A weather pattern based approach to evaluate Antarctic Mesoscale Prediction System (AMPS) forecasts: Part 2. Comparison to automatic weather station observations

A critical element in improving weather prediction models used for weather forecasting is a careful evaluation of the model forecasts, so that model errors can be identified and corrected. Typical model evaluation strategies evaluate the model over large periods of time (months, seasons, years, etc) or for single case study events of storms or other events of interest. Here we will discuss a new method of evaluating weather prediction models that assesses the model skill over long periods of time but subsets this time period into similar weather patterns, allowing us to determine if certain model errors occur under certain weather regimes and not others. The method of self-organizing maps (SOMs) is used to create a synoptic climatology of the weather patterns that occur in the Ross Sea sector of Antarctica. Using these weather patterns the model forecasts, from the Antarctic Mesoscale Prediction System (AMPS), that correspond to each weather pattern are identified and model errors are then determined for each weather pattern. In part 2 of this presentation we will present results from the SOM analysis when used to compare the AMPS forecasts to in-situ automatic weather station observations. The ability of the SOM technique to determine the variability in atmospheric state and model performance over different weather patterns and geographic locations will be presented. Results from this analysis show that the AMPS forecasts do not perform with equal accuracy for the different weather patterns considered. The results of this analysis will benefit both modelers and forecasters.