Abstract
Estimates of the net convergence of atmospheric moisture flux over the Amazon Basin, [C], derived using data products from three global reanalyses, the NCEP–NCAR reanalysis (NCEP-1), the NCEP/Department of Energy reanalysis (NCEP-2), and the 40-yr ECMWF Re-Analysis (ERA-40), are compared. Two types of uncertainty in these [C] estimates are distinguished and quantified: “model-associated uncertainty,” which necessarily arises from imperfections in the numerical weather models or data assimilation algorithms, and “postprocessing uncertainty” introduced by operations performed on the original reanalysis data products to compute [C], particularly the finite-difference approximation of divergence. Model-associated uncertainty is found to overwhelm the postprocessing error. A closer look at the time series of this field extending over the period 1980–2001, and their comparison to basin-averaged precipitation and runoff data, reveals the signatures of two potential sources of model-associated errors. 1) ERA-40 estimates of [C] exhibit an artificial shift in 1987, possibly produced by the start of assimilation of Special Sensor Microwave Imager (SSM/I) data. The estimates preceding 1988 are negatively biased relative to the remaining time series, and hence subsequent analysis is limited to the 14-yr period 1988–2001. 2) NCEP-1 and NCEP-2 estimates of [C] show a negative bias over the period 1992–98, which likely originates in biased Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) data assimilated by these reanalyses. A measure of the random error in the [C] time series produced by each reanalysis, computed using river discharge data as reference, indicates that ERA-40 gives the most accurate estimates of net atmospheric moisture flux convergence for the aforementioned 14-yr period.
Corresponding author address: Hanan Karam, 15 Vassar St., 48-212, Cambridge, MA 02139. Email: hnkaram@mit.edu
