Climate Risk Prediction System

Abstract

This paper presents ClimateGuard, a localized, explainable, multi-hazard climate risk prediction system integrating satellite-derived geospatial indicators, real-time meteorological data, and machine learning for climate hazard forecasting. The system combines Random Forest and XGBoost models with environmental and disaster datasets, including NASA FIRMS and EM-DAT, while incorporating Google Earth Engine for satellite-derived vegetation indicators and SHAP for model interpretability. A web-based architecture integrates real-time climate data, predictive analytics, explainable AI, and cloud-based logging to provide localized climate risk insights and support disaster preparedness and resilience.