abstract - wit press...figure 1. structure of oil spill modelling 3 gis and spatial data quality of...

10
Spatial data quality and coastal spill modelling Y. Li, A.J. Brimicombe & M.P. Ralphs Abstract Issues of spatial data quality are central to the whole oil spill modelling process. Both model and data quality performance issues should be considered as indispensable parts of a complete oil spill model specification and testing procedure. This paper presents initial results of research that will emphasise to modeller and manager alike the practical issues of spatial data quality for coastal oil spill modelling. It is centred around a case study of Jiao Zhou Bay in the People's Republic of China. The implications for coastal oil spill modelling are discussed and some strategies for managing the effects of spatial data quality in the outputs of oil spill modelling are explored. 1 Introduction In recent years a growing number of numerical models have been applied to oil spill research, contingency planning and risk assessment. Spatial data quality has become a critical theme in the development and utilisation of oil spill modelling^. For example, the quality of data outputs, especially in quantitative terms, is the decisive factor in determining the fitness for use of the results from oil spill modelling. Geographic Information Systems (GIS) are increasingly being used in conjunction with oil spill modelling as a tool for integrating and pre- processing spatial data inputs to the numerical modelling and for post- processing and visualisation of the modelling outputs. During the 1990's Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Upload: others

Post on 15-Oct-2020

0 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

Spatial data quality and coastal spill modelling

Y. Li, A.J. Brimicombe & M.P. Ralphs

Abstract

Issues of spatial data quality are central to the whole oil spill modelling process.Both model and data quality performance issues should be considered asindispensable parts of a complete oil spill model specification and testingprocedure. This paper presents initial results of research that will emphasise tomodeller and manager alike the practical issues of spatial data quality for coastaloil spill modelling. It is centred around a case study of Jiao Zhou Bay in thePeople's Republic of China. The implications for coastal oil spill modelling arediscussed and some strategies for managing the effects of spatial data quality inthe outputs of oil spill modelling are explored.

1 Introduction

In recent years a growing number of numerical models have been appliedto oil spill research, contingency planning and risk assessment. Spatialdata quality has become a critical theme in the development andutilisation of oil spill modelling^. For example, the quality of dataoutputs, especially in quantitative terms, is the decisive factor indetermining the fitness for use of the results from oil spill modelling.

Geographic Information Systems (GIS) are increasingly being used inconjunction with oil spill modelling as a tool for integrating and pre-processing spatial data inputs to the numerical modelling and for post-processing and visualisation of the modelling outputs. During the 1990's

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 2: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

54 Oil & Hydrocarbon Spills, Modelling, Analysis & Control

a considerable body of research has been developed around themeasurement, modelling and management of the uncertainty present inmost spatial data handled by GIS ' ' . Given the importance of suchfactors as shoreline and bathymetric representation, data resolution andmethod of interpolation in modelling coastal spills for example, it isessential that modelling of spatial data quality, especially propagation oferror, is applied to hydrodynamic, trajectory and fate modelling.

2 Oil spill modelling and spatial data quality

2.1 Oil spill modelling

Oil spill trajectory and fate models play an important role in oil spillresearch and application. Figure 1 shows the general structure oftrajectory and fate modelling. In technical terms, these models arepreceded by hydrodynamic modelling and themselves form the basis ofother specialised models, such as various risk analysis models, impactassessment models and decision making models. There is considerablescope for the propagation of uncertainty through the system. Therefore,the focus of this paper is on the effect of spatial data quality primarily onthe hydrodynamic modelling with a consideration of knock-on effects forcoastal oil spill trajectory and fate modelling.

Oil spill modelling is a complex spatio-temporal modelling process.Since a considerable range of spatial data need to be handled in oil spillmodels, the quality of spatial data is an issue. Modeller and managercannot confidently use models without knowing their accuracy andlimitations. The research and application of data quality techniques willbe one of the key issues for the next generation of oil spill models^.

2.2 Spatial data quality problems in oil spill modelling

Oil spill modelling is a complex process influenced by many spatialfactors. The input data resolution, coverage and structure should beanalysed to minimise error. Such error analysis is necessary forbathymetric and tidal data in the hydrodynamic modelling. It also appliesto current data, wind data and related environmental data in the trajectoryand fate modelling. Bathymetric data is given emphasis here because ofits predominant role in the models. Due to error combination andpropagation, relatively high quality data input cannot necessarilyguarantee that output will achieve similar levels of quality. Thus therelationship of spatial data input to output is to a great extent model

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 3: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

Oil & Hydrocarbon Spills, Modelling, Analysis & Control 55

dependent. This particularly needs to be understood where calibration is

required. It is also useful to compare different models. As users may not

be experts in oceanography or oil spill behaviour, spatial data qualityperformance specifications for models are required such as use

limitations and the best performance range. Thus the objective of spatialdata quality analysis in oil spill modelling is not only to improve theaccuracy and reliability of the results, but also to understand what levelsof accuracy and reliability can be achieved.

