High-resolution (250x250m) gridded daily mean wind speed dataset for Austria spanning from 1961 to 2023.

High-resolution gridded climate data derived from in-situ observations play a crucial role in global and regional climatology. The data are a valuable input for further climate impact studies, especially in ecological and energy modeling, and can be subsequently used for the wind power potential analysis.
Moreover, policymakers can make informed decisions based on accurate climate information derived from these datasets, enhancing the effectiveness of climate-related policies and interventions.

This study explores methods to enhance mean wind speed interpolation techniques over complex topography, resulting in the creation of a high-resolution (250x250m) gridded daily mean wind speed dataset for Austria spanning from 1961 to 2023. A two-step approach is tested, wherein climatologies for each month are computed using the best-performing interpolation technique. Subsequently, the optimal interpolation method would be employed to interpolate the model residuals (in case of machine learning (ML) superiority). In the subsequent stage, the same interpolation approach is applied to interpolate daily residuals to the monthly climatologies. Combining both fields produces the final gridded daily mean wind speed dataset.

Various spatial interpolation approaches, including Inverse Distance Weighting (IDW), 3D IDW (an Euclidian method, which accounts for elevation differences), Thin Plate Splines (tp_spline),  Local Polynomial Interpolation (loc_poly), and Kriging approaches (OK, OK_trend, UK, UK_poly) are evaluated. Additionally, the results would be compared to regression models, such as Ridge Regression (RR), Random Forest Regression (RFR), Decision Tree Regression (DTR), and Gradient Boosting Regression (GBR),  as well as ensembles of these models by combining different regressors in a pipeline. Each selected regression model is trained independently on the training data, and the final prediction is obtained by averaging the individual model predictions. Each model has the same set of predictors and is set up for each month separately.

Additionally, qualitative comparisons will be conducted with other high-resolution gridded datasets. The dataset will be made publicly available for download.