Spatially adaptive ensemble optimal interpolation of in-situ observations into numerical vector field models

Ensemble optimal interpolation (EnOI) is a well-known, having a relatively low computational cost yet powerful method for correcting outputs of numerical models in accordance with in-situ observations. Although, more advanced methods exist, e.g. variational analysis, this technique is widely used among different areas including meteorology and oceanography. Meteorological fields possess spatial inhomogeneity so as the quality of available measurements can vary between locations. This affects efficiency of the correction scheme and consequently motivates the need for adaptive choice of the correction parameters. In this paper we study how the ridge regularization influences the EnOI outcomes regarding statistical measures of fit between corrected and measured time series. Our numerical experiments for the wind field in southwestern Arctic region show that the optimal values of regularization parameter change from one group of observation points to another. We found also that these groups can be identified by clustering analysis based on estimated mutual covariances between time series of the observation points. As a result, we can adapt the EnOI scheme to each geographic sub-region and therefore to achieve more accurate correction results.