Geospatial Artificial Intelligence for Flood Disaster Mapping in Sumatra: A Review of Machine Learning Models, Data, and Computational Workflows

Abstract

Flood disasters pose persistent socio-economic and environmental challenges, particularly in tropical regions such as Sumatra, Indonesia. Traditional hydrological and GIS-based approaches often struggle to capture complex interactions among terrain, rainfall, land use, and human activities. This review critically examines recent applications of Geospatial Artificial Intelligence (GeoAI) for flood disaster mapping, focusing on machine learning models, geospatial data sources, and computational workflows. Analysis of selected studies highlights that satellite imagery and digital elevation models remain dominant data inputs, while Random Forest, Support Vector Machines, Convolutional Neural Networks, and hybrid models are most frequently applied. Workflow patterns reveal recurring stages of data preprocessing, model training, and post-processing, yet gaps persist in model explainability, feature selection, and generalization across regions. The study underscores the importance of integrating multi-source data, standardizing workflows, and fostering interdisciplinary collaboration to enhance operational flood risk management. Findings provide a foundation for advancing GeoAI research and translating methodological innovations into practical flood preparedness and mitigation strategies.