Effect of Elevation on Variation in Reference Evapotranspiration under Climate Change in Northwest China

Abstract

Through its effects on water and energy cycles, elevation plays an important role in modulating the spatial distribution of climatic changes in mountainous regions. A key hydrological indicator, reference evapotranspiration (ET₀) reflects the maximum amount of water transferred to the atmosphere from the land surface. The current scarcity of information regarding elevation’s impact on variation in ET₀ under climate change limits our understanding of the extent to which elevation modulates interactions between ET₀ and climate change and of the attendant processes involved. Drawing upon long-term (1960–2017) meteorological observations from 84 stations in Northwest China (NWC), we examined (i) spatial and temporal variations in ET₀; (ii) the sensitivity and contribution of air temperature (T), sunshine duration (SD), relative humidity (RH), and wind speed (WS) to ET₀; (iii) the existence of a relationship between elevation and ET₀ trends; and (iv) the major factor in controlling this relationship by using attribution analysis. Overall, annual ET₀ in NWC showed a declining trend between 1960 and 2017, though at a change point in 1993, the trend shifted from a decline to a rise. A significant correlation between temporal change in ET₀ and elevation confirmed the existence of a relationship between elevation and ET₀ variation. The effect of elevation on changes in ET₀ depended mainly on the elevation-based tradeoff between the contributions of T and WS: WS was the primary factor contributing to the decrease in ET₀ below 2000 m, and T was the dominant factor contributing to the increase of ET₀ above 2000 m. The rate of reduction in WS declined as elevation increased, thereby diminishing its contribution to variation in ET₀. The present study’s results can serve to guide agricultural irrigation in different elevation zones under NWC’s evolving climatic conditions.

1. Introduction

Reflecting the maximum quantity of water transferred to the atmosphere from a land surface drawing upon an unlimited water supply, reference crop evapotranspiration (ET₀) is a key hydrological indicator [1], widely used to derive actual crop evapotranspiration [2], explore the hydrological water balance [3], plan irrigation scheduling, and implement water resources management [4]. Global warming is expected to accelerate the hydrological cycle, thereby significantly influencing global water storage and distribution [5,6,7]. As one can reasonably assume that coming changes in key climate change variables (e.g., air temperature, relative humidity, wind speed, and solar radiation) will lead to changes in ET₀, a better understanding of these variables’ current influence on ET₀ is important in planning future regional agricultural practices, ecosystem services, and water resources management.

While one might expect an increasing trend in ET₀ under global warming [2,8], previous studies have reported both increases and decreases in ET₀ across different regions of the world [3,4,9,10,11,12,13,14]. For example, Pandey and Khare [15] identified a long-term increase in ET₀ over the humid tropical Narmada River in India. Prăvălie et al. [4] showed a similar increase in ET₀ in Romania. Increasing trends of ET₀ were also reported in Iran [16], the Mediterranean [17], Spain [18], and some parts of China [19]. However, decreasing trends in ET₀ have been recorded in certain regions of China: in the middle [12] and upper [20] reaches of the Yellow River basin, and in Xinjiang province [21]. Widely described as the ‘evaporation paradox’ [22], these counterintuitive relationships between increasing air temperature and declining ET₀ have nonetheless received little attention in previous studies, limiting our understanding of the relationship between ET₀ and climate change.

Under climate change, variations in ET₀ are based not only on the air temperature but also on other climatic variables influenced by global climate change [2,13]. However, these variables are also strongly affected by regional conditions such as topography and elevation [23]. Therefore, ET₀ has different responses to changing climatic variables in different regions and under different climatic conditions. Having divided the mainland of China into different climatic zones to evaluate the climate’s impact on ET₀, Fan et al. [2] found the primary driver of change in ET₀ to vary between the different zones. Focusing on the sensitivity of ET₀ to elevation in a small catchment within China’s Qilian Mountains, Yang et al. [24] noted that the response of meteorological factors (e.g., net radiation, soil heat flux, air temperature, relative humidity, and wind speed) to elevation varied widely, resulting in a decrease in ET₀ as elevation increased. While elevation plays an important role in modulating spatial changes in climate within mountainous regions, and some scholars have drawn attention to elevation-dependent warming in high mountain areas [25,26], the effect of elevation on variation in ET₀ under a changing climate remains unclear.

Due to its complex topography and fragile vegetation, Northwest China (NWC) is one of the regions in China most sensitive to climate change. Drawing upon long-term (1960–2010) data from 80 observation stations across NWC, Liu and Zhang [27] revealed the roles of changing wind speed and surface air temperature on variation in ET₀. Zheng and Wang [28] found most stations in NWC to exhibit a significant downward trend in ET₀ between 1960 and 2010, with the greatest contribution coming from wind speed. However, these studies did not consider the role of the elevation gradient on ET₀ changes under such complex topography. Therefore, more thorough studies are needed in this region, using a longer observational dataset and a systematic investigation of climate change’s impact on current variation in ET₀. Moreover, since the elevation gradient is critical to regional ET₀, identifying the main drivers of ET₀ variation in different elevation zones would allow a better understanding of the region’s hydrological cycle and support rational water resources management within its confines.

