD) AS34B1Dd(DMD) 9 9
50 AS35 B 2
Subernarekha & Other
River (So)
Subernarekha (SNK) AS35B2So(SNK) 4 14
51 Baitami & Brahmani (BHM) AS35B2So(BHM) 10
52 AS18 B 2 Mahanadi (Mn) Mahanadi (MHN) AS18B2Mn(MHN) 40 40
53
AS07 B 2 Godavari (Gv)
Weinganga (WGN) AS07B2Gv(WGN) 13
78 54 Penganga (PGN) AS07B2Gv(PGN) 12
55 Indravati (IRV) AS07B2Gv(IRV) 17
56 Godavari (GVR) AS07B2Gv(GVR) 36
57 AS36 B 2
Vamsadhara & Nagvati
(Vn)
Vamsadhara & Other (VDO) AS36B2Vn(VDO) 3 9
58 Netravati and Other (NTO) AS36B2Vn(NTO) 6
59
AS15 B 2 Krishna (Ks)
Bhima (BMA) AS15B2Ks(BMA) 19
56 60 Krishna (KSN) AS15B2Ks(KSN) 29
61 Tungabhadra (TBD) AS15B2Ks(TBD) 8
62 AS37 B 2 Penner (Palar) (Pn) Palar and Other (PLO) AS37B2Pn(PLO) 14 14
63 AS39 B 2 Cauvery (Cv) Cauvery (CVR) AS39B2Cv(CVR) 22 22
64 AS38 B 2 Ponnaiyar (Py) Ponnaiyar and Other (PNO) AS38B2Py(PNO) 16 16
65 AS40 B 2 Pamba & Vaippar (Pv)
Pamba and Other (PMO) AS40B2Pv(PMO) 6 12
66 Vaippar and Other (VPO) AS40B2Pv(VPO) 6
67 AS26 X 1 Tarim (Tr)
Shaksgam (SKG) AS26X1Tr(SKG) 1 2
68 Sulmar (SLM) AS26X1Tr(SLM) 1
69 AS12 X 1 Irrawaddy (Iw) Mangpui Lui & Other (MLO) AS12X1Iw(MLO) 6 6
70 AS33 X 1 Kaladan (Kd) Karnaphuli & Muhury (KPM) AS33X1Kd(KPM) 3 3
71 AS41 X 2 Dr. in Andaman &
Nicobar (An) Drainage in Andaman & Nicobar (AMN) AS41X2An(AMN) 1 1
72 AS42 X 2 Drainage in Lakshadweep
(Ld) Drainage in Lakshadweep (LKD) AS42X2Ld(LKD) 1 1
22 3 6 22 72 817
5.8 Micro-Watershed
Micro-watershed can be defined as natural hydrological element that covers a particular extent of terrestrial surface from which the
precipitation, and runoff flows into a well-defined river, stream, drainage, or channel at any specific point. Micro-watershed (MWS) boundary
has been delineated based on the stream ordering system (Strahler, 1952); it has defined on each of 3rd order stream, but limited to starting of
the 3rd order stream, with the area from 100 hectare to 300 hectare (Pareta, 2004; Pareta and Pareta, 2012). MWS code has followed the similar
system as proposed above, MWS code represented by numerals suffixed to sub-watershed code as 1, 2, 3, …, n. The completed code for a micro-
watershed is “AS06B1Gn(DSN)11k13” as example of a micro-watershed of Dhasan sub-basin.
5.9 Mini-Watershed
Mini-watershed (Mini-WS) boundary has been demarcated on each of 2nd order stream, but limited to starting of the 2nd order stream, with
the area of less than 100 hectare (Pareta and Pareta, 2011). The completed code for a mini-watershed with nine digits is symbolized as
“AS06B1Gn(DSN)11k13b”, as an example of a mini-watershed of Dhasan sub-basin (Figure 5), where “AS06” represents Indian Sub-
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 38
Continent Largest Transboundary, “B” for Water Division, “1” for Water Sub-Division, “Gn” for Basin, “DSN” for Sub-Basin, “11” for
Watershed, “k” for Sub-Watershed, “13” for Micro-Watershed, and “b” for Mini-Watershed.
Figure 3 Indian river basins and sub-basins boundary with code
Figure 4 Indian major watershed and watershed code in Dhasan sub-basin
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 39
Figure 5 Watershed, sub-watershed, micro-watershed and mini-watershed boundary with code
6. Conclusion
The codification systems of river basins are different in different countries. Consequently, the river basin/watershed boundaries do not match
with each other internationally. To overcome this limitation, a comprehensive uniform categorizations and codifications are proposed for Indian
River basins. A series of single nine-digit code is sufficient to uniquely identify mini-watershed (mini-WS). This paper suggests this codification
system as an essential spatial framework that can be used to reconcile the data and information from a variety of scales for better water resource
management, river basin planning, and sustainable water resource development.
7. Acknowledgement
We are profoundly thankful to our Guru Ji Professor J. L. Jain, who with his unique research competence, selfless devotion, thoughtful
guidance, inspirational thoughts, wonderful patience and above all parent like direction and affection motivated us to pursue this work.
References CartoDEM (2008). Cartosat-1 Digital Elevation Model (CartoDEM). Indian earth observation, National Remote Sensing Centre (ISRO):
http://bhuvan.nrsc.gov.in/data/download/index.php, India.
CWC (1997). India River Basin Atlas. Central Water Commission (CWC), New Delhi, India.
CWC (1989). Major River Basins of India - An Overview (1989). Central Water Commission (CWC), New Delhi, India.
GDEM (2009). ASTER Global Digital Elevation Model (GDEM). Japan space systems (J-space systems) Japan, cooperation with US:
http://gdem.ersdac.jspacesystems.or.jp/search.jsp.
GoI (Government of India) (1999). Integrated water resources development. A plan for action, Report of the commission for integrated water
resource development, Volume I. New Delhi, Ministry of Water Resources, India.
India-WRIS (2012). Water resource information system in India (India-WRIS): http://www.india-wris.nrsc.gov.in.
IWMI (2001). IWMI climate and water atlas. Colombo, Sri Lanka: IWMI CD-ROM.
“New Watershed Codification System for Indian River Basins” Pareta and Pareta
Journal of Hydrology and Environment Research 40
NCIWRD (1999). Water requirement for various uses. National Commission on Integrated Water Resources Development, Government of
India.
Pareta K (2004). Hydro-geomorphology of Sagar district (M.P.): A study through remote sensing technique, Proceeding in XIX M. P. Young
Scientist Congress, Madhya Pradesh Council of Science & Technology (MAPCOST) Bhopal, India.
Pareta K, Pareta U (2011). Hydromorphogeological study of Karawan watershed using GIS and remote sensing techniques, International Scientific Research Journal, 3(4), 243-268.
Pareta K, Pareta U (2012). Quantitative Morphometric Analysis of A Watershed: Based on digital terrain Model and GIS, LAP Lambert
Academic Publishing, Germany.
Strahler AN (1952). Dynamic basis of geomorphology, Bulletin of the Geological Society of America, 63, 923-938.
STRM (2006). Shuttle Radar Topography Mission (SRTM) 90m Digital Elevation Model: CGIAR (2006). http://srtm.sci.cgiar.org.
WAI (1990). Watershed Atlas of India (WAI). All India Soil and Land Use Survey (AISLUS), Ministry of Agriculture, New Delhi.
WRI (2003). Watersheds of the world. International Union for Conservation of Nature (IUCN), International Water Management Institute
(IWMI), Rasmar Convention Bureau (RCB), and World Resources Institute (WRI), Washington, DC.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 41
Assessment of Heavy Metal Contamination from Municipal Solid Waste Open
Dumping Sites in Bangladesh
Md. Rezaul Karim 1,*
, Megumi Kuraoka 2, Takaya Higuchi
2, Masahiko Sekine
2, Tsuyoshi Imai
2
1 Department of Civil and Environmental Engineering, Islamic University of Technology (IUT), Gazipur, Bangladesh
2 Department of Civil and Environmental Engineering, Yamaguchi University, 2-16-1 Tokiwadai, Ube 755-8611, Japan
Peer Review History1
Abstract: Co-disposal of the household originated hazardous materials with municipal solid waste (MSW) into the open dumping sites is the
usual practice in Bangladesh. In this paper, characterization of heavy metals in MSW in the open dumping sites in Bangladesh is presented.
MSW samples were collected from two open dumping sites at Matuail, Dhaka and Khulna and analyzed for total heavy metals content (Cd, Co, Cr, Cu, Mn, Ni, Pb and Zn) and also heavy metal fraction (water extractable, exchangeable and bio-fraction). Moreover, Toxicity Characteristic
Leaching Procedure (TCLP) extraction was done to evaluate pollution potential of heavy metals from the dumping sites. Leachate samples were
also collected and analyzed for heavy metals content. The results of the analysis showed that the total metals content in MSW at the Matuail dumping site is higher than Khulna dumping site and the metals are predominantly associated with fine soil fraction. The total heavy metals
content in MSW in the study sites are less than the total metals content in MSW at the dumping sites reported from Japan, India and Thailand.
The study results showed that both sites contain high bio-available fraction of metals, which may easily be entered into a food chain and may
cause health hazards. The result of TCLP extraction with the USEPA regulatory levels showed that the dumping sites are non-hazardous in nature in the context of heavy metal pollution. The runoff leachate also contains insignificant concentration of heavy metals. Under the present
condition prevailing at the dumping sites, the dissolution of the acid soluble metal and the associated risk of heavy metal contamination are
deemed very low. The findings may be useful as a first step in evaluating the heavy metals pollution potential from the open dumping sites in Bangladesh.
Keywords: Open dumping, municipal solid waste, heavy metal, leaching, leachate, Bangladesh.
1. Introduction In Bangladesh, urban population especially in city areas is increasing rapidly. This rapid increase in urban population causes a significant
pressure on urban services like municipal solid waste (MSW) management. MSW management systems in all cities in Bangladesh are very
much traditional and labor based and most of the solid wastes are disposed in the open dumping spaces due to lack of regulatory systems and
effective management. Open dumping of municipal solid waste without source segregation is the usual practice in developing countries like
Bangladesh. Co-disposal of household hazardous waste including batteries, paint residues, ash, treated woods and electronic wastes increases the
heavy metal content in MSW dumping sites (Para et al., 1999; Esakku et al., 2008; Ahsan et al., 2013).
One of the major environmental impacts of municipal solid waste disposal is the influence of heavy metals in the dumping site. The effects of
heavy metals are found to vary with the conditions prevailing in the dumpsites and its binding forms. The open dumpsite being exposed to the
atmospheric condition undergoes different effects due to oxygen diffusion. Under high redox condition, the binding of metals to Mn and Fe
oxide increases, whereas binding to carbonate, organic compound and sulfide tends to decrease (Prechthai et al., 2008). With more possibility of
oxygen diffusion through the upper layer of the dumpsite and with sufficient moisture content, the degradation rate and the acid buffer capacity
1 Paper JHER0201, submitted on 18/02/2014; accepted for publication after peer review and subsequent revisions on 20/10/2014
* Corresponding author may be contacted at [email protected]
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 42
of the dumpsite is highly influenced. Under this condition, there is a drop in alkalinity, pH and sulfide oxidation, where heavy metals are easily
available and released (Bozkurt et al., 2000; Prechthai et al., 2008).
Professionals in MSW management often require interpretation of the leachability of metals in order to assess the risk of dumping sites to
human health and environment (Scott et al., 1990). Leaching tests are often applied in assessing worse case environmental scenario, where
components of the samples become soluble and mobile. Various leaching methods have been cited in the literature. The mobility and toxicity of
heavy metals present in dumping sites depend on the chemical form of the metals. Knowledge of heavy metal content, their species and
leachability at various environmental conditions from the dumping site is a pre-requisite for assessing hazardous potential to the environment.
Therefore, the hazardous potential of dumpsites can be better evaluated by fractionation of the metal content into bio-available, exchangeable,
water soluble, reducible, oxidizable and residual fractions. The exchangeable and bio-available fractions are easily available for biological
functions and can easily be entered into the food chain.
The presence of toxic heavy metals in the dumping sites in Dhaka and other cities in Bangladesh may create an acute pollution of soil and water
and also may pose health hazards to the city people. However, no study has yet been conducted to assess the heavy metal content of the MSW
and its pollution potential to the environment in Bangladesh. The pollutants may spread with high monsoon rain in and around the dumping
sites. Leachate from dumping sites may also pose an important hazard for the environment. Several cases of groundwater pollution from landfill
leachate were reported (Arneth et al., 1989). This study was conducted on the samples collected from two open dumping sites (one is Matuail,
Dhaka and another is Khulna) in Bangladesh. The purpose of the present research work is to characterize the heavy metals content in the solid
wastes disposed of to the dumping sites and its mobility potential in different forms under various conditions. The concentration of the heavy
metals in the runoff leachate from the dumping sites was also analyzed in this study.
2. Materials and Methods 2.1 Description of the selected dumpsites
More than 3500 tons of MSW are generated per day by 12 million populations in Dhaka City, of which about 1200 tons of wastes are disposed
into the Matuail dumping site by Dhaka City Corporation (Yousuf, 2009). The total area of the Matuail landfill is about 40 ha, half of which is
a 15-year old dumping site. Before 2007, the site was an open dumping site receiving about 1800 tons of MSW daily. The site is now converted
into sanitary landfill having well managed daily operation, leachate and gas collection system. On the other hand, the daily generation of MSW
in Khulna City is about 520 tons, of which about 50% of the MSW is collected and dumped by Khulna City Corporation (KCC) from 1977
into the open dumping site at Khulna, located about 8 km away from the city center. The total area of this dumping site is 20 acres. Recently,
KCC has started a new dumping site and thus dumping to this old site is closed for the last few years. The site is unprotected and uncontrolled
with no leachate or gas collection system.
2.2 Basic characteristics of MSW in Bangladesh
The general nature of MSW in the urban areas of Bangladesh is almost the same and very much similar to that of the cities of least developed
Asian countries. The waste generation rate varies from 0.325 to 0.50 kg/cap/day (Alamgir, 2009; Rahman and Al-Muyeed, 2010), of which
major source of MSW are generated from residential (75-85%), commercial (11-22%), institutional (1.0-1.5%), municipal services (0.5-1.25%)
and other sources (0.4-2.5%). The typical composition of MSW is 68-81% food and vegetable waste, 7-11% paper and paper products, 3-5%
polythene and plastics, 1-7% textiles and 7-19% inert materials like glass, metals, ceramic and construction materials. Another study by Alamgir
and Ahsan (2007) reported an average composition of MSW in the cities of Bangladesh was about 74.4% organic matter, 9.1% paper, 3.5%
plastic, 1.9% textile and wood, 0.8% leather and rubber, 1.5% metal, 0.8% glass and 8% other waste. The MSW contains high volatile solids
(43-71%), while ash residue from 29-57% and high moisture (56-70%). The average value of pH are in the range 7.7-8.7, with average carbon,
nitrogen, potassium and phosphorus contents of 11.50%, 0.91%, 0.76% and 0.33%, respectively (Alamgir, 2009).
