Can discharge be used to inversely correct precipitation?

This study explores the feasibility of using the information contained in observed streamflow measurements to inversely correct catchment-average precipitation time series provided by reanalysis products at the continental scale. We explore this possibility by training LSTM ensemble networks to inversely predict precipitation by using the streamflow of catchments as additional input. The first model uses discharge as an input feature along with other meteorological variables, while the second model uses only the meteorological predictors. Analysing the performance of both models showed that the discharge information not only led to an average improvement overall, but also resulted in a significant improvement (around 30 %) on days with precipitation amounts greater than 5 mm. An out-of-sample test showed that the inversely estimated precipitation is better able to reproduce small-scale, high-impact events that are poorly represented in the reanalysis product. Further, using the inversely generated precipitation time series for classical hydrological “forward” modeling resulted in improved estimates for streamflow and soil moisture. Given that the wealth of streamflow gauges around the world is currently underutilised for meteorological applications, our findings have significant implications for achieving better estimates of precipitation associated with high-impact flood events.