Drawing upon long-term (1960–2017) observational data from NWC, we sought to (i) detect spatial and temporal variations in ET₀ at annual and seasonal scales; (ii) investigate the sensitivity of ET₀ to key climatic variables (e.g., mean air temperature (T), sunshine duration (SD), relative humidity (RH), and wind speed (WS)); (iii) evaluate the contribution of climatic variables to changes in ET₀ change; and (iv) discuss the role of elevational effects on ET₀ change under climate change in NWC. By helping to fill in the knowledge gap regarding how elevation influences variation in ET₀ under climate change, this study can provide a useful perspective on sustainable applications of regional water practices.

2. Materials and Methods

2.1. Study Area

Extending over a total area of approximately 2.23 × 10⁶ km² and varying in elevation from −154 to 6828 m, Northwest China (NWC), the site of the current study, is situated between 73.48–106.95° E and 34.44–47.93° N (Figure 1). Including Xinjiang province, large parts of Gansu and Qinghai provinces, small parts of Ningxia, and Inner Mongolia provinces, NWC accounts for 23.2% of China’s land area [29]. With a complex topography of interwoven high mountains, deserts, and oases, the region is structured by mountain-basin systems [8]. Affected by the northern hemisphere’s westerly winds, this region is characterized by an arid continental climate: hot in summer and cold in winter. Air temperature in this area can reach over 40 °C in summer and drop below −30 °C in winter [8]. Highly spatially heterogeneous across the region, the mean annual precipitation across the region is less than 200 mm. In desert areas such as the Tarim, Turpan, and Badain Jaran deserts, annual precipitation is even less than 50 mm, while in alpine regions such as Tianshan and Qilian Mountains, the annual precipitation can exceed 600 mm. The ratio of ET₀ to annual precipitation exceeds 1.5 and can be even larger in desert areas. The large water demand and shortage of water supply contribute to these ecosystems’ fragility [30].

2.2. Data Collection

To detect climatic change and estimate ET₀, continuous time series (1960 to 2017) of daily meteorological observations, rigorously controlled for quality by the Chinese Meteorological Administration, were obtained for 84 stations in NWC. Spatially well-distributed, these observation stations provided data reflecting the general characteristics of the regional climate and elevation gradient (Figure 1). The climatic variables used in this study are T, SD, RH, and WS. The average annual and seasonal ET₀ values for each station were aggregated by daily values. We defined the seasons as spring (March–May), summer (June–August), autumn (September–November), and winter (December–February).

2.3. Estimation of ET₀

The Penman–Monteith equation (PM), recommended by the Food and Agriculture Organization (FAO), is considered the best among various ET₀ estimation Equations [1]. The restrictive definition of ET₀ renders it a standard for estimating and comparing atmospheric evaporative demand in different climates worldwide [2,14]. The FAO PM equation can be divided into two components: aerodynamic (ET_0A) and radiative (ET_0R) evapotranspiration [1]:

$${\mathrm{ET}}_{0\mathrm{A}}=\frac{\gamma \frac{900}{T+273}{U}_{2}VPD}{\mathsf{\Delta }+\gamma \left(1+0.34U_2\right)}$$

(1)

$${\mathrm{ET}}_{0\mathrm{R}}=\frac{0.408\mathsf{\Delta }\left(R_n-G\right)}{\mathsf{\Delta }+\gamma \left(1+0.34U_2\right)}$$

(2)

$${\mathrm{ET}}_0={\mathrm{ET}}_{0\mathrm{A}}+{\mathrm{ET}}_{0\mathrm{R}}$$

(3)

where Δ is the slope of the saturation vapor pressure–temperature curve (kPa/°C); R_n is the surface net radiation (MJ/m²/day); G is the soil heat flux (M/m²/day), which can be ignored in daily simulation; γ is the psychrometric constant (kPa/°C); T is the average daily air temperature at a height of 2 m (°C); U₂ is the mean daily wind speed at a height of 2 m (m/s) and can be derived from the wind speed at a height of 10 m [31]; and VPD is the water vapor pressure deficit (kPa), calculated as [1]:

$$VPD=e_s-e_a$$

(4)

where e_s is the saturation vapor pressure (kPa) and e_a is the actual vapor pressure (kPa).

Net radiation is a function of solar radiation (R_s, MJ/m²/day). Based on the relationship between sunshine duration and R_s [1,32], solar radiation can be calculated as:

$$R_s=\left(a+b\frac{n}{N}\right)R_a$$

(5)

where $R_a$ is the extra-terrestrial radiation (MJ/m²/day), n is the sunshine duration (h), N is the maximum possible duration of daylight hours (h), and a and b are empirical coefficients with commonly recommended values of 0.25 and 0.50, respectively [1]. Considering the complicated territory and various climatic zones in the study area, the commonly recommended values are not suitable across the entire study area. The coefficients a and b were optimized using directly-measured solar radiation values obtained from China’s National Meteorological Information Center. Combining the results of Fan and Thomas [32] and Fan et al. [2], we adopted the mean values of the two studies, yielding a and b coefficients of 0.23 and 0.56, respectively, for NWC. More details about the calculation on ET₀ by using the FAO PM equation can be found in the reference of Allen et al. [1].

2.4. Statistical Analysis

2.4.1. Trend Estimation and Test of Its Significance

The Sen’s slope method was applied to the estimation of trends in the change of climatic variables and ET₀ (Sen, 1968). This technique calculates the slope as a change in measurement correlated with units of temporal change, thereby providing a real slope of the trend, which can differ slightly from the slope obtained by linear regression [13].

The rank-based Mann–Kendall (MK) statistical test developed by Mann [33] and Kendall [34] was employed to assess whether or not a trend is statistically significant (p < 0.05). This has the advantage of its significance being unaffected by the actual distribution of the data [35,36].

2.4.2. Abrupt Change Detection

The non-parametric Pettitt test [37] was used to detect any abrupt change point in the ET₀ times series. For a time series $X\left(n\right)$, the two sub-samples—before $X(x_1,\dots ,x_{\tau })$ and after $X(x_{\tau +1},\dots ,x_n)$ date $\tau$—have different distribution functions, such that the date $\tau$ can be regarded as a change point. The Pettitt statistic $k\left(\tau \right)$ can be calculated as [37,38]:

$$k\left(\tau \right)={\displaystyle \sum }_{i=1}^{\tau }{\displaystyle \sum }_{j=\tau +1}^{n}sgn(x_j-x_i)$$

(6)

where $x_i$ and $x_j$ are random variables with $x_i$ following $x_j$ in time. When the absolute value of $k\left(\tau \right)$ reaches a maximum, the abrupt change most likely takes place at the date $\tau$. The statistic is [39]:

$$K_n=\underset{1\le \tau <\mathrm{T}}{\max }\left|k\left(\tau \right)\right|$$

(7)

Note that the climatic variable series is identified to exhibit a significant abrupt change only when the result of a t-test is true. The significant probability associated with the value $K_n$ is given by [39]:

$$p=2\exp \left(\frac{-6{\mathrm{K}}_T^2}{T^3+T^2}\right)$$

(8)

where the approximation holds, accurate to two decimal places, for p < 0.05 [37].

2.5. Sensitivity Analysis

The sensitivity of ET₀ to T, SD, RH, and WS was assessed in the present study. Given that these variables have different dimensions and ranges, a dimensionless sensitivity coefficient was used to compare the sensitivity of ET₀ to each variable using partial derivatives. The sensitivity coefficient of ET₀ to a given variable $x_i \left[SC\left(x_i\right)\right]$ can be expressed as [13,28]:

$$SC\left(x_i\right)=\underset{\mathsf{\Delta }{x}_i\to 0}{\lim }(\frac{\mathsf{\Delta }E{T}_0/E{T}_0}{\mathsf{\Delta }{x}_i/x_i})=\frac{\partial E{T}_0}{\partial x_i}\frac{x_i}{E{T}_0}$$

(9)

A positive or negative sensitivity coefficient indicates that the ET₀ will increase or decrease along with an increase of the given variable. The larger the coefficient of a variable, the greater the impact the given variable, $x_i$, has on ET₀ [12,13].

2.6. Contribution Assessment

A differentiation equation method was used to investigate the contribution of different climatic variables (e.g., SD, T, RH, and WS) to change in ET₀. The contribution of the change of climatic variables to variation in ET₀ can be expressed as [12,19,40]:

$$\frac{dE{T}_0}{dt}=\frac{\partial E{T}_0}{\partial SD}\frac{dSD}{dt}+\frac{\partial E{T}_0}{\partial T}\frac{dT}{dt}+\frac{\partial E{T}_0}{\partial RH}\frac{dRH}{dt}+\frac{\partial E{T}_0}{\partial WS}\frac{dWS}{dt}+\delta$$

(10)

The equation can be simplified as:

$$\frac{dE{T}_0}{dt}=C\left(SD\right)+C\left(T\right)+C\left(RH\right)+C\left(WS\right)+\delta$$

(11)

where $\frac{dE{T}_0}{dt}$ is the long-term trend of ET₀ derived with the Sen’s slope method (observed variation); $C\left(SD\right)$, $C\left(T\right)$, $C\left(RH\right)$, and $C\left(WS\right)$ are the actual individual contributions of changes in SD, T, RH, and WS, respectively, to the long-term trend of ET₀; and $\delta$ is the systemic error term. The impact of a non-significant trend in climatic factors and the existence of simulation uncertainties can lead to the calculated ET₀ trend being slightly different from the observed one. In practice, the $\delta$ can be slightly smaller or larger than 0. Thus, a slight difference is acceptable. The estimation of the partial differential equation of ET₀ with respect to climatic factors is more fully described by Liu and Zhang [27].

