Assessing Flood Early Warning Systems for Flash Floods

Flash floods are an increasing concern, especially in regions with abrupt topography and small areas where floods are rapid and energy-filled. That is the case of the El Guamo stream basin located in Manizales, Colombia. It has been proposed a duration-independent rainfall threshold for flash floods in this basin, using a hydrodynamic method that links critical water stages to cumulative rainfall. This paper presents a systematic literature review of 19 case studies from 2016 to 2021 to compare and highlight complexities and differences in the methods used in rainfall threshold estimation in both the El Guamo stream basin as in other case studies. The results identified four types of methods: (i) empirical, (ii) hydrological/hydrodynamic, (iii) probabilistic, and (iv) compound. Each method identified the principal indicators and their predictor variables. Each method uses different indicators, such as accumulated rain, accumulated antecedent rainfall, intensity-duration of the rain event, maximum cumulative or cumulative rainfall depth for a specific duration, and critical rainfall within given time periods. Scenario analysis of the predictor variables is a common approach used in rainfall threshold estimation. Some predicting variables found are antecedent moisture conditions, inundation criteria, and synthetic hyetographs. Some case studies include a probabilistic analysis of the predictor variables. This article concludes that indicators and their predicting variables can be adjusted to local flood early warning systems depending on the rainfall threshold method selected. Hydrodynamic models are solid in rainfall threshold estimation. However, it is highly advisable to include uncertainty analysis and new data sources to have more robust rainfall thresholds. Furthermore, probabilistic methods, including uncertainty analysis with utility functions, are a valuable tool to improve decision-making in early warning systems, which can help to refine the rainfall threshold estimation.