Introduction

Understanding climate change and its impacts on agriculture requires access to long and reliable daily weather records. For robust estimation of crop yield levels and variability, daily climate data over at least 10–20 years are needed. Reanalysis datasets, which combine observations with numerical weather prediction models, have therefore become key inputs for agro-climatic applications, including climate change assessment, water-resource management and crop modelling. Products such as ERA5, MERRA-2, NCEP/NCAR and the agriculture-oriented AgMERRA provide spatially continuous information on temperature, solar radiation and other variables, and have been successfully used to drive models such as DSSAT and WOFOST in different regions. Previous studies generally report good performance of ERA5 and AgMERRA for temperature and, to a lesser extent, for solar radiation, although biases and spatial heterogeneity in skill are frequently noted. In Iran, and particularly in the three Khorasan provinces in the north-east, the evaluation and use of both datasets remain limited despite their potential to support climate-change impact assessment and agricultural planning. The study therefore aims not to prove the superiority of one product over the other, but to provide an unbiased assessment of the performance and practical usability of both for daily temperature and solar radiation in this data-sparse region.



Materials and Methods

The study area comprises the three Khorasan provinces in north-eastern Iran, extending from 30°50′ to 38°30′N and 55°35′ to 61°25′E, with an area of about 294,000 km² (Fig. 1). Daily data from 42 synoptic stations (Table 2) were used, including minimum, maximum and mean air temperature and incoming solar radiation. Gridded reanalysis data were obtained in NetCDF format from ERA5 and AgMERRA (Table 1). ERA5 provides hourly temperature and radiation at 0.25° spatial resolution for 1940–2022, while AgMERRA provides daily data at 0.5° resolution for 1980–2010; ERA5 data were aggregated to daily values. Because the evaluation periods differ but partly overlap, direct comparison of skill metrics is only approximate and long-term analysis focuses on ERA5. For each station, the nearest grid cell was extracted in R and daily reanalysis values were compared with observations. Performance was assessed using Pearson’s correlation coefficient (r), root mean square error (RMSE), normalized RMSE (NRMSE, %), mean bias error (MBE) and Willmott’s index of agreement (d). For solar radiation, both reanalysis products were first evaluated against pyranometer measurements at stations with sufficiently long records. In a second step, ERA5 and AgMERRA were compared with daily radiation estimated from the Angström–Prescott relationship at nine synoptic stations in Razavi Khorasan. This two-stage approach was adopted because direct radiation measurements are sparse in the region and Angström–Prescott estimates are widely used as a surrogate in data-scarce agro-climatic studies.



Results and Discussion

Daily minimum temperature (tmin) was generally well reproduced by both reanalysis products. Across the 42 stations, ERA5 showed r mostly around 0.96–0.98 (mean ≈0.97) and AgMERRA around 0.92–0.96 (mean ≈0.94). The index of agreement d was high for both datasets (mean ≈0.98 for ERA5 and 0.95 for AgMERRA). However, NRMSE for tmin in ERA5 was typically about 4–9% (mean ≈5%), compared with roughly 5–13% (mean ≈7%) for AgMERRA, and both products exhibited a predominantly cold bias. In ERA5, most stations showed MBE close to zero or down to about −2 to −3 °C, whereas in AgMERRA some stations underestimated tmin by about 3–5 °C, which can be critical for frost-related applications. For daily maximum temperature (tmax), performance was even more robust. In almost all stations, r was close to 0.98–1.00 in ERA5 and about 0.97–0.98 in AgMERRA, with d generally above 0.97 in both datasets. Relative errors were mostly within 3–7% for ERA5 and about 4–9% for AgMERRA. Despite this high skill, both products tended to underestimate tmax by roughly 1–4 °C at many stations, while a few sites showed mild warm biases. Mean daily temperature (tm) also showed very high correlations (r >0.98 for ERA5 and >0.97 for AgMERRA), but NRMSE reached about 3–9% in ERA5 and 3–16% in AgMERRA, with larger relative errors at some northern and southern stations. Bias in tm was again predominantly negative, implying that degree-day indices derived from these data may be underestimated if no bias correction is applied. From a climatic perspective, the spatial pattern of skill is consistent with the geography of the Khorasan region. In open, relatively homogeneous plains, where temperature variability is mainly controlled by large-scale synoptic systems and regional temperature gradients, both datasets—especially ERA5—reproduce daily temperatures with high r, low NRMSE and limited bias. In contrast, in mountainous areas, valley margins and desert fringes, complex topography, strong nocturnal radiative cooling and large diurnal ranges increase the discrepancy between point measurements and grid-cell means, particularly for tmin and tm. This reflects the limited spatial resolution and inherent smoothing of reanalysis fields and explains why some stations in northern highlands and southern arid zones show larger cold biases and relative errors. These temperature results are broadly consistent with previous evaluations of AgMERRA in North and Razavi Khorasan and with recent national-scale assessments indicating that ERA5 generally provides the most accurate air-temperature estimates over Iran, with larger errors in complex mountain and coastal regions. Considering the longer evaluation period for ERA5 compared with AgMERRA, direct comparison of skill metrics remains approximate. Nevertheless, the combined evidence indicates that both datasets offer acceptable accuracy for daily tmin, tmax and tm at station scale in the three Khorasan provinces, and that statistical bias correction can further enhance their suitability for climate, hydrological and agricultural impact studies.



Conclusion

This study showed that both datasets reproduce daily tmin, tmax and tm over the three Khorasan provinces with high correlations and generally small relative errors, although a cold bias remains at some mountainous and desert stations and for tmin and tm. For solar radiation, ERA5 performs acceptably at most sites, whereas AgMERRA has more limited spatial coverage and a tendency to underestimate. Considering its longer temporal coverage, finer spatial resolution and ongoing updates, ERA5 emerges as the preferred dataset for long-term climate, water and agricultural applications in the region, while both products should ideally be used with local bias correction.