Daily minimum and maximum surface temperatures have been estimated using data from the Spinning Enhanced
Visible and Infrared Imager (SEVIRI) onboard the Metosat Second Generation (MSG) satellite platform, which is
the operational weather satellite at 0 degrees longitude/latitude. Both the land surface skin temperature (LSTmin
and LSTmax) and and estimated air temperature (Tmin and Tmax) are provided in this data set, with Tmin and
Tmax being derived from the LSTmin and LSTmax data through an empirical regression-based approach. Fifteenminute LST data from SEVIRI are used to estimate daily LSTmin and LSTmax. These LST data are then regressed
with observed daily vegetation fraction, latitude, urban fraction, elevation and distance from coast against
collocated station Tmin and Tmax data from ECA&D (http://eca.knmi.nl/). Separate regression models are
produced for Tmin and Tmax for a rolling 11-day window to account for temporal variation in the relationships
between air temperature and the regression predictor variables. The derived regression coefficients are applied to
every available SEVIRI LSTmin/max observation for the central day in the analysis window, providing estimates
of Tmin and Tmax where station data are unavailable. Assessing both the model residuals and using independent
station data from the UK and Germany not used in the regression formulation suggests that for most days, at least
50% of the estimated LSATs are within 3 °C of collocated station observations, with around 80% within 4 °C and
90% within 5 °C. Results for Tmax are better than for Tmin. The mean bias of the satellite-estimated LSATs oscillates around zero and shows little seasonal variation, although the variance is noted to be lower during summer
months. The satellite data sets used for the regression are sourced from the Land Surface Analysis Satellite Applications Facility (LSA SAF; http://landsaf.meteo.pt/), with 300-m land cover data from GlobCover
(http://due.esrin.esa.int/globcover/) used to provide an estimate of the urban fraction and distance from coast
parameters used in the regression.