2.3 Solid waste sampling and analysis
Four MSW samples from the dumping site at Khulna and seven MSW samples from the dumping site at Matuail, Dhaka were collected during
August 2008. Partially decomposed MSW samples of about 4.0 kg from each sampling location from the two dumping sites were taken at a
depth of about 0.3 m by manual excavation. All the MSW samples were then air dried and the composition of each sample was determined.
Each air dried sample was then sieved through 2.0 mm sieve and about 100 gm of each samples passing through 2.0 mm sieve (fine fraction)
was collected into plastic containers for subsequent heavy metal analysis. The fraction over 2.0 mm of each sample was then thoroughly mixed,
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 43
grinded and sieved through 2.0 mm sieve and about 100 gm of each samples passing through the sieve (grinding fraction) was collected into
plastic containers for subsequent laboratory analysis.
2.4 Extractions
Acid extraction, water extraction, Ethylenediaminetetraacetic acid (EDTA) extraction and KNO3 extraction were done to assess the total heavy
metals content, water extractable, exchangeable and bio-available fractions of heavy metals, respectively in both of the fine and grinding fraction
of the collected samples. For the determination of total heavy metal content, aqua regia extract of both fractions of MSW samples were done
using HNO3 and H2O2 according to the standard method by USEPA (1996). The digests were filtered through 0.10 m membrane filter and
then filled to 50 mL with ultra pure water for element analyses. For water extraction, 5.0 gms of each fraction were mixed with 50 mL ultra-pure
water (L/S ratio is 10) and shaken for 4-hr using a mechanical shaker. Then, samples were filtered through 0.10 m filter paper and acidified to
pH 2.0 and stored at 40C for later analysis for heavy metals.
For EDTA extraction, 4.0 gm of samples of each fraction was added with 40 mL of 0.05 M DTPA, 0.01 M CaCl2, 0.1 M TEA (triethanolamine)
buffered at pH 7.3 and mechanically shaken for 2-hr. Samples were then filtered and stored for later analysis for heavy metals. For KNO3
extraction, samples were mixed with 0.5 M KNO3 at L/S ratio of 1:4 and mixed using a mechanical shaker for 16 hrs. Samples were filtered and
stored for subsequent heavy metals analysis. Moreover, TCLP extraction according to USEPA (1992) was done to evaluate pollution potential of
heavy metals from the dumping sites. The extract solution with a pH of 4.930.05 was used for the test. All the extractions were done twice for
accuracy and the concentrations of Cd, Cr, Pb, Cu, Ni, Zn, Fe, Mn and Mg were measured by using ICP-AES. The moisture content of solid
waste was analyzed by drying at 1050C, whereas the volatile content was determined by the method of ignition at 550
0C. The pH and electrical
conductivity (EC) of the samples were measured by adding 2.0gm of samples into 20 mL ultra-pure water (L/S ratio is 10) and shaken for 2-hr
(Prechthai et al., 2008). Samples were then filtered with 0.45 m filter paper after stabilizing for 20-30 minutes. The pH and EC of the samples
were then measured using pH and EC probes.
2.5 Leachate sampling and analysis
There is no leachate collection system in the Khulna dumping site. Therefore, four leachate samples were collected at the time of solid waste
sampling from the accumulated leachate into the depressions inside the dumping site. The Matuail dumping site has pipe and surface drain
networks for runoff leachate collection. The collected leachate is treated by the stabilization pond system (consists of four ponds) located nearby
the site. Two leachate samples were collected from the surface drains and three samples from three ponds. Each sample was collected into plastic
sampling bottle of 250 mL capacity. All the samples were acidified with 1.0 mL concentrated HCl, labeled properly and preserved well for
subsequent laboratory analysis.
The leachate analysis was done in duplicates following the standard methods for water and wastewater analysis (APHA 1998). The pH and EC
of the leachate samples were measured using pH and EC probes and total organic carbon (TOC) by TOC analyzer. For heavy metal analysis,
100 mL of the leachate sample was digested with 65% HNO3, then filtered and analyzed for metal contents (Cd, Cr, Pb, Cu, Ni, Zn, Fe, Mn
and Mg) using ICP-AES.
3. Results and Discussion 3.1 Basic characteristics of solid waste at the dump sites
Table.1 shows the average result of physical and chemical characteristics of the MSW samples tested. The composition of waste is presented in
Figure 1. The pH of the waste at both sites is within the neutral range indicating the possibility of waste to neutralize the organic acid that can
be generated from the anaerobic degradation of organic matter. The moisture content and the total solids in both sites are not varied
significantly; however, a significant variation in volatile solids is observed. In Matuail dumping site, every day fresh wastes are dumped
containing higher organic matter and plastic than the wastes at Khulna dumping site, the wastes are partially decomposed. The wastes at
Khulna dumping site are more stable as the last dumping was done only five years ago. The major identifiable components of the wastes were
soil, plastic, noncombustible and combustible matters (Figure 1). The composition of the samples varied significantly, indicating the
heterogeneous nature of the sites.
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 44
Table 1 Physical and chemical characteristics of the solid waste (mean and SD)
Location pH EC (mS/cm) Moisture (%) Volatile Solids (%) Total Solids (%)
Matuail,
Dhaka 7.76 0.57 1.107 0.68 46.51 6.25 46.04 6.25 52.35 6.25
Khulna 7.95 0.26 2.64 2.12 40.76 5.03 18.94 1.40 59.24 5.03
Figure 1 Composition of solid waste in the dumpsites. DSW and KSW mean the waste from Matuail, Dhaka dumping site and Khulna
dumping site, respectively
3.2 Heavy metal content
The mean concentration of heavy metals and its range (minimum and maximum values) in solid waste samples of both Matuail and Khulna
dumping sites are shown in Table 2. The average metal contents in both sites are also shown in Figure 2. In Matuail, heavy metals are
predominantly associated with the fine fraction; whereas in Khulna, heavy metals are predominated in grinding fraction. The presence of Cd
and Co in both sites is very insignificant, much Cr, Cu, Mn, Ni and Zn are present in Matuail than Khulna. However, in Khulna, the
concentration of Pb is higher than Matuail. In Bangladesh, co-disposal of household hazardous wastes, paint residues, ash, electronic wastes,
biomedical, plastic and non-ferrous metals with kitchen wastes are practiced. Moreover, a significant portion of the industrial wastes is also
disposed to the dumping sites with the MSW. These two factors mostly contribute to the presence of heavy metals in the MSW at the dumping
sites. Based on the average concentration, the heavy metal components in the MSW were found in the following order: Zn > Cu > Mn > Cr >
Pb > Ni > Co > Cd in Matuail, Dhaka dumping site and Pb > Zn > Mn > Cu > Cr > Ni > Co > Cd in Khulna dumping site. The result shows
contamination levels at Matuail dumpsite with Zn, Cu, Mn, Cr and Pb are at a higher level, whereas Pb, Zn, Mn and Cu are at a higher level at
the Khulna dumping site.
The heavy metal contents in the MSW in the study sites were compared with reported level of heavy metals content in the dumping sites in
Japan, India (Esakku et al., 2003; Esakku et al., 2008) and Thailand (Prechthai et al., 2008) as shown in Figure 3. It reveals that the heavy metals
content in MSW at the two study dumping sites is less than other dumping sites. In Bangladesh, domestic wastes are mostly disposed to the
dumping sites and the resource recycling from the MSW both at the secondary and the final disposal sites is the main factor behind the lower
content of heavy metals in the wastes at both the dumping sites examined here.
0%
20%
40%
60%
80%
100%
DSW 1 DSW 2 DSW 3 DSW 4 DSW 5 DSW 6 DSW 7 KSW 1 KSW 2 KSW 3 KSW 4
Plastic Metal Combustible Others Incombustible Soil
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 45
Table 2 Heavy metal contents in the MSW at the dumping sites (mean and range)
Parameters in mg/kg
dm
Matuail Dumping Site Khulna Dumping Site
Grinding Fraction Fine Fraction Grinding Fraction Fine Fraction
Cadmium (Cd) ND ND ND ND
Cobalt (Co) 0.0457
(0.088 - 0.23)
ND ND ND
Chromium (Cr) 12.06
(5.03 - 19.66)
25.23
(10.10 - 81.19)
2.66
(2.12-3.62)
2.28
(1.72-2.96)
Copper (Cu) 18.95
(4.58 - 30.30)
41.53
(14.41-137.70)
12.31
(5.66-17.70)
8.95
(4.72-14.66)
Manganese (Mn) 13.84
(8.54 - 22.20)
30.16
(9.66-82.89)
23.2
(20.00-25.20)
19.72
(16.28-24.20)
Nickel (Ni) 1.54
(0.78 - 2.86)
3.02
(0.84-9.89)
0.875
(0.68-1.22)
0.70
(0.42-0.90)
Lead (Pb) 8.36
(2.24 - 17.75)
21.14
(5.66-87.79)
43.49
(18.74-69.60)
31.55
(11.18-69.6)
Zink (Zn) 23.42
(18.05 - 32.14)
62.24
(19.41-163.80)
29.85
(15.98-45.80)
20.94
(12.46-29.4)
ND = Not Detected. Values into the parenthesis indicate the range (minimum and maximum value)
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 46
Figure 2 Average concentration of heavy metals in the MSW samples at the dumping sites
Figure 3 Comparison of heavy metals content in MSW at several dumping sites in Asia
3.3 Heavy metal leaching test
The mobility and thus toxicity of heavy metals in wastes depends largely on their binding forms. The water soluble, bio-available and
exchangeable fraction of heavy metals in the two dumping sites are presented in Table 3. For Matuail, these values are associated with the fine
fraction and for Khulna, these are associated with grinding fraction. The test results showed that the leaching of the metals with rainwater is
very low, mostly insoluble in water and metals are not expected to be released with rainwater under the conditions prevailing in the dumping
sites. However, bio-available fractions of the metals like Pb, Mn and Cu are significant and may easily be entered into the food chain. This high
bio-available fraction of the heavy metals in the dumping sites may cause significant health hazard and the sites must be restricted for vegetable
growing and other agricultural activities. Bio-availability of the metals has the following order: Pb > Mn > Cu > Ni > Zn > Cr in Matuail and
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 47
Pb > Cu > Mn > Zn > Ni > Cr in Khulna. The exchangeable fraction is mobile fraction that is also easily available for biological functions. The
test results indicated that an exchangeable fraction of the metals is less available for biological functions.
Table 3 Result of leaching tests (mean and SD)
Heavy
Metals
Matuail Dumping Site Khulna Dumping Site
Water Soluble
(%)
Bio-available (%) Exchangeable (%) Water Soluble
(%)
Bio-available (%) Exchangeable (%)
Cd 0 0 0 0 0 0
Co 0 0 0 0 0 0
Cr 0 3.772.72 0 0 0 0
Cu 0.0320.085 33.1124.80 0.110.194 0.8261.185 45.529.09 0.700.58
Mn 0 39.1624.96 0.0820.16 0 39.067.32 0.0740.15
Ni 0 24.7518.64 0.290.71 0 22.6511.24 0
Pb 0 54.1136.45 0.070.18 0.250.05 62.4733.96 0.1030.207
Zn 0 24.0715.54 0.1140.16 0 34.958.82 0.0580.115
The result of TCLP extraction with USEPA regulatory levels is shown in Table 4. The concentration of Zn, Mn and Cu analyzed in the TCLP
leachate were high. However, all the heavy metals especially Cd, Cr and Pb concentration in TCLP test were well below the regulatory levels.
Thus the degree of leaching and the risk of metal contamination and the environmental hazard associated with the dumpsites are very low with
respect to Cd, Cr and Pb contents (< 1, 5, 5 mg/L, respectively). However, more sampling and analysis should be done to confirm this
hypothesis.
Table 4 Heavy metal concentrations in the extracted leachate from TCLP test in mg/L (mean and range)
Heavy Metals Matuail Dumping Site Khulna Dumping Site USEPA Regulatory Level
Cd 0 0 1.0
Co 3.04 (2.95-3.25) 2.93 (2.9-2.95) NA
Cr 3.24 (3.2-3.3) 3.29 (3.2-3.4) 5.0
Cu 4.66 (3.3-6.4) 4.20 (4.1-4.4) NA
Mn 69.87 (46.35-111.0) 59.71 (30.35-77.5) NA
Ni 2.54 (0-7.20) 0 NA
Pb 2.06 (1.8-2.43) 3.35 (3.1-3.8) 5.0
Zn 70.23 (19.05-170.50) 11.88 (8.55-17.0) NA
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 48
3.4 Characteristics of Leachate
The characteristics of the leachate collected from the study dumping sites are presented in Table 5. It reveals that pH of the leachate samples is
within the neutral range and there is no significant variation. The EC and TOC levels of the leachate from Matuail dumping site are much
higher than Khulna; leachate in Matuail dumping site contains much organic matter, as the major portion of the MSW are partially
decomposed. Significant variations of these parameters are observed in both sites as a result of the influence of solid waste age heterogeneity on
the degradation rate.
Table 5 Characteristics of leachate from the dumpsites (average and range)
Parameter Khulna Dumping Site Matuail Dumping Site
pH 7.23 (7.03-7.54) 7.71 (7.35-8.09)
EC (mS/cm) 3.14 (1.57-5.89) 10.45 (3.77-16.16)
TOC (mg/L) 219.7 (74.3-420.0) 2351.0 (202.0-4390.0)
Cd (mg/L) 0 0
Co (mg/L) 0 0
Cr (mg/L) 0 0.02 (0.013-0.039)
Cu (mg/L) 0 0
Mn (mg/L) 0 0.01 (0-0.026)
Ni (mg/L) 0 0.01 (0.06-0.023)
Pb (mg/L) 0 0
Zn (mg/L) 0 0.01 (0-0.032)
The metal contents in the leachate samples from both dumping sites are insignificant. The concentration of the metals in the runoff leachate is
very much similar to water exchangeable fraction as presented in Table 3. Absent or very low level of Mn in the runoff leachate indicates very
high redox potential prevailing at the dumpsites. Under such condition, the binding of metals to Mn and Fe oxide increases, Mn precipitates
with carbonate and sulfide and retains at the dumpsite (Prechthai et al., 2008). Therefore, the dissolution of the acid soluble metal is probably
low under the condition that presently prevailing at the dumpsites.
4. Conclusions The municipal solid waste (MSW) at the Matuail dumping site at Dhaka is relatively fresh and is less decomposed than the waste at Khulna
dumping site. The heavy metals content in MSW at Matuail site is higher than Khulna, which is mainly associated with fine soil fraction. The
high bio-available fraction of the metals especially Pb, Mn and Cu in both the dumping sites may easily be entered into the food chain and may
cause health hazards. The TCLP extractable metal contents were well below the USEPA regulatory levels; thus the dumping sites are non-
hazardous in nature in the context of heavy metal pollution. The runoff leachate also contains insignificant concentration of heavy metals.
