Advancing flood risk estimation in ungauged basins with machine learning and climate data at the global scale

Abstract

Increasing flood risk due to urbanization and climate change poses a significant challenge to societies at global scale. Flood prediction across scales and more specifically in ungauged areas remains a great challenge that limits the efficiency of flood risk mitigation strategies and disaster preparedness. Machine learning (ML) based models have demonstrated a great potential for streamflow prediction. ML based procedures are relatively easier to apply and are less computationally demanding, especially for applications at regional scales, than traditional physics-based models. Thus, their application for hydrologic predictions have attracted a lot of interest from stakeholders in academia, industry and federal agencies.For streamflow prediction, these models perform very well at capturing streamflow variability but fail to accurately predict extreme values (i.e. peak flow) of flood events, which are important to be considered in flood design or for flood warning purposes. This work therefore examines an event-based predictive framework that is solely focused on peak flow prediction and considers the characteristics of the flood triggering precipitation, the catchment and antecedent wetness conditions. Analysis for multiple distinct hydroclimatic regions is presented across the contiguous US. Evaluation of the drivers of flood peaks noted distinct dependencies among the dynamic and static predictors considered in the models for flood peaks of different severity. Furthermore, a flood prediction framework is developed and tested in ungauged regions that relies on two fundamental components. First, meteorological data from satellite and reanalysis global datasets (IMERG and ERA5-Land, respectively) provide key input variables and second, ML models trained in the data-rich contiguous US, are applied in climatically similar regions in other parts of the world. Results indicate acceptable performance for both products providing a starting point from which more and improved ML procedures and precipitation datasets can be integrated to potentially address the PUB (Prediction in Ungauged Basins) problem at the global scale. Another remaining and critical need at the global scale is future flood risk estimation. Conventionally, hydrologic models are calibrated and used to facilitate mid- to long-term mitigation strategies. However, the disparity among the vulnerable populations and the resources required for this task calls for approaches that are simple to access and use. The advantageous properties of ML models, motivate an ML-hydrological model comparison to evaluate the capability of ML models to spatially provide information on future flood risk under various climate scenarios. A proposed ML model has been assessed as viable for this purpose; a result that could serve communities at global scale that lack the resources to access, develop and operate hydrologic models.