An Artificial Intelligence Lightning Threat Algorithm

Sydejko, J., Mercer, A., & Brown, M. E. (2010). An Artificial Intelligence Lightning Threat Algorithm. Northern Gulf Institute Annual Conference. Mobile, Alabama.

Abstract

Thunderstorms are a common occurrence in Mississippi in the late summer and fall months, but the lightning strikes associated with these storms are difficult to predict. According to the National Weather Service, cloud to ground lightning strikes are the second most deadly weather phenomenon nation-wide, killing up to 58 people per year. However, current lightning threat forecasts are limited to output from numerical weather prediction simulations. This method of lightning threat assessment is not useful when determining the threat for ground-based lightning activity, since the model identifies thunderstorms by their capacity to produce lightning, regardless of type (i.e. cloud to ground, cloud-to-cloud, or inter-cloud). In order to improve lightning threat assessment for Mississippi, an artificial-intelligence based lightning detection algorithm is under development. The algorithm uses several common thunderstorm forecast parameters (i.e. surface dewpoint, CAPE, CIN, surface divergence, etc.) from Weather and Research Forecasting model simulations of 10 different thunderstorm and non-thunderstorm days in Mississippi. This output, combined with data from the National Lightning Detection Network, is used to train a support vector machine classification algorithm. Individual probabilities of cloud-to-ground lightning strikes at each grid point in the entire state are computed, resulting in a spatial probability of cloud-to-ground lightning strikes. Overall, initial development and testing has met with modest success, though the algorithm seems to be highly biased towards high lightning probability in the southern third of the state, an issue likely a result of sample size.