A Deep Learning Model of Lightning Stroke Density

Abstract

Lightning plays a crucial role in the Earth's climate system, yet existing parameterizations for use in forecasting and earth system models show room for improvement in capturing spatial and temporal variations in its frequency. This study develops deep learning-based parameterizations of lightning stroke density using meteorological variables from the ERA and IMERG datasets. Convolutional neural networks (CNNs) with U-Net architectures are trained using World Wide Lightning Location Network (WWLLN) data from 2010 to 2021 and evaluated on WWLLN lightning observations from 2022 and 2023. The CNNs reduce the average domain mean bias by an order of magnitude and produce significantly higher Fractions Skill Score (FSS) values across all lightning regimes compared to the multiplicative product of CAPE and precipitation. The CNNs show skill relative to previously published parameterizations over the oceans especially, with r2 values as high as 0.93 achieved between the best performing modeled and observed lightning stroke density climatologies. The CNNs are also able to accurately capture the 12-hourly evolution of lightning spatial patterns on an event-scale with high skill. These results show the potential for deep learning to improve on lightning parameterizations in weather and earth system models.