Short-term Forecasting Using Advanced Physical Modelling - The Results of the Anemos Project Results from mesoscale, microscale and CFD modelling

A possible solution to the problem of forecasting wind farms output in complex terrain sites comes in the form of high-resolution, advanced numerical flow models trying to improve on the NWP models shortcomings. These models can be linear flow models like Riso’s WAsP, or AriaWind, meso-scale models like the well-known MM5 community model, MeteoFrance’s MesoNH or IASA’s RAMS model, or full-blown CFD models (Computational Fluid Dynamics) like Fluent or Mercure. The idea of all models is the same: use higher resolution calculation and input data bases plus a more complete physics descriptions than the NWP model to try to capture the local air flows, be it in the mountains or at a land-sea border. While NWP models typically have a horizontal resolution of 5-10km, the meso-scale models employed here can go down to 500m. The new approaches were tested at three sites: Alaiz, a complex terrain site in northern Spain, Ersa-Rogliano, a two-cluster wind farm on the narrow tip of Corsica, and four wind farms in the eastern end of Crete. For MM5, several Planetary Boundary Layer parameterisations were tried out, and it was found that the Blackadar scheme performed not as good as the MRF or ETA PBL schemes. The last bit in horizontal resolution might not be necessary, the same accuracy can be gained with a larger finest nested area. A higher number of vertical levels in the lowest 100m above the surface helps. MM5 could improve on the simple HIRLAM forecasts in Alaiz. The accuracy of the MM5 forecasts seems to depend a lot on the accuracy of the driving model (NCEP 6-hourly or GFS hourly). KAMM could explain the turning effects of the wind for the Spanish test case. A domain size of 400x400 km 2 was needed. However, a MOS system (where data is available) might do as good. For RAMS in Corsica and Crete, the second model level (46m a.g.l.) performed usually better than the 10 m wind. Using 500 m horizontal resolution helped here (probably due to the much better orography description used in comparison to MM5). In general, the models revealed the problem of representativity of a single measurement for a whole region. We are comparing model output valid for an area with a measurement in one particular point. Another issue which has to be solved from case to case is whether it is worth to use the calculation power needed.