Congress Resources: Papers, posters and presentations

< Return to abstract list

Rapid Flood Inundation Modelling For Emergency Response

Congress: 2015
Author(s):
AbstractIntroduction Regional scale flood inundation has significant impacts wherever it occurs. In developed countries, the financial costs can be astronomical, yet the physical impacts tend to be short-lived and fatalities are relatively uncommon. In contrast, regional scale events elsewhere often have more significant and longer-lasting physical impacts. A good example are the widespread floods currently spreading across Pakistan, which have already affected 2.3 million people, caused 312 fatalities and devastated 6900km2 of crops. Whilst it is clear that little can be done about the drivers behind regional scale events, and resource constraints often limit defence options, more accurate and timely flood predictions could increase preparedness levels and help minimise flood impacts. However, simulating regional scale flood inundation is problematic, both in terms of simulation accuracy and computational run times, e.g. using a fine resolution mesh is computationally demanding whilst coarser meshes can reduce simulation accuracy. This paper details the use of a computationally efficient modelling approach to the simulation of regional scale flood inundation, and in so doing illustrates that such approaches can provide the type of near-real-time flood prediction vital for emergency response purposes. Methodology The RFSM-EDA model [1] was applied to a specific regional scale flood event, namely the northern avulsion element of the 2010 River Indus flooding in Pakistan which inundated ~8,000 km2 of agricultural land to depths of 1--3 m [2]. The computational approach underpinning RFSM-EDA is specifically designed to overcome current modelling limitations for large scale and probabilistic applications. Irregular grids are automatically defined around 'topographic depressions' such that any high points are always fully represented on the element edges. In the past decade, this approach has been incorporated into a number of models of increasing sophistication, with the most recent version (RFSM-EDA) using a diffusive-type scheme with adaptive time stepping, which provides both improved stability and computational efficiency. The extent of the Indus flooding in 2010 meant that there was a wealth of, primarily remote sensed, data, for simulation purposes, including: Topographic data for the sites of interest were available from two sources, namely SRTM:Shuttle Radar Topography Mission [3] and ASTER:Advances Spaceborne Thermal Emission and Reflection Radiometer [4]. Each source has its advantages and disadvantages (e.g. SRTM data is only freely available at 3" (3 arc-second) resolution, whilst the 1" ASTER data is generally more "noisy"), and so both sets of data were trialed with RFSM-EDA. Similarly a number of different data "sampling" approaches were trialed to determine the optimum balance between maximising resolution and minimising the impact of "noisy" (erroneous) topographic data points. Flood extent imagery was available from the MODIS suite of instruments on board NASA's Aqua and Terra satellites [5]. As the standard false colour composites were not sufficient for an accurate GIS flood extent, a NDVI (Normalised Difference Vegetation Index) approach was used to "classify and clean" the MODIS data. Flood Hydrographs were available for the River Indus from the Dartmouth Flood Observatory using AMSR-E: Advanced Microwave Scanning Radiometer data. However, as there was some concern with the accuracy of the AMSR-E generated data, an alternative flood hydrograph was developed using three key initial flow parameters (initial distance travelled, flow width and flow depth) and qualitative observations from satellite imagery. Results and Discussion Initial results using both the ASTER and SRTM topographic data were disappointing, with Flood Area Index (FAI) values indicating that the model output only matched the observed data in only 20-30% of instances (depending on time). Further model runs, using "averaged" topographic data led to significant improvements in simulation accuracy, resulting in FAI's over 40% when the underlying 1" (ASTER) and 3" (SRTM) data was resampled to a 20" resolution. Importantly, the RFSM-EDA results with resampled 20" topographic data was found to be of comparable accuracy to the results from using industry standard (full SWE) 2D models, but at a fraction of the computational cost (typically 2.5%). Irrespective of modeling approach, the deficiencies in the topographic data used to drive the models were the primary reasons for the differences between the observed and simulated flood extents. These deficiencies ranged from poor representation of vital infrastructure (e.g. extensive irrigation canal network), effectively no representation of urban areas and erroneously low or high topographic data points. In all cases these deficiencies resulted in over-estimation of flood extent throughout the modelled domain, and it was this over-estimation that led to the relatively low FAI values. Accordingly, the topographic data was manually "corrected" to incorporate some of the key infrastructure and urban areas, resulting in increased FAI values up to 60%. The impact of the assumed flood hydrographs and floodplain Manning's n values was also assessed. As expected for a topographically driven and simulated system, realistic changes to the flood hydrograph did not have any significant impact on simulation accuracy. And whilst a global decrease in Manning's n resulted in reduced simulated flood extent, and hence marginal improvements in FAI values, it was again not overly significant. Similarly, the use of localized Manning's n values was shown to be evident, though not particularly significant. Conclusion The overall results indicate that the type of rapid flood inundation model represented by RFSM-EDA can provide comparable accuracy relative to conventional full SWE 2D models, at a fraction of the computational runtime. This indicates that, given appropriate topographic data, RFSM-EDA is suitable for regional scale flood inundation modeling, and thus has clear application to real-time flood prediction, as well as national flood risk assessment. However, the results do again highlight the importance of both accurate topographic data and appropriate DEM sampling techniques. Recommendations for future work include