A factor’s relative contribution can be estimated as [27]:

$$P\left(x\right)=\frac{C\left(x\right)}{\left|C\left(SD\right)\right|+\left|C\left(T\right)\right|+\left|C\left(RH\right)\right|+\left|C\left(WS\right)\right|}\times 100\%$$

(12)

where $P\left(x\right)$ is the relative contribution of a given climatic variable and $C\left(x\right)$ is the actual contribution of the given climatic variable.

3. Results

3.1. Spatial Variability and Temporal Trends of Climatic Variables

Spatial variation plots of long-term trends of change in annual T, RH, WS, and SD for 84 stations in NWC were explored. An increasing trend in T was dominant in the study area, with 82 out of 84 stations presenting non-decreasing trends of between 0 and 0.09 °C/yr and 79 out of the 82 stations showing significant changes over time (p < 0.05; Figure 2). Trends in RH showed no obvious pattern, with 37 stations presenting increasing trends and 47 stations decreasing trends. In terms of WS, 75 stations showed decreasing trends, 62 of which showed significant changes (p < 0.05). Decreasing trends in WS mostly ranged between −0.015 and −0.048 m/s/yr. SD also predominantly showed a decrease over time, with 59 stations presenting a decreasing trend, of which 40 showed significant changes (p < 0.05). The change in the magnitude of the decreasing SD trends was in the range of −9.14 to 0 h/yr. On an annual basis, T showed a significantly (p < 0.05) increasing trend of 0.032 °C/yr⁻ RH was stable, and WS and SD decreased significantly (p < 0.05) by 0.013 m/s/yr and 1.646 h/yr, respectively, from 1960 to 2017 in NWC.

A seasonal statistical analysis of changes in climatic variables within NWC (

Table 1. Seasonal and annual statistics of climatic variable trends from 1960 to 2017 in NWC.

	Spring	Summer	Autumn	Winter	Annual
T (°C/yr)	0.034 *	0.028 *	0.032 *	0.030 *	0.032 *
RH (%/yr)	−0.051 *	0.005	0.021	0.029	0.000
WS (m/s/yr)	−0.015 *	−0.014 *	−0.012 *	−0.009	−0.013 *
SD (hour/yr)	0.487	−0.580	−0.607 *	−0.895 *	−1.646 *

* represents a significance level of p < 0.05.

) showed T to significantly (p < 0.05) increase in all four seasons, with the increases in the spring and autumn being greater than those in the summer and winter. In contrast to the other three seasons, RH decreased and SD increased in the spring. The WS showed decreasing trends in all four seasons.

3.2. Spatiotemporal Variation in ET₀ and Its Dependency on Elevation

In order to verify the ET₀ that was calculated by the FAO PM equation and forced by observational meteorological variables from 84 stations in NWC, we compared our ET₀ calculation results with other ET₀ calculations [27], products [41], and site-based evaporation observations (ET_pan) (Figure S1). It is observed that the calculated ET₀ in this study was highly correlated with the results of Liu and Zhang [27], the global ET₀ product, and ET_pan, with the coefficient of determination (R²) all above 0.81. This means that the selections of the equations and parameters in this study were reasonable and the calculated ET₀ was reliable. Despite the similar use of forcing observational meteorological variables in our study, some differences existed between our results and those of Liu and Zhang [27]. There may be two reasons for this: (1) different stations used and regional extent and (2) different empirical coefficient selections in Equation (5).

The annual ET₀ varied from 609 to 1578 mm within the period of 1960 to 2017 in NWC, and was significantly influenced by elevation (Figure 3a). Lower annual ET₀ values were mainly distributed at high elevations (>3000 m) in the cold air of the Tianshan and Qilian Mountains’ alpine regions. Higher ET₀ (>1200 mm) values were mainly distributed at low elevations such as in the Tarim, Turpan, and Badain Jaran deserts, where the average elevation is below 1000 m and air temperature is high. The simulated ET₀ results were similar to those of Fan et al. [2] and Liu and Zhang [27] and were also consistent with the output of a regional climate model [8]. The spatial distribution of seasonal ET₀ components (Figure 3b) shows that for all 84 stations, summer ET₀ contributed the dominant proportion of annual ET₀, followed by spring, autumn, and winter ET₀. Regionally, the proportions were 41.37% for summer, 29.92% for spring, 16.92% for autumn, and 11.79% for winter.

