*4.2. Spatial and Temporal Variations in Future ET0*

As a result of global climate change, the hydrometeorological elements and hydrological environment of the YRB have been significantly affected [67]. Although many studies have examined the trends and attribution of *ET*<sup>0</sup> in the YRB [36,68], previous studies were based on site-scale and historical data. Studies on future water balance and hydrological cycles in the YRB are relatively weak owing to the lack of studies on *ET*0.

This study predicted *ET*<sup>0</sup> trends in the YRB for different periods under four future emission scenarios based on CMIP6 temperature data with high spatial and temporal resolution generated by the delta statistical downscaling method and the Hargreaves model. This study revealed an overall significant increase in *ET*<sup>0</sup> in the YRB from 2022 to 2100 (*p* < 0.01) (Figure 8), without the "evaporation paradox" [69], which is similar to the future *ET*<sup>0</sup> trends predicted by many studies [28,69,70]. Based on the Hargreaves model (i.e., Equation (2)), the change in *ET*<sup>0</sup> is proportional to tas and the difference between tasmax and tasmin. With general global warming, tas, tasmax, and tasmin in the YRB in the future period showed sudden increases relative to the historical period (Figure S1). Therefore, the abrupt increase of *ET*<sup>0</sup> can be attributed to the abrupt increase of temperature-like variables in the future period [27]. Radiative forcing is expected to stabilize at 2.6 W/m2 by 2100 for SSP126 and 8.5 W/m<sup>2</sup> by 2100 for SSP585, and radiation values and temperature are positively related to *ET*<sup>0</sup> [4]. The positive effects of increasing climatic factors, such as temperature and radiation, on *ET*<sup>0</sup> in the YRB were greater than the negative effects of other factors, and therefore, the latter trend was greater than the former as emission concentrations increased in the YRB (Figure 8). The *ET*<sup>0</sup> changes in the YRB in the near-, mid-, and long-term future under different future scenarios exhibited high spatial heterogeneity (Figures 10 and 12), with a spatial distribution high in the west and low in the east, and the *ET*<sup>0</sup> increase became more significant as the radiative forcing scenario increased. Consistent with the results of Ding and Peng [31], the increase in *ET*<sup>0</sup> was generally greater at higher elevations than at lower elevations in the basin, with the most pronounced change in *ET*<sup>0</sup> in the western part of the basin, reaching a maximum variation of 112.91% compared to that in the historical period (Figure 12). According to the *ET*<sup>0</sup> equation (i.e., Equation (2)) and Figure S2, this change can be attributed to the largest temperature difference between tasmax and tasmin in the western part of the basin. In addition, as shown in the change in future precipitation in the YRB relative to the historical period (Figure S3), precipitation in the western part of the basin showed less growth overall and even negative growth in some phases. Warming and decreasing precipitation caused an increase in dryness in the western part of the basin, and the warm-dry trend intensified. Wang et al. [71] found that the evapotranspiration process was more sensitive to relative humidity in the western part of the basin, and a decrease in relative humidity caused an increase in evapotranspiration. Therefore, *ET*<sup>0</sup> predictions based on the Hargreaves model

were greater in these areas. Water loss in the YRB is likely to accelerate in the future than in historical periods, which will threaten the food and ecological security of the region; thus, developing flexible mitigation strategies tailored to local conditions is critical to coping with climate change [72].

#### *4.3. Climate Model Uncertainty Analysis*

Because of the differences in the feedback processes of different GCMs, a certain degree of uncertainty exists in their response to future greenhouse gas emissions, and the actual generalized optimal climate models and *ET*<sup>0</sup> models cannot be determined [73]. Inevitable uncertainties exist in future *ET*<sup>0</sup> predictions stemming from climate scenarios, climate models, and *ET*<sup>0</sup> models [74], which greatly affect the confidence of the prediction results.

In response to the uncertainty of climate scenarios, this study selected the CMIP6 data, which had the largest number of participating models, the richest design of numerical experiments, and the largest amount of simulated data available than other CMIP generations for more than 20 years of the CMIP [37,75], initiated by the current Working Group on Coupled Models (WGCM). Compared to previous generations (CMIP3, CMIP5, etc.), CMIP6 uses a new scenario combining shared socio-economic pathways and typical concentration pathways to constrain multi-model predictions of key climate change indicators such as global surface temperature and ocean heat content based on historical observations, climate simulations, and climate sensitivity awareness, reducing uncertainty in predictions and providing higher resolution and reliability [76], thereby making the results more informative and time-sensitive than those based on CMIP5 for future *ET*<sup>0</sup> studies, such as in Ahmadi and Baaghideh [25], Ding and Peng [31], Kundu et al. [6] and Le and Bae [77]. In response to the uncertainty of climate models, this study selected 24 GCMs with historical and future emission scenarios, which is more extensive than the studies of Li et al. [70], which only used one climate model (HadCM3) under two emission scenarios (A2 and B2), and Nooni et al. [1], which used only one climate model (CNRM-CM6). In addition, this study reduced the uncertainty of future temperature data by preferentially selecting climate models based on multiple interpolation methods, multiple evaluation indicators, and equal weight sets on a downscaling basis. The MAE was controlled within 2.5 mm, S and SS were approximately 1, and TS was approximately 0, indicating very high simulation accuracy (Table 3). Wang and Chen [24] reduced the spatial resolution of GCMs' data to 0.5◦ based on the delta method, and the MAE of tas was in the range of 1.6–5.7 ◦C. However, the MAE of tas in this study was controlled in the range of 2.2–2.6 ◦C and had a higher spatial resolution of 1 km.

