Project selection
Short-term power forecasting at run-of-river power plants - part 1: development
At the Austrian hydropower plant operator Verbund, a conceptual, spatially distributed hydrological model with a following hydrodynamic module is operationally used for the prediction of power produced by run-of-river power plants. Potential for improvement was identified especially in times of steeply increasing and decreasing flow rates.
The performance of machine learning methods in short-term power forecasting (up to 4 hours) was determined at the run-of-river power plants Braunau-Simbach (river Inn), Aschach (Danube) and Greifenstein (Danube). Measured runoff and power values from upstream run-of-river power plants and gauges were used as input for the machine learning models.
The achieved results are promising and the developed machine learning models will be incorporated into operational planning at Verbund after an evaluation phase. The documentation of the project was published in the open-assess ÖWAV journal. Link to article
Short-term power forecasting at run-of-river power plants - part 2: operation
In this project we applied the developed approach from the previous project (see reference above) on the whole Austrian Danube run-of-river power plant chain (10 power plants: Jochenstein, Aschach, Ottensheim-Wilhering, Abwinden-Asten, Wallsee-Mitterkirchen, Ybbs-Persenbeug, Melk, Altenwörth, Greifenstein, Freudenau | average total power approx. 1500 MW) and on 7 Inn run-of-river power plants (Oberaudorf, Nussdorf, Braunau-Simbach, Ering-Frauenstein, Egglfing-Obernberg, Schärding-Neuhaus, Passau-Ingling | average total power approx. 350 MW).
In contrast to the previous project, not the individual but the total power of 10 respectively 7 run-of-river power plants is forecasted. This procedure led to a faster project accomplishment and also to advantages in initializing a forecast. Prior to the model training, a preparation of the provided input time series was conducted. The created application for periodically calculating the forecasts includes several checks. In case of missing or unrealistic input values the application switches automatically to a back-up model. In conclusion, both forecast models for the Danube and the Inn has shown high accuracy in an independent test period.
Austria-wide prediction of runoff characteristics
The aquaZoom project, which was commissioned by the European Marine and Fisheries Fund and the Austrian Federal Ministry BMLRT, estimated the nationwide potential for aquaculture flow-through systems in Austria. Information about runoff characteristics, e.g. average minimum runoff, are not only an important basis for planning aquaculture flow-through systems, but also for various other applications. The calculation of runoff characteristics is simple if a runoff gauge with a correspondingly long observation time series is in the immediate vicinity of the area of interest. However, since the number of gauge stations is limited, baseflow AI was commissioned to predict six different runoff characteristics for approximately 8000 Austrian surface water bodies, including an estimate of the prediction uncertainty.
Machine learning methods have often shown significant improvements compared to conventional methods in a wide variety of hydrological applications. Therefore, after initial tests of several models, we used the XGBoost (eXtreme Gradient Boosting) algorithm to predict the runoff characteristics. Uncertainties in the predictions were determined with Quantile Random Forests (QRF). In addition to the data from all publicly available runoff gauges from Austria and its hydrological upstream regions, we also used a large number of (catchment area) properties with approximately 90 predictors for training the models.
The applied models achieved with a median deviation of around 20% between model predictions and observations in as ungauged treated basins a high performance. Even local artificial characteristics such as the increase in low water flows due to large reservoirs or the influence of cross-basin water transfers could be reproduced. Detailed information and results were published in the ÖWAV journal. Link to article. Our client has also agreed that the predicted runoff characteristics and the degree of uncertainty are publicly available free of charge. Link to dataset
“The cooperation with baseflow AI solutions was very successful and pleasant. The professional and solution-oriented processing by baseflow AI led to exactly the results that we needed.”
Dr. Florian Borgwardt (Senior Scientist)
Analysis of a chain of flood retention reservoirs
In the catchment area of a river in Amstetten (Lower Austria), four flood retention reservoirs were built. During the 2020 and 2021 flood events, the two retention reservoirs located in the lower part of the river exhibited significantly different behaviour than those two in the middle part. The reasons could be determined by detailed hydrological modelling and systematic analysis of measured data.
Based on the findings of the first part, structural and control modifications were worked out to optimize the retention reservoirs.
EFAS runoff forecasts
The pan-European flood warning system EFAS (European Flood Awareness System, spatial coverage 5000 x 4750 km, link) supports the EFAS partners with runoff forecasts for operational use. The runoff forecasts are initialized twice a day and have forecast ranges of +10 days (meteor. forcing ECMWF) respectively +7 days (DWD) with a six-hour increment. EFAS forecasts are provided in binary file format (NetCDF or GRIB).
viadonau investigates the possibility of using EFAS. Therefore, we were contracted to provide the following services: 1) Automated retrieval of EFAS raw data for the relevant spatial domain and the last 4 years via API. 2) Extracting and processing the runoff forecasts with 40 (ECMWF) respectively 28 (DWD) forecast levels per initialization date for two locations. This step was done by an algorithm including checks for data completeness. 3) Creating the output files with a format suitable for viadonau.
Meteorological forecasts for building automation
As part of the Clean Energy Hackathon, we developed for STIWA a concept for the continuous integration of meterorological forecast data into building automation. This enables to optimize the building automation on several levels. For example, the heating / cooling of the building can be controlled more efficiently and a more accurate peak shifting of the machines would be possible.
Workshop about Machine Learning in Hydrology
The association “Digital Engineering” organizes an annual boot camp to educate young engineers in the field of Data Science and Earth Observation. We were invited to conduct a workshop about Machine Learning in Hydrology. This workshop included theoretical as well as practical elements, where the participants were encouraged to optimize the prediction of water temperature of an Austrian river with Machine Learning methods using a prepared JupyterLab.
“The founders of baseflow AI solutions conducted a workshop on Machine Learning in Hydrology at the Digital Engineering Bootcamp 2022. The complex topic was explained in an understandable way accompanied by numerous exercises. The high expectations of our participants were exceeded. We thank Moritz and Christoph for their great commitment and hope to welcome them again at our next bootcamp.”
Oliver Konold, BSc (Association President)
Workshop about Machine Learning in runoff forecasting
Machine Learning (ML) models can effectively represent complex relationships, structures and patterns in data due to their internal structural flexibility and consequently their ability to generalize and transfer learned patterns. This flexibility can lead to clear improvements in various hydrological tasks.
To order the in-house understanding at viadonau, two workshops were conducted addressing the topics 1) theoretical background of ML, 2) possible applications of ML in runoff forecasting, and 3) current research findings.
Publications
Max Preiml, Christoph Klingler, Hubert Holzmann, Petr Lichtneger, Christine Sindelar, Helmut Habersack (2024): Improving the weir operating rules of Lake Mondsee, part I – Hydraulics. Österreichische Wasser- und Abfallwirtschaft, 76(3), 182-192, Link
Christoph Klingler, Max Preiml, Helmut Habersack, Hubert Holzmann (2024): Improving the weir operating rules of Lake Mondsee, part II – Hydrology. Österreichische Wasser- und Abfallwirtschaft, 76(3), 193-202, Link
Wei Zhi, Christoph Klingler, Jiangtao Liu, Li Li (2023): Widespread deoxygenation in warming rivers. Nature Climate Change, 1758-6798, Link (only for personal use)
Hanna Zeitfogel, Moritz Feigl, Karsten Schulz (2023): Soil information on a regional scale: Two machine learning based approaches for predicting saturated hydraulic conductivity. Geoderma, 116418, Link
