Challenges in simulating the wind over a coastal complex-shaped site: Madeira Island

Abstract

Abstract The wind behaviour on Madeira Island is shaped by the intricate coastline and mountainous terrain, features that are not easily represented in numerical models. Thus, this study addressed some challenges and possible solutions in setting up a meso-to-microscale model chain, with the WRF (1-km mesh) and VENTOS®/M (50-m mesh in the high-resolution area) models, in Madeira’s coastal and complex terrain. Wind measurements from four meteorological towers (located in the eastern peninsula–T0978/E–and on the SE–T0521/SE, S–T0522/S, and N–T0960/N–coasts) served as references to assess the simulations. First, due to WRF’s landmask position, land areas were misclassified as sea (and vice versa). This issue was addressed by altering the domain position to better allocate the landmask on the east peninsula, resulting in improved near-surface wind simulations at T0978/E (reducing RMSE and bias by 19% and 67%). Secondly, WRF’s default interpolation of the SST variable did not account for missing and masked data. As such, a different SST interpolation method was employed, leading to improved near-surface wind simulations at T0960/N (reducing RMSE and bias by 11% and 84%) and T0522/S (10% and 16% reduction) masts, but higher errors at T0978/E (7% and 45% increase). The negative influence arose from an incorrect speedup with the new interpolation method. Thirdly, the impact of SST_SKIN, which influences the temperature distribution at the skin level, was evaluated in WRF. Activating SST_SKIN led to a slight improvement in the near-surface wind simulation only at T0521/SE (2% and 6% RMSE and bias reduction), probably due to the dominant smaller-scale nature of the atmospheric circulation in the area, which contrasts with the circulation at the other towers, dominated by the trade winds (N and E masts) and the Island’s wake (S mast). When using the WRF outputs as boundary conditions, these effects on the microscale runs were less pronounced than on the mesoscale results. Nonetheless, the RMSE and bias of the near-surface wind simulation in VENTOS®/M were reduced by 6% and 9% at T0978/E.