Data Science Internship 2 with CERC/SLF
Data Science Internship with external pageCERC/external pageSLF
Simulating streamflow and hydrologic extreme events such as floods and droughts is challenging, particularly in basins with few or no streamflow observations. Machine learning based models represent an alternative to classical hydrological models and have been shown to have good predictive performance in catchments both with and without observations. Specifically, long short-term memory (LSTM) models are well suited for the problem because they can learn long-term dependencies between the input and output of the network, which is essential for representing the long memory of hydrological systems. LSTMs have previously also been successfully used to predict streamflow, especially when they were trained not just locally on meteorological time series but regionally using catchment attributes in addition to meteorological time series. Such regionally trained models do not just improve predictive performance locally but also enable predictions in ungauged basins, i.e., catchments without streamflow observations. The project aims to set up an LSTM for Switzerland using a recently published large-sample dataset of streamflow, meteorological time series, and catchment attributes for Switzerland. It sets up the model structure, trains the model on available data, and evaluates model performance for catchments with and without streamflow observations.
Qualifications required: MSc or BSc (ETH students only); excellent machine learning skills; strong programming skills (preferably R or Python); experience in working with large datasets, ideally with climate and hydrological data.
Earliest project start: November 2023.
Duration: 6 to 8 weeks; time in Davos to be discussed.
Financials: Train travel to/from and accommodation in Davos are covered. A CHF 800 stipend will be provided.
More information and application: Dr. Raul Wood, .