We propose an extension of the non-homogeneous Gaussian regression (NGR)
model by Gneiting et al. (2005) that yields locally calibrated probabilistic
forecasts of tem- perature, based on the output of an ensemble prediction
system (EPS). Our method represents the mean of the predictive distributions as
a sum of short-term averages of local temperatures and EPS-driven terms. For
the spatial interpolation of temperature averages and local forecast
uncertainty parameters we use a Gaussian random field model with an
intrinsically stationary component that captures large scale fluctuations and a
location-dependent nugget effect that accounts for small scale variability.
Based on the dynamical forecasts by the COSMO-DE-EPS and observational data
over Germany we evaluate the performance of our method and and compare it with
other post-processing approaches such as geostatistical model averaging. Our
method yields locally calibrated and sharp probabilistic forecasts and compares
favorably with other approaches. It is reasonably simple, computationally
efficient, and therefore suitable for operational usage in the post-processing
of temperature ensemble forecasts