The Abrupt Change in Potential Evapotranspiration and Its Climatic Attribution over the Past 50 Years in the Sichuan–Chongqing Region, China

Abstract

Potential evapotranspiration (PET), as an indicator of atmospheric evaporative demand, is a critical hydrological and meteorological factor to reflect regional and global hydrological cycles and environmental change. Understanding these nuanced responses of PET to environmental changes is important for agricultural production and water demand estimation. This study rigorously evaluated fluctuations in PET using the Penman–Monteith model over a 50-year span from 1970 to 2020 in the Sichuan–Chongqing region, an area notably susceptible to climate change. The changing characteristics of PET and local meteorological factors were detected by integrating the Mann–Kendall method and Pettitt test. Furthermore, the contribution and sensitivity of key meteorological variables to the observed variation in PET were also thoroughly investigated. Breakpoint analysis revealed that abrupt changes appeared in 1996 for annual PET. The detrending method indicated that substantial decreases in net radiation and wind speed (p < 0.01) were responsible for the decrease in annual PET from 1970 to 1996. Marked increases in minimum temperature and wind speed were the driving forces behind the uptick in annual PET in 1997–2020. At seasonal scales, wind speed and net radiation predominantly influenced PET in 1970–1996 in general. However, from 1997 to 2020, the factors controlling PET fluctuations displayed considerable seasonal variation. Sensitivity analysis showed that W_s and T_min were the second-most sensitive factors. By exploring the impacts of PET changes and shifts, attention must be paid when allocating water resources reasonably under the background of ongoing climate change and likelihood of future drought.

1. Introduction

The global climate has changed in recent decades [1,2]. From the latest report of the Intergovernmental Panel on Climate Change (IPCC), there is no doubt that the global warming trend will continue [3]. Huge changes in streamflows, rainfall, evapotranspiration, base flows, and soil moisture in recent years indicate that ongoing climate change will remarkably affect the hydrological cycle [4,5]. Water scarcity and drought have become the main constraints on agricultural production because of population rise and associated increased water demand [6].

Evapotranspiration (ET) is an essential component of ecosystem water and serves as a principal hydrometeorological term for understanding global and regional climate changes [7,8]. Predicting long-term ET time series has emerged as a focus of the research into regional-scale hydrological processes [9,10,11]. However, quantifying actual ET directly remains challenging.

A commonly used method to calculate actual ET is through quantifying potential evapotranspiration (PET) [8,12]. PET was defined by Penman in 1956 as “the amount of water transpired in a given time by a short green crop, completely shading the ground, of uniform height and with adequate water status in the soil profile” [13]. PET is considered a crucial indicator of hydroclimatic change [14,15]. Understanding changes in PET and its driving factors are critical for accurately evaluating the likely environmental impacts on agricultural irrigation (especially under the drought condition) and climate moderation [5,15].

PET fluctuation is affected by multiple climatic factors, instead of air temperature alone [10,12]. Changes in PET, related to the dramatic changing climate, are directly involved in regional and global water resources. Therefore, PET is a key determinant in water resource assessments and has profound implications for the estimation of water demand in agricultural systems [16,17]. Consequently, the analysis of the long-term dynamic in PET is significant to evaluate agricultural water demand, regional water balance, and environmental change.

Many methods for computing PET have been developed. They vary in their data demands from the simple to the more complex. Tang et al. (2021) used the Penman–Monteith (P-M) model, the Priestley–Taylor model, and the Hamon model to predict PET in Siberian River Basins, and found that the P-M model was the most accurate for humid sites [4]. Moreover, the P-M model is highly recommended by the Food and Agriculture Organization of the United Nations (FAO) for evaluating PET [18]. Since PET is affected by many climatic elements, exploring the trend of PET using linear regression analysis alone cannot reflect the combined effects of different meteorological variables. The rank-based Mann–Kendall (MK) method [19,20] is widely used for trend testing hydrological and meteorological data because it does not need any distribution assumption for the data [21]. The nonparametric Pettitt test was developed by Pettitt (1979) and has been regarded as one of the most classical methods for detecting the change points in long-term hydrological as well as meteorological records [22]. Additionally, lots of previous literature provided a detrending method that could effectively quantify and estimate the contributions of key climate factors to the increasing or decreasing trends in PET [23,24].

