Answers are grounded in this paper and cite their sources.
Machine learning has recently emerged as a promising approach for predicting lightning activity by learning patterns from historical data. This study focuses on the prediction of lightning density using machine learning techniques over Northern India, region with high monsoon lightning activity. Models were trained using LIS/OTD 2.5° LRMTS and LRTS lightning data archived at GHRC as the target variable, with K-index, Convective Available Potential Energy (CAPE), Surface Sensible Heat Flux (SSHF), and 2-m air temperature (T2M), were sourced from the ERA5 reanalysis (ECMWF), as predictor variables for the period 2000-2011 and tested for 2012-2013. Three machine learning models, namely Decision Tree, Random Forest, and XGBoost, were developed separately for the monthly dataset using k-fold cross-validation with each year treated as one-fold. Before tuning, Random Forest (R 2 = 0.7781) and XGBoost (R 2 = 0.7482) outperformed the Decision Tree (R 2 = 0.6283). Ensemble performance was greatly enhanced by hyperparameter tuning, with XGBoost achieving R 2 = 0.84, followed by Random Forest (R 2 = 0.82). Similar hierarchy was seen in baseline performance for the daily dataset, with XGBoost (R 2 = 0.72) and Random Forest (R 2 = 0.72) significantly outperforming the Decision Tree (R 2 = 0.50), while a Long Short-Term Memory (LSTM) model was designed specifically for daily dataset achieving R 2 = 0.62. These findings show potential of machine learning, especially ensemble approaches for lightning prediction over Northern India. To improve operational lightning forecasting across India's various climatic regions, future research may use hybrid NWP–ML frameworks, multi-step deep learning architectures like Transformers, and higher-resolution lightning observations.
No takes yet. Share an insight, caveat, or question.
Jangid et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: