With the recent addition of the Alaska Region Operational Network (ARONET), the operational forecaster has the opportunity to access a wealth of real-time meteorological data. However, the sheer volume of information available on ARONET raises the problem of how to assimilate and analyze this data in an expedient and consistent manner. Two successful Artificial Intelligence techniques offer a possible solution to this problem. The expert system technique captures subjective forecast knowledge from weather experts and allows the consistent application of this knowledge. In contrast, neural network techniques excel at pattern recognition and do not require any prior knowledge of a solution. Used in combination, these techniques appear to provide a highly useful forecaster decision assistance tool.
The Partners Project between the University of Alaska and the NWS Fairbanks forecast office was designed to develop such a tool to be used in forecasting lightning strikes over the interior of Alaska during the summer. A prototype neural network (NN) system was developed to automatically extract data from various NGM files, filter the data, and supply it to the trained neural network. Four neural networks for four different regions were designed, developed and tested for lightning prediction. From the test results, it appeared that the NN was able to classify patterns to a great degree of accuracy. However, for certain cases, the NN was not able to perform well, probably because it had not been trained with enough days of lightning strike data.