Evaluation of Global Reanalysis Strengths and Weakness using Surface Observations

Global reanalyses (e.g. ERA-5, MERRA-2, NCEP/NCAR, NAV-GEM) are commonly used as substitutes for observations to assess numerical weather prediction model forecast skill and the accuracy of climate model historical realizations. However, the quality of reanalysis output is not uniform across all variables, times of day, seasons, and geographic settings. Situations where a model struggles are often not improved much by data assimilation since assimilated observations are usually given less weight in circumstances where there are large differences between the observation and the model output.

This study evaluates the strengths and weaknesses of several reanalysis products over a multi-year period by comparing them to worldwide in-situ surface observations from over 1200 stations and buoys. We examine biases between reanalysis and observations across diverse climate zones, seasonal and diurnal cycles, and environments.

By identifying how reanalysis performance varies by variable, weather condition, and geography, this study will document where and when these products are more reliable and less reliable substitutes for observations. This information can also be used to support the training of machine learning models as well as assessments of adequacy for purpose.