Evaluating Precipitation Datasets Using Surface Water and Energy Budget Closure

Abstract

Evaluation of global gridded precipitation datasets typically entails using the in situ or satellite-based data used to derive them, so that out-of-sample testing is usually not possible. Here we detail a methodology that incorporates the physical balance constraints of the surface water and energy budgets to evaluate gridded precipitation estimates, providing the capacity for out-of-sample testing. Performance conclusions are determined by the ability of precipitation products to achieve closure of the linked budgets using adjustments that are within their prescribed uncertainty bounds. We evaluate and compare five global gridded precipitation datasets: IMERG, GPCP, GPCC, REGEN, and MERRA-2. At the spatial level, we show that precipitation is best estimated by GPCC over the high latitudes, by GPCP over the tropics, and by REGEN over North Africa and the Middle East. IMERG and REGEN appear best over Australia and South Asia. Furthermore, our results give insight into the adequacy of prescribed uncertainties of these products and shows that MERRA-2, while being less competent than the other four products in estimating precipitation, has the best representation of uncertainties in its precipitation estimates. The spatial extent of our results is not only limited to grid cells with in situ observations. Therefore, the approach enables a robust evaluation of precipitation estimates and goes some way to addressing the challenge of validation over observation scarce regions.

Robust historical evapotranspiration trends across climate regimes

“…However, unlike several other merging techniques (Mueller et al, 2013;Paca et al, 2019;Rodell et al, 2015;Stephens et al, 2012), it accounts for performance differences between parent estimates using in situ data as the observational constraint rather than assigning weights based on the ability to match another gridded dataset that is deemed more reliable or the ensemble mean of a selection of datasets (Munier et al, 2014;Sahoo et al, 2011;Wan et al, 2015;Zhang et al, 2018). The efficacy of using in situ measurements for constraining much larger-scale gridded estimates has also been shown explicitly (Hobeichi et al, 2018(Hobeichi et al, , 2020c. Next, most available merging techniques do not account for dependence between parent estimates, where redundant information in different parent products is likely to bias the hybrid estimate (Abramowitz et al, 2019;Herger et al, 2018).…”

Robust historical evapotranspiration trends across climate regimes

“…There are of course some notable limitations to the approach we have taken here, some of which were previously discussed in Hobeichi et al (2018). First, the weighting approach adopted here relies heavily on flux tower observations, which can suffer from a range of technical issues (Burba and Anderson, 2010;Fratini et al, 2019), as well as temporal gaps during particular weather conditions such as extremes (Van Der Horst et al 2019), which can affect our results. Next, unresolved land surface processes in the parent datasets due for example to the absence of a proper representation of snow and permafrost dynamics, or the heterogeneity of the land surface are likely to lead to uncertain ET estimation in DOLCE V2, since it is only a combination of its parent data sets.…”

GRACEfully Closing the Water Balance: A Data‐Driven Probabilistic Approach Applied to River Basins in Iran

“…A separate group of studies focuses on bringing together estimates of the different water balance variables and modifying the original estimates so as to close the water balance (Aires, 2014; Allam et al., 2016; Hobeichi, Abramowitz, Contractor, & Evans, 2020; Munier et al., 2014; Pan, Sahoo, et al., 2012; Pan & Wood, 2006; Pellet et al., 2019; Rodell et al., 2015; Sahoo et al., 2011; Simons et al., 2016; Wang et al., 2015; Y. Zhang, Pan, Sheffield, et al., 2018; Y. Zhang, Pan, & Wood, 2016). In closing the water balance, variables with large errors are adjusted more than variables with small errors, a process that can be formalized by what Pan and Wood (2006) called a constrained Kalman filter.…”

“…A separate group of studies focuses on bringing together estimates of the different water balance variables and modifying the original estimates so as to close the water balance (Aires, 2014;Allam et al, 2016;Hobeichi, Abramowitz, Contractor, & Evans, 2020;Munier et al, 2014;Pan, Sahoo, et al, 2012;Pan & Wood, 2006;Pellet et al, 2019;Rodell et al, 2015;Sahoo et al, 2011;Simons et al, 2016;Wang et al, 2015;Y. Zhang, Pan, Sheffield, et al, 2018;Y.…”