Evaluation of snow simulations in SHyFT
MetadataShow full item record
Snow is a very important component in the hydrological cycle in Norway and crucial for determining reservoir operation during the spring flood to ensure full reservoir and as little flood spill as possible. The Statkraft Hydrological Forecasting Toolbox (SHyFT) is a newly developed hydrological toolbox that is used for forecasting inflow in the Statkraft system. This is a flexible system in which model can be custom designed for various purposes. This study calibrates, runs and evaluates the three snow routines implemented in SHyFT against field observed snow data provided by Statkraft. The three routines are Gamma Snow, HBV Snow and Skaugen Snow. Part of the study is based on the process of finding the closest and most representative cells of the grid in the catchment of the study area (the Nea-Nidelva river basin), in order to later make the comparison between observed and simulated snow data. To do that, the snow transects where Statkraft made the snow field measurements are analysed. Next, the calibration of the model and their validations are performed by simulating runoff and comparing them to the unregulated observed discharge data from Aune gauging station (SeNorge). The calibration period, where the calibration of the model takes place, is from 01/09/2012 to 01/09/2014 and the validation, where the runoff is simulated without having calibrated the model for that period, is from 01/09/2014 to 01/09/2015. The calibration results showed a Nash-Shutcliffe efficiency criteria (R2) of 0.733 for Gamma Snow, 0.755 for HBV Snow and 0.784 for Skaugen Snow. However, the results show that Gamma Snow performs better simulations (or closer to the observed data) that HBV Snow and Skaugen Snow. Regarding the snow results, SHyFT codes are used to extract the SWE from the specific grid cells. All the models show both similarities and discrepancies between the observed SWE data and the simulated results. The comparison was carried out by observing the percentage of the difference between observed and simulated discharge per transect, and it is concluded that all of the three models present flaws and that the simulations were sometimes poor. Some other reasons why these simulations were not too good are commented, such as redistribution of the snow by wind, elevation, or orientation. The models behaved in a way in 2013 and 2015 that they presented close difference values of SWE, whereas in 2014, the results of each model differ significantly from each other. It can be said that the calibration of the model was successful and the R2 values were good, the runoff simulations were acceptable and the snow simulations were somehow poor.