Towards a high-resolution wind climatology for Switzerland – a comparison of different methods

This presentation addresses the challenge of generating high-resolution spatial wind climatologies for Switzerland, a region characterized by complex mountainous terrain and, for the purpose of mapping wind, a sparse measurement network. Accurately mapping wind patterns in such areas is thus inherently difficult but of great importance for supporting applications such as risk assessment or wind energy planning. 

In this study we explore three different and complementary approaches to tackling this challenge:  

• Model-only: a new test data set, the Swiss ICON Reanalysis-Light1-CH1 (REA-L), has recently been produced by MeteoSwiss for the period 2005-2024 at 1km mesh-size over the ICON-CH1-EPS domain. Long-term climatologies for mean wind and wind gusts have been produced from this data set. 

• Machine-Learning: in an attempt to further downscale the ICON REA-L surface fields to sub-kilometer scale, a ML approach based on Gaussian Processes has been developped within obsweatherscale, an open-source Python package that integrates station measurements, high-resolution topographic descriptors, and auxiliary atmospheric predictors to produce a continuous spatial distribution of the target variable. The method captures nonlinear wind-terrain interactions, provides uncertainty estimates, and remains physically interpretable. 

• Station Transfer: this approach follows classical statistics to estimate the full statistical distribution for mean wind and wind gusts at each grid point based on station measurements from 1981 – 2025 and topographic information. 

The resulting outcomes are intercompared, verified, and explored regarding their consistency and accuracy, and to gather the pros and cons for each method. A special focus is on their suitability to accurately describe extremes. We present the resulting climatological fields as well as some case-study verification of selected wind events over Switzerland.  

 

References: 

Lloréns Jover, I., & Zanetta, F (2024). obsweatherscale: observation-conditioned ML downscaling of surface weather fields. GitHub repository: https://github.com/MeteoSwiss/obsweatherscale 

Zanetta, F., Nerini, D., Buzzi, M., & Moss, H. (2025). Efficient modeling of sub-kilometer surface wind with Gaussian processes and neural networks. Artificial Intelligence for the Earth Systems.