Combining observations and model data for short-term storm forecasting

This paper describes the use of a machine learning data fusion methodology to support development of an automated short-term thunderstorm forecasting system for aviation users. Information on current environmental conditions is combined with observations of current storms and derived indications of the onset of rapid change. Predictor data include satellite radiances and rates of change, satellite-derived cloud type, ground weather station measurements, land surface and climatology data, numerical weather prediction model fields, and radar-derived storm intensity and morphology. The machine learning methodology creates an ensemble of decision trees that can serve as a forecast logic to provide both deterministic and probabilistic estimates of thunderstorm intensity. It also provides evaluation of predictor importance, facilitating selection of a minimal skillful set of predictor variables and providing a tool to help determine what weather regimes may require specialized forecast logic. This work is sponsored by the Federal Aviation Administration's Aviation Weather Research Program. Its aim is to contribute to the development of the Consolidated Storm Prediction for Aviation (CoSPA) system, which is being developed in collaboration with the MIT Lincoln Laboratory and the NOAA Earth System Research Laboratory's Global Systems Division. CoSPA is scheduled to become part of the NextGen Initial Operating Capability by 2012.