15arspc submission 210

18
FLOOD INUNDATION MAPPING OF TROPICAL RIVER CATCHMENTS IN NORTHERN AUSTRALIA USING OPTICAL AND ALOS-PALSAR DATA Renee Bartolo 1,3 , Doug Ward 2,3 and David Jones 1,3 Author affiliation 1 : Supervising Scientist Division, ERISS GPO Box 461, Darwin NT 0801 Ph (08) 8920 1392, Fax (08) 8920 1199 [email protected]  Author affiliation 2 : Australian Rivers Institute, Griffith University 170 Kessels Road, Nathan Qld 4111 Ph (07) 37353543 [email protected]  Author affiliation 3 : Tropical Rivers and Coastal Knowledge (TRaCK) Commonwealth Environmental Research Facility, Australian Government, Canberra, ACT Abstract Defining the extent of wet season inundation in floodplain and riverine environments is an important component of the annual catchment surface and groundwater budgeting process. This paper reports on a project to map the flood inundation extent of the Daly River (Northern Territory) catchment and was undertaken as part of a larger flood mapping project for Theme 4 (catchment water budgets and water resource assessment) through the Tropical Rivers and Coastal Knowledge (TRaCK) program. Determining the extent of flooding in tropical catchments using remote sensing is dependent on a number of factors: local conditions at the time of image acquisition (e.g.: cloud cover and flooding under vegetation); sensor selection (e.g.: optical and SAR); and definition of flood extent (interannual and intrannual analysis). Using historic rainfall data for the Daly catchment, 2009 was identified as a suitable year for the mapping of a wet year maximum inundation extent. Optical remote sensing imagery (Landsat 5 TM) were captured during the 2009 wet season, coincident with ALOS PALSAR ScanSAR scenes. Classification of extent of flooding for a single flood event (March 7-8, 2009) was conducted using Geographic Object Based Image Analysis (GEOBIA) approaches, whereby the use of both optical and SAR data reduced the potential for confusion between vegetation types and enabled issues of cloud to be addressed in the optical image. The results from this project will be used in a number of TRaCK projects: Theme 4 (catchment water budgets and water resource assessment); Theme 5 (food webs and biodiversity) as part of the assessment process for defining the biodiversity and biomass components of floodplain and riverine ecosystems; and Theme 1 (scenario evaluation).  

Upload: reneebartolo

Post on 10-Apr-2018

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 1/18

FLOOD INUNDATION MAPPING OF TROPICAL RIVERCATCHMENTS IN NORTHERN AUSTRALIA USING OPTICAL

AND ALOS-PALSAR DATA

Renee Bartolo1,3

, Doug Ward2,3

and David Jones1,3

Author affiliation1:

Supervising Scientist Division, ERISSGPO Box 461, Darwin NT 0801

Ph (08) 8920 1392, Fax (08) 8920 [email protected] 

Author affiliation2:

Australian Rivers Institute, Griffith University170 Kessels Road, Nathan Qld 4111

Ph (07) [email protected] 

Author affiliation3:

Tropical Rivers and Coastal Knowledge (TRaCK) Commonwealth Environmental Research Facility, AustralianGovernment, Canberra, ACT

Abstract

Defining the extent of wet season inundation in floodplain and riverineenvironments is an important component of the annual catchment surface andgroundwater budgeting process. This paper reports on a project to map theflood inundation extent of the Daly River (Northern Territory) catchment andwas undertaken as part of a larger flood mapping project for Theme 4

(catchment water budgets and water resource assessment) through theTropical Rivers and Coastal Knowledge (TRaCK) program. Determining theextent of flooding in tropical catchments using remote sensing is dependent ona number of factors: local conditions at the time of image acquisition (e.g.:cloud cover and flooding under vegetation); sensor selection (e.g.: optical andSAR); and definition of flood extent (interannual and intrannual analysis).

Using historic rainfall data for the Daly catchment, 2009 was identified as asuitable year for the mapping of a wet year maximum inundation extent. Opticalremote sensing imagery (Landsat 5 TM) were captured during the 2009 wetseason, coincident with ALOS PALSAR ScanSAR scenes. Classification of

extent of flooding for a single flood event (March 7-8, 2009) was conductedusing Geographic Object Based Image Analysis (GEOBIA) approaches,whereby the use of both optical and SAR data reduced the potential forconfusion between vegetation types and enabled issues of cloud to beaddressed in the optical image.

