Remotely sensed ensembles of the terrestrial water budget over major global river basins: An assessment of three closure techniques

Abstract

Remote sensing is a useful tool for observing the water cycle. However, combining remote sensing products over any major river basin will result in a residual error in the overall water balance. Previous studies have either quantified this error without correcting it, or have merged observations together with land surface models (LSMs) to produce a single “best” estimate of the water balance. Here, we present a new approach in which combinations of remote sensing and in situ observations are constrained to enforce water balance closure. Rather than a single estimate, this produces an ensemble of unique water balance estimates intended to characterize uncertainty and to avoid biases implicit in LSMs. We evaluate three techniques of varying complexity to enforce water balance closure for individual ensemble members over 24 global basins from Oct. 2002 - Dec. 2014, resulting in as many as 60 realizations of the monthly water budget, contingent upon data availability. Compared with a published climate data record, the ensemble shows strong agreement for precipitation, evapotranspiration and changes in storage (R2: 0.91–0.95), with less agreement for streamflow (R2: 0.42–0.47), which may be indicative of LSM biases in the climate data record. Water balance residual errors resulting from combinations of raw products vary significantly (p < 0.001) with latitude, with a tendency for positive biases for low- and mid-latitude basins, and negative biases elsewhere. Overall, residual errors are equivalent to 15% of total precipitation when averaged across all data products and basins. This study shows that closure constraints provide additional value outside of closing the water budget, including reduction of uncertainty and transfer of closure constraints in time to provide skillful estimates of mean annual basin discharge. We also showed that a simpler closure technique, proportional redistribution, performed better than more complex ones in decreasing uncertainty and for transfer through time to estimate basin discharge when a rigorous analysis of errors for each data product is not accounted for. This observation-based dataset is distinct from modeled estimates and therefore has the potential to preserve important information of anthropogenic effects on the water balance.