Intercomparison of global reanalysis precipitation for flood risk modelling

Abstract

Abstract. Reanalysis datasets are increasingly used to drive flood models, especially for continental and global analysis and in areas of data scarcity. However, the consequence of this for risk estimation has not been fully explored. We investigate the implications of four reanalysis products (ERA-5, CFSR, MERRA-2 and JRA-55) on simulations of historic flood events in five basins in England. These results are compared to a benchmark national gauge-based product (CEH-GEAR1hr). The benchmark demonstrated better accuracy than reanalysis products when compared with observations of water depth and flood extent. All reanalysis products predicted fewer buildings would be inundated by the events than the national dataset. JRA-55 was the worst by a significant margin, underestimating by 40 % compared with 14 %–18 % for the other reanalysis products. CFSR estimated building inundation the most accurately, while ERA-5 demonstrated the lowest error in terms of river stage (29.4 %) and floodplain depth (28.6 %). Accuracy varied geographically, and no product performed best across all basins. Global reanalysis products provide a useful resource for flood modelling where no other data are available, but they should be used with caution due to the underestimation of impacts shown here. Until a more systematic international strategy for the collection of rainfall and flood impact data ensures more complete global coverage for validation, multiple reanalysis products should be used concurrently to capture the range of uncertainties.