Abstract
The application of artificial intelligence (AI) and machine learning in geohazard modelling has been rapidly growing in recent years, a trend that is observed in several research and application areas thanks to recent advances in AI. As a result, the increasing dependence on data driven studies has made its practical applications towards geohazards (landslides, debris flows, earthquakes, droughts, floods, glacier studies) an interesting prospect. These aforementioned geohazards were responsible for roughly 80% of the economic loss in the past two decades caused by all natural hazards. The present study analyses the various domains of geohazards which have benefited from classical machine learning approaches and highlights the future course of direction in this field. The emergence of deep learning has fulfilled several gaps in: i) classification; ii) seasonal forecasting as well as forecasting at longer lead times; iii) temporal based change detection. Apart from the usual challenges of dataset availability, climate change and anthropogenic activities, this review paper emphasizes that the future studies should focus on consecutive events along with integration of physical models. The recent catastrophe in Japan and Australia makes a compelling argument to focus towards consecutive events. The availability of higher temporal resolution and multi-hazard dataset will prove to be essential, but the key would be to integrate it with physical models which would improve our understanding of the mechanism involved both in single and consecutive hazard scenario. Geohazards would eventually be a data problem, like geosciences, and therefore it is essential to develop models that would be capable of handling large voluminous data. The future works should also revolve towards interpretable models with the hope of providing a reasonable explanation of the results, thereby achieving the ultimate goal of Explainable AI.