Some Conclusions on Applying Statistical Design of Experiments to Numerical Weather Prediction

Numerical Weather Prediction (NWP) is the science of forecasting weather or climatic conditions based on past and present weather observations using computational methods applied to mathematical representations of the atmospheric physical processes. Temporally, weather forecasts range from a few hours to several days in the future, while climate forecasts range from several weeks to years (or decades) into the future. Spatially, forecasts can cover small scale, highly resolved “local” weather conditions over small domains to large scale global weather features and climatic patterns. A key feature in NWP is the range of possible paramenterization schemes available to the user to configure the NWP for a specific domain and forecast regime. We present results from our initial work applying statistical design of experiments, to understand and attribute error in the forecast to parameterization schemes. By applying statistical design of experiments we are able to judiciously sample the NWP to obtain data at well chosen points from which we can make the requisite statistical inferences regarding the tendencies of the parameterization schemes.