Under the present condition prevailing at the dumping sites, the dissolution of the acid soluble metals and the associated risk of environmental
pollution by heavy metals are very low. However, there may be other organo-chemicals that may contaminate the soil and water in and around
the dumping sites. This research work was based on limited samples. More sampling and also depth wise sampling and analysis as well as
sequential extraction of metals are thus suggested in order to obtain a firm conclusion regarding the binding of heavy metals and their pollution
potential from the dumping sites in Bangladesh.
References Alamgir M (2009). Environmental Sustainability of Municipal Solid Waste Management in Bangladesh, Proceedings of the International
Conference on Solid Waste Management: Technical, Environmental and Socio-economical Contexts-WasteSafe, 9-10 November 2009,
Khulna, Bangladesh, Vol-I, 1-12.
Alamgir M, Ahsan A (2007). Municipal solid waste and recovery potential: Bangladesh Perspective, Iran Journal of Environmental Health
Science and Engineering, 4.2, 67-76.
“Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in Bangladesh” Karim et al.
Journal of Hydrology and Environment Research 49
Ahsan A, Ismail N, Rahman MM, Imteaz M, Rahman A, Mohammad N, Salleh M (2013). Municipal solid waste recycling in Malaysia: present
scenario and future prospects, Fresenius Environmental Bulletin, 22, 12a, 3654-3664.
APHA (1998). Standard Methods for the Examination of Water and Wastewater, 20th edition, Washington, D.C., USA.
Arneth JD, Midle G, Kerndoff H, Schleger R (1989). Waste in deposits influence on ground water quality as a tool for waste type and site
selection for final storage quality. Landfill reactions and final storage quality, Baccini, P (ed), Springer Verlag, Berlin, 339 pp.
Bozkurt S, Moreno L, Neretnieks I (2000). Long-term processes in waste deposit, Science of Total Environment, 250, 101-121.
Esakku S, Karthikeyan OP, Joseph K, Nagendran R (2008). Heavy metal fractionation and leachability studies on fresh and partially
decomposed municipal solid waste, Practice Periodical of Hazardous, Toxic and Radioactive Waste Management, ASCE, 12-2, 127-132.
Esakku S, Palanivelu K, Joseph K (2003). Assessment of Heavy Metals in a Municipal Solid Waste Dumpsite, Workshop on Sustainable Landfill
Management, Chennai, India, 139-145.
Para T, Diel H, Schmitzer M (1999). Extractability of trace metals during co-composting of biosolids and municipal solid waste, Biology and
Fertility of Soils, 29-1, 31-37.
Prechthai T, Parkpain P, Visvanathan C (2008). Assessment of heavy metal contamination and its mobilization from municipal solid waste open
dumping, Journal of Hazardous Materials, 156(1-3), 86-94.
Rahman MH, Al-Muyeed A (2010). Solid and hazardous waste management, First Edition, ITN-BUET, Dhaka, Bangladesh.
Scott J, Beydown D, Amal R, Low G, Cattle J (1990). Land-fill management, leachate generation and leach testing of solid wastes in Australia
and overseas, Critical Reviews in Environmental Science, USEPA, Washington, DC, USA
USEPA (1992). Toxicity characteristic leaching procedure, Method 1311, U.S. Environmental Protection Agency, Washington, DC, USA.
USEPA (1996). SW-846 Method 3050B Acid Digestion of Sediments, Sludges and Soils, Revised 2, Office of Research and Development,
Washington, DC, USA.
Yousuf TB (2009). Construction and operation of landfill: Experience of Dhaka City Corporation, proceedings of the International Conference
on Solid Waste Management: Technical, Environmental and Socio-economical Contexts-WasteSafe, 9-10 November 2009, Khulna,
Bangladesh, Vol-II, 517-524.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 50
Rating Curve Uncertainty in Flood Frequency Analysis: A Quantitative Assessment
Md Mahmudul Haque 1,*
, Ataur Rahman1, Khaled Haddad
1
1 School of Computing, Engineering and Mathematics, University of Western Sydney, New South Wales, Australia
Peer Review History1
Abstract: River discharge is one of the fundamental data in flood frequency analysis. Accuracy of these discharge data is crucial as
uncertainty in these data is directly translated into flood quantile estimates which have significant impact in flood risk assessment and
engineering design. Generally, these discharge data reported by the gauging authorities are not measured directly, rather estimated through a
rating curve which represents a state-discharge relationship at a particular river section. Consequently, this causes uncertainties in the
discharge data as the true rating curve is unknown and the established rating curves are generally most likely to be associated with some
degrees of errors due to several factors. Despite the fact that rating curve uncertainty can introduce errors in discharge data, it is often
disregarded in the flood frequency analysis. This paper examines the impacts of rating curve uncertainty on flood quantiles estimates for a set
of New South Wales catchments in Australia, which have been assembled as a part of Australian Rainfall and Runoff Project 5 ‘Regional
Flood Methods’. The results indicate that a higher assumed value of rating curve uncertainty in flood frequency analysis inflates the
uncertainty bounds of the estimated flood quantiles (i.e. increases the width of the 90% confidence limits). This is more noticeable for smaller
annual exceedance probability floods. Based on results from the 96 catchments examined here, it has been found that the difference in flood
quantile estimates for different assumed rating curve uncertainty values do not depend on standard deviation and skew of log-space annual
maximum flood series data. It is noted that the rating curve uncertainty issue needs to be recognised in flood frequency analysis as this
represents a significant source of uncertainty in flood frequency analysis, which is often ignored in practice.
Keywords: Rating curve, uncertainty, flood quantiles, flood frequency analysis, ARR, FLIKE.
1. Introduction
River discharge is one of the fundamental data in flood frequency analysis. These data need to be of high quality to have reliable estimates of
design floods. However, in most of the cases these data are not directly measured as continuous direct measurements of discharges are time
consuming and expensive (Nihei and Kimizu, 2008). Moreover, in many cases it is infeasible to take the direct measurement of streamflow, in
particular during high floods due to practical difficulties, such as high cost of velocity measurement, safety issues during high flow velocities and
accessibility of the site during flood event. Generally, a relationship is developed between the discharges and water levels (stage) based on a
series of concurrent stage and discharge measurements at a gauging station to estimate the discharges. This stage-discharge relationship is
generally designated as rating curve, which provides a means to generate discharge time series (Petersen-Øverleir and Reitan, 2009; Haddad et
al., 2010). Various issues on rating curves are presented in details in WMO (2008).
Since the reported discharges are not generally measured directly but estimated from a rating curve, some errors are likely to be associated with
the reported discharge data, which consequently influence the results of flood frequency analysis and introduce uncertainty in the estimated
flood quantiles. The errors in the discharges derived from a rating curve may be occurred due to several reasons: (i) errors in stage and
discharge measurements at the gauging stations used to build the rating curves, (ii) the assumptions regarding a suitable form of stage-discharge
relationship and the quality of the fit of the curve, (iii) extrapolation of the curves beyond the maximum gauging points and (iv) changes in the
cross section of the river due to vegetation growth or bed movement (McMillan et al., 2010; Jalbert et al., 2011) due to erosion or deposition.
1 Paper JHER0207, submitted on 31/10/2014, accepted for publication after peer review and subsequent revisions on 24/12/2014
* Corresponding author may be contacted at [email protected]
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 51
All these factors introduce uncertainty into discharge estimation through a rating curve. Several studies have investigated the uncertainties
present in the river discharge due to rating curve errors, such as Di Baldassarre and Montanari (2009), Di Baldassarre and Claps (2011),
Domeneghetti et al. (2012); they concluded that errors in river discharge data due to rating curve errors were significant and could notably
influence the results of hydrological and hydraulic studies. An extensive literature review of the methods for estimating uncertainty in the rating
curves can be found in Le Coz (2012).
Despite the fact that river discharge data are affected by a significant uncertainty, it is often neglected in the calibration of hydrological models
and assumed that these data to be accurate (Morlot et al., 2014; Haque et al., 2014). The calibration results of the models could be notably
improved if uncertainty in the river discharge data were considered. Moreover, these uncertainties in river discharge data derived from the rating
curves can induce significant uncertainty in flood risk assessment and flood forecasting which have direct impact on the safety of life and
property. Since different degrees of extrapolation of the rating curve are required in practice as the range of observed flood levels generally
exceed the range of historical “measured” flows. This implies that all the discharges estimated by rating curve are subject to uncertainty
particularly during flood events (Kuczera, 1996; Pappenberger et al., 2006; Di Baldassarre and Montanari, 2009). Moreover, as extreme flood
discharges are found at the very end of the rating curve, they are likely to be highly affected by this extrapolation uncertainty. The use of these
estimated discharges from the extrapolated rating curve in flood frequency analysis may result in inaccurate flood estimates especially for smaller
annual exceedance probabilities (AEPs).
Several studies have investigated the uncertainty in the design flood estimates caused by error in the river discharge data derived from a rating
curve. For example, Kuczera (1996) showed that uncertainties associated with the extrapolation of the rating curve can vary substantially and
can introduce notable uncertainty in design flood estimates. Petersen-Øverleir and Reitan (2009) found that the rating curve imprecision can
widen the estimation variability of the flood quantile estimates. Haddad et al. (2010) showed that the rating curve uncertainties have a
significant impact on smaller AEP flood quantiles and without taking into account the rating curve uncertainties, the estimated confidence limits
are underestimated. Lang et al. (2010) showed that ignoring the rating curve uncertainty can produce biased estimation of flood quantiles. Di
Baldassarre et al. (2012) found that the rating curve uncertainty has a significant impact on the uncertainty of design flood estimates. Ozbey et
al. (2008) discussed the uncertainty in the flow data reported for 80 sites in Gippsland in relation to Australian Standard 3778.2.3 (Australian
Standards International, 2001). Results from this study indicated that the uncertainty in the annual mean flow at most monitoring sites in
Gippsland ranges from +/- 5% to +/- 15% in 2005-2006. However, they did not assess the uncertainty of annual maximum (AM) floods in flood
frequency analysis, which is expected to be higher.
This paper focuses on the impacts of rating curve uncertainty in flood frequency analysis using a large number of catchments. To our
knowledge, no previous study has examined the impacts of rating curve error using a large dataset like this study. The findings of this study will
be useful to hydrology practice in Australia and other countries of the world.
2. Study area and data
This study uses data from New South Wales (NSW) State in Australia to assess the impacts of rating curve uncertainty on design flood
estimates. A total of 96 catchments, with the best available data were selected to examine the range of possible rating curve extrapolation in
practice. These 96 catchments are a subset of the Australian Rainfall and Runoff (ARR) Project 5 Regional flood methods’ database. These
catchments are unregulated, and are not affected by major urbanisation or any large storage/dam. From these 96 catchments, 12 were selected
for in-depth investigation (Table 1). As can be seen from Table 1, these twelve catchments range from 66 km2 to 900 km
2 and the annual
maximum flood record length ranges from 32 years to 60 years. The skew of loge (Q), where Q is annual maximum flood series, is presented in
the last column, which shows that 8 of these catchments have negative skew, including one having a value very close to zero, and 4 have
positive skew values. These different skew values are useful to assess whether the impact of rating curve uncertainty on flood quantile estimates
is affected by skew of the flood series for the catchment.
3. Rating curve and rating ratio
A rating curve is generally constructed based on the assumption that a one to one correlation exists between the river discharge and stage,
which is generally referred to as the “true rating curve”. However, the true rating curve is unknown and the standard method of constructing a
rating curve consists of taking field measurements of water stage, h, and river discharge, Q. These measurements help to identify discrete points
(Q, h) that are subsequently interpolated through an analytical relationship that generates the rating curve (Figure 1). Then the rating curve
extension is needed to get the discharge value for the larger floods, which can introduce systematic uncertainty, either over or under estimation
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 52
of true river discharge (Figure 1). The rating curve uncertainty is generally unknown but can be expected to increase as the water level rises
above the highest measured flow. Potter and Walker (1981) suggested it could be as high as 30% in the extrapolation zone. In the interpolation
zone, the uncertainty would be smaller (e.g. 1-5%) where the fitted rating curve is well supported by discharge-stage measurements (Kuczera,
1996; Reis and Stedinger, 2005).
Table 1 Selected 12 catchments from the state of New South Wales in Australia
Station
ID
Maximum
rating ratio
Average
Rating
ratio
Catchment
area (km2)
Record
length
(years)
Period of
record mean SD skew
203030 1.25 0.60 332 32 1980-2011 4.736 0.280 -0.607
204037 4.01 0.89 62 40 1972-2011 3.087 1.594 -0.854
204906 2.63 1.06 446 39 1973-2011 5.569 0.924 -0.995
207006 20.73 5.85 363 36 1976-2011 6.493 0.536 -0.159
209001 30.11 11.35 203 34 1946-1979 5.513 0.445 0.083
212008 2.33 0.29 199 60 1952-2011 4.290 0.907 0.262
218005 1.91 0.52 900 47 1965-2011 6.791 0.664 -0.553
219025 2 0.49 717 35 1977-2011 5.155 1.615 -0.263
222016 5.10 2.36 155 35 1976-2010 2.411 0.459 -0.004
410038 5.11 1.47 411 43 1969-2011 4.108 0.441 0.993
416008 8.89 2.84 866 40 1972-2011 5.776 0.387 0.515
419051 47.29 6.18 454 35 1977-2011 3.702 1.570 -0.429
In this study, a “rating ratio” (RR) (Haddad et al., 2010) was used to identify the stations which would have annual maximum flood data
associated with a high degree of rating curve extrapolation uncertainty. The RR is estimated by dividing the annual maximum flood series data
point for each year (estimated flow QE) by the maximum measured flow (QM) at that station. The RR can be expressed as:
M
E
Q
QRR (1)
Since the rating curve for a gauging station is usually updated with the availability of new measured flow data, a station may have several rating
curves each with a unique QM value applicable for a set period of time. Therefore, the appropriate QM value applicable for the respective rating
curve for a given year was used to estimate the RR value in this study. If the RR value is smaller than 1, the corresponding annual maximum
data points may be considered to be free from rating curve extrapolation uncertainty. The annual maximum flood data points are considered to
be associated with a higher degree of rating curve uncertainty when the RR values are well above 1. These data points can cause significant
uncertainty in flood frequency analysis.
As an example, potential rating curve uncertainty of the annual maximum flood data points for station 201001 in NSW is presented in Figure 2.
It can be seen that, 34 out of 54 data points (63% of total data points) have RR values greater than 1 and the maximum RR value is 6.47. The
largest measured flow has an approximate AEP of 50%. These data points with RR >> 1 are associated with a higher degree of rating curve
uncertainty, which will translate into flood frequency estimates with a higher degree of uncertainty, especially for smaller AEP floods such as
2% and 1% AEPs.