Over the past several decades, many studies [24,25,26,27,28,29,30] have quantified regionally averaged PET in different periods and found that it can show positive, negative, or stable trends (see Table 1). The trigger mechanisms behind PET changing across different climatic regimes have been examined. Escalating mean air temperature, associated with enhanced greenhouse gases, has resulted in the distinct increase in evaporation requirements globally; examples are in western Iran in 1966–2005 [29], the Yellow River Basin in 1961–2006 [23], and the Taohe River Basin in 1981–2010 [31] in China. However, decreased PET has been detected in several regions, despite the prevalence of warming trends around the world [2,24,32]. There exist two widely accepted explanations for the recent decrease in PET. One is caused by the decrease in available solar radiation due to increased cloud coverage, a proposition supported by studies in many regions worldwide, such as the northern hemisphere [33], the United States [34], and the Hai River Basin in China [35]. The second explanation invokes a decrease in vapor pressure deficit (VPD) or wind speed (W_s), as observed in India [36], the Siberian River Basin [4], and the Jing River Basin in China [37]. Cohen et al. (2002) detected that PET does not change considerably in West Iran (1964–1998). They attributed this phenomenon to the impacts of increased VPD and W_s offset by reduced solar radiation.

This research centers on the overall investigation of PET variations in the Sichuan–Chongqing (SC-CQ) region, a humid and semi-humid zone in Southwest China. On the one hand, the local government launched national strategies, such as the Development of Western Regions and the Development of the Yangtze River Economic Belt. Therefore, this region has experienced massive changes in land use and population in recent decades [38,39]. Human activities (e.g., urbanization, industrialization, and cultivation) have severely influenced local water balance, presenting a severe water shortage challenge for local administrations [8,10,40]. The Yangtze River, one of the largest and longest rivers in Asia, flows through the SC-CQ region, providing abundant water [41]. However, available water resource accounts for less than 1% of annual surface water [42]. Furthermore, rapid population growth and economic development could potentially double the water requirement for agricultural activities [43]. Hence, an intensifying conflict between urban expansion and the shortage of irrigation water is anticipated in this region. Compared with studying the interannual variation in PET alone, assessing seasonal behaviors will offer a more comprehensive understanding of agricultural water management. Moreover, the SC-CQ region shows distinct seasonal changes in meteorological factors [44]. Thus, it is essential to estimate climate elements and their relative significance to PET changes at a seasonal scale. Consequently, estimating seasonal patterns of PET, rather than annually alone, provides a more reliable reference for decision makers to better manage water resources in this region.

On the other hand, countless studies have been conducted in the arid and semi-arid regions, revealing that PET is sensitive to changes in moisture conditions [2]. However, PET and its related water cycle processes demonstrate different responses to changing environmental variables across different climatic conditions and regions. Few studies are available about PET variation and their implications for the regional hydrologic cycle in humid regions [12,45]. As a result, a more detailed analysis is necessary to address these limitations mentioned above. Thus, further investigation into the causes of the variations in PET will enhance our understanding of how environmental controls (e.g., urban expansion and climate change) relate to the local water balance. Particularly, in the important rice-producing regions, analyzing the driving forces of PET is critical for water resource management practices and agricultural systems.

Therefore, the objectives of this study are as follows: (1) to quantify seasonal and annual PETs over the past 50 years (1970–2020) in the SC-CQ region with the P-M model; (2) to investigate the seasonal and annual trends in PET and seven major climate variables (net radiation, R_n; wind speed, W_s; VPD, relative humidity, RH; mean temperature, T_mean; maximum temperature, T_max; and minimum temperature, T_min) using the MK test and Pettitt test; and (3) to identify the principal causes of variations in PET at seasonal and annual scales using the detrending method.

