This paper describes the use of Esri software to process airborne SAR data for a GIS flood model of a major UK river catchment. The model is used for insurance and reinsurance business purposes. The paper describes the background to the project, introduces the SAR product, and describes the ARC GRID and ARC TIN processes and the ERDAS techniques involved. The presentation will include examples of the ArcView GIS front end produced for user operation. The paper concludes with an analysis of the experimental project and an analysis of the suitability of Esri software and SAR data for insurance modelling.
1 Introduction
This paper describes the development of an innovative flood risk assessment model for the River Thames, the most important river in the UK. The model was built using ArcInfo and ERDAS tools and is based upon airborne SAR data. The model was initially aimed at the insurance and reinsurance market in the UK but the techniques are applicable to any organisation which is exposed to flood risk. The project is entirely commercial, completely funded by the needs of the UK insurance industry.
1.1 Development partners
The project was carried out by the consultancy division of Willis, a leading world-wide risk practice. Willis uses a blend of hazard and GIS expertise allied to actuarial and financial engineering to provide risk consultancy to the insurance, reinsurance and financial markets world-wide. The SAR data was supplied by Willis�s partner, Intermap Technologies. Intermap provides intelligent solutions to major companies, governments and development agencies around the world in the extraction of information from images. Intermap provides products and services using GIS and image processing tools based on image data from optical and radar satellites, interferometric airborne radar (STAR-3i) and aerial photography. Orthorectified images (ORIs), Digital Elevation Models (DEMs) and Thematic Maps are Intermap's primary outputs. These products are derived from remote sensing sources which include airborne interferometric radar (known as STAR-3i), RADARSAT and other satellites and air photo surveys.
1.2 River Thames
The River Thames is the UK�s most important river. The river is 215 miles long and has a discharge rate of 680 cumecs for a 50 year flood event at Teddington lock, the point where the river changes from tidal to non-tidal. The river effectively ends at the Thames barrier at Woolwich; this was completed in 1982 to prevent flooding from coastal surge with a design criteria of the 1000-year event. The catchment covers an area of approximately 14,000 square kilometres. There are 1.4 million residential and 100,000 commercial properties in the immediate vicinity of the river, with a population of approximately 3 million people. For most of its length the river can be described as semi-managed. Although the non-tidal section does not feature banks or levees, management is carried out using alleviation techniques such as diversion channels and ponding reservoirs.
2 Development
The background to the development can be split into three areas, the physical and insurance conditions that drove the project commercially, the data innovations that drove the project technically and the tools and processes that determined the methodology.
2.1 Physical and insurance conditions
The river Thames is a flood prone river. It has suffered serious flooding 20 times during the last 200 years. Major events occurred in 1894 and 1947, both of which caused severe flooding, particularly in London. Minor floods occur more frequently, flooding smaller areas, and areas of lower economic value. In the recent past, there has been considerable property development on the flood plains, particularly �out-of-town� commercial development, and new residential development in response to the UK government�s call for 4 million new homes to be built during the next 10 years. Property values have also increased greatly during the past 15 years, by 1.5 times the annual inflation rate adding to the value of property at risk.
Flood insurance is normally provided as part of buildings and contents insurance in the UK. For residential property, approximately 70% of property has buildings cover, and 85% has contents cover. Commercial flood insurance is similarly provided as part of normal cover. Flood insurance in the UK covers all types of flood.
The consequence of the increasing value of property at risk, and the insurance conditions, has resulted in very large exposures to potential flood risk for insurance companies. This exposure is also of interest to reinsurance companies and other organisations interested in risk. Recent estimates of insurance loss from a 100-year return period event for the Thames have ranged from �5billion to �20 billion. Willis has been asked to study the feasibility of providing loss estimates for Thames flood for insurance companies in the past, but due to data limitations had not been able to identify a suitable methodology which combined regional coverage, detailed property location and a feasible flood risk area delineation technique.
2.2 Innovative technology
The main data set used in flood risk assessment is the Digital Elevation Model (DEM). There have been a number of elevation models available in the UK, principally from the national mapping organisation, the Ordnance Survey. These have been used in the past for coastal flood models, but neither their accuracy nor their resolution were suitable for riverine models. Other data sets are becoming available, including Lidar and SAR. Satellite SAR from ERS1 & 2 and Radarsat does not have sufficient accuracy or resolution and requires data from multiple passes, leading to problems with coherence, whilst Lidar has very high collection and processing costs, and is normally collected at a much higher resolution than is required for regional models. Airborne SAR has proved to be the ideal source for regional flood modelling. The resolution of 5 or 10 metres, the accuracy of up to 2 meters* and the addition of the 2.5 metre resolution SAR image, provide a data set which can be handled with reasonable ease, and can provide sufficient vertical and horizontal detail for most requirements. The use of airborne SAR for a commercial, large area flood risk assessment has not previously been carried out.
2.2.1 Intermap STAR-3i
The STAR-3i system is an interferometric radar system mounted on a Learjet36. The system generates DEM�s and Ortho-Rectified Images (ORI�s) simultaneously. The system consists of two X-band radar antennae mounted on the Learjet. Data collection from the two antennae occurs simultaneously. The set of acquired data is interfered by a digital correlation process to extract terrain height data, which is used to geometrically correct the radar image. STAR-3i uses post-processed Differential Global Positioning Systems (DGPS) data, together with on-board laser-based inertial measurement data, to attain highly accurate positioning control. Precise terrain height and positioning data are enhanced by careful calibration of the baseline (the distance between the two antennae). Due to the accuracy of the positioning information and the careful baseline calibration, no in-scene control points are required. The only restriction is that a ground-based GPS receiver must be located within 200km of the data collection site for DGPS processing. In the typical collection mode, the system is flown at a height of 12,000m and acquires a 10km wide swath of 2.5m resolution radar data. The system has been designed to collect <2m* vertical accuracy DEMs at a rate of 100 km�/min.
Ortho Rectified Images (ORI�s) are images that have had all distortions caused by platform instability, radial distortion (airphoto) and terrain displacement removed. A DEM is used to remove the distortions due to terrain and elevation. Correcting for these distortions results in the imagery becoming a true scale representation of the ground, which can be used in a computerised system for measurements of length, area, and azimuth.
The Airborne Derived Global Terrain™ data which was used on this project was the GT2 product from Intermap. The specifications are as follows:
Digital Elevation Model (DEM) Data:
vertical RMS 2m*
(vertical RMS 1m, optional)
posting every 5m
available in 7.5' tiles
first surface elevation measured
DEM data in ASCII, BIL or GEOTIFF, 32 bit format
Orthorectified Image (ORI) Data:
image pixel size 2.5m
horizontal RMS 2.5m
available in 7.5' tiles
image data is 8 bit
image data in BIL, GEOTIFF or TIFF format
* The accuracy of the data, which has been used on this project, is likely to have been better than that quoted. SAR gives higher accuracy on flat or homogenous surfaces, such as flood plains. An extreme example of relative vertical precision has been obtained by obtaining multiple readings from the surface of a large reservoir, the results differing by not more than ten centimetres. The University of Stuttgart has produced an independent report on the accuracy of STAR-3i; see references, Kleusberg A. and Klaedtke H. G.
2.2.2 Flood data
The other essential data for the flood risk assessment is information regarding the flood itself. Conventional hydrological modeling is normally both data intensive and processing intensive. The complex multiple inputs interact in a way which provides a suitable result for local areas. However, application to areas larger than a single reach is problematic. The process, which has been developed for regional flood risk assessment is simpler than a full hydrological model, but is suitable for the purpose. Hydrological modelling is useful for local studies with good quality, site specific data; the new flood risk assessment process is suitable for regional areas with minimal data. The data used in this project were raw flood heights for events with various return periods, produced by HR Wallingford, a leading hydraulic engineering consultant, specialising in river and coastal flooding in the UK and overseas.
2.3 Tools and processes
The processes used on this project were closely linked with the tools selected. If advanced hydrological modelling had been feasible, tools such as Mike11 (from Danish Hydraulics Institute) or ISIS (from HR Wallingford) would have been used in addition to the GIS and image processing tools.
2.3.1 Tools
The primary data processing and editing tools for this project were ArcInfo and ERDAS Imagine. ArcInfo 7.2.1 runs on a Sun Enterprise 450 and accesses a DEC Raid5 storage works for data storage. The GRID and TIN modules were used extensively in this project.
ERDAS Imagine version 8.3.1 runs on a Compaq SP700 PC, which is networked to the UNIX equipment. The Essentials and Advantage modules were used. ArcView 3.1 was also used in the project, running alongside ERDAS. The extensions Spatial Analyst, Cad Reader, 3D Analyst and Geoprocessing were particularly useful.
2.3.2 Processes
The Thames survey flown by Intermap in August 1998 produced 120 tiles of DEM data, each 7.5� x 7.5�, in area about 120 square kilometres, consisting of approximately 600Mb of data each, in an ASCII format. 34 tiles were utilised in the flood risk assessment project.
