Deriving high-resolution climate series at points: The StationTransfer method and its application in Switzerland

Many applications in civil engineering, ecology and agriculture require climate data at high temporal resolution (e.g. resolving the diurnal cycle), physically consistent between a set of variables (e.g. between radiation, temperature and humidity) and representative at points (e.g. at the location of a building). Conventional grid data rarely meet these requirements, and it is a long tradition to use measurement series from a “suitable” weather station for such applications. However, choosing a suitable station is not trivial and the “most suitable” may still be poorly representative for the location of interest (e.g. in complex terrain).

Here, we propose a statistical method that is capable of deriving point-scale climate series at high temporal resolution. The method, denoted StationTransfer, builds on the idea of translating the data from a station to a location of interest. But it formalizes the selection of the source station and enhances the representativity at the target location by adopting corrections to the station series. StationTransfer builds on a statistical model where both, the prediction errors and the corrections are parameterized flexibly to deal with complex spatio-temporal variations. All the model parameters are estimated simultaneously, using maximum likelihood estimation.

We have applied StationTransfer to hourly surface air temperature in Switzerland and estimated the model over a 30-year time period from more than 100 stations. The presentation illustrates the configuration and results from this application. We show that the method can account for complex representativity issues, such as those related to cold-air pooling and the peculiarities of lake shores. These effects are found both in the proposal of source station and in the diurnal/seasonal pattern of the corrections. We also discuss how StationTransfer can be extended to multiple variables in order to meet all the above requirements targeted with this somewhat unconventional approach.