Multi-decadal variability in a centennial reconstruction of daily wind

A wind clustering methodology capable of dynamically characterizing and long-term reconstructing daily surface wind series is introduced and tested for six meteorological towers at different wind farms in Spain, for the period 1871–2009. On this basis this paper provides for the first time a centennial surface wind reconstruction with a daily resolution without the need of numerical simulations. Thus, several soft-computing algorithms are developed, with public domain Sea Level Pressure (SLP) Reanalysis data as the only input. These algorithms are constructed by tackling an Euclidean distances’ problem at the geostrophic speeds’ space. Once the wind-independent classifications are obtained, the methodology is calibrated by linking the obtained classifications with observed wind data, thus allowing to estimate and characterize the daily surface wind speed and direction. A cross-validation is then performed in order to obtain several measures of goodness of the method, such as its wind speed estimation uncertainty in terms of Mean Absolute Error (MAE) and Pearson correlation (r) for both the wind module and vectorial values. Regarding previous approaches, this statistic downscaling shows an outstanding performance: Wind speed module estimates produce a MAE of 1.12m/s (0.32m/s) in some towers for a daily (monthly) scale, as r reaches values of 0.78 (daily scale) and 0.91 (monthly scale).