ET₀ consists of radiative and aerodynamic components; the former is primarily affected by solar radiation, whereas the latter is affected by aerodynamics. Accordingly, we analyzed the ET₀ components (ET_0A and ET_0R) during the four seasons (Figure 3c–f). In mountainous areas, ET_0R for most stations was the dominant component during spring, summer, and autumn. In the plains and desert areas, the dominant component of ET₀ was ET_0A. In winter, ET_0A was dominant in eastern NWC, while ET_0R was dominant in western NWC.

Long-term (1960 to 2017) Sen’s slope derived ET₀ trends for the 84 stations ranged from −6.70 to 12.20 mm/yr (Figure 4), with 61 of 84 stations showing significant changes (p < 0.05). The Pettitt test method indicated abrupt change points in the annual ET₀ series of 54 out of 84 stations (Figure 4b,e). These ET₀ change points occurred in the 1980s and 1990s. Annual ET₀ trends differed in the pre-change and post-change periods (Figure 4c,d). During the pre-change period, 52 of 84 stations showed decreasing trends in annual ET₀, with 30 of the 52 stations’ showing significant changes (p < 0.05). During the post-change period, 61 out of 84 stations showed increasing trends, and for 41 of the 61 stations, these changes were significant (p < 0.05). The break point of the annual ET₀ time series was detected by Pettitt and MK abrupt detection methods; the results of the two methods indicate the break point of the annual ET₀ time series was 1993 in NWC (Figure 4e and Figure S2). The regional seasonal statistics results (

Table 2. Regional seasonal and annual statistics of ET₀ trends in NWC (mm/yr).

ET0 Trends	Spring	Summer	Autumn	Winter	Annual
Pre-change	−0.395	−0.274	−0.087	−0.099	−2.180 *
Post-change	1.153 *	0.888 *	0.280	0.131	4.692 *
Whole period	0.047	−0.323 *	−0.139	−0.017	−0.508

* represents a significance level of p < 0.05.

) indicated that ET₀ decreased during all seasons in the pre-change period and increased in all seasons in the post-change period. Annually, the ET₀ trend in the pre-change period was −2.18 mm/yr, whilst in the post-change period, it was 4.69 mm/yr (p < 0.05). Over the entire study period, the trend in annual ET₀ decreased at a rate of −0.51 mm/yr.

We investigated the relationship between elevation and annual ET₀ components, as well as ET₀ trends from 1960 to 2017 in NWC (Figure 5). The ET₀ and ET_0A significantly decreased (p < 0.01 and p = 0.013, respectively), while ET_0R showed no significant increase (p = 0.65) as elevation increased. The elevation gradient for annual ET₀ was −56.76 mm/km, for ET_0A, it was −62.33 mm/km, while for ET_0R, it was 5.57 mm/km. Therefore, the elevation gradient of ET₀ mainly results from that of the ET_0A. As ET_0A represents the aerodynamics-influenced portion of ET₀, combining the effects of WS and RH [14], it contributed a large proportion of ET₀ in NWC (Figure 3). Variation in WS showed a positive correlation with elevation, indicating the potential impact of WS on ET₀ (Figure S3). Yang et al. [24] reported that annual ET₀ in the Qilian Mountains had a significant elevation dependency and decreased as elevation increased across five meteorological stations with elevations ranging from 2980 to 4484 m. We also found the relationship between the trend in annual ET₀ and elevation was significantly correlated (p < 0.01) from 1960 to 2017 at an increasing rate of 0.75 mm/yr/km. This indicates the existence of the impact of elevation on the variation of ET₀ in NWC.

3.3. Sensitivity of ET₀ to Climate Parameters and How It Depends on Elevation

The sensitivity coefficient (SC) of ET₀ represents the relative change of ET₀ with respect to a given change in a climatic variable. The greater the absolute value of the SC, the greater the effect of the given variable on ET₀. The SC values for 84 stations for the four climate variables are plotted relative to the annual ET0 value (Figure 6), which shows that the SC for SD was highest in northern and eastern NWC, followed by T, WS, and RH. In the Tarim basin, the SC was greatest for T, followed by SD, WS, and RH. The SC for RH was negative, indicating that ET₀ would decrease as RH increased; however, the portion of the variation in ET₀ derived from RH change was slight. Across the entire study area, the long-term annual ET_0′s SC was 0.28 for T, −0.08 for RH, 0.27 for WS, and 0.52 for SD (

Table 3. Annual and seasonal statistics of ET₀ sensitivity coefficients (SC) for climatic variables in NWC.

Variables	Annual	Spring	Summer	Autumn	Winter
T	0.28	0.38	0.65	0.24	−0.33
RH	−0.08	−0.09	−0.20	−0.12	0.15
WS	0.27	0.28	0.23	0.31	0.38
SD	0.52	0.43	0.31	0.57	0.79

).

We evaluated the seasonal ET_0’s SCs for the four climatic variables in NWC (Figure 6) and found significant differences among the four seasons. In spring, the ET_0’s SC for SD was largest in northern and southeastern NWC, while the ET_0’s SC for T was largest in the deserts (e.g., Tarim and Badain Jaran). In summer, the ET_0’s SC for T was dominant, followed by SD, WS, and RH. The SC of ET₀ for SD was dominant in autumn and winter. The regional statistics of seasonal ET_0’s SCs for different climatic variables during the pre-change and post-change periods are shown in Table 3.

Plots of the relationship between elevation and SCs of annual ET₀ to climatic variables (Figure S4), show the SC of T and WS for annual ET₀ significantly decreased (p < 0.01 and p = 0.02, respectively), while the SC of annual ET₀ to RH and SD showed moderate (p = 0.087) and highly (p < 0.01) significant increases, respectively, with an increase in elevation. In NWC, elevation gradients of SC were −0.076 km⁻¹ for T, 0.028 km⁻¹ for RH, −0.028 km⁻¹ for WS, and 0.077 km⁻¹ for SD. Consistent with our findings, Liu et al. [42] concluded that the SC of maximum temperature and wind speed decreased as elevation increased in the Yellow River basin. However, Yang et al. [24] found that the sensitivities of ET₀ to RH and WS both increased as elevation increased on a slope of Qilian Mountain. Accordingly, the sensitivity of ET₀ to meteorological factors appears to vary by region, because climatic conditions and meteorological factors arising from the different climate zones differ with the regions [43,44]. The PM equation captured the information of the four major meteorological variables governing the evaporative process [14]; however, changes in those four variables did not always reflect changes in climate zones, locations, and land surface characteristics consistently. Accordingly, the dominant meteorological variables varied by different climate zones, locations, and land surfaces. Therefore, the sensitivity of ET₀ to meteorological factors varies based on a region’s conditions.

3.4. Contribution of Climatic Variables to ET₀ Variation

In order to ensure the reliability of the contribution to the calculated ET₀ changes, we compared the annual and seasonal dET₀/dt directly evaluated by Sen’s slope method with the dET₀/dt accumulated by the individual contributions of the four climatic variables (Figure 7). It was found that the regression slopes of the scatter plots were almost equal to one for the entire period. For the four seasons, the accumulated dET₀/dt were also correlated with those evaluated directly by Sen’s slope. The scatter plots closely approached an ideal fitted 1:1 regression line, indicating that the cumulative dET₀/dt fitted well with the directly evaluated dET₀/dt. Accordingly, it was concluded that the adopted contribution method was appropriate, and the calculated contributions were reliable.

Figure 8 shows the contribution of changes in climatic variables to changes in ET₀ variation from 1960 to 2017 in NWC. The range of the contribution of T change to ET₀ variation was between −0.33 and 5.92 mm/yr, with most stations showing positive contributions. The contribution of change in RH to variation in ET₀ ranged between −2.00 and 0.46 mm/yr; at 79 out of 84 stations, contributions were below 0.5 mm/yr, which indicated only a small contribution to ET₀ variation. The contribution of WS change to ET₀ variation ranged between −6.15 and 7.52 mm/yr, and the contribution of SD change to ET₀ variation varied between −2.29 and 1.16 mm/yr.

To more easily compare the contributions of climatic variables to ET₀ change, we performed zonal statistics of annual and seasonal contributions (

Table 4. Annual and seasonal statistics of the contributions of climatic variable changes to ET₀ trends in NWC (mm/yr).

	Annual	Spring	Summer	Autumn	Winter
T	1.300	0.443	0.395	0.226	0.087
RH	−0.016	0.036	−0.024	−0.019	−0.004
WS	−1.508	−0.485	−0.599	−0.299	−0.103
SD	−0.288	0.062	−0.093	−0.076	−0.055
Total	−0.513	0.057	−0.319	−0.169	−0.076

). Changes in T, RH, WS, and SD led to changes in annual ET₀ of 1.30, −0.02, −1.51, and −0.29 mm/yr, respectively, from 1960 to 2017. The WS contributed to a decrease in ET₀, which offset a significant increase due to T, resulting in a net annual ET₀ decrease of 0.51 mm/yr. Annually, WS was the dominant climatic variable for ET₀, whereas the contribution of RH to ET₀ change was negligible. The dominant variable of ET₀ change in spring was T. The positive contributions of T, RH, and SD offset the decrease from WS and resulted in a spring ET₀ increase of 0.06 mm/yr. The dominant variable of ET₀ in the remaining three seasons was WS, and the negative contributions of WS, RH, and SD offset the positive contribution of T, resulting in ET₀ decreasing by 0.32, 0.17, and 0.08 mm/yr in the summer, autumn, and winter during the whole period, respectively.