2. Data Processing

2.1. Study Area

The SC-CQ region (26°03′–34°19′ N, 97°21′–110°11′ E) encompasses an area of about 568,402 km² in Southwest China (see Figure 1). The SC-CQ region contains 22 major cities, including two megacities, Chengdu and Chongqing, which are core growth areas in China. The SC-CQ region has a typical East Asia summer monsoon climate. Due to alternating seasonal air mass movements and their accompanying winds, this region has a distinct moisture transfer pattern characterized by a wet summer and dry winter [16,25]. As one of the most important areas for rice cultivation in China, the yield of rice in this region increased by more than 60% during the past few decades, which is extremely important to the development of agriculture across East and Southeast Asia [41,46]. The national policy of Development of Western Regions was enacted and implemented in the late 1990s. During the period of rapid urban development, local ecosystems and microclimates are altered resulting from intense human activities. As a result, this region is increasingly vulnerable to the effects of climate changes [47]. In addition, a persistent drying and warming climate has restricted the development of agricultural production in this area [48].

2.2. Materials and Methods

2.2.1. Data Sources

Daily meteorological observations (1970–2020) from 56 standard weather stations within the SC-CQ region were obtained from the China Meteorological Data Sharing Service System (http://data.cma.cn/, accessed on 20 June 2024). The key meteorological variables used to quantify PET were humidity (RH), radiation (R_n), vapor pressure, sunlight duration, wind speed at 2 m height (W_s), and maximum, minimum, and average air temperatures (T_max, T_min, and T_mean, respectively). Missing data were filled using linear interpolation if time gaps were <5 d, and if time gaps were >5 d, data were filled with the multiyear mean values of those days [8]. The digital elevation model (DEM) was derived from the Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 20 June 2024). Data were analyzed by six time periods: spring (March–May), summer (June–August), autumn (September–November), winter (December–February in the next year), growing season (May–October), and annual (January–December).

2.2.2. FAO Penman–Monteith Method to Quantify PET

There are several methods of quantifying PET, such as the Thornthwaite equation [49], the Hargreaves method [50], the Priestley–Taylor method [51], and the P–M model [13]. The P-M model is more accurate than other methods for humid areas [4,12] and is highly recommended by the Food and Agriculture Organization of the United Nations (FAO) for evaluating PET [18].

The FAO56 P-M model is shown in the following equation [18]:

$$\mathrm{PET}=\frac{0.408\mathsf{\Delta }\left(R_n-G\right)+\gamma \frac{900}{T+273}{U}_2(e_s-e_a)}{\mathsf{\Delta }+\gamma (1+0.34U_2)},$$

(1)

where PET is the daily potential evapotranspiration rate (mm/d); Δ is the slope of the saturated vapor pressure curve (kPa/°C); R_n is net radiation (MJ/m²/d); G is soil heat flux density (MJ/m²/d; zero on a daily scale); γ is the psychrometric constant (kPa/°C); T is the mean daily air temperature (°C); U₂ is the mean daily wind speed at 2 m height (m/s); e_s is saturated vapor pressure (kPa); e_a is actual vapor pressure (kPa); and e_s − e_a is the vapor pressure deficit (kPa).

Net radiation at the crop surface, R_n, is a function of solar radiation, R_s. Solar radiation (R_s, MJ/m²/day) is an infrequently measured climatic variable and is often estimated from sunshine data.

$$R_s=(a_s+b_s\frac{n}{N})R_{\alpha },$$

(2)

where $R_{\alpha }$ is the extraterrestrial radiation (MJ/m²/day), N is the maximum possible sunshine duration (h) and n is the actual sunshine duration (h), and $a_s$ and $b_s$ are empirical coefficients with recommended values of 0.25 and 0.5, respectively.

Due to the availability of relative humidity data, the actual vapor pressure is calculated using the following equation [11]:

$$e_a=\frac{Pq}{0.378q+0.622},$$

(3)

where q is the specific humidity (kg/kg), and P is atmospheric pressure (kPa).

2.2.3. Trend Analysis

The MK test [19,20] was used to test trends in PET and related climate data. This method is commonly used to detect whether a long time series is significant or not [52].

