A machine learning framework and Granger causal links of climatic variables over Jammu & Kashmir (1972–2024)

This study analyzes long-term meteorological records (1972–2024) from Srinagar, Gulmarg, Qazigund& Jammu, Jammu & Kashmir to investigate climatic anomalies using statistical diagnostics, causal inference, and frequency extraction techniques. Multiple machine learning (ML) models—including Gradient Boosting (GBM), Extreme Gradient Boosting (XGBoost), Random Forest (RF), Multi-Layer Perceptron (MLP), and Linear Regression—were applied to quantify sunshine duration. Model comparison revealed that Linear Regression provided the most consistent performance (R 2 =0.57, RMSE = 2.20; CV mean R 2 =0.65), highlighting the dominance of linear dependencies with diurnal temperature range, relative humidity, and mean sea level pressure.Beyond predictive modeling, Granger causality tests demonstrated significant linkages between scalar momentum flux (derived from MSLP data) and key hydroclimatic variables. Scalar flux showed no predictive power for rainfall at lags 1–3, Rolling cross-correlation further revealed that the strength and direction of flux–rainfall associations varied across decades, with low-frequency variability dominating outside imputed data periods. Together, these findings show that sunshine duration and rainfall variability in the Himalayan climate are primarily governed by simple meteorological relationships and scalar flux dynamics, rather than complex nonlinear or autocorrelated processes. The results have potential applications in the renewable energy sector, particularly for improving solar energy generation forecasts by leveraging robust sunshine duration estimates and causality-based diagnostics of atmospheric fluxes.Granger causality applied to ENSO and NAO indices with NOAA OLR data demonstrates the temporal direction of influence on tropical convection, while Kullback–Leibler (KL) divergence quantifies the extent to which OLR distributions differ between positive and negative phases. Together, these diagnostics provide complementary insights into both predictive relationships and structural impacts of teleconnections on the radiation field.