Machine Learning (LTM) and rainfall runoff modelling
Introduction
Are you interested in this MSc topic? Send an email with your CV to Jaap.Kwadijk@deltares.nl, m.j.booij@utwente.nl, ruben.dahm@deltares.nl
Participating Institutes
- Deltares
- University of Twente
Areas Under Investigation
- The Rhine basin
- Other basins
- A thesis addressing the research questions
- Python scripts and datasets used for analysis
- Reproducibility of the results
- Successfully completed courses on hydrology and/or hydrological modeling and/or water resources management
- A reasonable acquaintance with Python or other scripting languages for programming
- Kratzert, F., Nearing, G.S., & Klotz, S. (2019). Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets. Water Resources Research, 55(1), 478-502.
- CAMELS: catchment attributes and meteorology for large-sample studies. Retrieved from https://camelsscience.org/
- Nearing, G.S., Kratzert, F., & Klotz, S. (2020). What Role Does Hydrological Science Play in the Age of Machine Learning? Water Resources Research, 56(1), e2019WR026347.
Rationale and Objective
Hydrological modeling has traditionally integrated accumulated knowledge of water stores and physical processes into schematizations. However, recent research by Kratzert et al. [1] has demonstrated that an LSTM neural network, trained on the publicly available CAMELS dataset [2], can outperform classical hydrological models. This study suggests that machine learning can play a significant role in hydrological science.
The objective of this thesis is to explore the applicability of these machine learning methods in various hydrological contexts, including the Rhine basin and other basins. Specifically, the goal is to determine when and where these methods are most effective compared to traditional models, identify their limitations, and develop practical applications beyond the scope of existing training datasets.
Approach
The student will begin by reviewing relevant literature on the subject. Python scripts will be developed to prepare suitable training datasets for the LSTM model. Model performance will be evaluated against traditional models and observational data, focusing on both high and low flow conditions.