Statistical–dynamical downscaling of wind fields using self-organizing maps

In this work a temporally and spatially consistent method for the efficient long-term assessment of the wind resource is presented. It contributes to the field of statistical–dynamical downscaling of the wind resource by combining stratified sampling of long-term mean Sea Level Pressure (SLP) fields with a neural-network method called self-organizing maps (SOM). The objective of the method is to construct a synthetic year which can be considered representative of the long-term period (typically 30 years) in terms of its wind resource. Validation is performed in two ways. (1) A comparison of the long-term against the synthetic SLP field was conducted showing that the proposed approach is capable of reproducing the overall SLP long-term mean with an error of less than 1hPa. (2) The wind representativeness of the selected year was verified against 10 years of measured wind data from 22 automatic stations in Navarra (Northeastern Spain), covering a variety of different climate and terrain conditions. The error found in the prediction of a variety of wind speed parameters is of the order 1% for most stations. •The SOM-constructed synthetic year accurately predicts long-term wind statistics.•Synoptic climate patterns are well reproduced.•Non-contiguous data sets and seasonal variability can be accounted for.•The temporal variability of the synoptic climate patterns is accurately predicted.•The performance is superior to the industry-standard approach.