190-7 Explainable GeoAI Framework for Landslide Prediction and Decision Support

Abstract:

Machine learning (ML) has considerably advanced landslide susceptibility assessment over the past decade by improving predictive accuracy through ensemble learning, hybrid optimization algorithms, and the integration of multi-source geospatial data. Despite these advances, most existing models remain difficult to interpret, offering limited insight into the geologic and environmental processes governing landslide occurrence and thereby restricting their application in operational hazard management. As GeoAI becomes progressively integrated into geoscientific research, improving model transparency and interpretability has emerged as a critical challenge. Building on prior developments in GIS-based landslide susceptibility modeling and ensemble machine learning, this study proposes an Explainable GeoAI (XGeoAI) framework that shifts the emphasis from prediction alone to interpretable, actionable geohazard intelligence. The proposed framework integrates topographic, geological, hydrological, climatic, land-use, and remote-sensing datasets with historical landslide inventories into a unified geospatial workflow. ML models, including Random Forest, XGBoost, and LightGBM, are coupled with explainable AI techniques to quantify both global and local feature contributions, identify dominant triggering mechanisms, and characterize prediction uncertainty. Unlike conventional susceptibility mapping approaches that primarily assess model performance using statistical accuracy measures, the proposed framework embeds explainability throughout the modeling workflow to support scientific interpretation and evidence-based decision-making. The framework produces interpretable susceptibility maps, identifies spatially varying controlling factors, evaluates uncertainty, and enables scenario-based analyses under changing environmental and climatic conditions. The framework represents a transition from black-box prediction to transparent geoscientific reasoning. Therefore, providing a scalable decision-support framework for early warning, infrastructure planning, land-use management, and climate-resilient disaster risk reduction in mountainous environments while advancing the next generation of GeoAI applications in geoscience.

Geological Society of America Abstracts with Programs. Vol. 58, No. 5, 2026

doi: 10.1130/abs/2026AM-16880

© Copyright 2026 The Geological Society of America (GSA), all rights reserved.