<![CDATA[Geohazards such as earthquakes, landslides, and volcanic eruptions pose severe threats to human life and infrastructure, causing significant global losses every year. Existing hazard assessment methods are limited by single-hazard focus, high computational cost, sparse data integration, and poor real-time forecasting capabilities, which limit their operational use. This study aims to develop a unified artificial intelligence (AI) framework for multi-hazard forecasting by integrating convolutional neural networks (CNNs), long short-term memory (LSTM) models, random forest classifiers, and ensemble fusion techniques. A multi-source dataset consisting of seismic, geospatial, and geochemical data was processed using an 80/10/10 split train-validate-test, cross-validation, and spatial validation strategies. The results show strong performance, with earthquake classification AUC-ROC of 0.961, magnitude prediction RMSE of 0.23 Mw, landslide sensitivity AUC of 0.957, and volcanic classification accuracy of 91.2%, outperforming several state-of-the-art benchmarks. Ensemble fusion improved performance by 2.1–3.7% over individual models. The key contribution is a scalable ensemble-based AI framework that enables integrated multi-hazard forecasting on heterogeneous datasets. However, limitations include information heterogeneity and reduced cross-regional generalizability. The framework supports real-time early warning systems, disaster risk management, and land-use planning, especially in hazardous areas.]]>
Copyrights © 2026