The initial processing carried out on the DEM data involved producing a physically limited and horizontally corrected DEM, which could be used for the flood processes. The generalised process for producing the DEM is as follows:
Generalised process for producing DEM
Process for producing landuse map from ORI
The next stage in the process was to take the raw flood levels, which were produced by HR Wallingford, and produce a flood surface. The points were located upstream of weirs and locks, to produce a representation of the flood that was relatively unaffected by the structures. There were 55 points used in the tidal and non-tidal Thames. The flood heights at each point were extended perpendicular to the river, a TIN was built using these points, and the subsequent surface was used as an estimate of the slope of the river and the flood. There were two processes used in the creation of the flood surfaces:
a) Intersecting the flood surface with the DEM was the first step towards producing flood maps. Polygons non-adjacent to the river, i.e. areas which the terrain model indicates are not logically flooded at a given water depth, were removed. This results in a logically correct flood risk envelope.
b) A basic propagation model was designed to propagate the water across the DEM surface and approximate the actual progress of water. The flood was then propagated out from the river in a series of iterations until the limit of the surface likely to be flooded was reached. As with the previous method (a) this took into account embankments and solid features (trees and other �non-solid� features were removed by filter). Both processes were carried out using ArcInfo GRID tools and processes. A visual comparison of techniques a) and b) showed that both methods provided similar flood extents. This would suggest that the results obtained from a) provide a good approximation of risk.
3. Risk analysis
Flood modelling allows the delineation of areas at risk, as well as indicative flood depths to be expected at any location within the area affected. This data is then used with relevant locations in order to assess risk to property. Property locations for the UK, at a unit postcode level (see table), are intersected with the flood depth data using a suitable interpolation algorithm, which results in a measure of flood intensity (depth) for each location.
Post codes in the UK
|
Postcode |
Number in the UK |
Approximate number of properties |
|
Outcode |
||
|
KT (Area) |
123 |
200,000 |
|
KT4 (District) |
2,807 |
9,000 |
|
Incode |
||
|
KT4 8 (Sector) |
9,114 |
2,500 |
|
KT4 8X (Sub Sector) |
106,000 |
220 |
|
KT4 8XW (Unit) |
1,600,000 |
15 |
4. Loss estimation
The object of the project is to provide an estimate of the loss to an insurance company portfolio of property, from flood events of various return periods. The process is carried out using a proprietary data warehouse system, built using an Oracle RDBMS, and called the Integrated Catastrophe Modelling Platform (ICMPTM). This process takes the inputs of flood intensity, client portfolios, and suitable loss curves, and outputs an estimated loss, for each location, and/or for the whole portfolio. The initial use of the outputs from the system is to allow insurance companies to accurately calculate their reinsurance purchase requirements. The data can also be used by insurers as a rating tool for properties and to provide hazard intensity for potential development sites.
5 User interface
An important aim of this project was to provide the results in a user friendly form to clients. A customised ArcView 3.1 project was developed which allows easy visualisation of the flood maps in conjunction with associated demographic and financial information. A startup script loads the relevant shapefiles and grids and assigns legends to them, based on a pathname defined by an environment variable. A main view window and a 'control' view are opened automatically and resized according to the screen resolution. The control view can be used by the client to navigate around the flood plain. Mapping of the relevant themes is done via a modeless dialog. This allows the client to toggle flood map themes on/off, search by geographic locators or grid coordinate, and to interrogate the flood depth and geo-demographic information. A toggle to full functionality ArcView is also provided. The data (which takes up approximately 350 Mb) can be accessed by CD-Rom with the apr file loaded onto the computer. The project is designed to operate on a machine with either 800 by 600 or 1024 by 780 screen sizes, with at least 16 million colours.
6 Suitability of tools and data
6.1 Suitability of tools
The geoidal correction of the DEM was found to be beyond the capabilities of the GIS software used in-house. This process was outsourced to University College London in order for specialist software and expertise to be applied.
There were a number of tasks which were initially attempted using one tool, and which proved easier using another, although generally the tools proved suitable for the task. Should we have decided to carry out detailed hydrodynamic investigations, specialist tools would have been required, however the Esri and ERDAS tools chosen were very suitable for the methodology which has been detailed. Further use could have been made of ArcView Spatial Analyst, however the available processing resource available on UNIX was greater than that on PC so ArcInfo was used. ArcInfo GRID was successfully used for propagation across the flood plain, by using multiple iterations of the flooding process. This process although simple, used GRID to its full advantage and does not have serious resource impacts.
