Hybrid Machine Learning-Geostatistical Framework for Sustainable Spatiotemporal Temperature Forecasting

Temperature plays a key role in climate systems, with significant implications for environmental sustainability, agricultural productivity, and water resource management. Accurate spatiotemporal forecasting of temperature is, therefore, essential for supporting sustainable development and informing climate adaptation strategies. However, traditional modeling approaches often treat spatial and temporal dimensions separately, thereby limiting their ability to capture the full spectrum of spatiotemporal dependencies. This paper proposes a hybrid approach that integrates a machine learning model with a geostatistical prediction technique to forecast daily mean air temperature over the study area. The best-performing spatiotemporal correlation model is selected among various time series and machine learning models including Holt–Winters, Seasonal Autoregressive Integrated Moving Average (SARIMA), Neural Network Autoregression (NNAR), and Artificial Neural Networks (ANNs). The findings demonstrate that the proposed hybrid approach consistently outperforms traditional, non-integrated methods. Importantly, this study contributes to sustainability by enabling high-resolution temperature forecasting that supports climate-resilient agriculture, efficient water resource allocation, and evidence-based environmental management. The generated predictive maps provide actionable insights for policymakers and stakeholders, enhancing adaptive capacity and promoting sustainable resource management under changing climate conditions.