On the Application of Machine Learning into Flood Modeling: Data Consideration and Modeling Algorithm

This article reviews the literature on the application of Machine Learning (ML) to identify flood-prone areas, covering studies published since 2013. The review focuses on data considerations, such as the specifics of the study area and conditioning factors, as well as the ML algorithms used to identify flooding areas. 100 scientific articles were analyzed through a wide scope of geographical areas, ranging from arid to tropical climates and from small catchments to large river basins, to evaluate the influence of geographical features, historical flood occurrences, climatic patterns, urbanization, and data availability on flood susceptibility modeling (FSM). Iran, India, China, and Vietnam are the most frequently studied locations. The slope of the land, topographic wetness index, land use and land cover, rainfall levels and distance to rivers were key conditioning factors in at least 61% of the reviewed articles. Furthermore, the employed ML algorithms can be categorized into various types: statistical, kernel-based, tree-based, Neural Network (NN)-based, ensemble, and hybrid approaches. NN-based models, such as Long Short-Term Memory (LSTM) and Recurrent Neural Networks (RNNs), excel in solving high-dimensional problems but face challenges related to reliability and overfitting. Kernel-based approaches require optimal configuration through a trial-and-error process, while tree-based models offer simplicity and are less prone to overfitting, although they may be less precise. Among these, ensemble and hybrid models generally outperform traditional ML methods, despite their own limitations. These methods primarily focus on event-based historical floods, limiting their ability to make real-time predictions due to the lack of time-series data. Additionally, most models face restrictions given data consistency and validity. They often use inconsistent data, where flood conditions and input parameter values are not aligned in time and space. This discrepancy undermines the models' reliability. Consistent and valid datasets are imperative for accurate model development.