The results from this project will be used in a number of TRaCK projects:Theme 4 (catchment water budgets and water resource assessment); Theme 5(food webs and biodiversity) as part of the assessment process for defining thebiodiversity and biomass components of floodplain and riverine ecosystems;and Theme 1 (scenario evaluation). 

Page 2: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 2/18

1. Introduction

The main focus of this paper is to determine whether mapping inundation on atropical floodplain in northern Australia using both optical and Synthetic

Aperture Radar (SAR) combined improves the ability to undertake suchmapping, compared with using just one data source (either optical or SAR).

There are a number of challenges to using remotely sensed data for such atask as mapping tropical floodplain inundation extent. In the tropics,atmospheric attenuation or interference resulting from cloud cover during the‘wet’ season, and smoke from fires during the ‘dry’ season (Schultz andEngman, 2000) are limiting factors for the use of optical sensors. Despiteaddressing issues of cloud cover and smoke, Bartolo et al ., (2005) in a studythat used ERS SAR data to map inundation limits in the Trans Fly Bioregion ofsouthern New Guinea, list instances where classification using thresholdtechniques can result in either the overestimation or underestimation ofinundated areas. Instances of overestimation of inundated areas include wheresoil moisture is sufficiently high enough to result in low backscatter in the rangeof open water. Conversely, instances of underestimation of inundated areasinclude: where wind results in an increase in surface roughness of standingwater thereby increasing the backscatter (van Leeuwen et al ., 2004); and withshort wavelength SAR data from a single scene (such as ERS), wherevegetation in standing water results in high backscatter (Laugier et al ., 2004).

Few studies have been published using both optical and SAR data for tropicalfloodplain mapping applications with a GEOBIA approach. Hamilton et al .,(2007), used Landsat 7 ETM+, JERS-1 and the Shuttle Radar TopographyMission (SRTM) DEM integrated using a GEOBIA approach to map fivefloodplain vegetation classes for a tributary of the Amazon River in Peru. Morerecently, Silva et al ., (2010), used MODIS, Radarsat-2 and SRTM data toclassify flooded vegetation and specifically macrophytes using a GEOBIAapproach also.

A key study that demonstrates the utility of SAR data for mapping flooding andvegetation on a tropical floodplain (Magela Floodplain, Kakadu National Park),was conducted by Hess and Melack (2003). They used multi-frequencypolarimetric SIR-C (dual C- and L-band, HH and HV) acquired in 1994 to mapwoody (Melaleuca ) and herbaceous macrophyte vegetation (Nelumbo nucifera ,Pseudoraphis grassland and Hymenachne-Eleocharis swamp) in both floodedand non-flooded states. The authors suggest that due to the similarity of theMagela Floodplain to other floodplains in northern Australia, that the methodscan be used in a regional context on other floodplains.

This paper presents the use of optical (Landsat 5 TM) and SAR (ALOS-PALSAR ScanSAR) imagery to map floodplain inundation in Daly Rivercatchment. The data processing methods, characterisation of spectral andbackscatter signatures, and development of a rule set for input into a GEOBIAclassification are presented.

Page 3: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 3/18

2. Study Area

The Daly River catchment is located in the Top End of the Northern Territory,Australia, and 200 km south of Darwin. It encompasses approximately 52 600km

2and is one of the largest catchments in the Top End. Figure 1 shows the

Daly River catchment and the Daly River Floodplain System. The Daly Riveritself is one of the largest rivers in the Northern Territory (Faulks, 1998) and hasa perennial flow component, which is an atypical water regime for northernAustralia. The major population centre within the catchment is Katherine andthe dominant land use type is grazing of natural vegetation followed bytraditional indigenous use (BRS: Land Use Mapping of the Northern Territory,2002).

Like most of northern Australia, the Daly River catchment is dominated by amonsoonal climate, with mean daily air temperatures exceeding 30º Cthroughout the year. The region is characterised by a distinctive wet and dry

season. The dry season extends from June to August and is a prolonged periodof minimal to nil rainfall and low humidity. April-May and September-Novemberare transitional months dominated by wet-dry and dry-wet weather respectively(Taylor and Tulloch, 1985). Mean annual rainfall for the Daly River catchmentvaries from around 700 mm in the southern section to over 1,300 mm in thenorth-western section (Erskine et al , 2003, Moliere, 2008) and mainly occursduring the four month wet season from December to March.

