*3.1. Simulation Accuracy Assessment of Regional Temperatures and Multi-Model Ensemble*

For the three temperature variables, tas, tasmax, and tasmin, out of all interpolation methods, IDW exhibited the lowest overall error for all four different evaluation metrics, and therefore, it was used to further assess the effectiveness of the temperature simulations for all climate models (Figures 3–5). For tas, the climate models ACCESS-ESM1-5, CMCC-CM2-SR5, and GISS-E2-1-G were ranked high for each evaluation index; for tasmax, ACCESS-CM2, NESM3, and ACCESS-ESM1-5 were ranked high for each evaluation index; for tasmin, ACCESS-CM2, ACCESS-ESM1-5, and GISS-E2-1-G were ranked high for each evaluation index. Because some climate models do not have data on future emission scenarios, considering the data integrity and actual fitting performance of climate models, tas selected ACCESS-ESM1-5, CMCC-CM2-SR5, and INM-CM5-0; tasmax selected ACCESS-CM2, ACCESS-ESM1-5, and MRI-ESM2-0; and tasmin selected ACCESS-CM2, ACCESS-ESM1-5, and MPI-ESM1-2-LR for the ensemble analysis of subsequent temperature patterns in the YRB. The fitting results of the simulated data of the selected climate model for each variable and MME of the 1995–2014 period to the observed data from 93 ground-based meteorological stations (Figure 6) show that, although the simulated tas, tasmax, and tasmin of the YRB by CMIP6 were slightly lower than the observed values, most of the points were near the 1:1 line, and the *R*<sup>2</sup> and regression coefficients were greater than 0.9, with all passing the 99% confidence level test. The simulated and observed values of the MME dataset were more concentrated than those of the three independent climate models, and the *R*<sup>2</sup> values of the three variables were 0.9418, 0.9226, and 0.9362, respectively (Figure 6d,h,l), which reduced the errors caused by outlier points and slightly improved the fit of the simulated data. The above analysis reveals that climate models have high application potential in the YRB, and the CMIP6 multi-model ensemble is a good reference value for predicting *ET*<sup>0</sup> trends in the YRB under future climate scenarios.


**Figure 3.** Fitting performance ranking of 21 climate models simulating tas monthly series of the Yellow River Basin from 1995 to 2014 under different interpolation methods and evaluation metrics. a, b, c, d, and e under each evaluation method are the evaluation rankings of the five interpolation methods: BI, IDW, kriging, NNI, and spline, respectively. The higher the ranking, the higher the model's fitting accuracy.

**Figure 4.** Fitting performance ranking of 19 climate models simulating tasmax monthly series of the Yellow River Basin from 1995 to 2014 under different interpolation methods and evaluation metrics. a, b, c, d, and e under each evaluation method are the evaluation rankings of the five interpolation methods: BI, IDW, kriging, NNI, and spline, respectively. The higher the ranking, the higher the model's fitting accuracy.

**Figure 5.** Fitting performance ranking of 21 climate models simulating the tasmin monthly series of the Yellow River Basin from 1995 to 2014 under different interpolation methods and evaluation metrics. a, b, c, d, and e under each evaluation method are the evaluation rankings of the five interpolation methods: BI, IDW, kriging, NNI, and spline, respectively. The higher the ranking, the higher the model's fitting accuracy.

**Figure 6.** Scatter density plots of measured and selected climate models and multi-model ensembles simulating monthly temperatures in the Yellow River Basin at stations from 1995 to 2014. (**a**−**c**) correspond to the fitting effects of ACCESS-ESM1-5, CMCC-CM2-SR5, and GISS-E2-1-G, respectively, under tas variable; (**d**) corresponds to the fitting effect of MMEtas; (**e**–**g**) correspond to the fitting effects of ACCESS-CM2, ACCESS-ESM1-5, and MRI-ESM2-0, respectively, under tasmax variable; (**h**) corresponds to the fitting effect of MMEtasmax; (**i**–**k**) correspond to the fitting effects of ACCESS-CM2, ACCESS-ESM1-5, and MPI-ESM1-2-LR, respectively, under tasmin variable; and (**l**) corresponds to the fitting effect of MMEtasmin.
