LAUSR

Abstract An adaptive high-temporal resolution interpolation scheme for meteorological observations is presented. It stems from a combination of linear regression, anomaly correction and clustering. A number of approaches to tackle this problem for monthly and daily data have been proposed in the past, but interpolation studies at sub-daily temporal scales are much more limited. Hourly and sub-hourly observational datasets use to present high variability that may be related to different weather conditions. In the proposed methodology, rather than considering the whole data set to perform the interpolation, data are divided in different clusters of variable size, separating regions with potential dissimilar behaviour. A linear regression model is calculated for each cluster and compared against a global model obtained considering all the observations. Only those clusters whose regression model yields a reduction of errors compared to the global model are selected. The adaptive condition lays on that several numbers of clusters are tested and the one that performs the best, in terms of Root Mean Square Error, is selected every time an interpolation is conducted. The methodology presented provides gridded analysis fields of hourly and sub-hourly intervals at 250 m of horizontal resolution. It was originally developed for a complex terrain region (Catalonia, NE Spain), and it was also demonstrated in the German Land of Baden-Wurttemberg and in the Italian region of Emilia-Romagna. Results show a reduction of cross-validation errors using the leave-one-out method for air temperature and dew point temperature fields and a proper representation of complex orography features. The scheme presented is implemented in Python as pyMICA and it is available as open-source software.