Evaluation of the Global Atmospheric Moisture Budget as Seen from Analyses

For the period 1987 to 1993, quantities central to the global moisture budget from the global analyses of the European Centre for Medium-Range Weather Forecasts (ECMWF), the U.S. National Meteorological Center (NMC), and NASA/Goddard have been computed and compared. The precipitable water is computed and compared with satellite data from the Special Sensor Microwave Imager (SSM/I). Fluxes of moisture and their divergence have been used to estimate the vertically integrated moisture budget and thus evaporation minus precipitation (E − P) as residuals. Results of several test computations show that small biases exist in precipitable water as vertical resolution and methods of computing vertical integrals are changed, but the impact is small on the moisture budget. In the Tropics and subtropics the moisture budget is dominated by the divergence field rather than the moisture amounts, and consequently initialization of the analyses has an impact on the perceived moisture budget. The diurnal cycle is shown to be important especially for E − P. For the vertically integrated moisture budget, the use of pressure coordinates instead of the model coordinates on which the data are analyzed produces acceptably small differences for the most part, although adequate resolution at low levels and proper treatment of the surface is important. Computations at different horizontal resolutions reveal the importance of adequate resolution in the vicinity of steep orography but also that resolution is not a big factor on large scales even where steep gradients in precipitation exist. The implication is that it is the veracity of the large-scale divergence and moisture fields themselves that contribute to problems, and these arise from the moist physics of the assimilating model used in four-dimensional data assimilation, which dominates the character of the analyses. However, in the subtropics large positive biases are present in precipitable water in the ECMWF and, to a lesser extent, NMC analyses and are partly due to the assimilation of biased TOVS retrievals. The effects of the large and spurious changes in analysis systems at ECMWF and NMC are manifested in the results. Differences between ECMWF and NMC E − P locally are 60%–75% of the values themselves in the Tropics at T31 resolution and about one-half as much at TI5 resolution, and they are not diminishing with time. By making use of estimates of E and P from the NASA/Goddard reanalysis, both spatial and temporal variability of E − P are found to be dominated by the P field. Accordingly, for part of 1987 and 1988, estimates of precipitation from the Global Precipitation Climatology Project are used to qualitatively assess the E − P estimates. The ECMWF estimates appear to be best at that time, but since then the NMC values have become less intense while the ECMWF estimates have greatly intensified. The large differences that still exist in such estimates are apt to be carried over to the NMC and ECMWF reanalysis products.