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 53
Figure 1 Illustration of rating curve extrapolation uncertainty (Haddad et al., 2010)
As seen in the histogram of rating ratios of annual maximum flood data points for 96 catchments in NSW (Figure 3), 60.5% of the RR values
are less than 1 and 39.5% values between 1 and 47.29 (Figure 3). A RR value well above 1 could amplify the uncertainty in flood frequency
analysis. However, eliminating all stations with RR value greater than 1 would affect the results in the RFFA as it would reduce the number of
stations below the minimum required for a meaningful RFFA.
Figure 2 Plot of rating ratios (RR) for station 201001 in the state of New South Wales, Australia
4. Results
In this study, log-log extrapolation of rating curve was explored as this is the most commonly adopted technique to extend the rating curve,
among many other techniques. In log-log extrapolation, the uncertainty from the true rating curve increases systematically as the river discharge
value increases beyond the range of discharge measurements (Figure 1). Therefore, an extrapolation zone is created as the rating curve is
extended. The extrapolation zone is characterized based on the distance from the anchor point and not from the origin. Thus the systematic
uncertainty is proportional to the distance from the anchor point (in log space). In this study, the flow that has the RR value just greater than
one was used as the “anchor point”. The flows with RR value greater than one are expected to be associated with rating curve extrapolation
uncertainty. The higher the RR value for a discharge data point, the greater the rating curve uncertainty associated with the data point.
In this study, the FLIKE software, which implements the principles outlined in Kuczera (1999), was adopted to fit the LP3 distribution using the
Bayesian parameter fitting procedure to assess the impact of rating curve uncertainty on flood quantile estimates. No prior information was used
True but unknown
rating curve
True but unknown
rating curve
Actual rating curve
(reported by gauging
authority)
Maximum
measured flow
Under estimation
error
Over estimation
error
Interpolation Zone Extrapolation Zone
Log stage
Log discharge
0
1
2
3
4
5
6
7
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53
Ra
tin
g R
ati
o (
RR
)
Annual maximum data point
Data points with rating ratio < 1
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 54
in the FLIKE with both the “no rating curve” and the “rating curve uncertainty” cases. The results of 12 selected stations are shown in Table 2.
In the “no rating curve” uncertainty cases, the uncertainty coefficient of variation (CV) value was considered to be 0% for simplicity. In the
“rating curve uncertainty” cases, three scenarios were considered where flows in the extrapolation zone were corrupted by a multiplicative
uncertainty assumed to be log-normally distributed with mean one and CV values equal to 10%, 20% and 30%.
The results show that with the increasing CV values, the uncertainty in quantile estimates increases, in some cases reaching over 50% for 2%
AEP, which indicates that the rating curve uncertainty has a notable impact on flood quantile estimates. The flood estimates for lower AEPs are
found to be more affected by the rating curve uncertainties. Interestingly, there is no notable relationship between the RR values of the stations
(see Table 1) and percentage differences in quantile estimates for different CVs, which is somewhat unexpected, and needs further investigation.
Table 2 Impact of rating curve uncertainty on flood quantile estimates based on ARR-FLIKE for the 12 selected catchments in the state of New
South Wales, Australia
2% AEP flood quantile (m3/s)
Station
No rating
uncertainty
(CV = 0%)
Rating uncertainty
(CV = 10%)
Rating uncertainty
(CV = 20%)
Rating uncertainty
(CV = 30%)
Expected Expected % change from
CV = 0% Expected
% change from
CV = 0% Expected
% change from
CV = 0%
203030 171 179 5 190 11 205 20
204037 250 268 7 296 19 330 32
204906 978 1076 10 1209 24 1352 38
207006 1953 2392 22 2538 30 2600 33
209001 645 687 6 753 17 845 31
212008 501 515 3 534 7 560 12
218005 2640 2891 10 3375 28 4036 53
219025 2340 2499 7 2770 18 3094 32
222016 29 31 5 34 15 39 33
410038 176 192 9 227 29 277 57
416008 790 831 5 890 13 967 22
419051 773 806 4 903 17 1014 31
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 55
Figure 3 Histogram of rating ratio (RR) of annual maximum flood data points from 96 catchments in the state of New South Wales, Australia
The estimated flood quantiles (expected, 5% and 95% confidence limits) values for different AEP’s flood for the site 203030 in NSW are
presented in Figures 4(a) and 4(b) considering no rating curve uncertainty (CV = 0%) and CV =20%, respectively. From the figures it can be
seen that for the same AEP flood the expected quantiles values are higher in CV =20% than in CV = 0% indicating the presence of uncertainty
in the estimated results. This scenario is more significant for the larger flood quantiles (i.e. smaller AEP floods). It can be also seen from the
figures that 90% confidence band is wider when CV = 20% is considered as rating curve uncertainty than CV = 0%. This result indicates that
uncertainty bound increases in flood frequency analysis when larger errors are associated with a rating curve.
Figure 4(a-b) Estimated flood quantile values for different AEP floods for the site 203030 in the state of New South Wales, Australia; (a)
considering no rating curve uncertainty (CV = 0%), (b) considering rating curve uncertainty with CV = 20%
Figure 5 plots the differences in flood quantile estimates (between CV of 0% and CV of 20%) with catchment size; this shows no linkage
between the degree of differences in flood quantile estimates for different CVs and catchment area. Figures 6 and 7 show no relationship
between differences in flood quantiles due to different CVs and skew and SD, respectively. Figure 8 shows that difference in flood quantiles
between no rating curve uncertainty (CV = 0%) and CV = 20% can vary up to 50% for 2% AEP flood. The median difference for different
AEPs (between CV of 0% and CV of 20%) based on 96 catchments in NSW are found to be 1%, 2%, 3%, 6%, 9% and 12% for AEPs of 50%,
20%, 10%, 5%, 2% and 1%. These results show that differences in quantile values between CV = 20% and CV = 0% are higher for smaller AEP
floods indicating that large floods are likely to be more affected by the rating curve uncertainty.
60.52
17.99 16.20
3.28 1.36 0.66
0
10
20
30
40
50
60
<1 1 to 2 2 to 6 6 to 10 10 to 15 15 to 50
% o
f to
tal
an
nu
al
ma
xiu
m d
ata
po
ints
Rating ratio (RR)
0
50
100
150
200
250
300
350
20% AEP10% AEP 5% AEP 2% AEP 1% AEP
Flo
od
qu
an
tile
s (m
3/s
)
5% Confidence limit Expected 95% confidence limit
0
50
100
150
200
250
300
350
20% AEP10% AEP 5% AEP 2% AEP 1% AEP
Flo
od
qu
an
tile
s (m
3/s
)
5% Confidence limit Expected 95% confidence limit
(a) (b)
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 56
Figure 5 Plot of the catchment size vs. differences in flood quantile estimates between CV of 0% and CV of 20% for 2% AEP flood in the 96
New South Wales catchments (the green dash line represent the probable trend line)
Figure 6 Plot of the SD of annual maximum floods vs. differences in flood quantile estimates between CV of 0% and CV of 20% for 2% AEP
flood in the 96 New South Wales catchments (the green dash line represent the probable trend line)
Figure 7 Plot of the skew of annual maximum floods vs. differences in flood quantile estimates between CV of 0% and CV of 20% for 2% AEP
flood in the 96 New South Wales catchments (the green dash line represent the probable trend line)
y = 0.008x + 10.694
R² = 0.0303
0
10
20
30
40
50
60
0 200 400 600 800 1000 1200Dif
feren
ce i
n f
loo
d q
ua
nti
les
(%)
Catchment area (km2)
y = -6.7294x + 19.834
R² = 0.0413
0
10
20
30
40
50
60
0 0.5 1 1.5 2Dif
feren
ce i
n f
loo
d q
ua
nti
les
(%)
SD
y = -6.5841x + 13.494
R² = 0.0332
0
10
20
30
40
50
60
-1.5 -1 -0.5 0 0.5 1 1.5
Dif
feren
ce i
n f
loo
d q
ua
nti
les
(%)
skew
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 57
Figure 8 Plot of difference in flood quantiles between CV of 0% and CV of 20% for 2% AEP flood quantiles (96 catchments from NSW)
5. Conclusion
Flood frequency analysis is based on flood data which often consists of river discharge data from the extrapolated zone of the rating curve.
These larger flood data are normally subject to considerable uncertainty due to the possible deviation of the constructed rating curve from the
true rating curve, which is largely unknown. This study examined the effects of rating curve uncertainty (represented by coefficient of variation
(CV)) on flood quantile estimates using data from the state of New South Wales (NSW), Australia. The results show that the rating curve
uncertainty can result in notably increased uncertainty in flood quantile estimates. The results indicate that a higher assumed value of rating
curve uncertainty (i.e. a higher CV in flood frequency analysis) increases estimated flood quantiles and inflates the uncertainty bounds around
the estimated flood quantiles (i.e. increases the width of the 90% confidence limits). This is more noticeable for smaller AEP floods. Based on
results from the 96 NSW catchments, no relationship has been found between the difference in flood quantile estimates for different assumed
CV values and catchment area, log-space standard deviation (SD), skew of annual maximum flood series data. It is noted that the rating curve
uncertainty issue needs to be recognised in flood frequency analysis as this represents a significant source of uncertainty in flood frequency
analysis, which is often ignored in hydrologic practice.
6. Acknowledgements
Authors would like to acknowledge Department of Water NSW for providing the streamflow data and Geosciences Australia and Engineers
Australia for providing funding for carrying out the research, to A/Prof James Ball, Mr Mark Babister, Dr William Weeks, Prof George Kuczera
and Mr Erwin Weinmann for their comments and input to the project. This is part of Australian Rainfall-Runoff (ARR) Project 5 which is
supporting the revision of ARR.
References
Australian Standards International (2001). Australian Standard, Measurement of Water Flow in Open Channels, Part 2.3: General—
Determination of the stage-discharge relationship, AS 3778.2.3-2001, Australian Standards International, Sydney, Australia.
Di Baldassarre G, Claps P (2011). A hydraulic study on the applicability of flood rating curves, Hydrology Research, 42(1), 10-19.
Di Baldassarre G, Montanari A (2009). Uncertainty in river discharge observations: a quantitative analysis, Hydrology and Earth System
Sciences, 13(6), 913-921.
Di Baldassarre G, Laio F, Montanari A (2012). Effect of observation uncertaintys on the uncertainty of design floods, Physics and Chemistry of
the Earth, Parts A/B/C, 42, 85-90.
Domeneghetti A, Castellarin A, Brath A (2012). Assessing rating-curve uncertainty and its effects on hydraulic model calibration, Hydrology and Earth System Sciences, 16(4), 1191-1202.
0 5 10 15 20 25 30
-12 to 0
0 to 4.99
5 to 9.99
10 to 19.99
20 to 29.99
30 to 39.99
40 to 53
Frequency
% d
iffe
ren
ce i
n f
loo
d q
ua
nti
les
“Rating Curve Uncertainty in Flood Frequency Analysis: A quantitative assessment” Haque et al.
Journal of Hydrology and Environment Research 58
Haddad K, Rahman A, Weinmann PE, Kuczera G, Ball J (2010). Streamflow data preparation for regional flood frequency analysis: lessons from
southeast Australia, Australian Journal of Water Resources, 14(1), 17.
Haque MM, Rahman A, Hagare D, Kibria G (2014). Parameter uncertainty of the AWBM model when applied to an ungauged catchment,
Hydrological Processes, doi: 10.1002/hyp.10283.
Jalbert J, Mathevet T, Favre AC (2011). Temporal uncertainty estimation of discharges from rating curves using a variographic analysis, Journal of Hydrology, 397(1–2), 83-92.
Kuczera G (1996). Correlated rating curve uncertainty in flood frequency inference, Water Resources Research, 32(7), 2119-2127.
Kuczera G (1999). Comprehensive at‐site flood frequency analysis using Monte Carlo Bayesian inference, Water Resources Research, 35(5),
1551-1557.
Lang M, Pobanz K, Renard B, Renouf E, Sauquet E (2010). Extrapolation of rating curves by hydraulic modelling, with application to flood
frequency analysis, Hydrological Sciences Journal, 55(6), 883-898.
Le Coz J (2012). A literature review of methods for estimating the uncertainty associated with stage-discharge relations. WMO, Lyon, France.
McMillan H, Freer J, Pappenberger F, Krueger T, Clark M (2010). Impacts of uncertain river flow data on rainfall-runoff model calibration and
discharge predictions, Hydrological Processes, 24(10), 1270-1284.
Morlot T, Perret C, Favre AC, Jalbert J (2014). Dynamic rating curve assessment for hydrometric stations and computation of the associated
uncertainties: Quality and station management indicators, Journal of Hydrology, 517(0), 173-186.
Nihei Y, Kimizu A (2008). A new monitoring system for river discharge with horizontal acoustic Doppler current profiler measurements and
river flow simulation, Water Resources Research, 44(4), W00D20.
Ozbey N, Scanlon P, Western A (2008). Investigating the uncertainty in flow at gauging sites in Gippsland using Australian Standard 3778.2.3,
31st Hydrology and Water Resources Symp., Adelaide, 15-17 April 2008.
Pappenberger F, Matgen P, Beven KJ, Henry JB, Pfister L (2006). Influence of uncertain boundary conditions and model structure on flood
inundation predictions, Advances in Water Resources, 29(10), 1430-1449.
Petersen-Øverleir A, Reitan T (2009). Accounting for rating curve imprecision in flood frequency analysis using likelihood-based methods,
Journal of Hydrology, 366(1), 89-100.
Potter KW, Walker JF (1981). A model of discontinuous measurement uncertainty and its effects on the probability distribution of flood
discharge measurements, Water Resources Research, 17(5), 1505-1509.
Reis Jr DS, Stedinger JR (2005). Bayesian MCMC flood frequency analysis with historical information, Journal of Hydrology, 313(1), 97-116.
WMO (2008). Guide to hydrological practice, volume I, Hydrology - from measurement to hydrological information, 6th Ed, World
Meteorological Organization (WMO), Geneva, Switzerland.
