Quantifying the Wind Direction Climate over the Baltic States using Principal Component Analysis

Wind direction is an important meteorological parameter, however, its analysis is made difficult by it being a circular variable that cannot easily be averaged. The goal of this study was to identify the main features of wind direction climate over the Baltic States in a methodical way. We used Principal Component Analysis (PCA) for this purpose.

Two data sets were used: UERRA re-analysis with 11 km horizontal resolution and surface wind direction observations from Latvian stations. We used PCA on both of these datasets and analyzed the results together. Such an approach enabled comparison of the wind direction climate of the reanalysis with the observations. However, preliminary results suggested applying PCA also on the subset of UERRA data that corresponds to observation stations. This eliminates effects caused by differences in spatial coverage between  gridded and station datasets.

To verify the quality of the reanalysis independently of the PCA method, Earth Mover’s Distance (EMD) was used to directly compare wind direction distributions at the station grid points with observations.

Results show good correspondence overall between the reanalysis data and the observations. The PCA method identifies SW as the prevailing wind direction, in good agreement with the expectations. The PCA results enable identification of the main wind direction features of the region, such as increased frequency of northern winds during the summer and increased frequency of southern winds during the winter that can be explained by synoptic scale processes. Additionally, the PCA method identifies coast parallel flows created by mesoscale interaction between the Baltic Sea and the dry land, and wind deflection around terrain (hills up to 300 m above sea level).

This approach could be generalized to other regions and help create a more systematic understanding about wind direction climate, as well as assist in quantifying the performance of reanalysis and identify meteorological processes that need to be investigated further.

Corresponding author is grateful to the project “Mathematical modelling of weather processes - development of methodology and applications for Latvia (1.1.1.2/VIAA/2/18/261)” for financial support.