Today, Statkraft uses different weather prognoses to plan power production. How can this planning be improved further? Through the Climate Futures collaboration, Statkraft and the Norwegian Computing Center have investigated what level of skill we can expect from long-term forecasts, and how we can improve them further. Furthermore, Statkraft will implement them into their hydrological forecasting systems, so they can better predict how much water they can expect to have in their reservoirs in the next few days and weeks. This in turn can have an effect on price prognoses and production planning.
Background
Hydropower is based on gathering water in reservoirs, which you drain based on demand, and you cannot control how much water is in the reservoirs at all times. If there is a lot of water in the reservoirs and you can expect heavy rains or snow melting in the near future, you have to drain the reservoirs regardless of demand to avoid them overflowing. Equally so, you need to hold back if dry weather is expected, even if the demand is high.
Energy producers like Statkraft want to make the most of limited water resources, which means they have to plan power production thoroughly. They need to predict when prices will be high or low and plan their production thereafter. This is not always an easy task.
What determines if the reservoirs are full or empty? The most obvious factor might be precipitation, but temperature is also important, mainly because it controls snow melting, which in turn affects the amount of water in the reservoirs. Consequently, predicting these variables (temperature and precipitation) in the best possible way is in Statkraft’s interest. They can then estimate as accurately as possible how much water they can expect to have in their reservoirs at all times.
Statkraft has a pretty robust system for evaluating hydrology. NR evaluates the results on a meteorological level, and Statkraft evaluates the same results and the effect they have on the hydrology systems. Together, Statkraft and NR came up with two practical issues to explore:
- How much more use can we get out of operative monthly- and seasonal forecasts that already exist?
- How can we merge these operative prognoses with climate predictions?
The Pilot Project
Statkraft needs good access to predictions to be able to best utilise its hydropower resources. In practice, this means that they hold back water when they think prices will be low (as a result of low electricity demand) and offer more water when they think prices will be high (as a result of high demand).
For now, Statkraft has access to good predictions up to 15 days ahead. The models used to make these predictions are called Numerical Weather Prediction Models (NWP), which are the same models that are used for example to make the weather forecast on yr. You feed the models information about the state of the atmosphere today, and they calculate how the state of the atmosphere, and therefore also the weather, will be like 15 days ahead. Beyond 15 days, Statkraft uses historical climatological weather scenarios, and these are not based on the current state of the atmosphere. Thus, they can also be more challenging to use to predict the weather for specific purposes, like Statkraft wants.
This pilot project within Climate Futures is therefore concerned with using long-term forecasts, with a time scale of 10 days to 10 years, to make predictions in exactly the time period that Statkraft needs information about. Predictions become more uncertain the longer into the future you predict, and exactly how good they are for specific purposes like these is hard to say. This is exactly what Climate Futures wants to find out. We investigate how much skill you can expect from these long-term forecasts, especially when it comes to precipitation and temperature, which are the variables Statkraft is most interested in. If these forecasts turn out to be useful, they can potentially be implemented in Statkraft’s hydrological warning systems, which can lead to better predictions of how much water they can expect to have in their reservoirs in the upcoming days and weeks.
Results
It turns out that the skills of the NWP systems decrease quite quickly with every extra day you include in the forecast. Beyond the first 15 days, the forecasting skill is relatively limited, both at a seasonal and monthly scale. However, it seems that at certain times of the year, the long-term forecasts are better in weeks 3 and 4 compared to the climatological weather scenarios Statkraft originally used, especially for temperature. These are interesting results that can be further investigated, which is what Statkraft and NR plan to
do. The next step is therefore to try to use this information even better using new statistical methods.
At the beginning of the project, Statkraft and NR spent a lot of time discussing relevant issues that are useful for Statkraft, which is one of the successful aspects of the pilot project. When the issue is sufficiently narrowed down, even small results can have practical significance. According to Statkraft, they have also learned important things that affect their deliveries already, which concerns the specific quality of the data sets that NR has looked at for different horizons, and how Statkraft can use them when they make prognoses for price and production. Statkraft mainly delivers prognoses to optimize the production and price of hydropower, and the work that NR has done in this pilot project will affect both.
Climate Futures creates a dialogue between research and practice, and a pilot project like this one can be a springboard for bigger and more long-term collaborations. These kinds of platforms will be essential for a greener future.
About the Partners Involved
Statkraft is a leading energy producer and the biggest supplier of renewable energy in Europe. They produce hydropower, wind power, solar power, in addition to supplying district heating. The company has been producing renewable energy since it started with Norwegian hydropower in 1895.
Norsk Regnesentral (NR) does research within statistical modelling, machine learning, and ICT. NR has one of Europe’s biggest research communities within statistics, and work within several sectors, including climate and environment.