A statistical method to downscale temperature forecasts. A case study in Catalonia

Abstract

The aim of this work is to present a methodology to downscale weather forecasts (i.e. to give regional-scale forecasts based on synoptic-scale forecasts from a numerical model). This methodology consists of the regionalisation of a large zone into thermally homogeneous meteorological regions followed by a study of how different types of weather affect them. Once the meteorological behaviour of each region is determined, cokriging is used to predict its temperature knowing the current temperature and type of weather expected (i.e. based on forecasts from a numerical model). Thus, from a methodological point of view, it is shown that geostatistical prediction techniques can be used in the meteorological sciences by combining classical multivariate statistics and space-time prediction techniques. To illustrate this methodology, the maximum and minimum temperatures at 54 observatories in Catalonia in winter and synoptic data from Barcelona's airport have been used. The method proposed in this paper, which represents an improvement in temperature prediction compared to that given by classical statistical methods, allows the forecasting of temperature. This case study shows that forecasts can be easily downscaled and that, to produce daily forecasts, just a pocket calculator is needed. The results are presented for a one-day forecast, but this technique can be applied to forecasts over a four-day period.