AI-powered digital twin of India’s climate using multi-sourcenational datasets: A machine learning-based prototype

Abstract

Climate variability and extreme weather events create significant challenges for agriculture, water resources, infrastructure, disaster management, and public safety in India. Effective climate intelligence requires the integration of historical meteorological observations, machinelearning-based prediction, geospatial visualization, and decision-support capabilities. This paper presents an AI-powered climate Digital Twin prototype that integrates Indian Meteorological Department (IMD) climate datasets with machine-learning prediction and an interactive webbased visualization environment. The prototype uses long-term IMD gridded rainfall and temperature datasets and provides an AI-assisted temperature prediction component using meteorological variables including humidity, rainfall, and wind speed. The system is implemented as a web-based climate intelligence platform with interactive geospatial visualization and report-generation capabilities. An experimental Random Forest regression model achieved an MAE of 2.4285 °C, an RMSE of 3.4855 °C, and an R² of 0.5951 on the available test dataset. A preliminary rainfall regression experiment produced an R² of approximately 0.002, indicating that the current rainfall baseline requires substantial improvement before it can be considered suitable for operational prediction. Therefore, the proposed system is positioned as a prototype rather than a complete national-scale climate Digital Twin. The results demonstrate the feasibility of combining national climate datasets, machine learning, and geospatial decision-support interfaces while identifying important directions for improved spatiotemporal modelling, uncertainty quantification, and nationalscale deployment.