Determination of the spatial distribution of wind speed based on deterministic and geostatistical methods

DOI: 10.31673/2412-9070.2025.040304

Abstract

The aim of this paper is to determine the spatial distribution of wind speed using a deterministic interpolation method based on radial basis functions (RBF), as well as geostatistical interpolation methods Simple Kriging and Ordinary Kriging, and to analyze the accuracy of these interpolation models relative to the input data sample. The subject of the research is to encompass the statistical indicators and validation methods for the RBF, Simple Kriging, and Ordinary Kriging interpolation techniques. A dataset of wind speed measurements was generated from 533 locations uniformly distributed across the territory of Ukraine with a spacing of 50 km. This dataset was derived by correlating reanalysis data from the NASA Power MERRA-2 model with observational data from 70 stationary meteorological stations at a height of 10 meters. Both datasets covered a 10-year period (2011–2020) during the winter season, with eight wind speed records per day. The correlation process was performed using a machine learning model Random Forest. Interpolation of wind speed was applied to the correlated dataset using the RBF, Simple Kriging, and Ordinary Kriging methods. As a result, three output raster images were obtained, each depicting wind speed values assigned to every pixel. Based on the conducted analysis and cross-validation statistics of the interpolation methods, it was determined that Ordinary Kriging provides the highest accuracy in wind speed prediction among the evaluated methods. The modeling accuracy was assessed using RMSE, ME, MSE, RMSSE, and ASE metrics, along with visual comparisons of scatter plots of predicted versus observed values, distribution plots, and standardized error scatter plots for the geostatistical interpolation methods. Based on the RMSE metric, the relative accuracy of the interpolation methods was determined, with ordinary kriging taken as the reference at 100% accuracy. In comparison to the ordinary kriging, the RBF method reaches 92.6% accuracy, and simple kriging reaches 88.2% accuracy.

Keywords: spatial distribution of wind speed; reanalysis data; MERRA-2; meteorological data; radial basis functions; interpolation methods; Simple Kriging; Ordinary Kriging; cross-validation.