Auto-Nowcaster (ANC) is an automated algorithm that is capable of forecasting thunderstorm initiation. However, its parameters have to be tuned to regional environments, a process that is time-consuming, labor-intensive and quite subjective. When the National Weather Service decided to employ the ANC in all the forecast offices, therefore, a faster, less labor-intensive and objective mechanism to tune the parameters was sought.

In this paper, a genetic algorithm approach to tuning the ANC is described. The process consisted of choosing data sets, employing an objective forecast verification technique and a fitness function. The ANC was modified to operate create forecasts offline using parameters iteratively generated by the genetic algorithm. The parameters are generated by probabilistically combining parameters that result in better performance, leading to better and better parameters as the tuning process proceeds.

The forecasts created by ANC using the automatically determined parameters are compared with the forecasts created by ANC using parameters that were the result of human tuning. It is shown that forecasts created using the automatically tuned parameters are better than the ones created through human tuning. In addition, automated tuning can be done in a fraction of the time that it takes human forecasters to analyze the data and tune the weights.