The MK test first calculates the rank statistic, $T$, for a time series, $X=\left\{x_1,x_2,\cdots , x_n\right\}$

$$\mathrm{T}={\sum }_{i<j}{ \mathrm{p}}_{\mathit{ij}},$$

(4)

where:

$${\mathrm{p}}_{\mathit{ij}}=\mathrm{sign}\left(x_j-x_i\right)=\left\{\begin{array}{cc}1& x_i<x_j\\ 0& x_i=x_j\\ -1& x_i>x_j\end{array}\right.,$$

(5)

where $x_i$ and $x_j$ are observations in years j and k, respectively. The mean and variance in the $T$ statistic is expressed as follows:

$$E\left(T\right)=0,$$

(6)

$$\mathrm{var}\left(T\right)=\frac{n(n-1\left)\right(2n+5)}{18},$$

(7)

$${\mathrm{var}}^{*}\left(T\right)=\frac{n(n-1\left)\right(2n+5)}{18}-{\sum }_{j=1}^m\frac{k_i(k_i-1\left)\right(2k_i+5)}{18},$$

(8)

where $n$ is the length of observations, $m$ is the length of tied ranks in groups, and $k_i$ is the number of data values in the $i^{\mathrm{th}}$ group. The standardized variable $Z$ is provided by:

$$\mathrm{Z}=\left\{\begin{array}{cc}\frac{\mathrm{T}-1}{\sqrt{\mathrm{var}\left(\mathrm{T}\right)}}& \mathrm{T}>0\\ 0& \mathrm{T}=0\\ \frac{\mathrm{T}+1}{\sqrt{\mathrm{var}\left(\mathrm{T}\right)}}& \mathrm{T}<0\end{array}\right.,$$

(9)

If Z is positive, the time series of hydro-meteorological factors shows an increasing trend; if Z is negative, the time series shows a decreasing trend. |Z| > 1.28, |Z| > 1.64, and |Z| > 2.32, respectively, indicate that the trends are significant at >90% (p < 0.1), 95% (p < 0.05), and 99% (p < 0.01).

The Pettitt test [53] was used to detect the abrupt change points in hydro-meteorological records at both seasonal and annual scales. The Mann–Whitney statistic $U_{\mathrm{t},N}$ is given by:

$$U_{\mathrm{t}, N}=U_{\mathrm{t}-1, N}+{\sum }_{\mathrm{j}=1}^N\mathrm{sgn}\left(x_t-x_j\right), (t=2,3,\cdots ,N)$$

(10)

The statistic $k\left(t\right)$ and related probabilities used in significance testing are given by:

$$k\left(t\right)={\mathrm{Max}}_{1\le t\le N}\left|U_{t,\mathrm{N}}\right|,$$

(11)

and

$$P\cong 2\exp \left\{-6{\left[k\left(t\right)\right]}^2/({\mathrm{N}}^3+{\mathrm{N}}^2)\right\},$$

(12)

2.2.4. Quantification of Factor Contribution and Sensitivity Analysis

The influence of climate factors on PET at both annual and seasonal scales was quantified by detrending the time series of each factor [12,21]. The procedures of this method are as follows: (1) The trend of the climate factor was removed to render it stationary; (2) The detrended data series of each factor was used in the recalculation of PET while other factors remained unchanged until PET had been calculated for each detrended factor; and (3) Recalculated PET for each factor was compared with the original PET, and the difference was considered to be the effect of the factor on PET. The contribution of each factor was proportionalized by an indicator, R [54]:

$$R={\sum }_{i=1}^m\frac{{\mathrm{PET}}_{\mathrm{O}}-{\mathrm{PET}}_{\mathrm{R}}}{{\mathrm{PET}}_{\mathrm{O}}i},$$

(13)

where ${\mathrm{PET}}_{\mathrm{O}}$ and ${\mathrm{PET}}_{\mathrm{R}}$ are original and recalculated PET, respectively; $m$ represents the length of the dataset. R > 0 and R < 0 are the positive and negative effects of climate factors on the PET trend, respectively. R ≈ 0 indicates that the factor has little effect on the PET trend. A greater value of $\left|R\right|$ indicates a greater contribution of the climate factor to PET.

Sensitivity analysis was used to quantify and delineate relative changes in each climate factor against corresponding relative changes in PET. Sensitivity analysis is a simple but practical method that has been used by many previous researchers [21,22]. The two steps of the analysis are as follows: (1) Generate scenarios for each climate factor using the equation:

$$X\left(t\right)=X\left(t\right)+∆X ∆X=0, \pm 10\%, \pm 20\%, \pm 30\% \mathrm{of} X\left(t\right),$$

(14)

where $X$ is the climate factor and t is the time (d); (2) Recalculate PET separately for each variation in X, ΔX. For each climate factor, a greater relative change in the recalculated PET indicates the greater sensitivity of the PET to the factor.

