Journal Club: 18-04-2018

Article Title: Application of numerical weather prediction in wind power forecasting: Assessment of the diurnal cycle

This paper investigated the performance of different NWP models at forecasting the diurnal pattern of wind speeds at hub height in Germany. Analysis of data from German TSOs found that large biases tend to occur in the early morning. Output from 4 deterministic NWP models and a regional ensemble were compared to observations from masts in Germany/Netherlands/North Sea.

Forecast and observed wind speeds at different heights at Falkenberg mast, June 2015, for (a) ICON global, (b) ICON EU, (c) COSMO-EU and (d) COSMO-DE models. Red and grey shading represent positive and negative biases respectively at a height of 98m.

COSMO-EU and COSMO-DE were found to have larger negative biases overnight when forecasting a day ahead (larger area of grey shading above). This suggests that the nocturnal boundary layer in these models does not stabilise sufficiently allowing faster wind speeds to develop away from the surface. All models also produce positive biases in the early morning, indicating that there is a timing issue with the transition out of the nocturnal layer as convective activity begins to occur.

Ensemble bias for different lead times at (a) onshore and (b) offshore locations. Black line represents ensemble mean.

Additionally, analysis of ensemble output found a difference in performance of the model over land and over sea. Onshore sites exhibited a strong diurnal cycle in mean error (left plot) while the bias is more steady for offshore sites (right)

Applications to our research:

  • Use Eirgrid data to search for similar patterns in Ireland.
  • Good ideas for evaluating model performance.
  • Potentially use techniquees to identify some model deficiencies in WRF and/or HARMONIE forecasts.