Lightning Parameterization and Prediction: Conventional and Data-Driven Approaches

Abstract

Lightning is a key component of Earth’s weather and climate system, and its impact on our daily life will be increasingly important going into the future. An accurate lightning forecast is therefore critically important. However, lightning is still poorly represented in the numerical weather and climate models, because the physical mechanisms of lightning are not yet fully understood partially due to the lack of high-quality lightning observations. The recent progress in the lightning observations and machine learning (ML) techniques have provided a new opportunity to improve the understanding of lightning mechanisms and to improve the representation of lightning in numerical models. By taking advantage of the rich lightning observations from World-Wide Lightning Location Network (WWLLN), this study aims to better understand the relationship between lightning and environmental conditions and to explore the potential for improving lightning prediction using data-driven approaches. I investigate the relationship between convective available potential energy (CAPE), precipitation, the number and size of storms and overshooting tops, and lightning stroke density ($f$) over the Central America region. It is found that the storm size required to produce lightning appears to be disproportionately high in the low-CAPE regime, which leads to a CAPE-threshold feature for lightning over the ocean. I show that applying the CAPE threshold for lightning over the ocean can improve the performance of a CAPE-based lightning parameterization scheme. Various ML-based lightning parameterization methods are implemented by using grid-scale variables available from reanalysis data as inputs. By using 10 years of data, our results show that the ML-based lightning parameterization methods are able to outperform an empirical lightning parameterization method in terms of capturing the spatial and temporal variability of lightning. Lastly, I will show that implementing the ML-based lightning parameterization methods can also help improve the forecast skills of $f$ in the weather models.