Technical Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 59
Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the
Indus Upper Catchment
A. Sugiura 1,*
, S. Fujioka 2, S. Nabesaka
2, T. Sayama
1, Y. Iwami
1, K. Fukami
3, S. Tanaka
1, K. Takeuchi
1
1 International Centre for Water Hazard and Risk Management, PWRI, 1-6, Minamihara, Tsukuba, Ibaraki 305-8516, Japan
2 Japan Water Agency, Japan
3 National Institute for Land and Infrastructure Management, Ministry of Land, Infrastructure, Transport and Tourism, Japan
Peer Review History1
Abstract: The unprecedented floods of 2010 in Pakistan highlighted the necessity of a well-calibrated hydrological model of the Indus upper
catchment for a comprehensive flood risk assessment. However, this modelling was an extremely challenging exercise because of the lack of
hydrometeorological data, which are difficult to collect due to the geography of the catchment. In the study area (133,300 km2), there are 24
raingauges collecting sufficient daily data, which leads to an average area of Thiessen polygons well over (by 10 times) the WMO minimum
density network requirements of 250 km2 for hilly area. The lack of local data for soil and aquifer poses another challenge. Despite those
limitations, IFAS (Integrated Flood Analysis System) was run to conduct rainfall runoff analysis from the very upstream (in India and China)
to Taunsa (midstream Indus in Pakistan). The 30 sec Digital Elevation Model based on GlobalMap Elevation from ISCGM was upscaled to a
5 km grid model. The runoff analysis engine of IFAS was based on a 3-layered spatially distributed tank model. The daily precipitation data from 24 raingauges, discharge data from 9 river stations, barrages and dams and NCEP reanalyzed latent heat fluxes were considered as input
data. Global datasets for land cover (Global Map Land Cover, ISCGM) and soil textural types and depths (FAO/UNESCO DHSM) were used
for parameterization. The upper catchment was divided into sub-basins and calibration was conducted independently for each of them. As simulated discharges for mid-lower stream sub-basins were more reasonable than for more upstream sub-basins, parameters calibrated in the
mid-lower sub-basins were applied to the upstream ones. Then, the calibration was conducted for three flood events (1988, 1997 and 2010).
Finally, in order to validate the parameters and the model, Nash-Sutcliffe efficiencies (ENS) were calculated for discharges simulated for three other flood events (1992, 1994 and 2012). In average, ENS values were found over 0.80 at seven river stations and the model was considered
well calibrated.
Keywords: Flood, hydrological modelling, large river basin, Indus, Pakistan, IFAS.
1. Introduction
Flood occurs frequently in Indus River Basin due to heavy rainfall during the monsoon season (July-September), exacerbated sometimes with
increased snowmelt contribution to discharge (Inam et al., 2007; FFC, 2013). After 2010 unprecedented floods in Pakistan, Pakistani
authorities highlighted the need to develop improved flood forecasting models (FFC, 2013). This research is part of the “Strategic
Strengthening Flood Forecasting and Management Capacity in Pakistan” implemented by UNESCO from January 2012, and the final goal of
this research is the development of a flood forecasting system. This paper focuses on the development of a hydrological model using the
physically distributed rainfall-runoff model (Public Work Research Institute – Distributed Hydrological Model, 3-layer tank PWRI-DHM)
mounted in IFAS (Integrated Flood Analysis System) on mainstream Indus from its source to mid-stream Taunsa (133,300 km2, 37% of the
1 Paper JHER0203, submitted on 31/10/2014, accepted for publication after peer review and subsequent revisions on 22/12/2014
* Corresponding author may be contacted at [email protected]
“Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the Indus Upper Catchment” Sugiura et al.
Journal of Hydrology and Environment Research 60
whole river basin, Figure 1a) and the verification of its reproducibility for 6 past flood events. Based on the lack of data availability and on
the day order of magnitude lead-time between the different stations of the basin at which forecast and flood alert are issued, the study area
was divided into 6 sub-basins and the model was calibrated separately using discharges of upstream stations as boundary condition. The
objective of this paper is to present the methodology adopted to develop a calibrated hydrological model and verify its reproducibility despite
the lack of data in the Indus river basin. Snowmelt contribution to discharge is only considered indirectly by giving boundary condition since
snowmelt component modelling is out of the scope of this paper.
Figure 1a (left) Division of Upper-Indus into 6 sub-basins (river stations are shown as yellow and reservoirs as red).
Figure 1 b (right) Flow chart illustrates methodology.
2. Data availability in the target area
The Indus River system takes its source at 5182 m in Tibet (Inam et al., 2007) and crosses India, Afghanistan to reach the Arabian Sea in
Pakistan where it mostly lies (FAO, 2009). The Indus River basin comprises the Western rivers, Indus mainstream with its tributaries like
Shyok, Shigar, Gilgit and Kabul and Jhelum, and the Eastern Rivers, Chenab, Ravi, Beas and Sutlej (Inam et al., 2007). This section describes
the data availability for the study. Following a consistency test comparing measured rainfall data from different sources in Pakistan, Pakistan
Meteorological Department (PMD) daily rainfall data were selected as the most reliable source of measured rainfall. Hence, PMD daily rainfall
data collected from 03:00 to 03:00 GTM were the only complete source available and were therefore chosen as the input data for both
calibration and validation processes. Figure 2 describes the distribution of PMD raingauges and meteorological stations over Pakistan. The
number of stations for which data are available varies from year to year, illustrating the effort of PMD to develop an operational meteorological
data network with the number of stations almost tripled from 36 to 92 between 1988 and 2012 over the whole river basin in Pakistan. Secondly,
reference evapotranspiration ETo (FAO-Penman-Monteith (Allen et al., 1988)) can be calculated at 18 PMD meteorological stations in the
whole Indus river basin. There are two areas with no measured evapotranspiration even after distributing ETo using Thiessen polygons (Figure
2, circles). Therefore, National Centres for Environmental Predictions (NCEP) latent heat net fluxes, available globally, were chosen as an
estimator of evapotranspiration. NCEP latent heat net fluxes daily mean value (W/m2) are available monthly, at 1.9 ° grid, from 88.542 N-
88.542 S, 0 E-358.125 E and are calculated from 30 years data (1979-2009) (Kanamitsu et al., 2002). Thirdly, 6 hourly measured discharges
provided by WAPDA (Water and Power Development Authority, Pakistan) at the following seven stations’ discharges were considered: Partab
Bridge, Besham, Nowshera, Tarbela, Kalabagh, Chasma and Taunsa (refer to discharge stations in Figure 2). In addition, discharges data for
Skardu, the most upstream station and for Warsak dam, in the upstream of Nowshera station were also included as boundary conditions.
Fourth, available soil hydraulic properties data do not cover the whole target area. Data collected so far are very localized e.g. Kelleners et al.
(1999) study on soil hydraulic properties for unsaturated zone around Faisalabad in Punjab province.
In summary, both local PMD evapotranspiration and soil hydraulic properties data were found to be very limited and it was necessary to rely on
global datasets and PMD daily rainfall data distributed using Thiessen polygons, NCEP reanalysis data for latent heat net fluxes and
FAO/UNESCO globally available soil type distribution map (FAO, 2009) were selected as default input data.
Large basin with scarce data
Divide into 6 sub-basins
Calibration of sub-basins 2 and 3,
sub-basins with less uncertainty on
water balance
Feedback calibrated parameter
values to the other sub-basins
Validation by simulation on the 6
sub-basins simultaneously
“Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the Indus Upper Catchment” Sugiura et al.
Journal of Hydrology and Environment Research 61
Figure 2 Raingauges, meteorological stations, discharge stations and reservoirs distribution. Thiessen polygons areas are in km2 (mostly red or
over 900 km2 for raingauges, green dash for ETP). Circles indicate territories without any coverage of raingauges: Kabul river basin and very
upstream of the Indus.
Table 1 Description of the 3-layer model PWRI-DHM
3. Methodology
3.1 PWRI-DHM in IFAS 3-layer tank model parameterization
PWRI-DHM is based on Sugawara et al. (1956) tank model concept, used in a distributed and 3-layer tank configuration. The 3 layers are
(Figure 3 and Table 1): firstly, a surface layer tank presenting a set of parameters replicated into land use classes based on GlobalMap Land
cover (ISCGM) 30 sec resolution) categories grouped into five classes of parameters according to land use (i) forest and woodland (ii) shrubs,
herbaceous or bare land (iii) cropland, paddy field, and wetland (iv) urban and (v) snow/ice and water bodies. Secondly, an unsaturated layer
tank and an aquifer layer tank presenting a set of parameters replicated into soil textural classes, which flows directly or indirectly to the river
Model Function
Surface tank model Infiltration to unsaturated layer, surface runoff, surface storage, evapotranspiration,
rapid unsaturated subsurface flow
Subsurface tank model (for the 3 tank model) Infiltration to aquifer, subsurface runoff, subsurface storage, slow unsaturated
subsurface flow
Aquifer tank model Outflow from aquifer, aquifer loss
River tank model River course discharge
Thiessen Polygons for PMD raingauges
Min: 1,313 km2
, Ave : 33,000 km2
,
Max: 56,7782 km2
.
“Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the Indus Upper Catchment” Sugiura et al.
Journal of Hydrology and Environment Research 62
routing model in the river tank presenting a set of 10 parameters, with in total 130 parameters to tune. Details are available in Sugiura et al.
(2010) for IFAS, and Fujita et al. (2006) for the 3-layer tank model. In this study, a 5-km mesh model was set up. The parameterization was
performed by trial and error for the surface tank by essentially tuning infiltration capacity, maximum storage height, surface roughness
coefficient, for the aquifer tank by essentially tuning slow intermediate flow regulation coefficient and baseflow coefficient and river tank by
essentially tuning Manning’s roughness coefficient, those parameters being the most sensitive. For the unsaturated tank, all parameters were
fixed according to Maidment (1993) hydraulic properties corresponding to soil textural classes. Moreover, for the most northern area of the
Indus river basin, two uncertainties had to be overcome, firstly the lack of measured rainfall data and secondly the fact that precipitations for
this high elevation part (over 7000 m elevation for some parts) are snowfall and not rainfall. But snowfall estimates and snowmelt modelling are
not within the scope of this study. Hence, the solution retained was first to divide the study area into six sub-basins (Figure 1a). Then, for each
sub-basin, upstream station discharges were given as the boundary conditions and simulated discharges were compared to measured discharge at
the outlet (downstream) of each sub-basin for calibration and validation. By giving measured discharge as boundary conditions, it was expected
to account indirectly for snowmelt runoff. Hence, at Skardu, Partab Bridge and Nowshera river stations, discharges were taken as input to
compensate the lack of rain/snowfall and snowmelt data. To take into account Warsak, Tarbela dams and Kalabagh, Chashma, Taunsa barrages
operation, outflows measured at those points were also given as boundary conditions. Therefore, the discharges between the following points
were simulated: between Skardu and Partab Bridge (Sub-basin 6), Partab Bridge and Tarbela (Sub-basin 5), Tarbela and Kabul (Sub-basin 4),
Kabul, Tarbela and Kalabagh (Sub-basin 3), Kalabagh and Chashma (Sub-basin 2), then Chashma and Taunsa (Sub-basin 1).
3.2 Characteristics of the six sub-basins
The study area covered 133,300 km2 and was divided into 6 sub-basins according to the following characteristics (Figure 1a). Sub-basins 1 and 2
characteristics are the absence of raingauges on part of their area and the limited number of meteorological stations. Sub-basin 3 is characterized
by a greater but yet insufficient number of raingauges (minimum of 1812 km2 per station, maximum of 9958 km
2 per station and an average of
4791 km2 per station), all over the recommended minimum number of non-recording rain gauges network of 250 km
2 for mountainous area
(WMO, 2008). Sub-basins 4, 5 and 6 (Figure 2, circles) characteristics are the absence of raingauges on most of their areas, the absence of
meteorological station, and a significant contribution of snowmelt in their discharges (Inam et al., 2007).
Because of the principle of equifinality, it is always possible to find different sets of parameters giving good fits with observed data (Beven and
Freer, 2001). However, if the objective is to keep the model somehow with its physically distributed characteristics, calibration should take place
where uncertainty is lower. In our case, Tarbela and Chasma dams present more reliable inflow and outflow data than at barrages. Moreover,
the numbers of hydrometeorological data are more widely available in the mid-downstream part of the target area. Therefore, calibration was
performed on sub-basins 2 and 3 as explained in 3.1 and their tuned parameter values were then fed back to the other sub-basins. Moreover, the
aim of this modeling is to use the model as part of a flood forecasting system; therefore, the calibration needs to be performed for different
magnitude of floods. Therefore, considering also the data availability and because 1998, 1992, 1994, 2010, 2011 and 2012 were severe flood years
(FFC, 2013), with 2010 being an extreme flood, the three flood events (1988, 1997, 2010) were considered for calibration, and three others for
validation (1992, 1994, 2012) .
Moreover, because lead times in the Indus river basin can be counted in days, for instance lead time of 24 hours between P. Bridge and Tarbela,
26 hours between Tarbela and Kalabagh, 51-72 hours between Chasma and Taunsa; this configuration using upstream sub-basin measured
discharge as boundary condition to minimize uncertainty on a given sub-basin should allow forecasting discharges downstream.
4. Results
Nash-Sutcliffe Efficiency (ENS) (Nash and Sutcliffe, 1970) was selected to assess the performance of IFAS by inputting boundary conditions to
account for the contribution of upstream catchment feeding into each of the 6 sub-basins in the target area. The obtained ENS values are
reported in Table 2.
ENS values were calculated for each of the calibration flood events and sub-basins. The average value for all the stations and events confounded
is found to be -1.16 if no measured discharge data are given as boundary condition and 0.61 if they are given. Without boundary condition,
ENS are mostly negative, indicating the mean value of observed data is performing better as a predictor than our model. However, if measured
discharges are input as boundary condition, the performance of the model increases remarkably with ENS values over 0.9. The results show that
PWRI-DHM is able to simulate rainfall-runoff processes provided sufficient data are given. We will now focus on the discussion of the results
from the configuration with measured discharges input as boundary conditions to each sub-basin. For 1988, the performance of the model is
low in general (ENS even negative for Kalabagh). The limited number of raingauges for 1988 may explain these poor results. Hence, rainfall
input being insufficient, the simulated discharges are grossly inaccurate. Moreover, for Kalabagh, ENS is low (0.45) for 1988 and better for 1997
“Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the Indus Upper Catchment” Sugiura et al.
Journal of Hydrology and Environment Research 63
and 2010 (0.85 and 0.83, respectively). Figure 4 compares hydrographs at Kalabagh obtained from simulation with or without boundary
condition given in Tarbela. It appears that the two excessive peaks in the end of July 1988 are mainly due to uncertainties on rainfall data for
sub-basin 3 for 1988.
Table 2 Nash-Sutcliffe Efficiency (ENS) in calibration and validation, firstly with discharges given as boundary condition (white background)
and secondly without boundary condition (grey background). (Score under 0.5 are presented in red)
Calibration: average ENS = 0.61 with discharge input,
ENS = -1.16 without discharge input.
Validation: average ENS=0.67 with discharge input, ENS =
-1.60 without discharge input.