The Daly River catchment has the largest flow of all NT rivers, with an averageannual discharge of around 7 000 000 ML (Jolly, 2001). Dry season flowexceeds that of any other NT river, due to the large groundwater aquifers

supplying the river (Price et al., 2001).Two wetland systems within the Daly River catchment are listed in the Directoryof Important Wetlands in Australia (Department of the Environment and WaterResources, 2007): the 1593 km

2Daly-Reynolds Floodplain-Estuary System,

which includes the entire floodplain and estuary of the Daly River, and the 165km long Daly River Middle Reaches. A diverse range of wetland types arefound in the Daly-Reynolds Floodplain-Estuary System. It represents one of thelargest floodplains in the NT, and has the largest catchment of the majorfreshwater floodplains in the Top End (Department of the Environment andWater Resources, 2007). The focus of this study is on inundation mapping ofthe Daly River Floodplain System as shown in Figure 1.

Page 4: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 4/18

 

Figure 1: Location of the Daly River Floodplain System.

3. Methodology

The following sections describe the data used, pre-processing of data andclassification methodology.

3.1. Selection of “wettest” wet season from the satellite image record

In order to map the extent of maximum inundation of the Daly River FloodplainSystem, the wet season that received the highest rainfall was identified usingFoley’s precipitation deficit index (Foley, 1957). Foley’s precipitation deficitindex is the standardized monthly mean annual precipitation over a specifiedlag period relative to the long-term mean annual precipitation. Fensham andHolman (1999) found that 3 years is a significant lag period for precipitationdeficit to influence vegetation dynamics in Australia’s tropical savannas. Hence,we calculated Foley’s precipitation deficit index for a 3 year lag period. The yearthat matched the satellite record for both optical and L-band SAR data wasidentified to be 2009 (see Figure 2).

Page 5: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 5/18

 

Figure 2: Monthly rainfall and monthly Rainfall Deficit Index (3 year lag) for the DarwinRiver Gauge for 1988-2009.

3.2. Data Description

The extents of the imagery used in this study are illustrated in Figure 3. Theseimage data are Landsat 5 TM and ALOS PALSAR ScanSAR.

Figure 3: Extents of imagery used for the inundation mapping of the Daly RiverFloodplain System.

Page 6: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 6/18

3.2.1. Landsat 5 TM Data

A Landsat 5 TM image of the Daly River Floodplain System captured on the 7th

 March 2009 was acquired from the U.S. Geological Survey (USGS). Theproduct type (processing level) is Level 1T, Standard Terrain Correction. Thescene metadata and sensor characteristics are reported in Table 1. There wasminimal cloud cover over the floodplain region.

Table 1: Landsat 5 TM sensor characteristics and scene metadata

Date 07/03/2009

Path/Row 106/69

Centre Latitude 13°00'47" S

Centre Longitude 130°23'11" E

Swath Width 185 km

Pixel Size 30 m (Bands 1-5 & 7)

120 m (Band 6)

Orbit Number 133048

Sun Elevation 54.6°

Sun Azimuth 80.7°

Average cloud Cover (0-100) 32

3.2.2. PALSAR Data

PALSAR (Phased Array type L-band Synthetic Aperture Radar) data isacquired by the Japan Aerospace Explorations Agency’s (JAXA) ALOS(Advanced Land Observing Satellite). The PALSAR sensor acquiresbackscatter data at L-band (23.62 cm wavelength, 1.27 GHz), in four mainobservation modes: fine beam single; fine beam double; ScanSAR wide beam;and polarimetric.

A PALSAR ScanSAR wide beam mode (WB1) image was captured on the 8th

 

March 2009 to coincide as close to the Landsat 5 TM data capture as possible.The data were acquired as Level 1.5-G processing (map oriented image). Table2 lists the scene parameters. The data format was signed integer.