3. Results

3.1. Trends of PET in 1970–1996 and 1997–2020

Figure 2 shows the average annual PET from 1970 to 2020 for all 56 meteorological stations in the SC-CQ region. Annual PET varied from a minimum of 822 mm/year in 1989 to a maximum of 965 mm/year in 2013. The Pettitt test for annual PET showed that an abrupt change in PET occurred in 1996. On behalf of depicting the change pattern of PET, the annual PET sequence was separated into two segments: a significant negative trend (p < 0.01, −2.3 mm/year) before 1996, but followed a significant positive trend from 1997 to 2020 (p < 0.01, 2.8 mm/year). Based on the timings of abrupt changes, we defined these two periods as “period I” (1970−1996) and “period II” (1997−2020). Mean average PET was 869 mm/year in period I and 893 mm/year in period II; in each case, over 73% of PET occurred during the growing season (see Figure 3). Growing season PET pronounced a declining trend at a rate of −1.5 mm/year in period I and increased at a rate of 2.4 mm/year in period II (see Table 2).

The value of mean PET in period I was similar to that in period II in all seasons, with a slightly higher mean value in period II (see Figure 3). A significant decreasing trend was found in all seasons in period II (p < 0.1) (see Table 2). In period II, PET in autumn showed a nonsignificant negative trend (p > 0.1), and increasing trends were observed in the other three seasons. Mean PET was greatest in summer, accounting for approximately 37% of the annual PET. The rate of change in PET was greatest in summer, with a slope of −1.0 mm/year in period I and 2.3 mm/year in period II (p < 0.01), respectively.

3.2. Trends of Meteorological Factors in 1970–1996 and 1997–2020

Seasonal and annual climate factor trends were separately calculated in two periods (period I: 1970–1996; and period II: 1997–2020), similar to the calculation of PET trends described in the preceding section. Averaged RH, W_s, and R_n were greater in period I than in period II at both seasonal and annual scales (see Figure 3). The other four climate factors were less in period I than in period II. Increasing T_mean, T_max, and T_min indicated that warming in the SC-CQ region matched the global warming trend.

T_mean and T_max increased in spring and summer and decreased in autumn and winter in period I (see Table 2). In six time periods (spring, summer, autumn, winter, growing season, and annual), T_mean and T_max both showed nonsignificant trends (p > 0.1), which indicated that there were minor changes in these two factors, while T_min showed a significantly increasing trend (p < 0.1) in all time periods, except for spring. RH presented an upward trend in all six time periods and showed significantly changing trends in summer (p < 0.1), winter (p < 0.05), growing season (p < 0.05), and annual (p < 0.01), with respective slopes of 0.072%/year, 0.086%/year, 0.040%/year, and 0.062%/year. W_s and R_n all showed statistically significant decreasing trends at both seasonal and annual scales (p < 0.1). R_n showed the greatest decreasing slope of −0.030 MJ/m²/d/year in summer. The greatest rate of decrease in W_s was −0.011 m/s/year in spring. There was a slight decrease in VPD (nonsignificant and < −0.002 kPa/year) in all seasons. However, the decrease in annual VPD was significant at a rate of −0.0009 kPa/year.

T_mean, T_max, and T_min increased at both seasonal and annual scales in period II, and the trends were statistically significant, except for T_max in autumn and winter. Summer temperature showed the greatest rising rate in all seasons, and the slopes of T_mean, T_max, and T_min were positive in summer with respective values of 0.12, 0.14, and 0.12 °C/year. T_min showed a greater rate of increase (0.12 °C/year) on an annual scale than T_mean (0.09 °C/year) or T_max (0.07 °C/year). Seasonal and annual RH presented a nonsignificant increasing trend (p > 0.1), except during summer, when RH decreased significantly by −0.19%/year (p < 0.05). In contrast with period I, W_s showed a significant positive trend for all six time scales (p < 0.01), which suggested that W_s changed in the study area during the study period. The changing direction of annual average R_n in period II was opposite to that in period I. R_n increased significantly in summer (p < 0.1) by −0.013 MJ/m²/d/year and decreased significantly in autumn (p < 0.05) by 0.001 MJ/m²/d/year. An abrupt change was detected in annual VPD during the whole study period. VPD decreased significantly in period I (p < 0.1; 0.0009 kPa/year) but showed a significant increasing trend in period II (p < 0.01; 0.0035 kPa/year). At the seasonal scale, VPD showed a statistically significant trend during summer and growing season in period II (p < 0.01). The rate of increase in VPD was greatest in summer (0.0128 kPa/year) and in the growing season (0.0058 kPa/year), and the trend was significant for both (p < 0.01).

