Hydropower companies are benefiting from accurate streamflow predictions. This study proposes and tests a method to incorporate global climate indicators in these streamflow predictions.
Using climate indicators to condition operational streamflow forecasts
Supervisors: Joost Beckers (Deltares) en Paul Torfs
The Bonneville Power Association (BPA), a hydropower company that owns 31 dams in the North West US, is dependent on water availability for their energy production. Every year they need to give an assessment on the amount of energy they are able to deliver. Therefore, accurate operational streamflow predictions are required. BPA currently uses an ensemble streamflow prediction method (ESP), which constructs an ensemble of yearly streamflow traces based on output of a hydrological model fed with yearly meteorological data of all historical years available. This method disregards any additional knowledge about the climate state (e.g. ENSO related phenomena), while this state affects precipitation and temperature patterns. Attempts have been made to condition the ESP by removing the ensemble traces that shown the least similarity regarding a certain climate indicator. However, the drawback of this approach is the limitation in ensemble size.
To overcome this limitation, this study uses a nearest neighbour resampling procedure to create climate index predictions with coupled meteorological predictions. This procedure shuffles the historical meteorological record (precipitation and temperature) on a monthly scale, based on similarity in predicted climate index values. The newly constructed meteorological ensemble traces are used as input for the hydrological model, resulting in a new ensemble of streamflow predictions.
The results show that the proposed resampling method shows improvements compared to the ESP method in the prediction of cumulative winter precipitation and streamflows during the snowmelt period. These improvements were found for different test basins and for resampling the meteorological data conditioned with the different climate indices that are known to influence the precipitation patterns in the area of interest (NINO3, SOI, and PDO). Furthermore, conditioning the predictions with a combination of these three different indices, showed improvement compared to the ESP predictions.
From this research, it can be concluded that the used resampling approach showed promising results . However, this is a relatively new method of constructing meteorological and streamflow ensemble predictions, so there are plenty of possibilities to test other methods and statistical techniques in order to further improve and fine tune the chosen method.