Page 7: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 7/18

Table 2: PALSAR ScanSAR scene metadata

Date 08/03/2009

Path/Frame 60/3900

Centre Latitude 14°10'58" SCentre Longitude 129°03'25" E

Polarization HH

Incidence Angle 27.1°

Orbital Direction Descending

Swath Width 350 km

Pixel Size 100 m

3.2.3. Ancillary DataData delineating the extent of the Daly River System Floodplain were extractedfrom the GEODATA TOPO 250K Series 3 “flats” feature class. This wassubsequently spatially buffered to 2 kms for use as a clipping file for imagedata.

The wetland and riparian vegetation communities were extracted from theNational Vegetation Information System (NVIS) for the Daly River catchmentand used to aid in visual interpretation of the PALSAR ScanSAR data. Thewetland and riparian vegetation communities were extracted using the ‘environ’category within the data.

3.3. Pre-Processing of PALSAR Data

Figure 4 describes the processing steps for the PALSAR data. The imageswere received in signed integer data format. The data were converted tounsigned short integer format prior to applying a radiometric conversion usingthe Normalised Radar Cross Section (NRCS) equation for Level 1.5 product.This equation is available from the ALOS User Interface Gateway(https://auig.eoc.jaxa.jp/auigs/en/doc/an/20090109en_3.html) and is calculatedby:

NRCS = 10 x log10 (DN2) + CF

where CF for ScanSAR data is -83.

An Enhanced Lee Filter (3x3) was then applied, which reduces the speckle thatis inherent in radar imagery, whilst preserving texture and image sharpness.Finally the image was subset the floodplain with a buffer of 2kms.

Page 8: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 8/18

 

Figure 4: PALSAR ScanSAR processing flow diagram.

3.4. Landsat 5 TM Band Ratios

Two band ratios from the literature on water extraction from optical satelliteimagery were selected for highlighting open water (flooding) in the Landsat 5TM data. The Normalised Difference Water Index (NDWI) was developed todelineate open water in satellite imagery (McFeeters, 1996). The NDWI iscalculated as follows:

(Green – NIR) where Green = TM Band 2 and NIR = TM Band 4

(Green + NIR)

This equation results in an image with pixel values ranging between -1 and 1.Water is delineated by positive values greater than 0, and terrestrial vegetationand soil are represented by 0 and negative values.

The Modified Normalised Difference Index (MNDWI) is built upon the principleof the NDWI but further suppresses built up areas (Xu, 2006) and clearedregions in the imagery by substituting the NIR with the MIR band. The MNDWIis calculated as follows:

(Green – MIR) where Green = TM Band 2 and MIR = TM Band 5

(Green + MIR)

Apply NRCS conversion to the amplitude data

NRCS = 10 x log10 (DN2) + CF

where CF for ScanSAR data is -83.

Apply Enhanced Lee Filter to reduce speckle

Filter Size = 3 x3

Damping Factor = 1

Homogenous Areas (Cu) = 0.523

Heterogeneous Areas (Cmax) = 1.73

Subset data to floodplain file buffered to 2km

Page 9: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 9/18

3.5. Spectral Characteristics of Floodplain Classes

In order to develop rules for classifying inundated areas, the spectralcharacteristics and statistics of five floodplain classes (surface water; mixedwater/seasonal grasses sedges; standing woody vegetation; seasonal grasses

and sedges; and floodplain extent) derived from previously undertaken LandsatTM 5 classifications (Ward, unpublished), were assessed for the PALSARScanSAR, and Landsat TM 5 and associated band ratios (NDWI and MNDWI).Figure 5 shows the spectral characteristics of these classes for the Landsat 5TM data.

Figure 5: Spectral characteristics of the five predefined classes for Landsat TM bands,

March 2009.

Class statistics were collected for six newly generated classes (clear oceanwater, wind disturbed ocean water, turbid river water, billabong, flooded grassesand sedges, and flooded Melaleuca ) in order to develop rules for input into theclassification process.

3.6. Classification Method

A Geographic Object Based Image Analysis (GEOBIA) approach (Benz et al.,2004) was used to classify the Landsat 5 TM and PALSAR ScanSAR datausing Definiens Developer 8.0. A multi-resolution segmentation was conductedat two levels as follows:

i. Level 1- pixel level and scale parameter of 10 (Input bands: Green, NIR[Landsat 5 TM] and MNDWI and NDWI-band ratios); and

ii. Level 2- image object level and scale parameter of 50 (Input bands: Green,NIR [Landsat 5 TM] and MNDWI and NDWI-band ratios).

