Leonardo Santos, Luiz Fernando Satolo, Ricardo Oyarzabal, Elton Escobar-Silva, Michael Diniz, Rogério Negri, Glauston Lima, Stephan Stephany, Jaqueline Soares, Johan Duque, Fernando Saraiva-Filho, Luiz Bacelar

Background: Flash flood modeling faces many challenges since physically-based hydrological models are unsuitable for a small spatiotemporal scale. With the increased availability of hydrological observed data, an alternative approach is to use
Machine Learning (ML) techniques. This work conducts a Systematic Literature Review (SLR) to enhance our comprehension of the research landscape on ML
applications for modeling flash floods.
Methods: Starting with more than 1,217 papers published until January 2024 and indexed in Web of Science, SCOPUS/Elsevier, Springer/Nature, or Wiley databases, we selected 53 for detailed analysis, following the PRISMA guidelines. The inclusion/exclusion criteria removed reviews, retractions, and papers that were not in the scope of this SLR and included only papers with time resolution coarser than 12 hours. Data about forecasting horizon, area, method and input were extracted from
each study to identify which ML techniques and model designs have been applied to flash flood forecasting.
Results and Discussion: There has been a notable increase in publications investigating ML techniques for flash flood modeling over the last few years. Most studies focus on regions in China (36%) and the United States (11%). Of the total of selected papers, more than 90% used as input data just one or an exclusive combination of the following measurements: discharge, rainfall, and water level. From this set, the combination of discharge and rainfall appears in almost half of the papers.
Notably, almost 60% of the studies utilize the long short-term memory (LSTM) method. No one method always performs better than any other in the selected papers. Unfortunately, less than 10% of selected articles provide access to their data. To further explore the potential of ML approaches in flood forecasting, we recommend their integration into early warning systems, development and dissemination of benchmarks, publication of successful case studies, and multidisciplinary collaboration.