3.3. Causal Analysis

3.3.1. Original and Detrended Climate Variables at the Annual Scale

Seven annual original and detrended climate variables, T_mean, T_max, T_min, RH, R_n, VPD, and W_s, for period I and period II were compared in this paper (see Figure 4). During period I, only the original annual RH, T_mean, and T_min showed a positive trend, which resulted in lower values of the detrended annual RH, T_mean, and T_min. The four other climate variables all exhibited a decreasing trend and had greater annual detrended values. There was clearly a significant difference between the original and detrended time-series data for annual R_n, W_s, and RH during period I throughout the SC-CQ region. No significant differences were found in the other four climate variables at the annual scale.

The situation in period II was slightly different from that in period I. All climate factors showed increasing trends in period II, leading to lower values in the detrended series of these factors. In addition, the differences in values between the original series for the three temperature variables (T_mean, T_max, and T_min), VPD, and W_s and the detrended time series were large, which suggested that the changes in these climatic factors could influence annual PET variation to a greater extent in period II. Based on the example of annual series data (see Figure 4) and the trends of seven original basic climatic elements (see Table 2), the difference between the original climatic elements and detrended one at the seasonal scale could be distinguished.

3.3.2. Contribution of Climate Elements to PET Trend at the Annual Scale

PET was separately recalculated using the seven detrended climate factors for period I and period II, and it was used as an example to show the differences between the original and the recalculated data series (see Figure 5). Recalculated PET for period I with detrended R_n, W_s, RH, and VPD was higher than the original PET (see Figure 5). There was a small gap using the three detrended temperature variables (T_min, T_max, and T_mean) between the recalculated PET and original PET, which indicated that the changes in temperature variables had little influence on annual PET during period I. W_s, VPD, and R_n were the factors primarily responsible for the changing trend of PET at an annual scale in period I (see Figure 6).

The major climatic elements that dominated the trend of annual PET in period II were inconsistent with those in period I. Recalculated PET using the detrended RH was the only recalculated value to increase and was slightly greater than the original PET; the recalculated PET using the other climate factors was clearly less than the original PET (see Figure 5). This suggested that the changes in all climatic factors except RH drove the positive trend of annual PET during period II. Figure 6 shows that increased VPD, W_s, and T_min were the major factors responsible for the increasing trend of PET at the annual scale.

3.3.3. Influence of Climate Factors on PET at the Seasonal Scale

Seasonal behavior in period I was similar to behavior at the annual scale. Although T_min had a positive influence on PET, the factors RH, W_s, VPD, and R_n each had a distinct negative influence on PET, leading to a significant decrease in PET for all seasons (p < 0.1) (see Table 2 and Figure 6). The decrease in R_n was the most influential factor of PET variation in spring, summer, and growing season; the significant decreasing W_s and R_n were the main reason for decreasing trends of autumn PET; and changes in winter PET were heavily influenced by W_s. T_max and T_mean made little contribution to PET at a seasonal scale during period I when compared with the other five factors.

The response of PET to climatic factors varied seasonally during period II. In spring and winter, the negative influence of RH and R_n on PET was offset by the positive influence of VPD, W_s, and the three temperature variables (T_min, T_max, and T_mean). For summer and growing season, the seven climate factors all produced a positive effect on PET. However, in autumn, PET was mainly influenced by a significant decrease in R_n and an increase in RH, which resulted in decreasing PET at a rate of −0.13 mm/year. Increased PET was thus found in all seasons except autumn during period II.