Classification rules were derived from both the Landsat 5 TM and PALSAR

ScanSAR data. Figure 6 shows the classification scheme and rules to develop

0

10

20

30

40

50

60

70

80

90

1 2 3 4 5 7Landsat TM Bands

      D      N

Surface Water

Mixed Water/Seasonal

Grasses & Sedges

Standing Woody

Vegetation

Seasonal Grasses &

SedgesFloodplain Extent

Page 10: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 10/18

flooded classes. Firstly, “floodplain” and “upland areas” were classified from theLevel 2 segmentation. The flooded classes were then classified inheriting classrelated features from the “floodplain” class.

Figure 6: Process flow chart for classifying flooded classes.

4. Results and Discussion

4.1. Spectral Characteristics of Floodplain Classes

Appendix 1 summarises basic class statistics (minimum, maximum, mean andstandard deviation) of the six classes: clear ocean water; wind disturbed oceanwater; turbid river water; billabong; flooded grasses and sedges; and floodedMelaleuca . These class statistics enabled the generation of classification rulesthat enabled the separation of classes.

Figure 7 illustrates the class range values for the six classes for the PALSARScanSAR data and the band ratios, and also shows the spectral signature from

Multi-resolution segmentation

(Pixel Level)

Level 1

Multi-resolution segmentation

(Image Object Level)

Level 2

Level 2 (Floodplain/Upland Delineation)

Threshold classification:

Floodplain = Mean Landsat 5 TM Band <60

Upland = Landsat 5 TM Band ≥60

Level 1 (Flooded Class Delineation)Inherits Class Related Feature from

“Floodplain” class

Open Water = Mean MNDWI >0

Flooded Melaleuca = Mean ScanSAR -8 to -1

Flooded Grasses & Sedges = Mean ScanSAR-20.4 to -10

Open Water [2] = Mean ScanSAR <-20.4

Segmentation

Classification

Page 11: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 11/18

the Landsat 5 TM data. Wind disturbed ocean water has a higher range ofbackscatter values ( -13.9 up to -10.3) compared with the much lower range ofbackscatter values for clear ocean water (-23.9 up to -13.1). Surface waterbodies further inland also displayed different backscatter ranges with lower

backscatter values for turbid river water (-23.4 up to -17.5) compared to thehigher backscatter values for billabongs (-12.1 up to -5.9). The spectralsignatures derived from the Landsat 5 TM, indicate spectral overlap (similarsignatures) for the clear ocean and wind disturbed ocean classes. There is aslightly lower reflectance in the green and red bands for the billabong class,whilst conversely; there is comparatively higher reflectance in these bands inthe turbid river water class. The differences in SAR backscatter values(particularly with the higher backscatter range of the wind disturbed ocean andbillabong classes), and the spectral overlap in the Landsat TM 5 data indicatedthat the MNDWI threshold approach was the most appropriate method foridentifying the open water class in the first instance (i.e. first pass).

There is spectral overlap with a relatively high response in the NIR band whenexamining the Landsat 5 TM data (Figure 7) for the “flooded grasses andsedges” and “flooded Melaleuca” class signatures. The higher values in therange of the MNDWI values does indicate the presence of water, but the highervalues (0.05 and 0.12 respectively) are not significant enough to use these fordelineating this flooded vegetation classes. However, the SAR backscattervalues indicate good separability between these classes with the “floodedMelaleuca” class having a higher range of values (-6.0 up to -3.2) compared tothe lower values of the “flooded grasses and sedges” class (-19.8 up to -16.7).Therefore, once the open water class had been delineated, the SAR

backscatter values for these two classes could be used to define them.

4.2. Classification of Flooded Classes

The class hierarchy using class inheritance in developing class related featureswas developed as follows:

Floodplain

Open Water

Flooded Melaleuca

Flooded Grasses and SedgesOpen Water 2

Upland

The results of the Level 2 (floodplain and upland delineation) and Level 1(flooded classes) classifications are shown in Figures 8 and 9 respectively. Thefloodplain delineation classification corresponds well with the GEODATA TOPO250K Series 3 “flats” feature class, and the flooded classes correspond visuallywith the NVIS data. The ScanSAR data rule for classifying “Flooded Melaleuca”is in agreement with the Melaleuca mapped through NVIS.

