Precipitation has been one of the most difficult forecast fields for modern numerical weather prediction (NWP). The precipitation forecast skill in an NWP model has lagged considerably behind skills of many other meteorological fields, such as temperature, winds, etc. To improve this situation, we must first understand the characteristics of precipitation forecast errors better. The representation of precipitation physics in atmospheric models is typically categorized into resolvable and unresolvable processes. For the resolvable precipitation process, the key issue to improving the forecast relies very much on the improvement of data assimilation. For unresolvable physical processes, parameterization schemes must be used. Literally, these parameterization schemes will depend heavily upon a set of model parameters, which need to be tuned to best fit training observations. Two paradigms can lead to precipitation forecasts with a “spatial bias” and a “situational bias”, respectively, when using these parameterization schemes. First, because many convective parameterization schemes were developed on the basis of our limited knowledge of some particular convective or storm structure (e.g., a deepconvection process), the general application of this parameterization often fails (for instance, for precipitation due to a shallow cloud process). This scenario is what we referred to as the “situational bias”. Second, many convective parameterization schemes are developed and validated using observational data available for certain spatial locations, such as on continental areas where a relatively dense observational network exists. The question of whether such tested parameterization schemes are also suitable for precipitation forecasts over the oceans may give rise to an issue of spatial bias. Furthermore, the representation of resolvable and unresolvable precipitation processes in atmospheric models implies two different scales of forecast errors. Tracking down the model deficiency parties partly depends on how we are able to untangle the scales of those errors. In this study, precipitation forecasts from a global weather forecast model are verified against satellite observations. Forecast errors are computed and averaged for summer and winter seasons. A spatial 2-D wavelet is used to decompose these errors. The scaledependent forecast errors are analyzed.
This publication was presented at the following: