MeteoEurope1km: a high-resolution daily gridded meteorological dataset for Europe for the 1991–2020 period

Daily gridded meteorological datasets are an important source of information for analysis of historical weather and many other research areas since they have no gaps in the spatio-temporal domain they cover. Most of the daily gridded meteorological datasets represent reanalysis or estimations from different remote sensing sensors or are generated by downscaling procedures. There are none or a very few high resolution datasets on regional or global scale that are based on observational data and space-time geostatistics, i.e. that takes into account spatio-temporal correlation between observations. The only such dataset for Europe is the E-OBS dataset at 10 km spatial resolution, but it uses spatial correlation only. Therefore, a daily gridded meteorological dataset for Europe at 1 km spatial resolution, named MeteoEurope1km, is created, initially covering the 1991–2020 period. Spatio-temporal regression kriging (STRK), an interpolation method that combines multiple linear regression for trend modeling and space-time kriging for the estimation of the residuals, is used for interpolation of maximum, minimum, and mean daily temperature. Combination of GHCN-daily, ECA&D, and SYNOP observations from OGIMET service is used as an observational dataset, with previous removal of duplicated stations and outliers, while geometric temperature trend, digital elevation model and topographic wetness index are used as auxiliary variables. Accuracy assessment (leave-one-station-out cross-validation) shows high accuracy of the STRK models for daily temperature. Coefficient of determination for all three parameters is greater than 97% and root mean square error is less than 1.5°C. Future work will be oriented towards increasing the temporal extent of the MeteoEurope1km to the 1961–present period, interpolation of other daily meteorological variables, and improving performance of STRK models for daily temperature at higher altitudes, since the accuracy is lower due to a well known problem with lack of stations. The same methodology will be applied for interpolation of daily sea level pressure, while spatial machine learning methods, such as Random Forest Spatial Interpolation, will be used for interpolation of daily precipitation because of its complex nature.