An Artificial Neural Network–Based Predictive Framework For Forecasting Climate Change Impacts And Environmental Variability

Abstract

Climate change induces significant variability in meteorological parameters, impacting land use–land cover (LULC) dynamics and environmental sustainability. This study proposes an Artificial Neural Network (ANN)-based predictive framework to model and forecast climatic variables, including temperature, rainfall, humidity, and Air Quality Index (AQI), using long-term time-series data (1980–2022) obtained from the India Meteorological Department. The performance of ANN is comparatively evaluated against the Auto-Regressive Integrated Moving Average (ARIMA) model. Experimental results demonstrate that the ANN model achieves superior predictive accuracy, with lower residual errors across all climatic variables. For temperature prediction, the ANN model yields residual errors predominantly within ±2°C for most months, while rainfall prediction exhibits minimal deviations, with errors generally below ±3 mm except for peak seasonal variations. Quantitatively, the ANN model achieves reduced Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), along with a higher coefficient of determination (R² > 0.90), indicating strong model generalization. ARIMA shows higher variance in prediction accuracy due to its linear assumptions. LULC analysis reveals a consistent increase in built-up areas between 1988 and 2021, correlating with deteriorating air quality and altered climatic patterns. The integration of GIS-based spatial mapping with ANN forecasting provides enhanced capability for identifying climate-driven environmental transformations. Overall, the proposed ANN-based framework demonstrates robustness in capturing nonlinear climatic dependencies and offers an effective decision-support tool for climate impact assessment, environmental planning, and sustainable resource management.