Climate change and the growing intensity of its impacts have increased the need for accurately reproducing precipitation and temperature at the station scale, where local climatic characteristics are most relevant for hydrological assessments and regional risk management. Because CMIP6 climate model outputs have coarse spatial resolution and contain systematic biases, they cannot be directly used for local‑scale studies without applying statistical downscaling and bias‑correction techniques. Selecting an appropriate combination of climate model and correction method is therefore an essential prerequisite for reliable climate projection, especially in regions with complex climatic variability such as eastern Iran.

To address the lack of integrated and multi‑variable evaluations in this region, the present study introduces a station‑based, multi‑criteria framework for the simultaneous ranking of model–method combinations in reproducing precipitation, minimum temperature (Tmin), and maximum temperature (Tmax) across the three provinces of Khorasan. The analysis covers 19 synoptic stations over the period 1989–2014. Within this framework, outputs from eight CMIP6 models—five Earth System Models (ESMs) and three General Circulation Models (GCMs)—were evaluated using eight bias‑correction and downscaling techniques implemented in the CMHyd software. These methods include five approaches for precipitation and four for temperature, representing widely applied techniques in statistical downscaling.



Model–method performance was assessed using several complementary statistical metrics: normalized root mean square error (NRMSE), mean absolute error (MAE), Pearson correlation coefficient ®, and the Kling–Gupta efficiency (KGE). In addition, variable‑specific measures were used: percent bias (PBIAS) for precipitation and mean bias error (MBE) for temperature. After normalizing all criteria, weights were assigned using the Shannon entropy method, ensuring objective and regionally consistent weighting. Station‑level rankings were then produced using the TOPSIS decision‑making technique, and the aggregated results were summarized using median ranks, interquartile ranges (IQR), and the number of stations where each alternative performed best. Statistical significance of performance differences was tested using the Friedman and Nemenyi tests.

The results showed significant differences among alternatives for all three variables, with MPI|VarianceScaling emerging as the best overall option for Tmin and MPI|DistributionMapping for Tmax. However, according to the Nemenyi test, several additional combinations fell within the same statistical group as the top performer, forming an “excellent group” with similarly strong and stable performance indices. For precipitation, HadGEM|DistributionMapping achieved the best overall results, outperforming other members of the excellent group in most stations.

In contrast, combinations based on INM, CMCC, and GFDL models, as well as several correction techniques such as DeltaChange, LinearScaling, LocalIntensity, and PowerTransformation, consistently ranked lower and showed significant divergence from the excellent group.

Overall, the findings highlight that MPI combined with Distribution Mapping or Variance Scaling, and HadGEM combined with Distribution Mapping, provide the most reliable performance for reproducing temperature and precipitation in Khorasan. The proposed framework offers a structured basis for reducing uncertainty in climate model selection and for improving regional climate‑hazard assessment and future projection efforts.