Predicting Lightning from Near-Surface Climate Data in the Northeastern United States: An Alternative to CAPE

Lightning is a critical driver of natural wildfire ignition and ecosystem dynamics, but existing prediction models rely on upper-air predictors such as convective available potential energy (CAPE) that are absent from paleoclimate reconstructions. To enable long-term reconstructions of lightning activity, we developed and evaluated statistical models based solely on near-surface climate variables: temperature, precipitation, humidity, surface air pressure, wind, and shortwave radiation. Using ERA5 reanalysis and Vaisala Lightning Detection Network data (2005–2010) for the Northeastern United States, we compared linear regression, gamma generalized linear models, and Bayesian gamma models against CAPE-based benchmarks. While CAPE-based models outperformed models based on individual near-surface predictors, they showed limitations when predicting temporal anomalies. Models incorporating multiple near-surface predictors consistently outperformed CAPE-based models, reproducing observed spatial gradients, interannual variability, and strike rate distributions. Gamma generalized linear models achieved the strongest overall performance, balancing realistic, non-negative predictions with accuracy across error- and correlation-based metrics, while Bayesian models better captured the distribution of strike rates but sacrificed spatial precision. Our results demonstrate that near-surface predictors provide a viable alternative for lightning prediction when upper-air data are unavailable, providing a methodological pathway for reconstructing long-term seasonal lightning variability and its role in climate-fire interactions.