Estimating Evapotranspiration Over the Amazon Basin Using Multiple Data Sources

Estimates of regional evapotranspiration (ET) over the Amazon basin have been found to be highly dependent on the parameterization of transpiration. The inadequately understood and difficult to model vegetation control over ET under different conditions of energy and water availability leaves us with divergent estimates of the annual ET cycle in the basin. We estimate regional ET over the Amazon basin on a monthly timescale, independently of land surface models, as the residual of the atmospheric water balance. The control volume is defined to be the atmosphere overlying the basin and the variables involved are area-averaged ET, precipitation, vertically integrated moisture convergence, and change in total moisture storage. Other workers have shown that significant differences exist between measures of these water balance components obtained from different data sources, and thus the uncertainties associated with these available measures cannot be neglected. We apply a framework that utilizes multiple datasets to derive a `best' estimate of ET, minimizing its uncertainty. Multiple measures of each water balance component are used in a least squares estimator, where each input value is weighted by its variance. To further reduce the uncertainty in the derived ET estimate, we include a measure of ET and its variance and constrain the optimization with the water balance equation to ensure conservation of mass. This approach is tested over the 5-year period 1997-2001. We evaluate our ability to resolve the seasonal ET cycle with the available data, during regular years and during the 1998 ENSO-related drought that affected the region. The NCEP-NCAR Reanalysis-1, NCEP-DOE Reanalysis-2 and the ECMWF ERA-40 Reanalysis data products are used to derive alternative input estimates of moisture convergence and change in precipitable water. We use the TRMM 3B43 monthly precipitation rate along with its random error estimate, and the GPCP One-Degree Daily Precipitation Data Set that also includes a random error estimate, to obtain alternative estimates of monthly precipitation. Information on the uncertainty associated with each data product is extrapolated from the individual gridpoints to the area-averaged basin scale. An ET estimate computed as the mean over different model results and its corresponding variance are used in the constrained least square optimization.