3.5. Effect of Elevation on Variation in ET₀

To show the effect of elevation on variation in ET₀, we examined the contributions of the four climate variables to changes in ET₀ at elevational steps of 500 m (Figure 9). As expected, according to the warming trend, contributions of T to ET₀ changed with different elevation bands, but were all positive and within a range of 0 and 4 mm/yr. The contribution of RH to change in ET₀ was very slight, with values between ±0.4 mm/yr for different elevation bands. The contributions of WS to ET₀ change were generally negative across all elevation bands, with the largest negative contribution occurring at 0–500 m; the WS contribution was close to 0 at 3000–3500 m. The contribution of SD to change in ET₀ was mainly negative, with the value of the mean or median for different elevation bands all being below 0. There were no obvious correlations between elevation and the contributions of T, RH, and SD. However, the relationship between elevation and the contribution of WS showed a clear decline with rising elevation, except at 2500–3000 m.

In order to more clearly detect the relationship between elevation and the contributions of climatic variables, we examined the relative contribution of the four variables at different elevation bands (Figure 10). Clearly, changes to the relative contribution of RH and SD to change in ET₀ were small, with values of less than 3% and 15%, respectively. The major contributions to change in ET₀ were from changes in T and WS. The relative contribution of T generally increased, while that of WS generally decreased with rising elevation. Interestingly, the absolute relative contribution of T was less than that of WS below 2000 m and greater than that of WS above 2000 m. Thus, the significant increase of the trend in ET₀ with elevation in NWC represents a tradeoff between the contributions of T and WS. Below 2000 m, the negative contribution of WS offset the positive contribution of T, resulting in a decreasing trend in ET₀; above 2000 m, the positive contribution of T offset the negative contribution of WS, resulting in an increasing trend in ET₀.

4. Discussion

4.1. Temporal Trends of ET₀ in NWC

The evapotranspiration process is controlled by the integrated effect of climatic variables. Change in ET₀ is the result of a combination of amplitudes of variation among different climatic variables and ET_0’s sensitivity to them. The trend in annual regional ET₀ in NWC between 1960 and 2017 was −0.51 mm/yr, with a change point year in 1993. From 1960 to 1993, ET₀ significantly decreased (−2.18 mm/yr), while a significant increase of 4.69 mm/yr occurred after the change point (Table 2). Liu and Zhang [27] found that the contribution of decreasing wind speed offset the effect of increasing air temperature, leading to the decrease in ET₀ from 1960 to 1993, and that concomitant increases in air temperature and wind speed reversed the trend in ET₀ and led to its increase after 1994 in NWC. The WS played an important role in governing the evaporative demand. Reviewing a number of studies, McVicar et al. [14] showed that declining wind velocities (‘stilling’) were the primary factor contributing to declining rates of evaporative demand. This has recently served in explaining the ‘evaporation paradox’. Our study showed that the negative contribution of decreases in WS offsets the positive contribution of increases in T, resulting in the decrease of ET₀; therefore, WS is the primary factor contributing to the variation of ET₀ in NWC followed by T, SD, and RH. Zheng and Wang [28] also found WS and T were the two most important variables having an impact on variation in ET₀ in NWC. However, the responses of variation in ET₀ to changes in climatic variables varied in different regions. Having divided the mainland of China into different climatic zones to evaluate the impact of climate on ET₀, Fan et al. [2] found that the primary driver of change in ET₀ varied between different climatic zones. Although T has been rising globally [45], it is not the only factor affecting changes in ET₀ [14]. The PM equation is a physically based combination formula for calculating ET₀, which means it captures the variation of four climate variables governing the evaporative process. Therefore, an increase in T should lead to an increase in ET₀. However, this effect could be offset by a decrease in RH, SD, and WS, leading to a decrease in ET₀ [19,27]. Meanwhile, the contribution of climatic factors to ET₀ variation is modulated by the sensitivity of ET₀ to climatic factors and their magnitude of change. For example, ET₀ was particularly sensitive to SD (Table 3), but its corresponding change was relatively small; therefore, SD only contributed in a limited manner to variation in ET₀ in NWC (Table 4).

4.2. Effect of Elevation on the Variation of ET₀

Previous studies in NWC have paid little attention to the effect of elevation on variation in ET₀. Variation in annual ET₀ was significantly (p < 0.01) correlated with an increase in elevation in NWC (Figure 5). The WS dominated the decrease in ET₀ at low elevations, while T dominated increasing ET₀ at high elevations. Zhang et al. [46] found that increasing (decreasing) trends in ET₀ over the Tibetan Plateau and overall were negatively (positively) related to rising elevation over their study period (1971–2015), with the absolute trends in ET₀ becoming smaller along with rising elevation. Although there is growing evidence that the rate of warming is amplified with elevation, known as ‘elevation-dependent warming’ [47], especially in the Tibetan Plateau [25], elevation-dependent warming was insignificant in NWC; the relative contribution of warming to ET₀ variation was correlated to elevation (Figure 10). However, the trend in WS significantly correlated to elevation (Figure S1), resulting in the significant elevation-dependence of the contribution of change in WS to ET₀ variation. Unlike our findings, Guo et al. [48] found the rate of reduction in WS to be amplified with elevation in the Tibetan Plateau due to the increased surface roughness. We found that the rate of reduction in WS slowed with rising elevation in NWC, resulting in the slowdown of the negative contribution of WS to ET₀ variation. The positive contribution of warming was relatively amplified with elevation, leading to an increase in ET₀.