Moritz Feigl (2022): Machine Learning in Hydrological Modeling. Dissertation, Institute for Hydrology and Water Management, University for Natural Resources and Life Sciences Vienna, 211p, Link
Moritz Feigl, Benjamin Roesky, Mathew Herrnegger, Karsten Schulz, Masaki Hayashi (2022): Learning from mistakes – Assessing the performance and uncertainty in process-based models. Hydrological Processes, 36(2), e14515, Link
Moritz Feigl, Stephan Thober, Robert Schweppe, Mathew Herrnegger, Luis Samaniego, Karsten Schulz (2022): Automatic Regionalization of Model Parameters for Hydrological Models. Water Resources Research, 58(12), e2022WR031966, Link
Christoph Klingler, Moritz Feigl, Florian Borgwardt, Carina Seliger, Stefan Schmutz, Mathew Herrnegger (2022): Prediction of runoff characteristics in ungauged basins with machine learning. Österreichische Wasser- und Abfallwirtschaft, 74(11), 469-485, Link
Christoph Klingler, Moritz Feigl, Thomas Linsbichler, Simon Frey, Karsten Schulz (2022): Performance of Machine Learning in short-term power forecasting within a run-of-river power plant chain. Österreichische Wasser- und Abfallwirtschaft, 74(5), 224-240, Link
Paul Omonge, Moritz Feigl, Luke Olang, Karsten Schulz, Mathew Herrnegger (2022): Evaluation of satellite precipitation products for water allocation studies in the Sio-Malaba-Malakisiriver basin of East Africa. Journal of Hydrology: Regional Studies, 39, 100983, Link
Hanna Zeitfogel, Moritz Feigl, Karsten Schulz (2022): Prediction of soil hydraulic properties for the extent of Austria. Österreichische Wasser- und Abfallwirtschaft, 74(3), 166-178, Link
Moritz Feigl, Katharina Lebiedzinski, Mathew Herrnegger, Karsten Schulz (2021): Machine-learning methods for stream water temperature prediction. Hydrology and Earth System Sciences, 25(5), 2951-2977, Link
Moritz Feigl, Mathew Herrnegger, Robert Schweppe, Stephan Thober, Daniel Klotz, Luis Samaniego, Karsten Schulz (2021): Regionalization of hydrological models using function space optimization. Österreichische Wasser- und Abfallwirtschaft, 73(7), 281-294, Link
Moritz Feigl, Katharina Lebiedzinski, Mathew Herrnegger, Karsten Schulz (2021): Prediction of stream water temperatures in Austrian catchments using machine learning methods. Österreichische Wasser- und Abfallwirtschaft, 73(7), 308-328, Link
Christoph Klingler, Karsten Schulz, Mathew Herrnegger (2021): LamaH-CE: LArge-SaMple DAta for Hydrology and Environmental Sciences for Central Europe. Earth System Science Data, 13(9), 4529-4565, Link
Christoph Klingler, Karsten Schulz, Mathew Herrnegger (2021): LamaH | Large-Sample Data for Hydrology: big data for hydrology and environmental sciences. Österreichische Wasser- und Abfallwirtschaft, 73(7), 244-269, Link
Johannes Peter Koren, Sebastian Hafner, Moritz Feigl, Christoph Baumgartner (2021): Systematic analysis and comparison of commercial seizure-detection software. Epilepsia, 62(2), 426-438, Link
Moritz Feigl, Mathew Herrnegger, Daniel Klotz, Karsten Schulz (2020): Function Space Optimization: A symbolic regression method for estimating parameter transfer functions for hydrological models. Water Resources Research, 56(10), e2020WR027385, Link
Christoph Klingler, Matthias Bernhardt, Johannes, Wesemann, Karsten Schulz, Mathew, Herrnegger (2020): Local hydrological modelling containing global, alternative data sets. Hydrologie und Wasserbewirtschaftung, 64(4), 166-187, Link
Michael Weber, Moritz Feigl, Karsten Schulz, Matthias Bernhardt (2020): On the Ability of LIDAR Snow Depth Measurements to Determine or Evaluate the HRU Discretization in a Land Surface Model. Hydrology, 7(2), 20, Link
Stefanie Brezina, Moritz Feigl, Tanja Gumpenberger, Ricarda Staudinger, Andreas Baierl, Andrea Gsur (2020): Genome-wide association study of germline copy number variations reveals an association with prostate cancer aggressiveness. Mutagenesis, 35(3), 283-290, Link
