A Survey of Flood and Landslide Prediction Using Machine Learning

Floods and landslides are highly destructive natural disasters, causing severe damage to lives, infras- tructure, and economies. Their increasing frequency and intensity, driven by climate change, highlight the urgent need for advanced predictive systems to mitigate their impact. This survey examines the application of machine learning (ML) techniques for flood and landslide predic- tion, utilizing diverse data sources such as meteorological records, soil conditions, topography, remote sensing im- agery, and historical incidents. Various ML models, including Random Forest (RF), Convolutional Neural Networks (CNNs), and Attention- UNet, are reviewed for their effectiveness in risk assess- ment, spatial mapping, and prediction accuracy. Tradi- tional models like RF provide robustness and simplicity, while advanced architectures like Attention-UNet excel in capturing complex spatial dependencies, making them ideal for high-resolution disaster mapping. Hybrid and ensemble models further enhance prediction reliability by overcoming the limitations of individual techniques. The integration of real-time sensor data and transfer learning improves model adaptability to dynamic and data-scarce environments. These systems offer actionable insights, empowering policymakers and emergency respon- ders to optimize resource allocation, plan mitigation strate- gies, and enhance disaster preparedness. Moreover, ML applications in disaster management highlight the poten- tial of interdisciplinary approaches, combining geospatial analysis, environmental science, and artificial intelligence. This survey underscores the transformative potential of ML in advancing flood and landslide prediction. By addressing challenges like data scarcity and computational complexity, it aims to support the development of more accurate, scalable, and efficient disaster management solu- tions essential for building resilient communities in an era of increasing environmental risks