The talk starts at 12:15.

Please note that due to COVID-19, the participants can watch the streamed talk on Teams with a link (below).

Speaker: Michael Scheuerer

Location: NR and Teams link (below)

Join on your computer or mobile app

Click here to join the meeting

Title: Using statistical and machine learning techniques to improve the skill of subseasonal weather predictions over Norway

Abstract:

Hydrological forecasts are crucial for decision making in the context of hydropower production. Energy producers like Statkraft would like to produce electricity when demand (and thus prices) is high, but hydropower production is constrained by the available amount of water in rivers and reservoirs. Temperature, which controls snowmelt runoff, and precipitation are the key weather variables that drive streamflow, and thus there is a strong interest in high quality forecasts of these variables. Statkraft currently gets forecast information from numerical weather prediction (NWP) systems for forecast lead times up to 15 days into the future. In our ClimateFutures pilot project we explored to what degree forecasts with lead times beyond 15 days could benefit operational streamflow forecasting at Statkraft.

In this seminar I will give a quick overview over the basics of probabilistic NWP systems, I will explain how statistical techniques can be used to improve forecast quality, and I will present the results that we obtained for our statistically corrected, subseasonal (i.e. several weeks into the future) forecasts of temperature and precipitation over selected river basins in Norway. Since our findings confirm that forecasting at subseasonal lead times is rather challenging, I will also discuss new approaches like the application of convolutional neural networks (CNNs) to forecasts of large-scale weather patterns in an attempt to boost subseasonal forecast skill. I will end with an outlook on ongoing and planned further research on this topic.