The slowdown of the reduction in WS along with a rise in elevation may be the result of the following: (i) The increase in surface roughness at low elevations was faster than in high elevations. The increase in land surface roughness largely results from increasing vegetation cover [14]. In NWC, landscapes at low elevations are croplands, artificial shelter forests, and grassland, while at higher elevations, they are sparse alpine vegetations or bare lands. Therefore, low altitude vegetation cover is far greater than at high elevations. Vegetation cover increases can be monitored by following increases in the NDVI (normalized difference vegetation index). Xu et al. [20] found the increase in NDVI was mediated by elevation. Meanwhile, land surface changes caused by human activities (e.g., urbanization and large-scale afforestation) occur mainly at low elevations, leading directly to an increase in land surface roughness. (ii) Warming at high elevations is faster than in low elevations. The WS is a function of the horizontal temperature gradient [48], with a background of the mean air temperature decreasing along with elevation; therefore, the amplification of warming with elevation would result in a decline of temperature gradient force between high and low elevations. A more detailed mechanistic interpretation of the slowdown in the reduction in WS with rising elevation could be arrived at by considering the atmospheric circulation changes, surface energy partition, and land surface changes.

4.3. Application to Agricultural Ecosystems

Our results suggest a dominant contribution of a decline in WS to ET₀ variation below 2000 m. In the NWC, irrigated cropland is mainly located below 2000 m. Shelter forests have been installed among crop blocks to defend the cropland from the infringement of sandstorms by reducing the wind speed. In recent years, farmers have used water-saving irrigation systems (e.g., sprinkler, trickle irrigation, etc.) to resolve the conflict between the high rate of evapotranspiration capacity and limited available water [49]. For example, the Manas River basin has a drip irrigation quota of 5250 m³/ha, whereas broad irrigation (flood irrigation) has a quota of 6750 m³/ha [50]. However, unlike broad irrigation, these kinds of irrigation systems only deliver water to crops and the shelter forests cannot absorb enough moisture from deep soil to survive, evidenced by the decrease in the groundwater table [50]. Therefore, the use of such irrigation systems could damage the shelter forests, and its function of wind and sandstorm prevention would fade away. According to our study, the increase in wind speed could induce an increase in agricultural water demand. Consequently, the irrigation water demand would be increased correspondingly [51]. Zou et al. found the total irrigation water demand increased from 17.463 × 10⁸ to 30.599 × 10⁸ m³ in the Heihe River basin during 1985–2014 [52]. This is an internecine tragedy, and it can potentially lead to an increased risk of eco-environmental degradation [53]. Thus, a combination of broad and water-saving irrigation would best meet the relatively low agricultural water demand and allow sustainable oasis agro-ecosystems.

5. Conclusions

The changes in ET₀ and attributions of changes in climatic variables (e.g., air temperature (T), sunshine duration (SD), relative humidity (RH), and wind speed (WS)) to ET₀ changes were investigated using a long-term (1960–2017) dataset from 84 meteorological stations in Northwest China. A change point in annual ET₀ was detected to exist in the 1990s, following which an upward trend in ET₀ was detected. The annual ET₀ decreased at a rate of 2.18 mm/yr before the break point and then increased significantly at a rate of 4.69 mm/yr after the break. A partial derivatives method was employed to quantify the actual contributions of the changes in climatic variables to the changes in ET₀. The contribution of the decrease in WS offset the significant contribution of the increase in T, resulting in a decreasing trend in ET₀ from 1960 to 2017. Sensitivity analysis indicated that SD was the dominant variable affecting ET₀ followed by T, WS, and RH. However, the relatively large sensitivities and amplitude of change in WS and T led to their dominant contributions to ET₀ changes. The effect of elevation on changes in ET₀ depended mainly on the tradeoff between the contributions of T and WS with increasing elevation. Below 2000 m, the negative contribution of WS offset the positive contribution of T, resulting in a decreasing trend in ET₀ change, while above 2000 m, the positive contribution of T offset the negative contribution of WS, resulting in an increasing trend in ET₀. This study addressed the gap in knowledge regarding how elevation influences variation in ET₀ under climate change and proposed an application perspective to agriculture and ecological restoration.