Comparison of systematic errors in two forecast models -with similar dynamical frame-works

Abstract

FForecast errors exhibit the characteristics of approximations in simulating dynamical and physical processes in models. The models are very complex and hence it is not always possible to identify the approximations responsible for any particular error pattern in forecasts. A comparison between the models' forecast performances can be valuable in isolating the causes of error patterns. Here a comparison of forecast errors in the AFGL (Air Force Geophysics Laboratory) and COLA (Center for Ocean-Land-Atmosphere Interactions) models is made with the expectation of identifying the causes of forecast errors. The two models are based on identical approximations in simulating the dynamical processes and only minor differences in parameterizations of the physical processes. Nine ten-day forecasts are made to study the error characteristics in the two models. The errors in the 500 mb geopotential height are negative in tropics and positive in extratropics. The temperatures at 850 mb are colder than observed in tropics and warmer than observed in extratropics. At 150 mb the temperatures are warmer tan observed in tropics and colder than observed in extratropics. These qualitative error characteristics are not only common to these two models, but also to the NMC (National Meteorological Center), GFDL (Geophysical Fluid Dynamics Laboratory), and ECMWF (European Centre for Medium-Range Weather Forecast) forecast models. The difference in the error structure between the two models is the magnitude of the error in the tropics. The tropical error in the AFGL model is larger than that in the COLA model. Another difference is in the 850 mb relative humidity field. In the AFGL model, relative humidity errors are negative largely over the ocean and positive over land with minor exceptions. This error structure differs from that of the COLA model which consists of mostly positive errors everywhere with some small regions of negative errors. The major differences in the physical parameterizations between the two models are in the radiation interaction with deep convective clouds, the manner in which the sea surface temperature (SST) is prescribed and the vertical transport of heat and moisture by shallow convection. The magnitude of tropical errors in the geopotential height at 500 mb and temperature at 850 mb may be because the AFGL model does not include deep convective cloud-radiation interactions. The 850 mb relative humidity errors over oceans are probably due to the manner in which the SST is prescribed and the lack of proper vertical transport of moisture by the shallow convection parameterization.