1988 1997 2010 1992 1994 2012
Taunsa 0.89 -0.03 0.86 -2.40 0.96 -0.03 0.85 -1.46 0.93 0.55 0.10 -16.39
Chasma 0.82 -0.29 0.86 -0.61 0.95 0.03 0.92 -1.76 0.93 0.65 0.95 -6.15
Kalabagh -0.27 -0.34 0.85 0.01 0.83 -0.03 0.91 -1.80 0.86 0.63 0.89 -4.20
Kabul 0.34 0.73 0.69 0.69 0.28 0.48 NoData NoData 0.75 -0.57 0.71 -4.22
Tarbela 0.38 -0.45 0.78 -0.18 0.73 -0.59 0.8 -3.95 0.92 -0.23 0.92 0.06
Besham 0.2 -0.62 0.79 -0.25 0.76 -0.67 0.72 -4.49 0.9 -0.41 0.93 -0.05
P.Bridge 0.71 -0.25 0.85 -0.06 0.52 -1.18 0.56 -6.53 -1.18 -0.38 0.02 -0.02
For Kabul, ENS values are lower than for other stations, with an average of 0.56. This is mainly due to uncertainties on rainfall data. Indeed, as
reported in section 2, over the Kabul river basin, there is only one rain gauge station covering an area over 52,636 km2. Therefore, measured
discharges at Warsak dam were taken as boundary conditions. We did not try to fine-tune further as it would hide the model performance due
to the lack/uncertainties associated with the rainfall data. And for 2010, the raingauge was washed away during the highest stage of the flood
(PMD personal communication) and therefore, there is no data to compare the simulation with. The calibration results are deemed satisfactory
and we will now consider the performance of the model for the flood events 1992, 1994 and 2012 to validate the calibration process. For part
from P. Bridge, where snowmelt contribution is not negligible, ENS values are satisfactory (average of 0.82 without P. Bridge). The model
manages to simulate properly the trends (increase or decrease) according to rainfall input. In particular, peaks timing and intensities are correct.
For 2012, the poor performance of the model for ENS for Taunsa was unexpectedly low (0.10). However, after comparing the discharges at
Taunsa for 1992, 1994 and 2012 (Figure 5), it appears that in 2012, discharges were significantly lower (almost 50% lower) than those during
the other years. Moreover, the model response to rainfall is appropriately simulated and the trends are properly reproduced. This points out the
strong dependency of the model to the availability and quality of measured discharge data.
Figure 4 (left) Comparison between simulation of discharges in Kalabagh with (KALABAGH_w) or without (KALABAGH_wo) input of
Tarbela discharges (tarbela_up) against observed data (kalabagh_up ) for 1988.
Figure 5 (right) Comparison of observed discharges in Taunsa for the year 1992, 1994 and 2012 from the 15th of June until 1st of October of
each year.
“Challenges on Modelling a Large River Basin with Scarce Data: A Case Study of the Indus Upper Catchment” Sugiura et al.
Journal of Hydrology and Environment Research 64
5. Conclusion
This study presents the calibration and validation of a distributed physically based rainfall-runoff model known as PWRI-DHM for the Indus
river catchment in Pakistan. It has been found that the model performance is strongly dependent on the availability and quality of measured
discharge and precipitation data. Because the flow concentration times in Indus system are of ‘days order of magnitude’, dependence on
measured discharge data may not impede the model use efficiency as a flood forecasting system. The findings of this paper will be useful to
similar flood study projects in other countries.
6. Acknowledgements
This study is part of the UNESCO project “Strategic Strengthening of Flood Warning and Management Capacity of Pakistan” funded by
JICA. The authors acknowledge and thank the cooperation of all the project partners, in particular Pakistan Meteorological Department for
providing necessary data for the study.
References
Allen RL, Pereira, DR, Smith M (1988). Crop Evapotranspiration: Guidelines for Computing Crop Water Requirements. Paper 56, Irrigation
and Draianage, FAO, Rome, Italy, 300pp.
Beven K, Freer J (2001). Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems
using the GLUE methodology, Journal of Hydrology, 249, 1-4, 11-29.
FAO/IIASA/ISRIC/ISSCAS/JRC (2009). Harmonized World Soil Database (version 1.1). FAO, Rome, Italy and IIASA, Laxenburg, Austria.
FFC (2013). Annual Flood Report 2012. Federal Floods Commission (FFC).
Fujita K et al. (2006) Project Research Report 299: Watershed/Urban Regeneration in Accord with Nature Technical Report (II). River
Environment Laboratory, NILIM, Tsukuba, Japan: NILIM.
Inam A, Clift PD, Giosan L, Tabrez AR, Tahir M, Rabbani MM, Danish M (2007). The Georgraphic, Geological and Oceanographic Setting of
the Indus River. In Large Rivers: Geomorphology and Management, edited by A Gupta, John Wiley and Sons.
Kanamitsu M, Ebisuzaki W, Wollen S-K Y, Hnilo JJ, Fiorino M, Potter GL (2002). NCEP-DEO AMIP-II Reanalysis (R-2), Bulletin of the
American Meteorological Society, 1631-1643.
Kelleners TJ, Beekma J, Chaudhry MR (1999). Spatially variable soil hydraulic properties for simulation of field-scale solute transport in the
unsaturated zone, Geoderma, 92, 3-4, 199-215.
Maidment DR (1993). Handbook of Hydrology, McGraw-Hill.
Nash JE, Sutcliffe JV (1970). River flow forecasting through conceptual models, Part I - A discussion of principles, Journal of Hydrology, 10,
282–290.
Sugawara M, Maruyama F (1956). A method of prevision of the river discharge by means of a rainfall 555 models. Symposia Darcy. Dijon:
International Association Science Hydrological Publication, 42, 3, 556, 71-76.
Sugiura T, Fukami K, Fujiwara N, Hamaguchi K, Nakamura S, Hironaka S, Nakamura K, Wada T, Ishikawa M, Shimizu T, Inomata H, Itou Z
(2010). Experimental application of flood forecasteing system (IFAS) using satellite-based rainfall. 9th International Conference on
Hydroingormatics. Tianjin, China: HIC2010, 1786-1793.
WMO (2008). Guide to Hydrological Practices, Sixth Edition, 168, WMO, Geneva.
Review Paper
© EnviroWater Sydney, 2014 Journal of Hydrology and Environment Research, Vol 2, No 1 65
Uncertainty in Design Rainfall Estimation: A Review
Abdullah Al Mamoon 1,*
, Ataur Rahman 2
1 Ministry of Municipality & Urban Planning, Qatar
2 School of Computing, Engineering and Mathematics, University of Western Sydney, Australia
Peer Review History1
Abstract: Design rainfall is an essential input to a hydrologic model, which is used to estimate design discharge that is needed in the planning
and design of many engineering infrastructure projects. Design rainfall estimation is made using recorded rainfall data over many stations in a
given region. Uncertainties in design rainfall estimates arise from various sources such as data error, sampling error, regionalization error, model error and error due to climate change. This paper reviews various sources of uncertainties in design rainfall estimation. It has been found that
uncertainty in design rainfall estimates are hardly considered in design applications. Uncertainty in design rainfall estimation can be assessed
using Monte Carlo simulation and bootstrapping techniques. These techniques require significant computer power, which however is not a problem now a days. The biggest challenge in uncertainty estimation lies in the assessment of the impacts of non-stationarity in the rainfall data
on design rainfall estimates. The findings of this paper would be useful to future studies on design rainfall estimation.
Keywords: Design rainfalls, IDF, uncertainty, climate change, Monte Carlo simulation, Bootstrapping.
1. Introduction
Design rainfall is a probabilistic representation of rainfall intensity (depth of rainfall over a time period) at a given location for a given
duration and average recurrence interval (ARI). Design rainfall is an essential input to a hydrologic model, which is used to estimate design
discharge that is needed in the planning and design of many engineering infrastructure projects such as street drainage systems, culverts,
bridges and regulators. In design rainfall estimation, recorded rainfall data at many stations are used to develop intensity-duration-frequency
(IDF) curves by adopting statistical techniques such as regional frequency analysis.
Many countries in the world have carried out research on the derivation of IDF curves such as Australia (I. E. Aust., 1987; Haddad et al.,
2011; Johnson et al., 2012), U. K. (NERC, 1975), USA (Hershfield, 1961; Bonnin et al., 2006; Trefry et al., 2005), Hong Kong (Yu and Cheng,
1998), Italy (Baldassarre et al., 2006), Israel (Ben-Zvi, 2009), Denmark (Madsen et al., 2002, 2009), Malaysia (Zakarai et al., 2012), Iran
(Avolverdi and Khalili, 2010), Norway (Hailegeorgis et al., 2013) and Qatar (Mamoon et al., 2013, Mamoon et al., 2014). In many cases, the
quantity and quality of recorded rainfall data (in particular the continuous rainfall data) are inadequate, which results in a significant
uncertainty in the derived IDF curves. Also, climate change brings another dimension of uncertainty in the IDF derivation as in many cases
the past rainfall data may not satisfy the stationarity assumption (Ishak et al., 2013; Seidou et al., 2012; Leclerc and Ouarda, 2007).
As compared to humid region, design rainfall estimation in the arid region is more challenging mainly due to significant spatial and temporal
variability in rainfall and the limited availability of recorded rainfall data (Kwarteng et al., 2009; Zhang et al., 2005; Nasrallah and Balling,
1993). There can be long dry periods with little or no rainfall in the arid regions. Rainfall data time series must cover longer time periods to
capture the long term variability in rainfall to derive valid IDF curves that can be applied in the design with confidence. For example, shorter
rainfall data covering either dry or wet climatic regime would provide under- and over-estimation, respectively i.e. biased IDF curves.
1 Paper JHER0208, submitted on 14/10/2014; accepted for publication after peer review and subsequent revisions on 27/12/2014
* Corresponding author may be contacted at [email protected]
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 66
There have been many studies on design rainfall estimation; however, the uncertainty in design rainfall estimation has not been incorporated in
the final IDF curves in most of the previous applications. This paper focuses on the uncertainty in design rainfall estimation by identifying
various sources of uncertainty, reviewing various methods to account for the uncertainty and making recommendations on how uncertainty can
be incorporated in the final IDF curves.
2. Sources of uncertainties in design rainfall estimation
Zadeh (2005) defined uncertainty as an attribute of information. This definition can be applied to hydrology, where uncertainty has
traditionally been estimated using probability theory (Montanari, 2007). Uncertainty may be defined as a measure of the lack of accuracy
concerning observed data and modelling outcome.
Many attempts have been made to study different types of uncertainties in design rainfall estimation and rainfall runoff modelling (Yen and
Ang, 1971; Kavetski et al., 2006; Yu and Cheng, 1998; Ewen et al., 2006; Renard et al., 2010; Wu et al., 2011; Hailegeorgis et al., 2013 and
Tung and Wong, 2014). In general, the uncertainty in hydrological modelling can be divided into two main categories: (i) data and sampling
errors and (ii) modelling or structural errors (Haddad and Rahman, 2014). The data uncertainty is originated from measurements errors
resulting from instrumental and human errors and also due to inadequate representativeness of a data sample due to temporal and spatial
variability of the data.
The use of a limited quantity of rainfall data (such as data of short record length) in the frequency analysis introduces sampling uncertainty.
Due to sampling uncertainty, the estimates of higher order moments (such as skewness) become unstable, in particular due to the presence of
extremes/outliers data points. The sampling uncertainty is transmitted to the establishment of rainfall IDF model, model coefficients and,
eventually, to the design rainfall amount and adopted hyetograph (Tung and Wong, 2014). Uncertainty features of design rainfall via rainfall
IDF model coefficients in the risk-based design of urban drainage systems was addressed in an earlier study by Yen and Tang (1976). Wu et
al. (2011) addressed the cascade transmission of uncertainties starting from sampling errors to the analysis and design of drainage
infrastructures as illustrated in Figure 1.
The assumptions made during modelling may result in errors in the conceptual structure of the model. The choice of a model also introduces
errors in predicting quantile of interest. The uncertainty in the model parameters is attributed to inability in accurately quantifying the input
parameters for a model. The parameters obtained from the calibration process are also not free from uncertainty for various reasons including
data uncertainty (data used to calibrate the model contains errors), insufficient amount of data from which the parameters in an assumed
model are estimated, model uncertainty (the model structure used to calibrate the model is not adequate), and lack of sufficient data.
In modelling hydrologic systems, there are two sources of random variation that may affect the estimated system outputs: (i) the natural
temporal and spatial variability of climate and catchment factors being modelled; and (ii) the uncertainty in the definition of the model
structure, the model inputs and in the estimated model parameters (Nathan and Weinmann, 2013).
In general, hydrologic modelling is affected by four main sources of uncertainty (Renard et al., 2010): (i) input uncertainty, e.g., sampling and
measurement errors in catchment rainfall estimates; (ii) output uncertainty, e.g., rating curve errors affecting runoff estimates; (iii) structural
uncertainty (sometimes referred to as “model uncertainty”), arising from lumped and simplified representation of hydrological processes in
hydrologic models; and (iv) parametric uncertainty, reflecting the inability to specify exact values of model parameters due to finite length and
uncertainties in the calibration data, imperfect process understanding and model approximations.
Hailegeorgis et al. (2013) suggested that the regional frequency analysis of extreme precipitation events and hence derivation of IDF curves is
subject to the major uncertainties of different sources:
Data series used: data quality, which is related to the questions like is the data series stationary and independent; and sampling of data,
which are related to the time period and length of data series and the sampling type e.g. annual maximum series (AMS) and partial
duration series (PDS);
Selection of frequency distribution to describe the data;
Parameter estimation; and
Regionalization and quantile estimation.
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 67
Figure 1 Propagation of uncertainties in hydrological modelling
3. Uncertainty associated with regional rainfall frequency analysis
For estimation of design rainfalls (i.e. IDF curves), regional rainfall frequency analysis (RRFA) methods are generally used. The RRFA is
preferred over the at-site estimation to achieve consistency in estimation over the space. In RRFA, use of rain data from several sites and
grouping the rain gauges in homogeneous regions allow to trade space for time (Stedinger et al., 1993). Moreover, a regional approach allows
estimation of design rainfall at any arbitrary location within the region, in particular at ungauged locations. In RRFA approach, recorded rainfall
data within a ‘homogeneous region’ is pooled to compensate the scarcity of temporal data with the spatial data i.e. recorded rainfall data from
other stations in the region. In a homogeneous region, it is assumed that all the sites within the region have same regional growth curve/factors,
but the at-site scaling factor (e.g. mean or median value) is unique for each site which reflects the variation of at-site characteristics governing
rainfall generation. Principal steps of RRFA approach is illustrated in Figure 2. Types of uncertainties related to RRFA include the degree of
homogeneity of the assumed region, number of sites in the region, record lengths of the individual sites and how the regional data is pooled.
