ERA5-Land tracks Mato Grosso monthly temperatures with seasonal errors

A 15-year comparison with 14 automatic weather stations found that the ERA5-Land atmospheric reanalysis reproduced monthly mean air temperature across Mato Grosso, Brazil, with root mean square error below 1.0 °C in most locations. The evaluation, covering 2010 through 2024, also showed that errors were not uniform across the state: they varied by municipality and were most pronounced during the warm September-to-October period, when ERA5-Land generally estimated temperatures more highly than the station records.

The study examined station observations from the National Institute of Meteorology, or INMET. Hourly air-temperature readings from stations in 14 municipalities were checked for missing observations and periods of incomplete coverage. Daily means were then calculated from the available hourly records and aggregated into monthly means so that the observations could be compared with ERA5-Land at the same temporal scale. The evaluation was statewide and station-based, focusing on monthly mean temperature rather than daily extremes, hourly heat events or other near-surface variables.

ERA5-Land is produced by the European Centre for Medium-Range Weather Forecasts and combines observations, satellite information and numerical modeling to reconstruct atmospheric conditions. For this analysis, the researchers used monthly reanalysis data obtained from the Copernicus platform at a spatial resolution of 0.1°. Values originally expressed in Kelvin were converted to degrees Celsius, and the reanalysis series were extracted for the locations and periods corresponding to the weather stations. This approach tested whether a continuous gridded product could provide a useful alternative to the unevenly distributed observation network in Mato Grosso.

The local setting matters because Mato Grosso covers approximately 903,357 square kilometres and includes transitions among the Amazon, Cerrado and Pantanal biomes. The study described two broadly defined climatic periods, with a rainy season mainly from October to April and a dry season from May through September. Mean annual air temperature generally ranges from 23 to 27 °C, while annual precipitation varies from approximately 1,200 to 2,000 mm, with greater rainfall volumes in the northern part of the state. The combination of environmental heterogeneity, elevation differences and pronounced seasonality created a demanding test for a regional reanalysis estimate.

Observed temperatures varied with both location and elevation. Among the higher-altitude stations, Alto Araguaia at 753 metres above sea level recorded a mean annual temperature of 23.73 °C, Campo Verde at 747 metres recorded 23.70 °C and Alto Taquari at 862 metres recorded 22.39 °C. At lower elevations, Cuiabá averaged 27.21 °C and Sorriso 26.14 °C. The highest monthly temperatures were recorded in September and October at all the stations. Cuiabá recorded the highest values among them, reaching 29.85 °C in September, 28.87 °C in October and 28.12 °C in November, while the lowest means occurred from June to July, including 19.98 °C at Alto Taquari, 21.14 °C at Alto Araguaia and 21.88 °C at Salto do Céu.

Across Mato Grosso, the coefficient of determination for estimated against observed monthly temperature was greater than 70%, and most locations had values above 80%. Guiratinga recorded an R² of 92.5%, and Rondonópolis reached 91.9%. The smallest municipal RMSE was 0.48 °C at Sinop. Guiratinga and Tangará da Serra had the largest station-level RMSE values, at 1.02 °C and 1.08 °C, respectively, showing that the generally strong statewide performance did not eliminate local departures. The largest monthly deviations reported were −1.28 °C at Guiratinga in April and 1.29 °C at Tangará da Serra in October.

Seasonality was especially clear in the aggregate error statistics. Monthly RMSE was highest in August, September and October, at 0.95 °C, 1.03 °C and 0.94 °C, respectively. Mean bias was also most positive in September and October, when ERA5-Land estimates were 0.59 °C and 0.48 °C above the corresponding station observations. The numerical results indicate positive bias at 11 of the 14 stations, or 78.5%. Although the source article contains some inconsistent wording about the overall direction of bias, the reported values support a general pattern of overestimation in the warmest late-dry-season months, with both positive and negative departures at individual sites.

Principal component analysis was used to investigate patterns in the monthly errors between observed and estimated temperatures. The first two components explained 45.6% of the variability in the error data: 33.1% was attributed to the first component and 12.5% to the second. PC1 was associated with spatial differences among municipalities, indicating that the magnitude and direction of discrepancies were not homogeneous across the state. Sinop, Sorriso and Santo Antônio do Leste contributed strongly to this axis and generally had smaller differences between estimated and observed temperatures. PC2 was related to the seasonal cycle, with larger errors concentrated from June through October and mainly in negative PC2 values, while smaller errors from December through March occupied the positive region of the component space.

The largest positive mean departure among the stations occurred at Tangará da Serra, where ERA5-Land overestimated temperature by 0.90 °C, while Guiratinga showed the largest underestimation, at −0.93 °C. The investigators interpreted these contrasts in relation to topography, land surface characteristics and regional climatic conditions, all of which can affect agreement between a 0.1° reanalysis grid cell and the immediate surroundings of a weather station. The analysis did not independently isolate the contribution of elevation, soil cover or other local factors, so these should be treated as plausible influences identified through the spatial patterns rather than as experimentally demonstrated causes of the errors.

Several limitations constrain the findings. The validation used only 14 stations across a very large state, leaving many local environments without direct observations. Missing hourly records were particularly common in 2021 and 2022, when the authors identified possible operational, maintenance and power-related interruptions associated with the COVID-19 pandemic period. Consistency checks and monthly aggregation helped address the data gaps, but incomplete station records can still affect the observational reference used for comparison. The study also evaluated only monthly mean air temperature, so its results cannot automatically be extended to daily minima, hourly extremes, heat-wave duration or applications requiring kilometer-scale precision.

Within these limits, the study supports ERA5-Land as a useful tool for constructing spatially continuous historical temperature series in Mato Grosso. The reanalysis captured the broad monthly behavior recorded at the INMET stations, including the major influence of elevation, while its strongest systematic departures occurred during the warm September-to-October interval. It appears capable of filling observational gaps and supporting regional agroclimatic and environmental analyses, but the results do not justify treating the product as a universal replacement for direct measurement. Applications requiring finer precision should account for the 0.1° grid, local topography, seasonal bias, missing station data and the fact that some municipalities exhibited errors larger than the generally favorable statewide results imply.

Subject of Research: Evaluation of ERA5-Land monthly mean air-temperature estimates against 14 Mato Grosso weather stations from 2010 to 2024

Article Title: Spatio-temporal evaluation of air temperature estimated by reanalysis of ERA5-land in Mato Grosso, Brazil

Article References: e Silva, I. A. C., da Silva, T. J. A., Bonfim-Silva, E. M., de Oliveira, N. P. R., Meneghetti, L. A. M., & Custódio, A. S. C. (2026). Spatio-temporal evaluation of air temperature estimated by reanalysis of ERA5-land in Mato Grosso, Brazil. Theoretical and Applied Climatology, 157(11), Article 737. https://doi.org/10.1007/s00704-026-06676-z

Image Credits: AI Generated

DOI: 10.1007/s00704-026-06676-z

Keywords: ERA5-Land, air temperature, Mato Grosso, weather stations, atmospheric reanalysis, seasonal bias, climate monitoring, agroclimatology, Brazil, Spatio-temporal, evaluation, temperature