Determing key predictors for NCAR's Convective Auto-Nowcast system using climatological analyses

The NCAR Auto-Nowcast system is an expert system for forecasting thunderstorm initiation, growth, and decay. The focus of this paper will be to discuss an updated approach for determining which data fields will be used in the initiation forecast logic, along with how the membership functions that define the predictor fields and their associated weights can be optimized. The method utilizes longer term historical data to identify thunderstorm initiation areas and then explores the corresponding attributes of various data fields, ranging from larger-scale numerical model output to small-scale information deduced from radar observations, that could be used in the forecast logic.

Examples of how this approach can be used to narrow down the key input fields and also lead to improved membership function and weight definition will be presented. It will also be suggested that a single forecast logic rule set is not ideal since variable environmental conditions lead to different modes of convective development which may be represented by distinctly different responses in the various predictor fields.