Comprehensive analysis and predictive modeling of temperature and rainfall patterns in Southern Telangana

Abstract

This study presents a data-driven analysis of temperature and rainfall patterns across 12 districts in Southern Telangana using 43 years of gridded meteorological data (1981–2023). Employing descriptive statistics, correlation matrices, and six predictive models, Random Forest (RF), Artificial Neural Network (ANN), Support Vector Regression (SVR), Long Short-Term Memory (LSTM), ARIMA, and TBATS, we evaluated forecasting accuracy for maximum temperature, minimum temperature, and rainfall.The RF model demonstrated superior performance with the lowest Test RMSE (0.1178) and Test MAE (0.0601) across all parameters, outperforming traditional time series models. Correlation analysis revealed strong inter-location temperature synchrony (r  0.98–1.00), while rainfall exhibited high spatial variability (r = 0.15–0.77), indicating localized climatic influences. Feature importance analysis identified L332 and L333 as dominant predictors, with scores of 0.1209 and 0.0557, respectively. Novel contributions include: (1) a comparative evaluation of six models on long-term regional climate data, (2) integration of feature importance to enhance interpretability, and (3) prediction interval analysis confirming model stability with consistent upper bounds (~0.187) and zero lower bounds. These findings offer actionable insights for climate adaptation, agricultural planning, and resource management in semi-arid regions.