Projecting Cloud-to-Ground Lightning and Wildfire Ignition Risk in the Western United States

Cloud-to-ground lightning is a major source of summer wildfire ignition in the western United States (WUS). However, future projections of lightning are uncertain since lightning is not directly simulated by global climate models (GCMs). One approach is to use statistical parameterizations of lightning based on meteorological variables that are directly simulated by GCMs. We develop parameterizations using convolutional neural networks (CNNs) trained on observed data to project cloud-to-ground lightning occurrence in the WUS during June-September (warm season). CNN models are individually trained at each grid cell using spatial fields of three meteorological variables over the 1995-2022 period: the ratio of surface moist static energy (MSE) to the free-troposphere saturation MSE, 700-500 hPa lapse rate, and 500 hPa relative humidity to predict the occurrence of lightning (1 cloud-to-ground flash) at the daily scale. The CNNs achieve a domain-median correlation of 0.84 with observed lightning-day frequency. Using bias-corrected meteorological fields from CESM2-LENS, we find widespread increases in lightning days across much of the WUS by the late-21^st century under the SSP3-7.0 scenario. Projected increases are most pronounced across the interior Northwest and northern California, representing a doubling of warm-season lightning days in some locations, with decreases in some locations on the High Plains. We additionally project increases in Fire Weather Index values on future lightning days across most of the WUS, indicating a greater risk of lightning-caused wildfire ignition where sufficient fuels are available.