WaterBench-Iowa: a large-scale benchmark dataset for data-driven streamflow forecasting

Abstract. This study proposes a comprehensive benchmark dataset for streamflow forecasting, WaterBench-Iowa, that follows FAIR (findability, accessibility, interoperability, and reuse) data principles is prepared with focus on convenience utilizing in data-driven machine learning studies, provides performance state of art deep architectures the comparative analysis. By aggregating datasets streamflow, precipitation, watershed area, slope, soil types, evapotranspiration from federal agencies organizations (i.e., NASA, NOAA, USGS, Iowa Flood Center), we provided WaterBench-Iowa hourly forecast studies. has high temporal spatial resolution rich metadata relational information, which can be used variety research. We defined sample task predicting next 5 d future results this linear regression models, including long short-term memory (LSTM), gated recurrent units (GRU), sequence-to-sequence (S2S). Our model show median Nash-Sutcliffe efficiency (NSE) 0.74 Kling-Gupta (KGE) 0.79 among 125 watersheds 120 h ahead prediction task. makes up lack unified benchmarks earth science research accessed at Zenodo https://doi.org/10.5281/zenodo.7087806 (Demir et al., 2022a).