Privacy Policy Disclaimer
  Advanced SearchBrowse




Book Chapter

Open-Access Precipitation Networks and Machine Learning Algorithms as Tools for Flood Severity Prediction


Imagiire,  Luis O. K. M.
External Organizations;


Mester,  Benedikt
Potsdam Institute for Climate Impact Research;

Haun,  Stefan
External Organizations;

Seidel,  Jochen
External Organizations;

External Ressource
No external resources are shared
Fulltext (public)
There are no public fulltexts stored in PIKpublic
Supplementary Material (public)
There is no public supplementary material available

Imagiire, L. O. K. M., Mester, B., Haun, S., Seidel, J. (2022): Open-Access Precipitation Networks and Machine Learning Algorithms as Tools for Flood Severity Prediction. - In: Kolathayar, S., Mondal, A., Chian, S. C. (Eds.), Climate Change and Water Security, (Lecture Notes in Civil Engineering ; 178), Singapore : Springer Nature, 131-142.

Cite as: https://publications.pik-potsdam.de/pubman/item/item_26370
During the past decades, convective rainfall in the Echaz catchment, which is characterized by steep topography and a high degree of urbanization, led to recurring flash floods. The high spatial and temporal variability of precipitation, in combination with the small drainage basin, contribute to low predictability and to the considerable damage potential of such events. The aim of this study is the development of a simple model to predict flood severity in the Echaz catchment. This model is based on open-access precipitation data from a personal weather station (PWS) network and water level measurements from a low-cost ultrasonic sensor. Machine learning classification methods (logistic regression and decision tree) are trained with observational data to determine maximum rainfall thresholds for different accumulation periods, ranging from 5 to 60 min. Hence, the proposed model uses multiple triggers to predict the exceedance of critical water levels. As a result, severe floods can be recognized earlier and with higher reliability, providing more response time for local authorities. Although the limited data availability increases the risk of overfitting and lower performance for the first upcoming events, the model quality will increase with the incorporation of new measurement data in the future. The reduced complexity and high interpretability of the model allow for a fast decision-making process. Additionally, the model has high potential and can easily be adapted to similar small catchments.