Validation of 'Virtual' Wind Datasets from Mesoscale modelling

摘要

The application of limited area numerical weather prediction (NWP) models for wind assessment is not new to the wind industry. One of the most common applications is the so called "Virtual" wind data. MEGAJOULE has been using mesoscale models since 2007. An initial validation effort based on the "virtual" wind data approach was already published in 2011[1]. This paper presents a second validation effort, which includes developments made in the meantime. Validations are based on 3×3 km resolution estimates from mesoscale model WRF of 10 minute horizontal wind speed and direction time series. Input weather data used is the NCEP - FNL data set. Simulations are compared with local measurements from 13 sites located in Portugal. Observations are taken in typical wind assessment masts compliant with IEC and MEASNET recommendations. These observations are not used or assimilated in simulations, neither as inputs nor with MOS on any post-calibration of results. In what concerns mean wind speed, results show a good agreement with observations. Root Mean Square error is of 0.47 m/s while BIAS is of -0.04 m/s. Frequency distribution of absolute error shows that more than 50% of results are within +/-0.50 m/s deviation and that 80% of results show a deviation not above +/-0.56 m/s. These results compare well with previous validations from MJ, with evidence of improvements. Still, some limitations related with the geographical spread and the number of sample should be considered.