Diagnosis and Correction of Systematic Humidity Error in a Global Numerical Weather Prediction Model

Abstract

Accuracy of humidity forecasts has been considered relatively unimportant to much of the operational numerical weather prediction (NWP) community. However, the U.S. Air Force is interested in accurate water vapor and cloud forecasts as end products. It is expected that the NWP community as a whole will become more involved in improving their humidity forecasts as they recognize the important role of accurate water vapor distributions in data assimilation, forecasts of temperature and precipitation, and climate change research. As a modeling community, we need to begin now to identify and rectify the systematic humidity forecast errors that are present in NWP models. This will allow us to take full advantage of the new types of remotely sensed water vapor and cloud measurements that are on the horizon. The research reported in this paper attempts to address this issue in a simple, straightforward manner, using the Phillips Laboratory Global Spectral Model (PL GSM). It was found that significant systematic specific humidity errors exist in the much-used FGGE [First CARP (Global Atmospheric Research Program) Global Experimental] (initialized) analyses. However, when a correction on the analyses was imposed and the PL GSM forecasts rerun, forecast errors similar to the forecast errors generated from the uncorrected analyses were observed. The errors were diagnosed through an evaluation of the tendency terms in the model's specific humidity prognostic equation. The results showed that systematic low-level tropical drying and upper-level global moistening could be attributed to the convective terms and the horizontal and vertical advection terms, respectively. Alternative formulations of the model were identified in an attempt to reduce or eliminate these errors. In general, it was found that the alternative formulations that do not modify the convection parameterization of the model reduced the upper-level moistening, while those that do modify the convection scheme reduced low-level tropical drying but introduced low-level and midlevel moistening in the summer hemisphere extratropics. The authors conclude that the nonconvective modifications could be instituted in the model as is. However, more work is needed on improving the way that convective parameterizations distribute water vapor in the vertical.