Although this study provides a comprehensive theoretical basis for future *ET*<sup>0</sup> assessments, the uncertainties in the downscaling of GCMs [78] and in the selection and accuracy of *ET*<sup>0</sup> models [74] may impact the prediction results. To improve *ET*<sup>0</sup> estimates in future studies, consideration should be given to the long-term goal of the United Nations Framework Convention on Climate Change (Paris Agreement) to limit the increase in global average temperature to less than 2 ◦C compared to the pre-industrial period and to further efforts to limit it to less than 1.5 ◦C [79], as well as to achieve China's 2060 carbon neutrality and global carbon neutrality. Describing and quantifying the relative importance of various uncertainty sources and the risks they pose in the assessment is important in current and future climate change impact studies and water resource assessments that should be strengthened to reduce prediction uncertainty.


**Table 3.** Evaluation of the fitting error of downscaled climate models for monthly temperatures (tas/tasmax/tasmin) in the Yellow River Basin from January 1995 to December2014.


Note: MAE, S, SS, and TS are mean absolute error, Taylor diagram-based quantifiers, spatial skill scores, and temporal skill scores, respectively. MAE is measured in ◦C, and the others are dimensionless indicators. The closer MAE and TS are to 0, the better the simulation ability of the model; the closer S and SS are to 1, the better the simulation ability of the model.

#### **5. Conclusions**

Based on the 24 GCMs in CMIP6 and temperature data with high spatial and temporal resolution generated by the delta statistical downscaling model, this study assessed the evolution of *ET*<sup>0</sup> in the YRB under four emission scenarios (SSP126, SSP245, SSP370, and SSP585) for the near (2022–2040), mid- (2041–2060), and long (2081–2100) term future. The major conclusions are as follows:

The regionally high-precision climate data generated by delta statistical downscaling based on multiple interpolation methods reduced the uncertainty in the GCM dataset. For the YRB, tas selected the climate models 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. The equal-weighted multimodel ensemble had smaller mean absolute errors and higher correlation coefficients than single climate models, and CMIP6 efficiently simulated the temperature and *ET*<sup>0</sup> in the YRB.

Compared with that of the historical period (1901–2014), the annual *ET*<sup>0</sup> in the YRB under different emission scenarios (SSP126, SSP245, SSP370, and SSP585) in the future (2022–2100) substantially increased; the rate increased with the increase in emission concentration, and the *ET*<sup>0</sup> in 2100 under the SSP585 scenario reached 1170.39 mm. Morlet wavelet analysis revealed that *ET*<sup>0</sup> in the YRB had cyclic patterns of 34–38, 34, 39, and 27–32 years under the SSP126, SSP245, SSP370, and SSP585 scenarios, respectively.

Compared with that in the historical period, the *ET*<sup>0</sup> variation in the YRB in the near-, mid-, and long-term future under different future scenarios exhibited strong spatial heterogeneity. EOF analysis revealed that *ET*<sup>0</sup> had positive EOF1 values under all four emission scenarios, exhibiting a spatially consistent trend of *ET*<sup>0</sup> variation across the region. A maximum variation of 112.91% occurred in the western part of the YRB in the longterm future (2081–2100) under the SSP585 scenario. Without a scientific response, future increases in *ET*<sup>0</sup> could further reduce the shortage of water resources in the YRB.

**Supplementary Materials:** The following supporting information can be downloaded at: https:// www.mdpi.com/article/10.3390/rs14225674/s1. Figure S1: Interannual variations in tas (a)/tasmax (b)/tasmin (c) in the Yellow River Basin over the historical period (1901–2014) and under different future emission scenarios; Figure S2: Spatial variations in the near (2022–2040; (a,d,g,j)), mid- (2041– 2060; (b,e,h,k)), and long (2081–2100; (c,f,i,l)) term future difference between tasmax and tasmin of the Yellow River Basin relative to the historical period (1901–2014) under four SSP scenarios (SSP126, SSP245, SSP370, and SSP585); Figure S3: Spatial variations in the near (2022–2040; (a,d,g,j)), mid- (2041–2060; b,e,h,k), and long (2081–2100; (c,f,i,l)) term future annual precipitation of the Yellow River Basin relative to the historical period (1901–2014) under four SSP scenarios (SSP126, SSP245, SSP370, and SSP585).

**Author Contributions:** Conceptualization, S.J. and A.W.; methodology, C.S.; validation, S.J., A.W. and C.S.; investigation, A.W.; resources, S.J.; data curation, A.W.; writing—original draft, S.J. and A.W.; writing—review and editing, C.S. and K.W.; supervision, A.W. and K.W.; funding acquisition, S.J. and C.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Training Program for Young Backbone Teachers in Colleges and Universities of Henan Province (2021GGJS003), the Henan Natural Science Foundation (212300410413), the Henan Youth Talent Promotion Project (2021HYTP030), the China Postdoctoral Science Foundation (2020M672247), the Key Science and Technology Project of Henan Province (No. 222102320108), and the First-class Project Special Funding of Yellow River Laboratory (No. YRL22IR12).

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