TopographicData

SpatialDiscretisation

TemporalDiscretisation

Hydrodynamic

Model

Tidal current

Hydrodynamic Modelling

Envir

OilF

I

Mean Current

Wind Data

Wave Data

Other

onmental Data

'roperties Data

-

— »

t

u

Oil

Spill

Gridding

^

, ^

Oil Spill

Trajectory

Model

Oil Spill

Fate Models

^Outp

*

Trajectory and Fate Modelling

Figure 1. Structure of oil spill modelling

3 GIS and spatial data quality of oil spill

modelling

GIS is a spatial information technology which includes spatial datamanagement, analysis, handling and visualisation. There are a range ofsoftware packages on the market. GIS has enhanced ability to solve awide range of spatial problems when coupled with environmentalmodelling. The integration approaches for environmental modelling withGIS are separated to two general categories: loosely coupled and tightly

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 4: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

56 Oil & Hydrocarbon Spills, Modelling, Analysis & Control

coupled^. GIS can be used for mapping, model development and manydata related issues using in-built functionality or in conjunction with off-the-shelf tools. In oil spill modelling, GIS has been used for integratingand pre-processing spatial data inputs, analysing model outputs and forsupporting decision-making and impact assessment. This paper pointsout that GIS has considerable potential for studying the effects of spatial

data quality in oil spill modelling. Figure 2 presents a 'hierarchy ofneeds' for modelling error using GIS . Attainment of successivelyhigher-order needs is dependent on the satisfaction of the needs at all

lower levels in the hierarchy. This has been the approach in this study.

I strategies for error reduction |

strategies for error management

error propagation modelling

error detection and measurement

error source identification - - - »

Figure 2. A hierarchy of needs for modelling spatial data quality

3.1 GIS and spatial data quality in hydrodynamic modelling

Both data input inaccuracies and model performance uncertainty exist inhydrodynamic modelling. Input inaccuracies concern both bathymetricdata and tidal data. Modelling uncertainties are occur in griddevelopment and numerical computation of tidal current. Somemodelling processes will have different contributions to uncertainty,such as sampling, interpolation, smoothing and reshaping. Variousinaccuracies can be measured using traditional and spatial statistics suchas mean, standard deviation, spatial distribution and autocorrelation.

Furthermore, visualisation of the spatial distribution of error andconfidence levels could be provided. With such visualisation, it would bepossible to communicate the data quality between the researchers andusers from different disciplines.

Error propagation analysis offers much more information than errormeasurement. There are error propagation techniques in GIS such asMonte Carlo method, which are suitable for the complex dynamicmodels utilised in hydrodynamic modelling. Using error propagation, thereliability of tidal simulation could be derived and users could thenunderstand and apply the simulation results in the face of different levelsof data and modelling uncertainty. Tests on the contribution of error

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 5: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

Oil & Hydrocarbon Spills, Modelling, Analysis & Control 57

from each variable and how they propagate through the model would

greatly help error reduction and management. More effort could then be

directed to the data and modelling steps that have the largest error

contribution by using, for example, different interpolation algorithms orincreasing sampling density. Sensitivity analysis could determine the

limitations of various modelling techniques and assure the correctoperation of models in different conditions, for example, comparing

Finite Deferential and Finite Element methods of hydrodynamicmodelling. Finally, cost-benefit data resolution and required accuracycould be studied using error propagation in advance of a project.

At present, few of the more popular proprietary CIS have tools forhandling error and uncertainty. If some error propagation tools can becustomised for oil spill modelling within the GIS environment, bothresearch and application would benefit. Although various oil spill modelshave certain differences in principle and data format, the properties of

spatial data in a GIS environment are consistent and the rich resources ofspatial data handling in GIS could be made use of.

3.2 Spatial data quality in trajectory and fate modelling

The inputs to the oil spill trajectory model include the tidal current fieldand some related environmental data. The tidal current field is simulatedby hydrodynamic modelling, whilst the environmental data also involvesampling, interpolation and discretisation. The inaccuracies oruncertainties resulting from the hydrodynamic modelling will becascaded to the trajectory modelling through tidal current field data.Thus the trajectory simulation will be affected by any inaccuraciesarising from hydrodynamic modelling and trajectory modelling.Furthermore, fate models depend on the trajectory simulation to someextent and hence all of the accumulated errors in the modelling will bepropagated to the final output. It is a long way for any errors to propagatefrom hydrodynamic modelling, via trajectory modelling, to fatemodelling. In doing so, slight inaccuracies may propagate to form largeerrors in the final output. Therefore error analysis and management arevery important in order to be aware of the reliability of final results.

