**1. Introduction**

Climate change and its impacts on the water cycle, particularly on regional hydrological systems, are major global challenges in the 21st century [1–3]. As an important factor in the regional hydrological cycle and energy balance, reference evapotranspiration (*ET*0) can be used to make total energy estimates of actual evapotranspiration [4], and is the component of the water cycle that is directly affected by climate change. Changes in *ET*<sup>0</sup> have a significant impact on the global water cycle and water resources [5], thereby leading to droughts and floods, water scarcity, and ecosystem degradation. In the context of climate change, *ET*<sup>0</sup> is an important guide for understanding the hydrological cycle and formulating water resource plans in watersheds [6–8].

Although studies on *ET*<sup>0</sup> have been conducted recently [9–14], most existing studies focused on the historical period. With the development of global climate models (GCMs), exploring the future *ET*<sup>0</sup> of watersheds based on historical data has become a topic of research interest in the context of climate change. GCMs are the most powerful tools for

**Citation:** Jian, S.; Wang, A.; Su, C.; Wang, K. Prediction of Future Spatial and Temporal Evolution Trends of Reference Evapotranspiration in the Yellow River Basin, China. *Remote Sens.* **2022**, *14*, 5674. https:// doi.org/10.3390/rs14225674

Academic Editors: Alban Kuriqi and Luis Garrote

Received: 15 October 2022 Accepted: 7 November 2022 Published: 10 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

climate change modeling and future predictions [15,16], and the modeling results can provide valuable data to support studies on climate change-induced impacts at regional and continental scales. Nevertheless, low-resolution data will lead to large biases in the prediction of regional climate change when climate studies are conducted at regional scales. Downscaling is an effective method for transforming large-scale, low-resolution outputs from GCMs into small-scale, high-resolution regional ground information [17,18]. Current mainstream downscaling methods include dynamic downscaling [19] and statistical downscaling [20]. Compared to statistical downscaling, dynamic downscaling requires a large number of complex inputs and computational requirements [21,22], and sometimes fine and reliable climate data at regional scales are not available [22,23]. Statistical downscaling is the most widely used and established downscaling technique in basin climate change studies because of its low computational cost, easy model construction, multiple implementation methods, ease of operation, and lack of consideration of the influence of boundary conditions on prediction results [24].

Two methods are widely used for *ET*<sup>0</sup> prediction under future climate scenarios: (1) input of future meteorological data from GCMs into *ET*<sup>0</sup> models [25,26]; and (2) directly predicting future *ET*<sup>0</sup> via downscaling methods based on historical *ET*<sup>0</sup> [1,27]. Liu et al. [28] used Coupled Model Intercomparison Project Phase 5 (CMIP5) and Coupled Model Intercomparison Project Phase 6 (CMIP6) climate models to compare global potential evapotranspiration and found that both models could effectively simulate the increasing trend; they also revealed that CMIP6 multi-model results simulated higher values of global potential evapotranspiration than CMIP5 for the same emission scenario. Nistor et al. [29] assessed the impact of climate change on *ET*<sup>0</sup> in Turkey in the 21st century based on the Thornthwaite equation and the CMIP5 dataset. They revealed that *ET*<sup>0</sup> will increase in southern and southeastern Turkey and along the Mediterranean coast in the coming period owing to climate warming.

The Yellow River Basin (YRB) is an important component of China's strategic ecological security pattern, and most of the YRB is an arid and semi-arid region. Because of its unique geographical location, its environment is fragile and highly sensitive to global climate change [21,30], making it a good indicator of climate change. Despite the high sensitivity of the region to climate change, studies on the evolution of *ET*<sup>0</sup> in the YRB in the context of future climate are limited, and most of the existing studies on future *ET*<sup>0</sup> in other regions are at the CMIP5 stage [21,30,31], with no downscaling treatment [1], a single spatial interpolation method [21], or a single indicator for climate model preferences [32]. Therefore, against the backdrop of global warming, the *ET*<sup>0</sup> predictions in the YRB can provide a theoretical reference basis for water resource planning and management, as well as a scientific basis for relevant authorities to formulate future climate change response strategies.

This study used the YRB as the study area and developed a multi-model ensemble based on the delta statistical downscaling using multiple interpolation methods and multiple evaluation indicators to predict the spatial and temporal evolution characteristics of *ET*<sup>0</sup> in the YRB under different CMIP6 emission scenarios. Studies on *ET*<sup>0</sup> not only enhance the understanding of hydrological processes in the YRB but also provide data to support and guide future water resource management and drought mitigation. The specific objectives are to: (1) obtain monthly mean, maximum, and minimum temperature datasets in the YRB with a resolution of 1 × 1 km based on CMIP6 climate model data and delta statistical downscaling; (2) select the best simulated climate model and multi-model ensemble by evaluating and validating historical measured data; and (3) predict the spatial and temporal changes in *ET*<sup>0</sup> under different emission scenarios in the future based on the Hargreaves formula and downscaled temperature data from 2022 to 2100.
