Using triple collocation of precipitation and evapotranspiration products to reduce uncertainty and improve inferences of catchment-scale water budgets

Abstract:

Water budgets are a key tool to characterize undetermined fluxes and stores of water in catchment hydrology. Traditionally, water balances have been investigated using ground-based networks. Recently, advancements in scalable (e.g. remotely sensed, modeled, and reanalysis) data have become critical to understanding the terrestrial water cycle. Constraining uncertainty in ‘gridded’ remotely sensed, modeled, or reanalysis input variables such as precipitation (P) and evapotranspiration (ET) with ground-based validation networks remains a significant obstacle. Triple collocation (TC), and its variant Extended Triple Collocation (ETC), enable the estimation of root-mean-square-error (RMSE) and a correlation coefficient using three or more spatially and temporally collocated measurement despite not knowing the underlying true observations. Yet to the best of our knowledge, neither TC nor ETC has been applied to constrain water budget uncertainty. Using ~200 mountainous watersheds, we first apply ETC to P & ET and then use an objective methodology for optimally merging data products to provide merged output that weighted variables based on the ETC results. Using a simple inverted water balance analysis, we assess the performance of the optimally ET & P merged product against an unweighted ensemble mean and the best individual products. We also examine our results in the Budyko space—testing both an open and a closed water balance assumption. Our results indicate that ETC improves the water balance closure in some systems and dampens inter-annual fluctuations in storage and error when compared with both ensemble results and individual product results. These results highlight substantial P under-prediction in steeper, mountainous western U.S. catchments. ETC also improves input variable performance in the Budyko space when both ET and P estimates are used in an open water budget approach. Using the ETC merged product led to different inferences in hydrological processes, including the role of snow in streamflow generation. The application of ETC and independent physical checks on the system can meaningfully constrain uncertainty in catchment water balances, which gives insight into critical data limitations and changes inferences into hydrological sensitivity to climate.