4. Uncertainty due to climate change
Climate change can be defined as any change to atmospheric forcing resulting from human activities, including the emission of greenhouse
gases as well as anthropogenic aerosols, whereas, climate variability is defined as changes resulting from ‘natural’ features of the climate (Ishak
et al., 2013). Climate change has been affecting different aspects of hydrological cycle including rainfall and runoff (Wang et al., 2013). This
can eventually lead to increased occurrence of extreme events such as rainfalls, floods, droughts, heat waves, summer and ice storms
(Simonovic and Peck, 2009; De Paola et al., 2013; Laz et al., 2014; Mamoon and Rahman, 2014; Mamoon et al., 2014). Since there is a strong
link between the global climate system and the hydrological cycle, a change in a component of the climate system will have a notable impact
on the magnitude and frequency of hydrological extremes, including the potential for changes to rainfall. This challenges the assumption of
stationary, which is fundamental in frequency analysis of hydrological data (Milly et al., 2008; Westra and Sisson, 2011). Failure to take such
change into consideration can undermine the usefulness of the return period concept in hydrological frequency analysis (Khaliq et al., 2006).
Sampling errors
Frequency analysis
Design rainfall hyetograph
Runoff hydrographs
Design of hydro-system infrastructures
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 68
Figure 2 Principal steps in regional rainfall frequency analysis (RRFA)
The Intergovernmental Panel on Climate Change (IPCC) in its fourth Assessment Report AR4 predicts more extreme climate towards the
end of the century, which will impact the design of engineering infrastructure projects with a long design life. Since extreme rainfall data is
used in the derivation of design rainfalls, which is used to design future drainage infrastructure, it has become an important research question
whether the extreme rainfall at a given region would change in future due to climate change and how this change will happen. As
temperature increases, the evaporation will increase, which will result in an increase in intensity of heavy precipitation events in many regions
globally, including some regions where average precipitation may even show a decrease (Meehl et al., 2007).
Data collation
Quality control
AMS and PDS time series
At-site characteristics
Select optimum number of regions
Testing the regions for homogeneity
dAre the
regions
homogenous?
Choice of frequency distribution
Quantiles estimates
Frequency estimation ungauged sites
Adjustment of
heterogeneous
regions
No
Yes
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 69
The regional design rainfall estimates are made using the recorded rainfall data. It is however expected that the climate change will modify the
at-site and regional rainfall characteristics, which will undermine the use of the past data to make realistic long-term projections. For example,
the rainfall data statistics such as the mean or median may change due to climate change and hence the statistical distributional parameters or
the parent distribution itself would change. Due to the non-stationarity of rainfall under changing climate conditions, the change of intensity
for design storm under given duration and frequency has been observed in many regions (Willems and Vrac, 2011; Olsson et al., 2012).
Attempts have been made by researchers in various parts of the world to update IDF relationship under changing climate conditions. For
example, Simonovic and Peck (2009) applied two climate scenarios obtained as simulations outputs of global climate model (GCM) to assess
the impact of climate change on extreme rainfall events for the city of London in Western Ontario, Canada. Comparison of updated IDF
curves for climate change indicated that the rainfall intensity would most certainly increase under climate change scenario.
Wang et al. (2013) assessed the impact of climate change on IDF data at Apalachicola River basin using seven regional climate models under
emission scenario A2. Even though some models projected decreased rainfall intensity, the extreme rainfall intensity and frequency were
projected to increase by most models at the study area.
Mirhosseini et al. (2013) evaluated impacts on IDF curves for Alabama using high-resolution projections (for 2038–2070) derived from
dynamical downscaling of GCMs by regional climate models. Future IDF curves were constructed using 3-hourly precipitation data simulated
by six combinations of global and regional climate models being temporally downscaled using a stochastic method. The results of all six
climate models suggested that the future precipitation patterns for Alabama were expected to veer toward less intense rainfalls for short
duration events. However, for long duration events (e.g. 4 hours), the results were not consistent across the models.
It should be noted that significant uncertainty is associated with rainfalls generated by climate models (Wang et al., 2013). This could be
introduced by failure in improving long-term climate projection accuracy beyond what could be achieved by interpolating global model
predictions onto a finer-scale landscape (Pielke and Wilby, 2012). The rainfall observation is generally obtained from weather stations which
are point-based rainfall depth, but the rainfall depth from climate models is the average value at a very large spatial scale varying from 50 to
over 300 km. Due to large grid size, climate models provide coarse rainfall data, not easily comparable with point rainfall data.
5. Uncertainty analysis methods
Many studies in recent years suggest a range of methods for quantifying uncertainties (Hill et al., 2012; Renard et al., 2010; Pappenberger and
Beven, 2006; Aster et al., 2012; Gupta et al., 2005; Montanari, 2007). A few of the numerous approaches for understanding and quantifying
uncertainty are listed below:
Analytical methods (Tung, 1996);
Approximation methods e.g., first-order second moment method (Melching, 1992);
Simulation and sampling-based Monte Carlo methods (Kuczera and Parent, 1998; Burgman, 2005; Nathan and Weinmann, 2013);
Bayesian methods (Renard et al., 2010; Ye et al., 2008);
Methods based on the analysis of model errors (Montanari and Brath, 2004);
First-order variance estimation method (Tung and Yen, 2005) based on the Taylor series expansion;
Bootstrapping (Efron and Tibshirani, 1993);
Cross-validation approaches (Haddad et al., 2013); and
Methods based on fuzzy set theory (Maskey et al., 2004; Zadeh, 1978).
Uncertainty analysis methods in all of the above cases involve: (i) identification and quantification of the sources of uncertainty; (ii) reduction
of uncertainty; (iii) propagation of uncertainty through the selected model; (iv) quantification of uncertainty in the model outputs; and (vi)
application of the uncertain information in decision making process. However, Pappenberger and Beven (2006) noted that the practice of
uncertainty analysis and use of the results of such analysis in decision making is not widespread.
The Monte Carlo simulation technique is widely used in hydrology to assess uncertainty in the modeling (Abolverdi and Khalili, 2010;
Zakaria et al., 2012). It allows the quantification of the model output uncertainty resulting from uncertain model parameters. The Monte
Carlo simulation technique is based on the principle that model input variables are random instead of fixed values. The advantages of the
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 70
Monte Carlo simulation technique are that this allows examining the impacts of many possible combinations of the input variables and model
parameters in rainfall estimation.
Zakaria et al. (2012) used Monte Carlo simulation technique to evaluate the performance between the simulated and calculated rainfall
quantiles of specific recurrence intervals in Malaysia. Hailegeorgis et al. (2013) used non-parametric bootstrap resampling approach to
quantify the sampling uncertainty in terms of interval estimates of quantiles (i.e. 95% confidence bounds). The interval estimate showed that
there is a huge uncertainty in quantile estimation due to sampling of data which needs to be incorporated in any frequency analysis from
historical data. The updated estimated quantiles and IDF curves with uncertainty bounds obtained from this study were found to be more
reliable as compared to the existing IDF curves for the city of Trondheim, Norway.
Various sources of uncertainties and their methods of analyses are illustrated in Figure 3.
5.1 Monte Carlo simulation
Monte Carlo simulation is a technique for iteratively evaluating a deterministic model using sets of random samples as inputs. The term
Monte Carlo was coined by S. Ulam and Nicholas Metropolis to capture the random properties of the roulette wheel played at Casinos in
Monte Carlo, Monaco. This method is often used when the model is complex, nonlinear, or involves more than just a couple of uncertain
parameters and simulation involving numerous evaluations of the model (Wittwer, 2004).
Monte Carlo simulation has been widely used to determine the impacts of model and parameter uncertainty on simulation results; these are
generally expressed in the form of confidence limits on hydrologic estimates (Rahman et al., 2002; Nathan and Weinmann, 2013). In Monte
Carlo simulation, the inputs are randomly generated from probability distributions to simulate the process of sampling from an actual
population. The data generated from the simulation can be represented as probability distributions (or histograms) or converted to error bars,
reliability predictions, tolerance zones, and confidence intervals. The basic principle behind Monte Carlo simulation is schematically shown in
Figure 4.
The steps involved in undertaking a Monte Carlo simulation for analysing parameter uncertainty are outlined below:
Identify the probability distributions of the input variables and model parameters;
Generate random values of each of the variables from their respective probability distributions;
Run the model with each set of the generated input variables and generate a model output for the given set of model parameters;
Store the model outputs;
Repeat the steps until the convergence criterion is satisfied or total number of simulation is reached; and
Analyze the distribution of model outputs to derive cumulative distribution function and other statistical properties (e.g., mean and
standard deviation).
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 71
Figure 3 Classification and analysis of uncertainties in hydrological modelling
Some advantages of the Monte Carlo simulation technique are provided below:
Probabilistic results: Results show not only what could happen, but how likely each outcome is;
Graphical results: Because of the data a Monte Carlo simulation generates, it is easy to create graphs of different outcomes and their
chances of occurrence;
Sensitivity analysis: In Monte Carlo simulation, it is easy to see which inputs had the biggest effect on bottom-line results;
Scenario analysis: Using Monte Carlo simulation, analysts can see exactly which inputs had been combined to generate certain outcome.
This is invaluable for pursuing further analysis.
Correlation of inputs: In Monte Carlo simulation, it is possible to model interdependent relationships between input variables. It is
important for accuracy to represent how, in reality, when some factors go up or down.
Data and sampling
uncertainty
Uncertainties in
design rainfall
estimation
Uncertainty in
RRFA method
- Short record length
- Intrumental error
- Gaps in the data
- Manual error in entering the data
- Insufficient spatial data or gauging density
- Selection of AMS and PDS events
- Non-concurrence of the data
Uncertainty due to
climate change
- Delineation of regions
- Lack of homogeneity within the region
- Choice of probability distribution
- Input variables in regression equation
- Parameter estimation
- Quantile estimation
- Uncertainty at the ungauged location with
no measured data points
- Uncertainty due to spatial interpolation
to derive generalized IDF at any arbitrary
location.
- Trend analysis
- Selection of emissions scenario
- Selection of global climate models (GCM)
- Scale and resolution of GCM
- Parameterization of GCMs
- Downscaling of data to local scale
- Extrapolation of arial reduction factor
Uncertainty analysis
methods
- Analytical methods
- Approximation methods
- Monte Carlo simulation
- Bayesian methods
- Methods based on model errors
- First-order variance estimation methods
- Bootsrapping
- Cross-validation approaches
- Methods based on fuzzy set theory
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 72
Although flexible, robust and conceptually simple, Monte Carlo simulation methods tend to be among the most computationally demanding
(Hill et al., 2012) as they require a large number of model runs, time and resources to produce a reliable and meaningful uncertainty
estimation.
Figure 4 The principle of Monte Carlo simulation
5.2 Bootstrapping
Bootstrapping is a nonparametric statistical technique that allows computing estimated standard errors (bias variance), confidence intervals
and hypothesis testing (Efron and Tibshirani, 1993). Generally, it falls in the broader heading of resampling methods. It involves a relatively
simple procedure, but repeated many times and hence it is heavily dependent upon computer power.
This technique was introduced by Efron (1979a, 1979b) and further developed by Efron and Tibshirani (1993). The name “bootstrapping”
originated from an old saying the phrase, “To lift himself up by his bootstraps.” This refers to something that is unworkable and impossible.
In bootstrapping, the samples are drawn randomly from the original sample with replacement.
Generally bootstrapping involves the following basic steps:
Resample a given data set a specified number of times;
Calculate a specific statistic from each sample; and
Find the standard deviation of the distribution of that statistic.
Bootstrapping technique intends to be a more general and versatile procedure for sampling distribution problems without having to rely heavily
on the normality condition on which most of the classical statistical inferences are based (Tung and Wong, 2014). It is not uncommon to
observe non-normal data in hydro-system engineering problems. Although the bootstrapping technique is computationally intensive, such
concern is diminishing as the computer power has been increasing with time.
Sample distribution of the input variables (X1, X2, X3)
X1
X2
X3
Model function : Y = f (X)
Iterative simulation
Sample output distribution
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 73
6. Conclusion
Design rainfall is one of most commonly used inputs to hydrological models. Design rainfall estimation is made using recorded rainfall data
over many stations in a given region. Uncertainties in design rainfall estimates arise from various sources such as data error, sampling error,
regionalization error, model error and error due to climate change. This paper reviews various sources of uncertainties in design rainfall
estimation. It has been found that uncertainty in design rainfall estimates are hardly considered in design application. Uncertainty in design
rainfall estimation can be assessed using Monte Carlo simulation and bootstrapping techniques. These techniques require significant computer
power, which however is not a problem now a days. The biggest challenge in uncertainty estimation lies in the assessment of the impacts of
non-stationarity in the rainfall data on design rainfall estimates.
References
Abolverdi J, Khalili D (2010). Development of regional rainfall annual maxima for southeastern Iran by L-moments, Water Resources
Management, 24, 2501-2526.
Aster RC, Borshers B, Thurber CH (2012). Parameter Estimation and Inverse Problems, Academic Press, Amsterdam.
Baldassarre DG, Brath A, Montanari A (2006). Reliability of different depth-duration-frequency equations for estimating short-duration storms,
Water Resources Research, 42, W12501. doi:10.1029/2006WR004911.
Burgman MA (2005). Risks and decisions for conservation and environmental management. Cambridge University Press, Cambridge.
Ben-Zvi A (2009). Rainfall intensity-duration-frequency relationships derived from large partial duration series, Journal of Hydrology, 367, 104-
114.
Bonnin GM, Martin D, Lin B, Parzybokt T, Yekta M, Riley D (2006). Precipitation-Frequency Atlas of the United States, vol. 1, NOAA Atlas
14, NOAA, USA.
De Paola F, Giugni M, Bucchignani AE (2013). Stationary Vs. Non-stationary of Extreme Rainfall in Dar Es Salaam (Tanzania), Proceedings of
2013 IAHR World Congress.
Ewen J, O’Donnell G, Burton A, O’Connell E (2006). Errors and uncertainty in physically-based rainfall-runoff modelling of catchment change
effects, Journal of Hydrology, 330, 641– 650.
Efron B (1979a). Bootstrap methods: another look at the jackknife, Annals of Statistics, 3, 1189–1242.
Efron B (1979b). Computers and theory of statistics: thinking the unthinkable. SIAM Rev 21, 460–480.
Efron B, Tibshirani, R (1993). An Introduction to the Bootstrap. Boca Raton, FL: Chapman & Hall/CRC.ISBN 0-412-04231-2. software.
Gupta HV, Beven KJ, Wagener T (2005). Model calibration and uncertainty estimation. In: M. Andersen (ed.), Encyclopedia of Hydrological
Sciences, John Wiley & Sons, New York, NY, USA, 2015-2031.
Haddad K, Rahman A (2014). Design rainfall estimation and changes, In Handbook of Engineering Hydrology: Modeling, Climate Change and
Variability, Edited by Saeid Eslamian, CRC Press, 173-190
Haddad K, Rahman A, Green J (2011). Design rainfall estimation in Australia: a case study using L-moments and generalized least squares
regression, Stochastic Environmental Research and Risk Assessment, 25, 6, 815-825.
