A Two‐Dimensional Deep Learning Scheme With One Predicator Only to Parametrize Global Lightning

Ming Yin, Yong Wang, Yong Wang, Yun Fang, Feng Li, Xiushu Qie, Yeying Wang, Yeying Wang, Daoming Wei, Huihui Yuan, 冀永萍, Yu Zhang et al.

Geophysical Research Letters · 02 April 2026 · China Meteorological Administration (CMA), Fudan University (FuXi), Institute of Atmospheric Physics, Chinese Academy of Sciences, Tsinghua University (NowcastNet)

Paper

In one sentence. A lightweight 2D deep learning lightning scheme using only CAPE raises global lightning R² to 0.89, a 24% gain over prior ML schemes, and better captures extremes.

Abstract

Abstract Lightning simulation has long posed significant challenges. This study presents a novel two‐dimensional artificial intelligence‐based global lightning scheme that uses a single predictor, convective available potential energy (CAPE). The new scheme significantly improves global lightning simulation performance, achieving a determination coefficient of 0.89, which represents a 24% increase over an existing machine learning‐based global lightning scheme. Additionally, it achieves a 41% reduction in absolute bias and a 38% decrease in root mean square error. Crucially, the scheme effectively alleviates the underestimation of extreme lightning density predicted by all the current lightning schemes due to nonlocal feature information incorporated. This finding indicates that extreme lightning simulation hinges on the movement of a convective system from neighboring grid cells or the presence of a large convective system encompassing multiple grid cells. As a lightweight deep neural network, it shows promising potential for implementation in global climate models and broader applications.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.