Geostatistical downscaling of offshore wind speed data derived from numerical weather prediction models using higher spatial resolution satellite products

Abstract

Regional offshore wind assessment studies typically rely on forecasts from Numerical Weather Prediction (NWP) models. NWP products are typically available at fine temporal resolutions (e.g., on an hourly basis) but relatively coarse spatial resolutions (e.g., on the order of several kilometers) to be used directly for more detailed local assessments. Satellite data, e.g., SAR (Synthetic Aperture Radar) data, on the other hand have been widely used in the literature to reveal high spatial resolution wind fields along with their variations but are available only at a few instances within a month’s period. The C-Band SAR instrument onboard the Sentinel-1 platform, in particular, provides wind speed data at 10 m above sea surface with a repeat frequency of 6 days since 2016.Statistical downscaling techniques are often employed to obtain finer spatial resolution products from coarse NWP products for use in finely resolved impact assessment studies. This study investigates the application of a novel geostatistical approach for downscaling Regional Reanalysis wind speed data using SAR data in order to spatially enhance information captured by the former. The data used comprise Sentinel-1A and 1B VV-polarized SAR wind field measurements and Uncertainties in Ensembles of Regional Reanalyses (UERRA) data, both bias-corrected using in-situ data from local meteorological coastalstations. The reference data used for bias correction are generated via spatial interpolation and aggregation (upscaling) of the local meteorological station wind speed values within the closest Sentinel pixel(1 km) and UERRA cell (11 km).Prior to the downscaling procedure, Weibull distribution models are fitted to the wind speed time-series both at the coarse and fine spatial resolutions. Downscaled UERRA Weibull distributions parameters (scale (a) and shape (b)) are then generated via Area-To-Point Kriging with External Drift (ATPKED), whereby Weibull parameter values are computed at a finer spatial resolution as a weighted linear combination of neighboring coarse resolution attribute values. The fine resolution parameters are used as auxiliary variables. ATPKED is mass preserving, in that the average of the downscaled Weibull parameter values within a coarse cell reproduce the bias-corrected UERRA value at that cell. Once the fine scale parameters are estimated, the wind speed distribution at the pixel level can be extracted. Statistical comparison indicated that more than half of the wind speed variability in Sentinel images can be explained by the contemporaneous downscaled estimates. Geostatistical simulation is also employed to assess the uncertaintyin the fine resolution values.As an illustration of the methodology, offshore wind speed values are estimated at a spatial resolution of 1km for the coastal areas of the Republic of Cyprus at a 6-hour interval over a period of 1 year. The results imply that the downscaled products could furnish a basis for a more spatially resolved offshore wind power assessment for the region, provided the above procedure is generalized for a longer time period.