the use of alternative topographical data and follow-up sensitivity analysis, as well as refinements to the RFSM-EDA pre-processing to help deal with poor topographic data inputs. Introduction Regional scale flood inundation has significant impacts wherever it occurs. In developed countries, the financial costs can be astronomical, yet the physical impacts tend to be short-lived and fatalities are relatively uncommon. In contrast, regional scale events elsewhere often have more significant and longer-lasting physical impacts. A good example are the widespread floods currently spreading across Pakistan, which have already affected 2.3 million people, caused 312 fatalities and devastated 6900km2 of crops. Whilst it is clear that little can be done about the drivers behind regional scale events, and resource constraints often limit defence options, more accurate and timely flood predictions could increase preparedness levels and help minimise flood impacts. However, simulating regional scale flood inundation is problematic, both in terms of simulation accuracy and computational run times, e.g. using a fine resolution mesh is computationally demanding whilst coarser meshes can reduce simulation accuracy. This paper details the use of a computationally efficient modelling approach to the simulation of regional scale flood inundation, and in so doing illustrates that such approaches can provide the type of near-real-time flood prediction vital for emergency response purposes. Methodology The RFSM-EDA model [1] was applied to a specific regional scale flood event, namely the northern avulsion element of the 2010 River Indus flooding in Pakistan which inundated ~8,000 km2 of agricultural land to depths of 1–3 m [2]. The computational approach underpinning RFSM-EDA is specifically designed to overcome current modelling limitations for large scale and probabilistic applications. Irregular grids are automatically defined around ‘topographic depressions’ such that any high points are always fully represented on the element edges. In the past decade, this approach has been incorporated into a number of models of increasing sophistication, with the most recent version (RFSM-EDA) using a diffusive-type scheme with adaptive time stepping, which provides both improved stability and computational efficiency. The extent of the Indus flooding in 2010 meant that there was a wealth of, primarily remote sensed, data, for simulation purposes, including: Topographic data for the sites of interest were available from two sources, namely SRTM:Shuttle Radar Topography Mission [3] and ASTER:Advances Spaceborne Thermal Emission and Reflection Radiometer [4]. Each source has its advantages and disadvantages (e.g. SRTM data is only freely available at 3” (3 arc-second) resolution, whilst the 1” ASTER data is generally more “noisy”), and so both sets of data were trialed with RFSM-EDA. Similarly a number of different data “sampling” approaches were trialed to determine the optimum balance between maximising resolution and minimising the impact of “noisy” (erroneous) topographic data points. Flood extent imagery was available from the MODIS suite of instruments on board NASA’s Aqua and Terra satellites [5]. As the standard false colour composites were not sufficient for an accurate GIS flood extent, a NDVI (Normalised Difference Vegetation Index) approach was used to “classify and clean” the MODIS data. Flood Hydrographs were available for the River Indus from the Dartmouth Flood Observatory using AMSR-E: Advanced Microwave Scanning Radiometer data. However, as there was some concern with the accuracy of the AMSR-E generated data, an alternative flood hydrograph was developed using three key initial flow parameters (initial distance travelled, flow width and flow depth) and qualitative observations from satellite imagery. Results and Discussion Initial results using both the ASTER and SRTM topographic data were disappointing, with Flood Area Index (FAI) values indicating that the model output only matched the observed data in only 20-30% of instances (depending on time). Further model runs, using “averaged” topographic data led to significant improvements in simulation accuracy, resulting in FAI’s over 40% when the underlying 1” (ASTER) and 3” (SRTM) data was resampled to a 20” resolution. Importantly, the RFSM-EDA results with resampled 20” topographic data was found to be of comparable accuracy to the results from using industry standard (full SWE) 2D models, but at a fraction of the computational cost (typically 2.5%). Irrespective of modeling approach, the deficiencies in the topographic data used to drive the models were the primary reasons for the differences between the observed and simulated flood extents. These deficiencies ranged from poor representation of vital infrastructure (e.g. extensive irrigation canal network), effectively no representation of urban areas and erroneously low or high topographic data points. In all cases these deficiencies resulted in over-estimation of flood extent throughout the modelled domain, and it was this over-estimation that led to the relatively low FAI values. Accordingly, the topographic data was manually “corrected” to incorporate some of the key infrastructure and urban areas, resulting in increased FAI values up to 60%. The impact of the assumed flood hydrographs and floodplain Manning’s n values was also assessed. As expected for a topographically driven and simulated system, realistic changes to the flood hydrograph did not have any significant impact on simulation accuracy. And whilst a global decrease in Manning’s n resulted in reduced simulated flood extent, and hence marginal improvements in FAI values, it was again not overly significant. Similarly, the use of localized Manning’s n values was shown to be evident, though not particularly significant. Conclusion The overall results indicate that the type of rapid flood inundation model represented by RFSM-EDA can provide comparable accuracy relative to conventional full SWE 2D models, at a fraction of the computational runtime. This indicates that, given appropriate topographic
2011 IWRA - International Water Resources Association office@iwra.org - http://www.iwra.org - Admin

Warning: Unknown: write failed: No space left on device (28) in Unknown on line 0

Warning: Unknown: Failed to write session data (files). Please verify that the current setting of session.save_path is correct (/tmp) in Unknown on line 0