4 A preliminary case study

4.1 Objectives

Initial research has been carried out using oil spill modelling for a

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 6: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

58 Oil & Hydrocarbon Spills, Modelling, Analysis & Control

contingency plan of Qing Dao Port, Jiao Zhou Bay, north China . As an

oil export harbour, Jiao Zhou Bay is an important economic area ofShandong Province but environmentally sensitive. The length of theshoreline is 163 km, and the surface area of the bay is 423 km^. Thestudy area includes Jiao Zhou Bay and the contiguous water. Figure 3shows the bathymetry of the study area. The objective of this case studyhas been to improve the sampling quality of bathymetric data of JiaoZhou Bay. Error analysis methods were used to test the effect ofsampling density on tidal simulation. The results were balanced againstthe cost-benefit and resulted in a suggested maximum sampling interval.

This case study has shown that spatial data quality analysis is practical

and necessary in oil spill modelling.

Figure 3. Bathymetric chart of Jiao Zhou Bay

4.2 Methodology

During the analysis, some factors such as the boundary conditions andtidal data were held fixed. In addition the resampling process, reshapingprocess and related interpolation were not included as they mightinterfere with the main objectives. The hydrodynamic modelling errorscould be ignored through smoothing allowing the effects of bathymetricsampling interval on tidal simulation to be tested.

Only M2 tidal constituent was considered as it is the dominantconstituent in Jiao Zhou Bay. A developed tidal current field which hadbeen previously calibrated for the project was taken as the true tidalcurrent. Uniform sampling was carried out and the sampling interval wasvaried from 1.5 km to 4.0 km. Thus 26 sets of sampling data were

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 7: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

Oil & Hydrocarbon Spills, Modelling, Analysis & Control 59

obtained. Every sample data set had 109 common points which were

retained for analysing and comparing the simulation results. Finite

element tidal model (Tide2D) was applied to compute the tidal current.The finite element triangular meshes were created based on the differentsample data sets and the numerical computation for tidal simulation wascarried out for each triangular mesh. The simulation results give theamplitudes and initial phases of two orthogonal current components.Each result was compared with the original set of 109 common pointsand an series of residual errors were calculated.

Figure 4 illustrates the error distribution for two sample data sets.

Figure 4(a) shows the error in amplitude of the northward currentcomponent with 4.0 km sampling interval whilst Figure 4(b) shows theerror in amplitude of the eastward current component for the samesampling interval. The skewness of the error in Figure 4(a) and 4(b) is1.52 and 1.16 respectively. Some larger errors occur at some locationsalong the boundary because the shape of the triangles are too elongated.

The distributions of these errors were regarded as symmetrical and themean and standard deviation of the errors were computed for each of thesample data sets.

200 - 150"-100\050" 000 .050 .100 .150" .200" .250" .300"-175-125-075-025 025 .075 125 .175 225 275 325 -150 -100 -050 000 050 100 .150 200 250error2

Figure 4. Error distribution for two sample data sets

There are some interesting fluctuations in the error statistics whereinstabilities were caused by the sampling uncertainty, the irregular shapeof triangles or inconsistencies in the computation time step. The trend inerrors with change in sampling interval were smoothed as shown inFigure 5. As can seen, there are still some irregular fluctuations as aconsequence of the above mentioned factors.

4.3 Results

Figure 5 (a)-(d) show the mean error of na, ug, ra, vg, while Figure 5 (e)-(h) show the standard deviation of the errors in ua, ug, va, vg. Here, ua is

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 8: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

60 Oil & Hydrocarbon Spills, Modelling, Analysis & Control

008006004002

0000

Mean

error

of va

(rrts)

j 8

8

8 %

-"""

Sampling interval (km)

(a)

x-'"'--x ^- —

_ _ /

| 012

| .006

0000

Mean

error

of vg

(radia

n)_1 2 S 8

S

x'^/

Sampling interval (km)

(b)

/'

/''

Sampling interval (km)

(c)

Sampling interval (km)

(d)

gI «•| .01.

9 31 33 35 3.7 39Sampling interval (km)

(e)

S>"6

15 1.7 1.9 2.1 23Sampling interval (km)

(f)

Sampling interval (km)

(g)

Sarrpling interval (km)

00

Figure 5. Smoothed changes in current simulation error

for sampling intervals

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 9: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

Oil & Hydrocarbon Spills, Modelling, Analysis & Control 61

the amplitude of the northward current component, ug is the initial phase

of the northward current component, va is the amplitude of the eastward

current component and vg is the initial phase of the eastward component.