Page 12: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 12/18

 

Figure 7: Class data range values and spectral signatures

0

10

20

30

40

50

60

70

80

1 2 3 4 5 7

Landsat TM Bands

     D     N

Clear Ocean Water

PALSAR ScanSARReturn Response

-23.9 to -13.1

NDWI- DN Range0.42 - 0.56

MNDWI- DN Range0.42 - 0.75

0

10

20

30

40

50

60

70

1 2 3 4 5 7

Landsat TM Bands

     D     N

Wind Disturbed OceanWater

PALSAR ScanSARReturn Response

-13.9 to -10.3

NDWI- DN Range0.33 - 0.54

MNDWI- DN Range0.33 - 0.71

0

10

20

30

40

50

60

70

80

90

1 2 3 4 5 7

Landsat TM Bands

     D     N

Turbid River Water

PALSAR ScanSARReturn Response

-23.4 to -17.5

NDWI- DN Range0.22 - 0.37

MNDWI- DN Range0.55 - 0.78

0

10

20

30

40

50

60

1 2 3 4 5 7

Landsat TM Bands

     D     N

Billabong

PALSAR ScanSARReturn Response

-12.1 to -5.9 

NDWI- DN Range0.05 - 0.22

MNDWI- DN Range0.11 - 0.29

0

10

20

30

40

50

60

70

1 2 3 4 5 7

Landsat TM Bands

     D     N

Flooded Grasses& Sedges

PALSAR ScanSARReturn Response

-19.8 to -16.7

NDWI- DN Range-0.33 to -0.19

MNDWI- DN Range-0.24 to 0.05

0

10

20

30

40

50

60

70

1 2 3 4 5 7

Landsat TM Bands

     D     N

Flooded Melaleuca

MNDWI- DN Range-0.32 - 0.12

PALSAR ScanSARReturn Response

-6.0 to -3.2

NDWI- DN Range-0.41 - 0

Page 13: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 13/18

Classification results require validation with data collected from other TRaCKTheme 4 projects. We have data sets to be analysed for April, May and June2009 also, thereby producing a time series of flood or inundation extent. Thiswill require that at sensor calibration of the Landsat 5 TM data is undertaken to

determine whether the rule set developed for the March image data, can betransferred and applied to the other datasets. In addition, we have polarimetricand Fine Beam Double mode PALSAR data for the floodplain that can beanalysed to further improve mapping results.

Figure 8: Classification results for the Level 2 delineation of floodplain and uplandareas.

Page 14: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 14/18

 

Figure 9: Classification results for the Level 1 classes for flooded areas.

5. Summary

This paper has demonstrated that combining optical and SAR data improvesinundation mapping on tropical floodplains in northern Australia. Thecombination of the data effectively addresses areas of spectral or backscatterconfusion in the respective datasets. The use of SAR data also enables issuesaround potential cloud cover to be resolved. The GEOBIA approach allows for

sequences of rules to be applied using multiple datasets to produce meaningfulclasses and subsequent classifications. The development of such rule setsfacilitates standardised classifications for multi-date calibrated data from thesame site. Further work to compare PALSAR modes (ScanSAR, Polarimetricand Fine Beam Double imagery) will be conducted on the Daly River floodplainfor the 2009 year. In addition, PALSAR ScanSAR imagery and Landsat 5 TMdata for the Mitchell River catchment in Queensland captured in January 2009will be analysed and assessed against depth logger data. The incorporation ofDigital Elevation Data (DED) such as the SRTM data will also be investigated tofurther improve mapping outcomes. 

Page 15: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 15/18

Acknowledgements

TRaCK receives major funding for its research through the AustralianGovernment’s Commonwealth Environment Research Facilities initiative; theAustralian Government’s Raising National Water Standards Program; Land and

Water Australia; the Fisheries Research and Development Corporation and theQueensland Government’s Smart State Innovation Fund.

Appendix

Appendix 1: Basic class statistics (minimum, maximum, mean and standard deviation)of the six classes: clear ocean water; wind disturbed ocean water; turbid river water;

billabong; flooded grasses and sedges; and flooded Melaleuca .

