SmallFlex: Real-time hydrological forecasts for a flexible operation of an alpine run-of-river plant

The overall objective of this project was to show how a small alpine run-of-river power plant (here using the example of Gletsch-Oberwald) could produce peak energy and increased winter-energy for the Swiss grid by storing the water in the plant for short periods of time. This would allow the run-of-river power plant to be operated more flexibly and generate additional revenue.

The contribution of the Swiss Federal Research Institute WSL included an automated forecasting system with very short-term (< 6 hrs), medium-term (1-5 days), and long-term (1 month) discharge forecasts for the Rhone River at the Gletsch intake. A new nowcasting product from MeteoSwiss (INCA), as well as detailed snowmelt forecasts were assimilated into a hydrological model. The hydrological forecasts allowed operators to plan and implement temporary storage of water in the system.

The calculated discharges also allowed an initial estimate of bedload transport in the river at the water intake. This limits the flexible operation of the run-of-river power plant. However, it became apparent that the prediction of bedload transport in this alpine river is still associated with large uncertainties and would require additional measuring systems.

Further information on the forecast system

The basis of the implemented forecast system is the INCA-CH system, which merges all the meteorological information available in real-time, like station, radar and satellite data, and extrapolates this information into the future. These observation data are interpolated to a 1 km resolution grid and is updated every 10 minutes and yields predictions for the next 6 hours for the main meteorological variables like precipitation, temperature, wind. The hydrological model PREVAH, which calculates the streamflow taking these meteorological observed and forecasted variables as input, is running operationally with a spatial resolution of 100m and with hourly time steps. Thus, the INCA-CH forecasts have to be downscaled first. In the figure below an example of a INCA-CH temperature forecast for the Gletsch catchment is shown highlighting the different spatial resolutions of the 1km grid of the meteorological (in grey) and the 100m resolution after applying downscaling procedures within the catchment (in colours).

At WSL the coupling of the INCA –CH nowcasts and the PREVAH model is done hourly and results in stream-flow, resp. inflow forecasts at Gletsch for the next 6 hours. Beyond the 6 hours other meteorological forecasts provided by MeteoSwiss are assimilated seamlessly and transferred to streamflow forecasts.

In November 2018 the forecast system has been tested within the field campaign at Gletsch and WSL provided the project partners with forecasts in real-time twice a day.

Application of the forecast system

An example of a forecast during the week of the field campaign in November 2018 is shown above. Black dots show the observed streamflow at the gauging station and the blue line is the forecast. The dots in grey are the true values, which are not available in real-time, but have been included to see the difference between the forecast and the measurement. The red dots are observations of runoff, which are retrieved in the time gap between the initialization of the meteorological forecast and the completion of the hydrological forecasts.

Within the last project year different machine learning methods (e.g. Gradient Boosting and Random Forests) and forecast combination methods have been tested in order to reduce the errors in the forecast system. These post-processing methods show some significant improvements.

Further-development of a snow model for alpine catchments

The second part of WSL’s contribution to the SmallFlex project consists of improving of a high-resolution energy-balance-type snow model enabling a realistic representation of small-scale processes in alpine terrain. Accounting for spatial variability is key to accurately assess runoff in small mountain catchments. Our chosen snow model solves the energy balance at the snowpack surface directly. It uses downscaled Numerical Weather Prediction (NWP) in 250 m resolution as input to forecast snow melt. To provide best states of the snowpack conditions before the forecast starts, measured data were assimilated in the model. Snow density was modelled at the snow stations and the snow depth signal was transferred to solid precipitation. An optimal interpolation scheme was used to interpolate point measurements spatially and also account for measurements errors in both the precipitation gauges and snow depth measurements.

Since the snowpack is very variably distributed in complex terrain, patchy snowcover is a typical feature in the ablation season, which directly influences the available snow melt for runoff. Thus, a subgrid-paramerization of snow covered fraction (SCF) based on surface roughness was used. The model accounts for the ability of snow to store rain or meltwater in the snowpack, which influences the timing when water leaves the snowpack.

Finally, modelled surface water input (SWI) was provided to the hydrological model PREVAH to simulate the intake to a hydropower plant.

The forecasts of the combined model chain (i.e. energy-balance snow model and hydrological model PREVAH) will be visualized on our common website and thus can be used to assess the intake of the power plant for the next couple of hours in order to allow a flexible management of the power plant.