Earth observation informed modelling of flash floods

Abstract. More frequent extreme rainfall events in a changing climate increase the risk of flash flooding that is affecting populations globally. However, the flood hazard modelling required to reduce disaster risk in populated urban environments is often limited by the availability of data required for model calibration and validation. In this study, we use a historical flood event captured by 5 m resolution satellite imagery to quantify the effects of flood model complexity and inform flood hazards under future climate scenarios in the West Bank, Palestine. Flooding in January 2013 affected over 12,500 people and large areas of cropland. Vegetation loss and damage during the event were captured using satellite imagery and a normalised difference vegetation index (NDVI), and used as a reference flood extent. The physics-based HEC-RAS flood model best reproduced this NDVI-derived inundation extent (F1 score = 0.76), although the FastFlood model was able to produce a similar inundation pattern (F1 score = 0.74) over 300 times faster. Simulated flood depths from both models were similar; FastFlood displayed a mean difference of -0.03 m and a mean absolute error of 0.51 m when compared to HEC-RAS. Climate analysis revealed that the January 2013 rainfall corresponded to a historical return period of between 1 in 5 and 1 in 10 years. In comparison, a 1 in 100-year rainfall event (RX1day (maximum 1-day precipitation) of 148 mm) based on historical data (1985–2014) could increase by almost 40 % (to 205 mm) in the mid-future (2041–2060), which could cause 23 % (4 km²) greater inundation compared to the 2013 event. Although the patterns of future precipitation in the region are uncertain, our flood hazard maps can support urban planning and infrastructure development to manage storm water runoff, particularly where ephemeral channels, or wadis, intersect the road network.