Variability of Basin-Scale Terrestrial Water Storage from a PER Water Budget Method: The Amazon and the Mississippi

Abstract

In an approach termed the PER method, where the key input variables are observed precipitation P and runoff R and estimated evaporation, the authors apply the basin water budget equation to diagnose the long-term variability of the total terrestrial water storage (TWS). Unlike the typical offline land surface model estimate where only atmospheric variables are used as input, the direct use of observed runoff in the PER method imposes an important constraint on the diagnosed TWS. Although there is a lack of basin-scale observations of evaporation, the tendency of E to have significantly less variability than the difference between precipitation and runoff (P ? R) minimizes the uncertainties originating from estimated evaporation. Compared to the more traditional method using atmospheric moisture convergence (MC) minus R (MCR method), the use of observed precipitation in the PER method is expected to lead to general improvement, especially in regions where atmospheric radiosonde data are too sparse to constrain the atmospheric model analyzed MC, such as in the remote tropics. TWS was diagnosed using the PER method for the Amazon (1970?2006) and the Mississippi basin (1928?2006) and compared with the MCR method, land surface model and reanalyses, and NASA?s Gravity Recovery and Climate Experiment (GRACE) satellite gravity data. The seasonal cycle of diagnosed TWS over the Amazon is about 300 mm. The interannual TWS variability in these two basins is 100?200 mm, but multidecadal changes can be as large as 600?800 mm. Major droughts, such as the Dust Bowl period, had large impacts, with water storage depleted by 500 mm over a decade. Within the short period 2003?06 when GRACE data were available, PER and GRACE show good agreement both for seasonal cycle and interannual variability, providing potential to cross validate each other. In contrast, land surface model results are significantly smaller than PER and GRACE, especially toward longer time scales. While the authors currently lack independent means to verify these long-term changes, simple error analysis using three precipitation datasets and three evaporation estimates suggest that the multidecadal amplitude can be uncertain up to a factor of 2, while the agreement is high on interannual time scales. The large TWS variability implies the remarkable capacity of land surface in storing and taking up water that may be underrepresented in models. The results also suggest the existence of water storage memories on multiyear time scales, significantly longer than typically assumed seasonal time scales associated with surface soil moisture.