Calibration of uncertain flood inundation models using remotely sensed water levels

A traditional method of validating the performance of a flood model when remotely sensed data of the flood extent are available is to compare the predicted flood extent to that observed. The performance measure employed often uses areal pattern-matching to assess the degree to which the two extents overlap. Recently, remote sensing of flood extents using synthetic aperture radar (SAR) and airborne scanning laser altimetry (LIDAR) has made more straightforward the synoptic measurement of water surface elevations along flood waterlines, and this has emphasised the possibility of using alternative performance measures based on height. This paper considers the advantages that can accrue from using a performance measure based on waterline elevations rather than one based on areal patterns of wet and dry pixels. The two measures were compared for their ability to estimate flood inundation uncertainty maps from a set of model runs carried out to span the acceptable model parameter range in a GLUE-based analysis. A 1 in 5-year flood on the Thames in 1992 was used as a test event. As is typical for UK floods, only a single SAR image of observed flood extent was available for model calibration and validation. A simple implementation of a two-dimensional flood model (LISFLOOD-FP) was used to generate model flood extents for comparison with that observed. The performance measure based on height differences of corresponding points along the observed and modelled waterlines was found to be significantly more sensitive to the channel friction parameter than the measure based on areal patterns of flood extent. The former was able to restrict the parameter range of acceptable model runs and hence reduce the number of runs necessary to generate an inundation uncertainty map. A result of this was that there was less uncertainty in the final flood risk map. The uncertainty analysis included the effects of uncertainties in the observed flood extent as well as in model parameters. The height-based measure was found to be more sensitive when increased heighting accuracy was achieved by requiring that observed waterline heights varied slowly along the reach. The technique allows for the decomposition of the reach into sections, with different effective channel friction parameters used in different sections, which in this case resulted in lower r.m.s. height differences between observed and modelled waterlines than those achieved by runs using a single friction parameter for the whole reach. However, a validation of the modelled inundation uncertainty using the calibration event showed a significant difference between the uncertainty map and the observed flood extent. While this was true for both measures, the difference was especially significant for the height-based one. This is likely to be due to the conceptually simple flood inundation model and the coarse application resolution employed in this case. The increased sensitivity of the height-based measure may lead to an increased onus being placed on the model developer in the production of a valid model

Improving flood inundation models using remotely sensed data

The paper discusses the wide variety of ways in which remotely sensed data are being utilized in river flood inundation modeling. Model parameterization is being aided using airborne LiDAR data to provide topography of the floodplain for use as model bathymetry, and vegetation heights in the floodplain for use in estimating floodplain friction factors. Model calibration and validation are being aided by comparing the flood extent observed in SAR images with the extent predicted by the model. The recent extension of this to the observation of urban flooding using high resolution TerraSAR-X data is described. Possible future research directions are considered.

Hydraulic modeling of runoff over a rough surface under partial inundation

This study presents a numerical method to derive the Darcy- Weisbach friction coefficient for overland flow under partial inundation of surface roughness. To better account for the variable influence of roughness with varying levels of emergence, we model the flow over a network which evolves as the free surface rises. This network is constructed using a height numerical map, based on surface roughness data, and a discrete geometry skeletonization algorithm. By applying a hydraulic model to the flows through this network, local heads, velocities, and Froude and Reynolds numbers over the surface can be estimated. These quantities enable us to analyze the flow and ultimately to derive a bulk friction factor for flow over the entire surface which takes into account local variations in flow quantities. Results demonstrate that although the flow is laminar, head losses are chiefly inertial because of local flow disturbances. The results also emphasize that for conditions of partial inundation, flow resistance varies nonmonotonically but does generally increase with progressive roughness inundation.

Case study of the use of remotely sensed data for modeling flood inundation on the river Severn, UK.

A methodology for using remotely sensed data to both generate and evaluate a hydraulic model of floodplain inundation is presented for a rural case study in the United Kingdom: Upton-upon-Severn. Remotely sensed data have been processed and assembled to provide an excellent test data set for both model construction and validation. In order to assess the usefulness of the data and the issues encountered in their use, two models for floodplain inundation were constructed: one based on an industry standard one-dimensional approach and the other based on a simple two-dimensional approach. The results and their implications for the future use of remotely sensed data for predicting flood inundation are discussed. Key conclusions for the use of remotely sensed data are that care must be taken to integrate different data sources for both model construction and validation and that improvements in ground height data shift the focus in terms of model uncertainties to other sources such as boundary conditions. The differences between the two models are found to be of minor significance.

Calibrating flood inundation models: identifying and addressing uncertainty in satellite observed flood extents

