Review of flash flood susceptibility modeling derived from machine learning algorithms, with input data from remote sensing sources

Flash floods can have devastating effects on life and property, especially when in areas underdeveloped for flood prevention and mitigation. There have been efforts to improve the overall state of flash flood susceptibility assessment through modeling methods. In this article, we present the current state of literature on flash flood susceptibility modeling (FFSM). The approach is based on a literature review focusing on FFSM derived through machine learning (ML) algorithms, with input data from remote sensing (RS) sources. The literature review consists of an analysis of forty-four peer-reviewed articles searched using the keywords: remote sensing, flash flood, susceptibility, detection, potential, prediction, forecast, deep learning, machine learning, and artificial intelligence. Examples of ML algorithms include Random Forest and Extreme Gradient Boosting. RS sources include satellite and aerial imagery. This review shows that FFSM derived through ML algorithms is accurate and reliable in determining susceptibility to flash flooding. Through a combination of multiple ML algorithms, a higher accuracy and reliability of FFSM can be observed. Common conditioning factors for model development involve relief slope, elevation, and basin size. Although focusing on many different geographical areas, most of these articles focused on rural areas consisting of small towns and villages. It is possible to augment current flash flood warning systems with the use of flash flood susceptibility modeling derived from machine learning algorithms with input data from remote sensing sources. This can aid future development or awareness of flooding hazards for urban areas.