6.2 Suitability of SAR
The airborne SAR data was ideal for the task of regional flood risk assessment. The combination of resolution and accuracy, delivered a final product which is far superior to anything else currently available. Initial problems with data sizes, particularly with the DEM data, were overcome by �trimming� the area of interest to the flood plain only and careful disk space management. The SAR data provides a unique combination of elevation models and ORI image, which allows land use to be included in the estimation of flood propagation. This is a major advantage over other data sets and allows a great refinement of the flood estimation process, with little additional processing.
7 Conclusion
The flood risk assessment is now available to UK insurance companies, and enables them to estimate their exposure to flood risk on the Thames from a number of different return period events, both at a portfolio level and at a geocode level. The project was carried out over 12 months elapsed time at the Willis offices in London.
There are a number of potential refinements, which will further increase the value of the flood risk assessment project. The proposed increase in vertical accuracy which Intermap are planning for 2000 will greatly enhance the flood depth analysis. It is currently planned to improve vertical accuracy to 0.3 metres. The propagation of the flood can be enhanced by the addition of improved volumetrics for historic floods. This may prove problematic for the Thames, but will prove more successful for other rivers which have defined defences (levees) and for which a realistic limit can bet set for available floodwater volume.
The project has been extremely successful, the methodology is currently being assessed by local and national authorities in the UK and elsewhere, and is being proposed for use by the insurance industry in Europe and the Far East.
Acknowledgements
This work was carried out in partnership with Intermap Technologies Inc, and HR Wallingford. The author would particularly like to acknowledge the assistance provided by David Ramsbottom of HRW and Hugh Mackay of Intermap, and of colleagues at Willis.
Integrated Catastrophe Modelling Platform is a registered trademark.
References
Consuguera D, Joerin F and Vitalini F, "Flood Delineation and Impact Assessment in Agricultural Land using GIS Technology", pp. 177- 198 in Carrara A. and Guzzetti F. (ed.), "Geographical Information Systems in Assessing Natural Hazards", Kluwer Academic Publishers, 1993.
DETR Water Unit, "Surface Water: United Kingdom 1977-1980", Her Majesty�s Stationery Office, 1983.
Dingman S. A., "Physical Hydrology", Prentice Hall, 1993.
Dunsmore, S. J., "River Thames Flood Hydrology Design Curves", Journal of the Chartered Institution fo Watre and Environmental Management, Vol. 11, No.1, pp. 67-71, 1997.
Giammarco P. and Todini E., "A Control Volume Finite Element Method for the Solution of 2-D Overland Flow Problems", pp. 82-101 in Proceedings of the Speciality Conference, ENEL-DSR-CRIS, Giammarco P., Todini E., Molinaro P. and Natale L. (ed.) "Modellinf of Flood Propagation over initially dry areas, Milan, Italy, 1994.
Giammarco P. and Todini E., "Combining a 2-D Flood Plain Model with GIS for Flood Delineation and Damage Assessment", pp. 171-185 in Proceedings of the Speciality Conference, ENEL-DSR-CRIS, Giammarco P., Todini E., Molinaro P. and Natale L. (ed.) "Modellinf of Flood Propagation over initially dry areas, Milan, Italy, 1994.
HR Wallingford, "Hydraulic Factors in Flood Risk Mapping", Report EX 3574, January 1997.
HR Wallingford, "Thames Flood Levels", Report EX 4030, April 1999.
Hubbert G. D. and McInnes K. L., "A Storm Surge Inundation Model and Impact Studies", Journal of Coastal Research, Vol. 15, No. 1, 1999.
Ingle Smith D. and Stopp P., "The river basin", Cambridge University Press, 1978.
Institute of Hydrology, "Flood Risk Map for England and Wales", Report 130, October 1996.
J&Y Consultant, "Flood Hazard Prevention Systems in China", July 1999.
Kleusberg A. and Klaedtke H. G., " Accuracy Assessment for the Star-3i derived DHM in Baden-Württemberg", Universität Stuttgart, Institut für Navigation, May 1998.
Richards K., "Rivers: Form and Process in Alluvial Channels", Methuen & Co., 1982.
Author information
Richard SandersWillis
Ten Trinity Square
London
EC3P 3AX
UNITED KINGDOM
Telephone: +44 (0) 20 7488 8952
Fax: +44 (0) 20 7481 7193
Email: sandersr@willis.com