Class Sensor/Band/Ratio Minimum Maximum Mean Standard

Deviation

Clear OceanWater

PALSAR ScanSAR -23.903057 -13.080664 -18.814840 2.072771

Landsat TM 5 Band 1 66 74 70.4 1.2

Landsat TM 5 Band 2 29 36 32.3 0.9

Landsat TM 5 Band 3 22 28 25.4 1.0

Landsat TM 5 Band 4 10 13 11.2 0.5

Landsat TM 5 Band 5 5 13 9.3 1.0

Landsat TM 5 Band 7 3 9 6.0 0.8

NDWI 0.42 0.56 0.48 0.02

MNDWI 0.42 0.75 0.55 0.04

Wind DisturbedOcean Water 

PALSAR ScanSAR -13.888852 -10.311194 -12.260462 0.620305

Landsat TM 5 Band 1 60 73 66.5 1.3

Landsat TM 5 Band 2 27 32 29.3 0.7

Landsat TM 5 Band 3 21 27 23.3 0.9

Landsat TM 5 Band 4 9 14 11.2 0.8

Landsat TM 5 Band 5 5 15 9.4 1.2

Landsat TM 5 Band 7 3 9 6.0 0.8

NDWI 0.33 0.54 0.45 0.03

MNDWI 0.33 0.71 0.52 0.05

Page 16: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 16/18

Turbid RiverWater 

PALSAR ScanSAR -23.411917 -17.479185 -20.885834 1.595148

Landsat TM 5 Band 1 73 82 76.9 1.3

Landsat TM 5 Band 2 38 43 40.8 0.6

Landsat TM 5 Band 3 44 49 46.5 0.7

Landsat TM 5 Band 4 19 26 22.0 1.1

Landsat TM 5 Band 5 5 12 9.1 1.1

Landsat TM 5 Band 7 3 9 5.5 0.8

NDWI 0.22 0.37 0.30 0.02

MNDWI 0.55 0.78 0.63 0.04

Billabong PALSAR ScanSAR -12.098226 -5.900466 -9.997066 1.145060

Landsat TM 5 Band 1 54 59 56.5 1.0

Landsat TM 5 Band 2 20 23 21.3 0.5

Landsat TM 5 Band 3 16 19 17.2 0.6

Landsat TM 5 Band 4 14 20 15.8 1.1

Landsat TM 5 Band 5 12 17 13.7 1.1

Landsat TM 5 Band 7 5 9 6.8 0.8

NDWI 0.05 0.22 0.15 0.03

MNDWI 0.11 0.29 0.22 0.04

FloodedGrasses andSedges

PALSAR ScanSAR -19.753723 -16.708330 -18.379116 0.664025

Landsat TM 5 Band 1 58 68 63.1 1.2

Landsat TM 5 Band 2 27 33 29.5 0.8

Landsat TM 5 Band 3 23 29 26.5 0.9

Landsat TM 5 Band 4 43 60 47.7 2.3

Landsat TM 5 Band 5 27 49 35.0 3.3

Landsat TM 5 Band 7 11 19 14.6 1.5

NDWI -0.33 -0.19 -0.24 0.02

MNDWI -0.24 0.05 -0.08 0.05

Page 17: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 17/18

FloodedMelaleuca

PALSAR ScanSAR -6.008411 -3.210153 -4.637622 0.572976

Landsat TM 5 Band 1 55 65 59.5 1.4

Landsat TM 5 Band 2 22 34 26.1 1.1

Landsat TM 5 Band 3 18 27 21.6 0.9

Landsat TM 5 Band 4 24 65 44.4 6.3

Landsat TM 5 Band 5 19 52 36.3 5.8

Landsat TM 5 Band 7 8 20 13.9 1.8

NDWI -0.41 0 -0.25 0.05

MNDWI -0.32 0.12 -0.15 0.07

References

Bartolo, R.E., Forner, J. and McGinley, B., 2005, Mapping inundation limits on tropicalfloodplains for biodiversity conservation, North Australian Remote Sensing and GIS Conference , Charles Darwin University, Darwin, 4-7 July.

Benz, U.C., Hofman, P., Willhauck, G., Lingenfelder, I. and Heynen, M., 2004,

Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry and Remote Sensing , 58,pp. 239-258.

