Development of a rainfall-runoff-inundation model and flood monitoring system based on satellite imagery for Kalu ganga basin, Sri Lanka

Abstract

Floods are getting severe due to climate change and anthropogenic activities which neeimmediate response to lower the risk and decrease the human and financial losses. Floodinundation mapping for flood risk preparedness using satellite data has been widely used imany recent studies. However, satellite imageries may contain some uncertainties. Thereforflood inundation maps from satellite data need to be verified with flood inundation mapgenerated by hydrological models from observed data for accurate estimation of flood risk.Although satellite-generated flood maps are widely used to determine the inundation extentthere are certain challenges to their use such as inaccessibility of imagery due to satellite orbior cloud cover, which hampers accurate measurement of inundation risk. In this study, the rainfall-runoff inundation (RRI) model for the Kalu Ganga basin wasdeveloped, and its applicability to evaluate the discharge and flood inundation areas wasdiscussed. The RRI model could estimate discharge, water levels, and inundation areasimultaneously based on two-dimensional diffusion wave equations. The results and statisticaanalysis indicate that the RRI model could efficiently estimate extreme flood events. Formodel calibration, the R 2 value ranges from 0.72-0.80 and for model validation, the R valueranges from 0.75-0.90, which shows good performance of the model. The simulated inundation extents were verified and compared with Sentinel 1A SA(Synthetic Aperture Radar) satellite imagery data for 2016 and 2017 flood events. Sentinel 1AGRD-IW (Ground Range Detected - Interferometric Wide swath) mode of VV co-polarization,with a spatial resolution of 20 m was acquired and pre-processed using the SentineApplication Platform (SNAP) software toolbox. The pre-processed images were correcteand maximum likelihood supervised classification was performed to produce the floodinundation maps of the study area. The actual flooded area from RRI is found to be 291.9km 2 and that from satellite image is found to be 201.7 km 2 for the 2016 flood event. For th2017 flood event, the actual flooded area from RRI is found to be 371.14 km and that frosatellite image is found to be 297.42 km 2 . Hence, the flooded area difference was found to be35.54 % for 2016 and 22.13 % for 2017 flood events from the total area selected from thmodel. Most of the floodplains from the RRI model and satellite images were along the mairiver in the basin, including the city of Ratnapura (upstream), the city of Kalutar(downstream), and the areas in between. These results with an accuracy level of ~25 % - 30 % are deemed to be within an acceptable range for emergency evacuation and rapid flood damageassessment purposes. Future studies should further investigate and validate the flooinundation mapping ability of Sentinel 1A SAR using ground-based reference flood maps orother satellite data. This study reveals that satellite imagery can be one of the most coseffective ways to capture the flood disaster footprints, identify flood-prone areas, anunderstand the flooding problem in a better way. This methodology can be effectively usefor disaster risk management, where the time factor is very critical. 2

2