Estimating evapotranspiration from the Amazon Basin using the atmospheric water balance

Abstract

The spatio-temporal patterns of evapotranspiration (ET) in the Amazon basin are still poorly understood. Field studies in the Amazonian forest have shown that at some sites, deep roots allow trees to sustain elevated transpiration during several months of minimal rainfall, whereas at others, trees experience evident dry season water limitation. However, the few sites investigated are inadequate to characterize the conditions of transpiration throughout the basin. As a result of this uncertainty in modeling trees' access to deep soil moisture, land surface models cannot provide reliable estimates of transpiration in the region. From a basin-averaged perspective, it remains uncertain whether transpiration is water-limited, peaking during the basin's wet season, or energy-limited, peaking during the dry season when clearer skies allow for higher surface radiation. In this work, we investigate an approach to deriving a spatially-averaged ET estimate for the Amazon basin, which avoids modeling the forest's terrestrial hydrology. ET is computed as a residual of the atmospheric water balance, using basin-averaged convergence of atmospheric water vapor flux [C], precipitation [P], and tendency of total atmospheric water vapor [dw/dt] as inputs.

(cont.) As our resulting estimate of ET is only as good as the input estimates of the other hydrologic components, we analyze multiple cutting-edge datasets that may be used to compute these components. [P] data are obtained from GPCP and TRMM products. The three global reanalyses, NCEP/NCAR, NCEP/DOE and ECMWF ERA-40 provide data on atmospheric fields to compute [C] and [dw/dt]. The large discrepancies between [C] estimates produced by the different reanalyses, interpreted as uncertainty in these estimates, led to a thorough investigation of data on this field over a time period dating back to 1980. Concurrent time series of precipitation and Amazon river discharge are used to evaluate the accuracy of the various reanalyses in simulating [C] at the monthly and annual timescales. A measure of the random error associated with [C] estimates from each data source is derived, and used as a weighting factor to combine information from the three reanalyses. The resulting estimates of monthly basin-averaged ET are significantly lower in their long-term mean than estimates published in the literature. The resulting climatological annual cycle of basin-averaged ET suggests a switch between water and energy limited conditions for transpiration over a year's duration.