Bureau of Rural Sciences,. 2002, Land Use Mapping at Catchment Scale (Edition 2). Commonwealth of Australia, Canberra.

Department of the Environment and Water Resources 2007. A Directory of Important Wetlands in Australia .www.environment.gov.au/water/publications/environmental/ wetlands/database/.

Erskine, W,D., Begg, G,W., Jolly, P., Georges, A., O’Grady, A., Eamus, D.,

Rea, N., Dostine, P., Townsend, S. and Padovan, A., 2003, Recommended environmental water requirements for the Daly River, Northern Territory, based on ecological, hydrological and biological principles . Supervising ScientistReport 175, Supervising Scientist, Darwin NT. Faulks, J.J., 1998, An Assessment of the Physical and Ecological Condition ofthe Daly River and its Major Tributaries. Technical Report No TR99/10 ,Northern Territory Department of Lands, Planning and Environment.

Fensham, R. J. and Holman, J. E., 1999, Temporal and spatial patterns indrought-related tree dieback in Australian savanna. Journal of Applied Ecology ,36, pp.1035-1050.

Page 18: 15arspc Submission 210

8/8/2019 15arspc Submission 210

http://slidepdf.com/reader/full/15arspc-submission-210 18/18

Foley, J. C., 1957, Droughts in Australia. Review of records from earliest years of settlement to 1955. Bureau of Meteorology, Commonwealth of Australia,Melbourne, Australia.

Hamilton, S.K., Kellndorfer, J., Lehner, B. and Tobler, M., 2007, Remotesensing of floodplain geomorphology as a surrogate for biodiversity in a tropicalriver system (Madre de Dios, Peru). Geomorphology , 89, pp. 23-38.

Hess, L.L. and Melack, J.M., 2003, Remote sensing of vegetation and floodingon Magela Creek Floodplain (Northern Territory, Australia) with the SIR-Csynthetic aperture radar. Hydrobiologia , 500, pp. 65-82.

Laugier, O., Felleh, K., Tholey, N., Meyer, C. and Fraipont, P., 2004, Hightemporal detection and monitoring of flood zone dynamics using ERS dataaround catastrophic natural events: The 1993 and 1994 Camargue floodevents. http://earth.esa.int:80/symposia/papers/laugier

McFeeters, S.K., 1996, The use of the normalized difference water index(NDWI) in the delineation of open water features. International Journal of Remote Sensing , 17, pp. 1425-1432.

Moliere, D., 2008. Hydrology. In A Compendium of ecological information on Australia’s Northern Tropical Rivers. Sub-project 1 of Australia’s Tropical Rivers – an integrated data assessment and analysis  Lukacs, G. and Finlayson, M.(Eds.) A report to Land & Water Australia. (Australian Centre for TropicalFreshwater Research, National Centre for Tropical Wetland Research:Townsville, Qld).

Price, O., Milne, D., Connors, G., Harwood, B., Woinarski, J.C.Z. and Butler,

M., 2003, Draft conservation plan for the Daly Basin Bioregion. Parks andWildlife Commission of the Northern Territory, Darwin.

Schultz, G.A. and Engman, E.T., 2000, Remote Sensing in Hydrology and Water Management . (Springer, Verlag Berlin Heidelberg: New York).

Silva, T.S.F., Costa, M.P.F and Melack, J.M., 2010, Spatial and temporalvariability of macrophyte cover and productivity in the eastern Amazonfloodplain: A remote sensing approach, Remote Sensing of Environment , 114,pp. 1998-2010.

Taylor, J.A. and Tulloch, D., 1985, Rainfall in the wet-dry tropics: Extreme

events at Darwin and similarities between years during the period 1870-1983inclusive, Australian Journal of Ecology , 10, pp. 281-295.

van Leeuwen, H., Martin.m T., Haque, I., Hassan, A., Werle, D., and Tittley, B.,2004, Flood monitoring study in the Jamuna and Amp Ganges Floodplain inBangladesh using ERS-1, http://earth.esa.int:80/symposia/papers/vanleeuwen1

Xu, H.Q., 2006, Modification of normalised difference water index (NDWI) toenhance open water features in remotely sensed imagery. International Journal of Remote Sensing , 27, pp. 3025-3033.