Global‐Scale Prediction of Flood Timing Using Atmospheric Reanalysis

Abstract The annual timing of flood events is a useful indicator to study the interaction between atmospheric and catchment processes in generating floods. This paper presents an assessment of the seasonal timing of floods for 7,894 gauging locations across the globe over a common period from 1981 to 2010. The averaged ordinal date of annual maximum streamflow is then estimated for ungauged locations following a two‐stage prediction scheme. The first stage identifies regions that share a common climatic predictor of flood timing by analyzing the similarity of flood timing with seven climate variables. These variables represent precipitation timing and snowmelt dynamics and are derived from a global climate reanalysis data set. Homogeneous regions in terms of the dominant predictor are generalized in the second stage through a rule‐based classification. The classification partitions the world into five hydroclimate classes, where each class has flood timing predicted using the most relevant climate predictor. Using this relatively simple and interpretable model structure, flood timing could be predicted with a global mean absolute error of approximately 31 days while maintaining consistency across large regions. Potential applications of the developed map include better understanding of climatic drivers of flooding and benchmarking the performance of global hydrological models in simulating the processes relevant to flooding.