Matthias Bernhardt, Stefan Härer, Moritz Feigl, Karsten Schulz (2018): The importance of Alpine research catchments for model evaluation and for the improvement of remote sensing products. Österreichische Wasser- und Abfallwirtschaft, 70(9), 515-528, Link
Presentations
Moritz Feigl, Christoph Klingler (2024): AI in the sector of water management – Short introduction and use cases. 7th Get-together of the group “Young Water Management” from the Austrian Water and Waste Management Association (ÖWAV), Vienna
Moritz Feigl (2023): Austrian-wide prediction of runoff characteristics with machine learning. Meeting of the Austrian Society for Hydrology (ÖGH), When water becomes scarce – science meets practice, Vienna
Moritz Feigl, Christoph Klingler (2022): Machine Learning in the field of Hydrology. Workshop for Digital Engineering – Association for networking and supporting digitization in engineering, Hirschwang Rax
Max Preiml, Christoph Klingler (2022): Simulation of water retention by lakes to reduce the risk of flooding – Use case lake Mondsee. DRD22 – Disaster Research Days, Innsbruck
Moritz Feigl (2022): AI and machine learning in hydrological sciences. BOKU Big Data Day, Vienna
Moritz Feigl, Robert Schweppe, Stephan Thober, Mathew Herrnegger, Luis Samaniego, Karsten Schulz (2021): Catchment to model space mapping – learning transfer functions from data by symbolic regression. EGU2021 – European Geosciences Union General Assembly, Link
Christoph Klingler, Frederik Kratzert, Karsten Schulz, Mathew Herregger (2021): LamaH: Large-sample Data for Hydrology in Central Europe. EGU2021 – European Geosciences Union General Assembly, Link
Moritz Feigl, Benjamin Roesky, Mathew Herrnegger, Karsten Schulz, Masaki Hayashi (2021): Learning from mistakes – Assessing the performance and uncertainty in process-based models. KGML21 – 2nd Workshop on Knowledge Guided Machine Learning, Link
Benjamin Roesky, Moritz Feigl, Mathew Herrnegger, Karsten Schulz, Masaki Hayashi (2021): Learning from mistakes – Assessing the performance and uncertainty in process-based models. EGU2021 – European Geosciences Union General Assembly, Link
Hanna Zeitfogel, Moritz Feigl, Karsten Schulz (2021): Variability across scales – exploring methods for predicting soil properties from multiple sources. EGU2021 – European Geosciences Union General Assembly, Link
Frederik Kratzert, Daniel Klotz, Martin Gauch, Christoph Klingler, Grey Nearing, Sepp Hochreiter (2021): Large-scale river network modeling using Graph Neural Networks. EGU2021 – European Geosciences Union General Assembly, Link
Moritz Feigl, Robert Schweppe, Stephan Thober, Mathew Herrnegger, Luis Samaniego, Karsten Schulz (2020): Automatic Estimation of Parameter Transfer Functions for Distributed Hydrological Models – Function Space Optimization Applied on the mHM Model. AGU2020 – American Geophysical Union Fall Meeting, Link
Moritz Feigl, Stephan Thober, Mathew Herrnegger, Luis Samaniego, Karsten Schulz (2020): Automatic estimation of parameter transfer functions for distributed hydrological models – a case study with the mHM model. EGU2020 – European Geosciences Union General Assembly, Link
Moritz Feigl, Mathew Herrnegger, Daniel Klotz, Karsten Schulz (2019): Function Space Optimization (FSO): A novel method for estimating parameter transfer functions for hydrological models. AGU2019 – American Geophysical Union Fall Meeting, Link
Interviews
Moritz Feigl (2022): baseflow.ai: Applying Machine Learning in Hydrology. Austrian AI Podcast by Manuel Pasieka, Link to podcast
Datasets
Christoph Klingler, Moritz Feigl, Florian Borgwardt, Mathew Herrnegger (2022): Prediction of runoff characteristics in ungauged basins in Central Europe with machine learning – files. Version 1.0, Zenodo, Link
Christoph Klingler, Frederik Kratzert, Karsten Schulz, Mathew Herrnegger (2021): LamaH-CE: LArge-SaMple DAta for Hydrology and Environmental Sciences for Central Europe – files. Version 1.0, Zenodo, Link