Clustering-based spatial interpolation of parametric post-processing models

Parametric approaches to post-processing methods are widely used today, as they provide full predictive distributions for the weather variable of interest. These methods rely on training data consisting of historical forecast-observation pairs to estimate their parameters. Consequently, post- processed forecasts are generally restricted to locations with accessible training data. To overcome this limitation, we introduce a general clustering-based interpolation technique that extends calibrated predictive distributions from observation stations to any location within the ensemble domain where ensemble forecasts are available. Using the ensemble model output statistics (EMOS) post-processing technique, we conduct a case study based on 10-m wind speed ensemble forecasts from the European Centre for Medium-Range Weather Forecasts.  The results illustrate the effectiveness of the proposed method, demonstrating its advantages over both regionally estimated and interpolated EMOS models as well as raw ensemble forecasts.

Reference:  Baran, S. and Lakatos, M. (2024) Clustering-based spatial interpolation of parametric post-processing models. Wea. Forecasting  9, 1591-1604.

Research was supported by the Hungarian National Research, Development and Innovation Office under Grant No. K142849.