Statistical intercomparison of gridded weather datasets in Europe

Observation datasets are invaluable data source for climate studies and a range of applications, including land surface process modelling, extreme event analysis, and crop yield forecasting. In the MARS (Monitoring of Agricultural ResourceS) programme at the European Commission’s Joint Research Centre (JRC), long time series of spatially contiguous weather observations are a staple dataset underpinning a suite of tools for agro-meteorological risk monitoring and quantitative yield forecasting – updated monthly throughout the growing season – for the major crops cultivated in Europe. Optimal representation of both climatological anomalies (deviations from long-term mean conditions) and extreme events is essential for weather risk analysis for early warning and accurate in-season crop yield forecasts.

While gridded observation datasets do not provide an absolute 'true' measure of a given meteorological variable due to sampling errors, spatial support size, and retrieval, estimation or spatial interpolation algorithm assumptions, an understanding of their relative bias is essential to guide dataset choice and interpretation of skill and reliability in the context of a specific application. Here, we evaluate the performance of gridded meteorological observations on temperature, precipitation, radiation, and wind speed relative to in situ observations from several thousand stations in Europe through a comprehensive suite of statistical metrics. We report on the skill and bias characteristics of gridded observations at different spatial resolutions, as well as re-analysis datasets, with particular attention to regions with varying density of stations. We discuss the implications of the statistical outcomes in the context of analysing anomalies and representation of extremes, as well as their feasibility for crop yield forecasting.