Haddad K, Rahman A, Zaman A, Shrestha S (2013). Applicability of Monte Carlo Cross Validation Technique for Model Development and
Validation in Hydrologic Regression Analysis Using Ordinary and Generalised Least Squares Regression, Journal of Hydrology, 482, 119-
128.
Hailegeorgis TT, Thorolfsson ST, Alfredsen K (2013). Regional frequency analysis of extreme precipitation with consideration to uncertainties
to update IDF curves for the city of Trondheim, Journal of Hydrology, 498, 305–318.
Hershfield DM (1961). Rainfall frequency atlas of the United States for durations from 30 minutes to 24 hours and return periods from 1 to 100
years,Weather Bureau Paper No.40. US Department of Commerce, Washington D.C.
Hill MC, Kavetski D, Clark M, Ye M, Lu D (2012). Uncertainty Quantification for Environmental Models, SIAM News, Volume 45.
Institution of Engineers Australia (I. E. Aust.) (1987). Australian Rainfall and Runoff: A Guide to Flood Estimation. Edited by D. H. Pilgrim,
Vol. 1, I. E. Aust., Canberra.
Ishak EH, Rahman A, Westra S, Sharma A, Kuczera G (2013). Evaluating the non-stationarity of Australian annual maximum flood, Journal of
Hydrology, 494, 134-145.
Johnson F, Haddad K, Rahman A, Green J (2012). Application of Bayesian GLSR to estimate sub-daily rainfall parameters for the IFD Revision
Project. Hydrology and Water Resources Symposium, 19-22 Nov 2012, Sydney, Australia.
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 74
Khaliq MN, Ouarda TBMJ, Ondo JC, Gachon P, Bobée B ( 2006). Frequency analysis of a sequence of dependent and/or non-stationary hydro-
meteorological observations: A review, Journal of Hydrology, 329 (3-4), 534-552.
Kwarteng AY, Dorvlob AS, Vijaya Kumar GT (2009). Analysis of a 27-year rainfall data (1977–2003) in the Sultanate of Oman, International
Journal of Climatology, 29, 605–17.
Kavetski D, Kuczera G, Franks S (2006). Bayesian analysis of input uncertainty in hydrological modeling: 1. Theory, Water Resources Research,
42, W03407.
Kuczera G, Parent E (1998). Monte Carlo assessment of parameter uncertainty in conceptual catchment models: the Metropolis algorithm,
Journal of Hydrology, 211, 69-85.
Laz OU, Rahman A, Yilmaz A, Haddad K (2014). Trends in sub hourly, sub daily and daily extreme rainfall events in south-east Australia,
Journal of Water and Climate Change, doi:10.2166/wcc.2014.035.
Leclerc M, Ouarda BMJT (2007). Non-stationary regional flood frequency analysis at un-gauged sites, Journal of Hydrology, 343, 254– 265.
Madsen H, Mikkelsen PS, Rosbjerg D, Harremoes P (2002). Regional estimation of rainfall intensity duration curves using generalised least
squares regression of partial duration series statistics, Water Resources Research, 38, 11, 1-11.
Madsen H, Arnbjerg-Neilsen K, Mikkelsen PS (2009). Update of regional intensity-duration-frequency curves in Denmark: tendency towards
increased storm intensities, Atmospheric Research, 92, 343-349.
Mamoon AA, Joergensen NE, Rahman A, Qasem H (2013). Estimation of Design Rainfall in Arid Region: A Case Study for Qatar Using L
Moments. Proceedings of the 35th IAHR World Congress, September 8 to 13, 2013 Chengdu, China, 1-9.
Mamoon AA, Jeorgensen NE, Rahman A, Qasem H (2014). Derivation of new design rainfall in Qatar using L-moments based index frequency
approach, International Journal of Sustainable Built Environment, 3, 111-118.
Mamoon AA, Rahman A (2014). Identification of Rainfall Trends in Qatar, International Conference on Environmental Systems Science and
Engineering, 15 -16 Dec 2014, Sydney, Australia, 8, 12, Part VI, 735-739.
Mamoon AA, Jeorgensen NE, Rahman A, Qasem H (2014). An Exploratory Study on the Impact of Climate Change on Design Rainfalls in the
State of Qatar, International Conference on Environmental Systems Science and Engineering, 15 -16 Dec 2014, Sydney, Australia, 8, 12,
Part VI, 727-734.
Meehl G et al (2007). Global climate projections, in Climate Change. The Physical Science Basis. Contribution of Working Group I to the
Fourth Assessment Report of the Intergovernmental Panel on Climate Change, edited by S. Solomon et al., Cambridge Univ. Press,
Cambridge, U. K, 747–845.
Mirhosseini G, Srivastava P, Stefanova L (2013). The impact of climate change on rainfall Intensity–Duration–Frequency (IDF) curves in
Alabama, Regional Environ Change (Suppl 1): S25–S33
Milly PCD, Betancourt J, Falkenmark M, Hirsh RM, Zbigniew W, Kundzewicz ZW, Lettenmaier DP, Stouffer RJ. (2008). Stationarity is Dead:
Whither Water Management? Science, 319, 573-574.
Melching CS (1992). An improved first-order reliability approach for assessing uncertainties in hydrological modelling, Journal of Hydrology,
132, 157-177.
Montanari A (2007). What do we mean by ‘uncertainty’? The need for a consistent wording about uncertainty assessment in hydrology,
Hydrological Processes, 21, 841–845.
Montanari A, Brath A (2004). A stochastic approach for assessing the uncertainty of rainfall-runoff simulations, Water Resources Research, 40,
W01106, doi: 10.1029/ 2003WR002540.
Maskey S, Guinot V, Price RK (2004). Treatment of precipitation uncertainty in rainfall runoff modelling: a fuzzy set approach, Advances in Water Resources, 27, 889-898.
Nasrallah HA, Balling Jr RC (1993). Analysis of recent climatic changes in the Arabian Gulf region, Environmental Conservation, 20(3), 223–
226, doi:10.1017/S0376892900023006.
Nathan R, Weinmann PE (2013). Australian Rainfall and Runoff, Discussion Paper: Monte-Carlo Simulation Techniques, Engineers Australia.
Natural Environmental Research Council (NERC) (1975). Flood Studies Report, Vol. II Meteorological studies. NERC, London.
Olsson J, Willén U, Kawamura A (2012). Downscaling extreme short-term regional climate model precipitation for urban hydrological
applications, Hydrology Research, 43 (4), 341–351. http://dx.doi.org/10.2166/nh.2012.135.
Pielke RAS, Wilby RL (2012). Regional climate downscaling: what’s the point? Eos Trans. AGU 93, PAGE 52. doi:201210.1029/2012EO050008.
Pappenberge F, Beven KJ (2006). Ignorance is bliss: Or seven reasons not to use uncertainty analysis, Water Resources Research, 42, W05302,
doi: 10.1029/2005WR004820.
Rahman A, Weinmann PE, Hoang TMT, Laurenson EM (2002). Monte Carlo Simulation of flood frequency curves from rainfall, Journal of
Hydrology, 256, 196-210.
“Uncertainty in Design Rainfall Estimation: A Review” Mamoon and Rahman
Journal of Hydrology and Environment Research 75
Renard B, Kavetski D, Kuczera G, Thyer M, Franks SW (2010). Understanding predictive uncertainty in hydrologic modeling: The challenge of
identifying input and structural errors, Water Resources Research, 46, W05521, doi:10.1029/2009WR008328.
Seidou O, Ramsay A, Nistor I (2012). Climate change impacts on extreme floods I: combining imperfect deterministic simulations and non-
stationary frequency analysis, Natural Hazards, 61, 647–659.
Simonovic SP, Peck A (2009). Updated Rainfall Intensity Duration Frequency curves for the City of London under the changing climate, Report
No. 065, The University of Western Ontario.
Stedinger JR, Vogel RM, Foufoula-Georgiou E (1993). Frequency analysis of extreme events. In: Maidment D. R. (ed) Handbook of Hydrology,
McGraw-Hill, New York.
Trefry CM, Watkins Jr DW, Johnson D (2005). Regional Rainfall Frequency Analysis for the State of Michigan, Journal of Hydrologic
Engineering, 10, 437-449.
Tung YK (1996). Uncertainty and reliability analysis. In: L.W. Mays (ed.), Water Resources Handbook, McGraw-Hill, New York, NY, USA,
7.1-7.65.
Tung YK, Yen BC (2005). Hydrosystems engineering uncertainty analysis. McGraw-Hill, New York.
Tung Y, Wong C (2014). Assessment of design rainfall uncertainty for hydrologic engineering applications in Hong Kong, Stochastic
Environmental Research and Risk Assessment, 28,583–592.
Yen BC, Tang WH (1976). Risk–safety factor relation for storm sewer design, Journal of Environmental Engineering, ASCE, 102, 509–516.
Yen B, Ang A (1971). Risks analysis in design of hydraulic projects. In: C. Chiu (ed.), Stochastic Hydraulics, Proc. of First International
Symposium, University of Pittsburgh, Pittburgh, PA, USA, 694-709.
Ye M, Meyer PD, Neuman SP (2008). On model selection criteria in multi-model analysis, Water Resources Research, 44, W03428,
doi:10.1029/2008WR006803.
Yu P, Cheng C (1998). Incorporating uncertainty analysis into a regional IDF formula, Hydrological Processes, 12, 713-726.
Wang D, Hagen SC, Alizad K (2013). Climate change impact and uncertainty analysis of extreme rainfall events in the Apalachicola River
basin, Florida, Journal of Hydrology, 480, 125–135.
Westra S, Sisson SA (2011). Detection of non-stationarity in precipitation extremes using a max-stable process model, Journal of Hydrology,
406, 1-2. doi:10.1016/j.jhydrol.2011.06.014. 2.
Willems P, Vrac M (2011). Statistical precipitation downscaling for small-scale hydrological impact investigations of climate change, Journal of
Hydrology, 402 (3–4), 193–205. http://dx.doi.org/10.1016/j.jhydrol.2011.02.030.
Wittwer JW (2004). Monte Carlo Simulation Basics, Vertex42.com, http://www.vertex42.com/ExcelArticles/mc/MonteCarloSimulation.html
Wu SJ, Yang JC, Tung YK (2011). Risk analysis for flood control structure under consideration of uncertainties in design flood, Natural
Hazards, 58(1), 117–140.
Zakaria ZA, Shabri A, Ahmad UN (2012). Regional frequency analysis of extreme rainfalls in the west coast of Peninsular Malaysia using partial
L-moments, Water Resources Management, 26, 4417-4433.
Zadeh LA (1978). Fuzzy set as a basis for a theory of possibility, Fuzzy Sets and Systems, 1(1), 3–28.
Zadeh LA (2005). Toward a generalized theory of uncertainty (GTU)—an outline, Information Sciences, 172, 1–40.
Zhang X et al (2005). Trends in Middle East climate extreme indices from 1950 to 2003, Journal of Geophysical Research, 110
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
Guide to Authors
Guide to authors for preparing manuscript for submission in Journal of Hydrology and Environment Research (JHER):
1. The paper should be prepared using Microsoft Word and emailed to the Editor of JHER ([email protected]).
No pdf file is accepted.
2. The margins of the papers are 2.5 cm on top, 2 cm at bottom, 2.2 cm on left and 1.5 cm on right sides.
3. The paper should be written using Times New Roman font, 1.5 lines spaced and justified on both sides.
4. The title of the paper is 18 point Times Roman font size, centred, with the first letters of keywords as capital.
5. The text in the body of the paper is of 11 point Times Roman font size.
6. The major headings are of 12 point font size and bold, the sub-headings are of 11 point font size and bold and the
sub sub-headings are 11 point font size and italics.
7. The headings and sub-headings in the papers are to be numbered.
8. The structure of the paper may vary depending on the type of the paper, but the typical structure of a paper in the
JHER consists of Abstract: 250 words, Keywords: Three to five keywords, Introduction, Methods, Data, Results,
Conclusion, Acknowledgements, References and Appendix.
9. The figures and tables should be numbered by Arabic numerals (e.g. 1, 2, 3 …) and titles should be centred. The
table texts are of 10 font size.
10. The figures and tables should be placed in the centre of the page.
11. The equations should be numbered by Arabic numerals (e.g. 1, 2, 3). Equations should be left justified, with
equation numbers to be written at the far right end of the margin. The equations should be quoted in the body of
the paper following the style: It is found in equation 2 that temperature changes inversely with wind velocity.
12. References: (i) In the body of the paper: depending on the number of authors or organisations, the following styles
are acceptable: Smith (1980), Smith and Jones (2008), Smith et al. (2010), Sydney Water (2011), ABS (2012). (ii)
All the quoted references are to be listed in the reference section of the paper. Make all journal names in italics. The
following styles are acceptable:
Smith KL, Rahman A, Jones PR, Haddad K (1999). Towards a new regional flood methods in Australia, Australian Journal
of Water Resources, 4, 23-32.
Dawson JL, Rahman A (2012). A new regional flood estimation technique for north Australia, Technical report, University
of Western Sydney, 46 pp.
Yang X, Lu K, Henry R (2011). Water data analysis, 3rd edition, Wiley, 306 pp.
Imrul K, Jones D (2013). Water price in Australia, Accessed via www.waterprice.australia on 22 Jan 2012.
USE OF WORD TEMPLATE:
Authors are encouraged to use the word template in preparing the manuscripts which is available in the download page
(see www.jher.org).
Published by EnviroWater Sydney, New South Wales, Sydney, Australia, © EnviroWater Sydney, 2014
Journal of Hydrology and
Environment Research
Vol 2 No 1
______________________________________________________
Contents
Technical papers:
Assessment of the Priestley-Taylor Parameter Value from ERA-Interim Global Reanalysis Data 1
J. Szilagyi, M. B. Parlange, G. G. Katul
Estimation of Water Surface Elevation on Inundated Area Using Satellite Data 8
A. Yorozuya, H. Kamimera, T. Okazumi, Y. Iwami, Y. Kwak
Impacts of Outliers in Flood Frequency Analysis: A Case Study for Eastern Australia 17
A. S. Rahman, K. Haddad, A. Rahman
New Watershed Codification System for Indian River Basins 31
K. Pareta, U. Pareta
Assessment of Heavy Metal Contamination from Municipal Solid Waste Open Dumping Sites in 41
Bangladesh
M. R. Karim, M. Kuraoka, T. Higuchi, M. Sekine, T. Imai
Rating Curve Uncertainty in Flood Frequency Analysis: A Quantitative Assessment 50
M. M. Haque, A. Rahman, K. Haddad
Challenge on Modelling a Large River Basin with Scarce Data: A Case Study of the 59
Indus Upper Catchment
A. Sugiura, S. Fujioka, S. Nabesaka, T. Sayama, Y. Iwami, K. Fukami, S. Tanaka, K. Takeuchi
Review Paper: Uncertainty in Design Rainfall Estimation: A Review 65
A. Mamoon, A. Rahman