Precipitation-runoff models are used for operational flood forecasting at local, national, continental and global scales. In Norway, The Norwegian Water Resources and Energy Directorate (NVE) is responsible for the national flood forecasting published at varsom.no and xgeo.no. The forecast is based on the conceptual HBV and DDD models applied to 145 indicator catchments across Norway. These models use precipitation and temperature observations and forecasts as input.
Mathematical representation of physical processes aims to mimic how the catchments transform precipitation into runoff. Important processes that are represented include snow accumulation and melt, infiltration to soil moisture zone, evaporation, percolation to deeper groundwater, and runoff. These models are in most applications tuned to individual catchments. For application in ungauged catchments, the performance of these models worsens.
In recent years machine Learning and deep learning models have evolved as a promising alternative the traditional precipitation-runoff models.
In particular Long Short-Term Memory networks (LSTM)-models have been shown to outperform traditional hydrological models in ungauged catchments (Arsenault et al., 2023) and represent well floods. LSTM-models were introduced by Hochreiter and Schmidhuber (1997). These models have built-in feedback connections that make the model capable of learning the dependency structure of streamflow time series and are therefore well suited for simulation and forecasting. LSTM-models were introduced to the hydrological community by Kratzert et al. (2018, 2019) where the improvement over traditional hydrological models is demonstrated.
This master project will apply LSTM-models to simulate and forecast streamflow at both gauged and ungauged locations in Norway. The models will be applied at hourly and/or daily time scales. Their performance will be evaluated with a special focus on floods and compared to the performance of traditional hydrological models and the LSTM-based flood forecast provided by Google. Connections between the internal structure of LSTM models and hydrological processes can also be investigated (Lees et al., 2022).
The modelling will be based on Python packages from neural hydrology (https://neuralhydrology.github.io/). Datasets will be established in collaboration with NVE.
- Arsenault, R., Martel, J.-L., Brunet, F., Brissette, F., and Mai, J.: Continuous streamflow prediction in ungauged basins: long short-term memory neural networks clearly outperform traditional hydrological models, Hydrol. Earth Syst. Sci., 27, 139–157, https://doi.org/10.5194/hess-27-139-2023, 2023.
- Hochreiter, S. and Schmidhuber, J.: Long Short-Term Memory, Neural Comput., 9, 1735–1780, https://doi.org/10.1162/neco.1997.9.8.1735, 1997.
- Kratzert, F., Klotz, D., Brenner, C., Schulz, K., and Herrnegger, M.: Rainfall–runoff modelling using Long Short-Term Memory (LSTM) networks, Hydrol. Earth Syst. Sci., 22, 6005–6022, https://doi.org/10.5194/hess-22-6005-2018, 2018.
- Kratzert, F., Klotz, D., Herrnegger, M., Sampson, A. K., Hochreiter, S., and Nearing, G. S.: Toward improved predictions in ungauged basins: Exploiting the power of machine learning, Water Resour. Res., 55, 11344–11354, https://doi.org/10.1029/2019WR026065, 2019a.
- Lees, T., Reece, S., Kratzert, F., Klotz, D., Gauch, M., De Bruijn, J., Kumar Sahu, R., Greve, P., Slater, L., and Dadson, S. J.: Hydrological concept formation inside long short-term memory (LSTM) networks, Hydrol. Earth Syst. Sci., 26, 3079–3101, https://doi.org/10.5194/hess-26-3079-2022, 2022.