Reconstructing of the German summer floods 2021 using the data-driven methods

Ronak Rahmati

Bachelor's thesis

Abstract:

Heavy rains caused floods in the Eifel/Ahr region of Germany on July 14–15, 2021, killing 184 people and severely damaging infrastructure such as houses, roads, and communications. Under current climatic conditions, a similar event is expected to occur approximately every 400 years in certain parts of Western Europe. This thesis aims to quantify the ability of two existing data-driven (Neural Network) and one hydrological (GR6J) model to reconstruct the record-breaking flood events, which was caused by daily rainfall accumulation exceeding 100 mm. We have used observed precipitation and temperature data derived from different meteorological data sets and observed discharge time series on the Ahr river basin to set up and run the Neural Network (BRNN and LSTM) and hydrological (GR6J) models for the period 1992–2021. The first half was used for calibration/- model training and the latter half was used for independent evaluation. The comparison of simulated results to observed discharge data has shown that the GR6J results outperform the simulations of the BRNN and LSTM setups, particularly in the validation period across all four meteorologic forcing data sets. However, if the data-driven models (BRNN and LSTM) are trained on high daily accumulations and exceptional flood peaks, they can capture flood signals and, in some cases, the flood apex.

Keywords

model setup, GR6J, brnn, floods, neural network, data analysis, July 2021, Ahr

Language used: English

Date on which the thesis was submitted / produced: 2. 4. 2023

Thesis defence

• Date of defence: 22. 5. 2023
• Supervisor: Ing. Oldřich Rakovec, Ph.D.
• Reader: Roman Kožín, externi

Czech University of Life Sciences Prague

Faculty of Environmental Sciences

Bachelor programme / specializace:
Environmental Data Science / Informatics