Comparison and calibration of terraclimate climatological variables over the Brazilian territory

Air temperature (T), rainfall (P) and reference evapotranspiration (ETo) are key data for hydrological and climatic changes studies, crop water demand estimates, crop zoning and water resource management activities. Although free global datasets of these variables have become more accessible in the last decade, their performance at some region needs a better investigation, mainly, if there is a limited validation in data-sparse region. Thus, this study aimed to analyze the performance and calibrate the climate variables derived from TerraClimate dataset using the automatic weather stations (AWS) of the National Institute of Meteorology (INMET) of Brazil. To achieve this goal, we used data from 2000 to 2017 (18 years data) of 589 AWS, which served as reference for the evaluation and calibration of the dataset. The data was acquired on an hourly basis and therefore has been aggregated for monthly periods to become compatible with TerraClimate dataset. To evaluate the results, we utilized a group of statistical metrics (root-mean-square error-RMSE, Nash–Sutcliffe efficiency NSE, percentage bias - PBIAS and coefficient of determination - R2). Based on NSE results, our research shows that before calibration, the minimum (NSE = 0.60), mean (NSE = 0.69) and maximum (NSE = 0.59) air temperature have good agreement with observed data. Rainfall (NSE = 0.45) and reference evapotranspiration (NSE = 0.29) show reasonable concordance. Among the air temperatures, the mean air temperature fitted better with AWS data, with a NSE equal to 0.69 and R2 equal to 0.72. In Brazil, TerraClimate underestimated data of air temperature and reference evapotranspiration, while overestimated data of rainfall. For all variables, the calibration reduced the bias of TerraClimate for under or overestimate the AWS data. All the statistical metrics were improved after the calibration process. After the evaluation of the data and to show the impact of calibration on datasets, we carried out applications to the Terraclimate data with and without calibration. Based in the findings, we have confirmed that it is crucial the evaluation of global datasets with observed data before use it, because sometimes the data may not represent well some specific regions of the territory. Not knowing the data provided can lead us to make wrong decisions, mainly when the information will be derived from the dataset, which is the case of the water balance application showed in the present study. •Evaluation of s