This review describes recent development in operational and research limited-area numerical weather prediction models in middle latitudes. The current skill of limited-area models is summarized through the use of conventional measures of verification such as S1 scores, root-mean-square errors and correlations between forecast and observed changes. Additional measures of verification, which measure the skill or realism of regional models in reproducing atmospheric structure on those scales, are discussed. Use of a uniform set of verification measures such as those discussed here would facilitate model comparisons and assessment of the impact of changes in model components on short-range (0–48 h) forecasts.
Three major components of regional models are discussed. These include numerical aspects (e.g., the grid structure, boundary conditions and the approximations to the analytic differential equations), physical aspects (modeling surface and boundary layer processes, condensation and evaporation, and radiation) and the analysis and initialization procedure. The paper emphasizes the impact of these components on the forecast rather than the details of each component.
The main conclusion of the paper is that further increases in the overall skill of operational regional forecasts are likely to occur through improvements in all the components of limited-area models. Improvements in various components developed and tested in research models are currently being incorporated in several operational models, and some modest but significant improvements in regional forecast skill are likely over the next five years.