Can Neural Networks aid the Week 3-4 forecasting challenge?

The Climate Prediction Center (CPC) issued its first experimental Week 3-4 temperature and precipitation outlooks in September 2015. These forecasts are released every Friday for the entire United States. Forecasters use both statistical and dynamical forecast guidance and CPC uses ensemble regression to further post process the numerical model output to improve forecast skill. Ensemble regression allows us to fit a probability distribution to each real-time ensemble member, which results in calibrated probability forecast maps from each numerical model. This technique shows promise in elevating forecast skill, but overall and especially for precipitation, skill needs to be further improved.

Artificial Neural Networks have been successfully used to make predictions in other fields, but their usage in meteorological applications has been somewhat limited, especially at subseasonal time scales such as required here. This presentation will provide a brief overview of how Neural Networks work and their utility in meteorological forecasting. We’ll present results from our attempts to apply Neural Networks tothe Weeks 3-4 forecast problem and compare those results to our previous efforts based on post processing dynamical model forecasts using ensemble regression techniques.