ERA5L Temperature validation in the Extended European Alpine Region

Reanalysis products, like ERA5-Land, offer user-friendly, high-resolution gridded climate data (9 km) by combining ground observations, remote sensing, and model estimates. However, they inevitably contain uncertainties due to data gaps and modelling. Validating these datasets with land-based measurements is essential, though these observations also suffer from errors and inconsistencies. For this reason, this study validates ERA5-Land Temperature over the Extended European Alpine Region using the EEAR-Clim dataset, which includes only observational data records that meet strict reliability and temporal-consistency criteria.

The validation process involves 159 land-based meteorological stations, along with their corresponding nearest grid points in the ERA5-Land dataset. These grid points meet two criteria: a maximum elevation difference of ±100 meters and a maximum horizontal distance of ±0.5°. The selection procedure is designed to avoid repetition. The 159 grid points are different from each other. The stations are located between 504 and 2,965 meters above sea level and cover the period 1980–2020. We compared the daily temperature probability distributions for each station, grouping the stations into five elevation bands as well as considering the entire dataset. Our analysis examined temperature bimodality, the autocorrelation function, the ‘near-0°C probability’, and the ongoing issue of elevation-dependent warming trend.

The analysis shows that ERA5-Land generally underestimates temperature, with a global mean bias of –0.94 °C, and overestimates the standard deviation by +0.24 °C. The mean absolute error ranges from +1.37 °C in the lowest elevation band to +2.19 °C in the highest. The EEAR-Clim dataset provides clear evidence that low-elevation stations exhibit a bimodal temperature probability distribution, while stations above 1,500 m show a transition toward a unimodal distribution. ERA5-Land does not reproduce this transition, as even the highest grid points retain two main modes. The autocorrelation function of the observations decreases with elevation, whereas ERA5-Land shows increasing errors in its estimates, particularly at high elevations. The ‘near-0 °C probability’ is overestimated at low elevations and underestimated at high elevations. Despite this, the two datasets show good agreement in their estimates of the mean annual temperature trend rate, irrespective of elevation. However, the EEAR-Clim dataset indicates that lower elevations have warmed faster than the highest ones. These results are influenced by high variability and the limited number of stations above 2,000 m, which may affect or obscure the true temperature behavior. This underscores the urgent need for additional instrumentation, particularly at high elevations.