Dynamic model performance-driven weighting system for nowcasting

This paper describes a dynamic model performance-driven weighting system for real-time, short term weather forecasting. The selected data sources include three numerical weather prediction models and observations. The performance of each NWP model for the past 6 hours or a specified period is statistically analyzed and continuously evaluated. Based on the performance of the three models, a weight is derived and assigned to each model. A new integrated forecast model is then generated by blending forecasts from the three models with corresponding dynamic weights plus a bias correction. Five major forecast parameters from 22 American and Canadian airports are selected as case studies for development and application of the system. Dynamic model verification is also carried out. The integrated model shows enhanced nowcasting accuracy. The resulting components for this system can be widely used for processing data from different models, different locations, and different weather parameters in real time.