Research

The study aims to develop a model system that uses artificial intelligence (AI) to predict the output of VERBUND's hydroelectric power plant chains along the Danube and Inn rivers. The AI system will use discharge forecasts from several operational precipitation-runoff models, all based on VERBUND's internal precipitation-runoff system COSERO. The input data and the measured power plant outputs are provided by the client at hourly intervals. The project aims to predict output up to a forecast horizon of 72 hours. Attempts will also be made to interpret the AI model, quantifying the individual contribution of a COSERO discharge forecast to the overall output forecast.

The study as part of the ‘Digital Transformation in the BMLRT’ is intended to drive forward the future digitalisation, networking and automation of Austria-wide flood risk management. In the first phase, modern geodata and AI technologies were analysed and integrated into an overall model for area-wide monitoring. An AI-based prototype for flood forecasting was developed and successfully tested, improving accuracy compared to existing models. Applications for event analysis and documentation were also developed. Phase 2 aims to extend the AI modelling system to predict exceedance probabilities and hazard indices for entire watercourse sections. This will be implemented and tested using a selected catchment area such as the Mur. It is also being investigated which updated data sets are required for the continuous operation of the modelling system and how these can be provided. The ‘deliverables’ will be a description of the methods developed and an analysis of the necessary data updates and data streams in a report.

In many environments, snow constitutes an important part of the hydrological cycle. A substantial portion of the planet’s population relies on freshwater resources seasonally stored as snow. Especially in the face of climate change, these resources are increasingly under pressure. The knowledge of the spatial and temporal distribution of water stored in snow (snow water equivalent, SWE) is therefore crucial for improved water resources management and flood forecasting in snow-fed river basins. A number of approaches with specific advantages and drawbacks exist to monitor SWE including in-situ measurements, remote sensing, and hydrological modelling. In particular, spatial representativeness is a major problem for SWE monitoring in mountain regions as snow depth and SWE can vary substantially over small distances. The key benefit of aboveground cosmic-ray neutron sensing (CRNS) for estimating SWE is that the signal contains information of a wider area and is thus insensitive to small-scale variations in snow accumulation. Previous work confirmed the general suitability of CRNS for monitoring snow water resources in mountain regions. Currently, research gaps regarding the spatial transferability of the above-mentioned results still exist. Therefore, measurements at other locations are planned to cover different site characteristics and to refine the estimation of SWE from CRNS data. Measuring in different elevation and climate zones allows to cover various SWE amounts and changing environmental conditions within one campaign. Alpine sites in Austria will be complemented by a cascade of lower-lying sites in Germany. Continuous stationary field measurements will be supported by campaign-based mobile measurements of the spatial SWE distribution in cooperation with the module Roving & Airborne (RA). The obtained field data will be further analyzed using physically-based simulations of the neutron response. This will enable an improved understanding of the use of CRNS for snow monitoring in different environments. Validation data will be generated using laser scanning based snow cover observations and terrestrial photography. In contrast to measurements at single points, laser scanning produces a 3D point cloud of the current surface elevation. Together with snow-free measurements during the summer period, a high-resolution representation of the snow depth can be calculated. Local measurements of the snow density allow to convert snow depth variations to SWE. Terrestrial photography provides high-resolution data on snow-covered areas. All field work and simulation experiments will be conducted in close cooperation with other modules of the research unit Cosmic Sense II. In particular, the modules Roving & Airborne (RA), Neutron Simulations (NS), Hydrological Modelling (HG), Vegetation (VG), Smart Coverage (SC), and Root Zone Water (RZ) will be involved.