Because the mean errors are at or near zero and the standard deviations

of errors are relatively higher, the standard deviation of the error is the

more significant parameter to be studied. It means that the prevailingeffect of the sampling interval is on uncertainty, not on system error.

From Figure 5, it can be seen that the general trend of the error istowards small and stable for smaller sampling intervals. Below a 2.1 kmsampling interval, the standard deviations of errors in both ua and ug areinclined to be stable. Meanwhile the standard deviations of errors in vaand vg decrease gradually in a stable, progressive manner. Therefore aninitial maximum sampling interval can be set at 2.1 km. However, the

break in mean error for ua took place at a 2.2 km sampling intervalwhilst for tig it was at a 1.7 km sampling interval. Nevertheless the meanerror for ug at a 2.1 km sampling interval is still small, and the initial

maximum sampling interval of 2.1 km can still be accepted. There weresome fluctuations in the mean error of va and vg because ofhydrodynamic modelling errors. If such fluctuations are considered, thebest sampling interval range should be 2.2 to 2.4 km for va mean error,and 2.0 to 3.0 km for vg mean error. The initial maximum samplinginterval of 2.1 km is around the lower limit of these two ranges. On theother hand, if these fluctuations are ignored, the maximum sampling

interval could be 2.4 km using va mean errors. In this situation asampling interval of 2.1 km would be acceptable.

To sum up , the analysis of errors has indicated appropriate values ofmaximum sampling interval. In this project the maximum recommendedsampling interval should be 2.1 km and hence an interval near 2.0 kmcould be regarded as the most effective from a cost-benefitconsideration. Obviously, this is an initial research result based on aspecific water body and model with many factors fixed. More advancedtechniques using GIS could be applied to improve the test procedure, andmore variables will be considered in forthcoming research.

From a project perspective, increases in bathymetric sampling dataare time consuming and costly. They also result to more complexcomputation which could cause instability in modelling (for instance,some models need smoothed data in order to avoid divergence). Clearlyan optimal bathymetric sampling is desirable. The analysis presentedhere also indicates that reference specifications can be provided forbathymetric data sampling in oil spill modelling. With an improvementin the simulation, time, manpower and cost could be saved.

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541

Page 10: Abstract - WIT Press...Figure 1. Structure of oil spill modelling 3 GIS and spatial data quality of oil spill modelling GIS is a spatial information technology which includes spatial

62 Oil & Hydrocarbon Spills, Modelling, Analysis & Control

5 Conclusion and further research

As discussed in this paper, there are a number of important spatial dataquality problems in the oil spill modelling process. It is important toassess, model, manage and visualise these spatial data quality problems.

GIS technology has considerable potential to assist in this area. Theinitial research, centred on the case study, was carried out forbathymetric sampling quality improvement as one of the main spatialdata quality issues in oil spill modelling. The results indicate that aquantitative analysis of data quality can be achieved to make modellingmore effective and that data quality performance specifications can be

provided to benefit modeller and user.Further research will be carried out. As it is difficult to separate

relevant variables in real data sets, a synthetic modelling method will beapplied to the bathymetric sampling to achieve more precise results. Theresearch will also be extended to include analysis of uncertainty in there-sampling process and the interpolation methods used in differentstages of oil spill modelling. Research on other spatial data quality issuessummarised in this paper will be carried out, particularly error

propagation. GIS will provided the tools for a range of advanced spatialanalysis and handling techniques necessary for such work.

References

[1] Chinese National Marine Bureau, Marine Science Bulletin 11(3),P.R.China, 1992.

[2] Goodchild, M. et al., GIS and environmental modelling: progressand research issues, GIS World, 1996.

[3] Hunter, G. & Goodchild, M., Managing uncertainty in spatialdatabases, L7MSW JowrW 5(2), pp. 55-62, 1993.

[4] Heuvelink, G.B.M., Error propagation in environmental modellingwith GIS, Taylor & Francis, London, 1998.

[5] Karimi, II. & Houston, B., Evaluating strategies for integratingenvironmental models with GIS, Computers, Environment and(TrW? 6y,yffm 20(6), pp. 413-425, 1996.

[6] Lehr, W. et al., The next generation in oil weathering modeling,7997 /MfemafmW O/V 6)9/'// Cow/brewcc, pp. 986-987, 1997.

[7] Veregin, H., Error modeling for the map overlay operation, inGoodchild, M. & Gopal, S. Accuracy of spatial databases, Taylor &Francis, London, 1989.

Transactions on Ecology and the Environment vol 20, © 1998 WIT Press, www.witpress.com, ISSN 1743-3541