Automated Verification Of Mesoscale Forecasts Using Spatial Statistics

摘要

The verification of high-resolution mesoscale numerical weather predictions presents unique challenges. Traditional verification metrics - root mean square error, etc., which rely on single point verification often give incomplete or misleading assessments of model performance. Small-scale features are often miss-represented (aliasing) or, due to much lower predictability than large-scale features, cause an unwarranted penalty by conventional verification measures due to small spatial or temporal errors. Both the model developer and the operational user need better metrics in order to assess the performance of very high-resolution models. Our long-term goal is to contribute to better high- resolution model development and selection by developing a suite of verification tools to assist both the model developer and the model user.