Introduction
Climate change, through altering precipitation patterns, increasing temperatures, and intensifying extreme events, poses a serious threat to water resources and ecosystems in arid and semi-arid regions. The limited coverage of meteorological stations in these regions hinders accurate climate monitoring. In recent years, reanalysis datasets such as ERA5-Land and TerraClimate have gained attention as effective alternatives to observational data. Previous studies have highlighted their varying performance across climate variables and emphasized the need for bias correction. This study focuses on the Maharloo Wetland Basin to evaluate the accuracy and bias correction of ERA5-Land and TerraClimate compared to station data, aiming to enhance their application in climate change analysis and water resource management.
Materials and Methods
Daily temperature and precipitation data from three synoptic stations—Shiraz, Doroodzan, and Zarghan—were collected for the period 1991–2022. Simultaneously, monthly climate data from ERA5-Land and TerraClimate for the same period were extracted. The Pearson correlation coefficient was applied to evaluate the agreement between reanalysis and observational data. Error metrics including PBIAS, RMSE, and NSE were then used to assess the accuracy of the reanalysis datasets. To reduce systematic errors, a simple linear regression–based bias correction method was applied. Finally, changes in the error indices before and after correction were compared to evaluate the effectiveness of the method.
Results
The results of Pearson correlation coefficient analysis showed a strong and statistically significant correlation (p-value = 0, r>0.9) between ERA5-Land and TerraClimate datasets and station observations at Shiraz, Doroodzan, and Zarghan. Error evaluation using PBIAS, RMSE, and NSE indicated that ERA5-Land performed better than TerraClimate in reproducing monthly precipitation, with higher NSE values and lower PBIAS across all stations. For temperature variables, both datasets showed relatively good performance, with generally high NSE values and low RMSE, although notable PBIAS was observed in some variables, particularly minimum temperature. After applying the simple linear regression–based bias correction, error indices improved considerably. For example, for ERA5-Land monthly precipitation, NSE increased from 0.9352 to 0.9716 and RMSE decreased from 6.92 to 4.58 mm. In ERA5-Land minimum temperature, PBIAS was reduced from −51.5% to 0.026%. TerraClimate temperature variables also showed reduced RMSE and PBIAS and higher NSE values after correction.

Discussion
The results of this study showed that ERA5-Land and TerraClimate performed well in representing the climatic patterns of the study area. Pearson correlation coefficient values above 0.9 indicate the strong potential of these datasets for climate analysis in data-scarce regions. Evaluation of NSE, RMSE, and PBIAS indices revealed that both datasets showed good accuracy in representing mean and maximum temperature, while higher errors were observed for minimum temperature. For precipitation, ERA5-Land demonstrated better performance compared to TerraClimate. Bias correction using a simple linear regression model led to a significant improvement in statistical indices, including reduced PBIAS and RMSE and increased NSE for both temperature and precipitation variables. However, the limitation of linear methods in representing simultaneous relationships between variables such as temperature and precipitation should be considered. The use of nonlinear approaches, such as Quantile Mapping and machine learning algorithms, can better capture extreme events and further reduce errors. Overall, the findings indicate that reanalysis datasets—particularly ERA5-Land—have a strong baseline accuracy and, with simple bias correction methods, can be effectively used in climate studies in arid and semi-arid regions. Nevertheless, applying nonlinear methods and incorporating more extensive observational data can substantially enhance their accuracy and applicability.