Improvements in the resolution of satellite imagery have enabled extraction of water surface elevations at the margins of the flood. Comparison between modelled and observed water surface elevations provides a new means for calibrating and validating flood inundation models, however the uncertainty in this observed data has yet to be addressed. Here a flood inundation model is calibrated using a probabilistic treatment of the observed data. A LiDAR guided snake algorithm is used to determine an outline of a flood event in 2006 on the River Dee, North Wales, UK, using a 12.5m ERS-1 image. Points at approximately 100m intervals along this outline are selected, and the water surface elevation recorded as the LiDAR DEM elevation at each point. With a planar water surface from the gauged upstream to downstream water elevations as an approximation, the water surface elevations at points along this flooded extent are compared to their ‘expected’ value. The pattern of errors between the two show a roughly normal distribution, however when plotted against coordinates there is obvious spatial autocorrelation. The source of this spatial dependency is investigated by comparing errors to the slope gradient and aspect of the LiDAR DEM. A LISFLOOD-FP model of the flood event is set-up to investigate the effect of observed data uncertainty on the calibration of flood inundation models. Multiple simulations are run using different combinations of friction parameters, from which the optimum parameter set will be selected. For each simulation a T-test is used to quantify the fit between modelled and observed water surface elevations. The points chosen for use in this T-test are selected based on their error. The criteria for selection enables evaluation of the sensitivity of the choice of optimum parameter set to uncertainty in the observed data. This work explores the observed data in detail and highlights possible causes of error. The identification of significant error (RMSE = 0.8m) between approximate expected and actual observed elevations from the remotely sensed data emphasises the limitations of using this data in a deterministic manner within the calibration process. These limitations are addressed by developing a new probabilistic approach to using the observed data.

Calibrating flood inundation models: methods for assessing model performance

Satellite observed data for flood events have been used to calibrate and validate flood inundation models, providing valuable information on the spatial extent of the flood. Improvements in the resolution of this satellite imagery have enabled indirect remote sensing of water levels by using an underlying LiDAR DEM to extract the water surface elevation at the flood margin. Further to comparison of the spatial extent, this now allows for direct comparison between modelled and observed water surface elevations. Using a 12.5m ERS-1 image of a flood event in 2006 on the River Dee, North Wales, UK, both of these data types are extracted and each assessed for their value in the calibration of flood inundation models. A LiDAR guided snake algorithm is used to extract an outline of the flood from the satellite image. From the extracted outline a binary grid of wet / dry cells is created at the same resolution as the model, using this the spatial extent of the modelled and observed flood can be compared using a measure of fit between the two binary patterns of flooding. Water heights are extracted using points at intervals of approximately 100m along the extracted outline, and the students T-test is used to compare modelled and observed water surface elevations. A LISFLOOD-FP model of the catchment is set up using LiDAR topographic data resampled to the 12.5m resolution of the satellite image, and calibration of the friction parameter in the model is undertaken using each of the two approaches. Comparison between the two approaches highlights the sensitivity of the spatial measure of fit to uncertainty in the observed data and the potential drawbacks of using the spatial extent when parts of the flood are contained by the topography.

Estimating discharge from inundation imagery

Remote sensing from space-borne platforms is often seen as an appealing method of monitoring components of the hydrological cycle, including river discharge, due to its spatial coverage. However, data from these platforms is often less than ideal because the geophysical properties of interest are rarely measured directly and the measurements that are taken can be subject to significant errors. This study assimilated water levels derived from a TerraSAR-X synthetic aperture radar image and digital aerial photography with simulations from a two dimensional hydraulic model to estimate discharge, inundation extent, depths and velocities at the confluence of the rivers Severn and Avon, UK. An ensemble Kalman filter was used to assimilate spot heights water levels derived by intersecting shorelines from the imagery with a digital elevation model. Discharge was estimated from the ensemble of simulations using state augmentation and then compared with gauge data. Assimilating the real data reduced the error between analyzed mean water levels and levels from three gauging stations to less than 0.3 m, which is less than typically found in post event water marks data from the field at these scales. Measurement bias was evident, but the method still provided a means of improving estimates of discharge for high flows where gauge data are unavailable or of poor quality. Posterior estimates of discharge had standard deviations between 63.3 m3s-1 and 52.7 m3s-1, which were below 15% of the gauged flows along the reach. Therefore, assuming a roughness uncertainty of 0.03-0.05 and no model structural errors discharge could be estimated by the EnKF with accuracy similar to that arguably expected from gauging stations during flood events. Quality control prior to assimilation, where measurements were rejected for being in areas of high topographic slope or close to tall vegetation and trees, was found to be essential. The study demonstrates the potential, but also the significant limitations of currently available imagery to reduce discharge uncertainty in un-gauged or poorly gauged basins when combined with model simulations in a data assimilation framework.

The impact of uncertainty in satellite data on the assessment of flood inundation models

The performance of flood inundation models is often assessed using satellite observed data; however these data have inherent uncertainty. In this study we assess the impact of this uncertainty when calibrating a flood inundation model (LISFLOOD-FP) for a flood event in December 2006 on the River Dee, North Wales, UK. The flood extent is delineated from an ERS-2 SAR image of the event using an active contour model (snake), and water levels at the flood margin calculated through intersection of the shoreline vector with LiDAR topographic data. Gauged water levels are used to create a reference water surface slope for comparison with the satellite-derived water levels. Residuals between the satellite observed data points and those from the reference line are spatially clustered into groups of similar values. We show that model calibration achieved using pattern matching of observed and predicted flood extent is negatively influenced by this spatial dependency in the data. By contrast, model calibration using water elevations produces realistic calibrated optimum friction parameters even when spatial dependency is present. To test the impact of removing spatial dependency a new method of evaluating flood inundation model performance is developed by using multiple random subsamples of the water surface elevation data points. By testing for spatial dependency using Moran’s I, multiple subsamples of water elevations that have no significant spatial dependency are selected. The model is then calibrated against these data and the results averaged. This gives a near identical result to calibration using spatially dependent data, but has the advantage of being a statistically robust assessment of model performance in which we can have more confidence. Moreover, by using the variations found in the subsamples of the observed data it is possible to assess the effects of observational uncertainty on the assessment of flooding risk.