3.4. Sensitivity Analysis

The climate factors (such as RH in winter during period II) showed nonsignificant trends, but had the greatest influences on changes in PET (see Table 2 and Figure 6). Sensitivity analysis of the climate factors was used to determine the major contributor of PET variation at annual and seasonal scales. The sensitivity analysis is plotted in Figure 7. PET was negatively correlated with RH and positively correlated with the other six factors at all time scales, signifying that PET increased with the decrease in RH but increased with the increases in VPD, T_max, T_min, T_mean, R_n, and W_s. There were some slight differences in sensitivity for each factor from period to period, but PET was most sensitive to RH and R_n. In most periods, VPD was the third most sensitive factor, and the three temperature factors were the least sensitive.

4. Discussion

4.1. Temporal Characteristics of PET and the Abrupt Change Point

There is a consensus among researchers that an increasing trend in PET is to be expected due to recent decades of persistent global warming. Gradual increases in PET have been detected in many parts of the world, including the arid areas of Iran [29] and India [7] to support this view, while a significant downward trend (p < 0.01) in PET was observed in period I in this area (see Figure 2), which was in contrast to the elevating air temperature (especially T_min). Numerous researchers have observed an identical phenomenon around the world and they have identified it as the evaporation paradox [8,24,25,27]. Nevertheless, global warming has continued unabated, so the evaporation paradox is unlikely to occur in this study area in the future.

Our results indicate that annual PET tends to decrease before 1996, and then increase significantly afterward in the study area (see Figure 2). The substantial trend of abrupt change existed in many regions in China, such as the Hai River Basin [35], Jiangsu province [24], Jing-Jin-Ji region [10], Xinjiang [55], and the Wei River Basin [22], although with differences in the timing of abrupt changes.

It is of interest to know what caused the abrupt change in PET in this study area. Although, many previous studies have investigated the controls on PET around the world and concluded that influential factors may have differed in different regions and periods. Han et al. (2018) analyzed PET in the Jing-Jin-Ji region, north China, located in semi-humid and semi-arid areas, and found that the decreasing W_s and sunshine duration were the main reasons for the decrease in PET from 1961 to 1991, and T_mean was the primary variable that increased the PET from 1992 to 2015 [10]. Dong et al. (2020) found that the abrupt change occurred in 1993 in annual PET time series for Xinjiang, northwest China, which has an arid-desert cold climate [55]. They attributed the decrease in PET to decreased W_s before 1993 and identified an increasing T_mean and decreasing RH as the principal factors of increased PET after 1993. This study differs from the preceding studies in that we attempt to explain the driving forces of abrupt change in PET in a humid region in China. In period I, we contributed the PET change to the decreasing R_n and W_s, while in period II, the significant increase in PET (p < 0.01) was due to the increasing T_min, W_s, and VPD (see Figure 6).

4.2. Sensitivity of PET Changes to Climate Factors

It is clear that changes in long-term PET are not caused by one climate factor but are the combined effects of multiple climate variables. Thus, to reveal the controlling factors for the detected PET abrupt change more easily, we identified the results of the magnitude of basic climate factors, the contributions and sensitivity of climate factors to PET at annual and seasonal scales in 1970–2020 (see Figure 6 and Figure 7).

In period I, the significant decrease in W_s (p < 0.01) appeared to be most influential in the decrease in PET in autumn and winter. This observation was consistent with the results of studies conducted in many parts of the world [2,37,56]. The decrease in R_n in period I was the leading factor for PET variation in spring, summer, and the growing season, a result that was consistent with the studies of the Taohe River Basin [31], the Hai River Basin [35], the entire northern hemisphere [33], and northeast India [56]. In light of the results presented above, one can appreciate that the driving factors that influenced PET dynamics differed according to periods. At an annual scale, although, rising air temperature should lead to increases in PET. However, this positive effect could be offset by the decreasing R_n, leading to the distinct “evaporation paradox” phenomenon.

In period II, the dominance of PET dynamics differed from period I. Changes in T_min, W_s, and VPD had the greatest influence and resulted in increased PET, and RH and R_n had less influence on annual PET. There is in general agreement with the literature concerning the principal climate factors that influence annual PET. Tang et al. (2021) identified that the combined effects of VPD and W_s changed PET in the Siberian River Basin in 1975–2014. Research into the arid region of Beijing–Tianjin over a longer period (1959–2011) showed that a decrease in W_s together with a decrease in air temperature were responsible for a significant decrease in PET [57]. Mosaedi et al. (2017) found that PET was influenced by RH, W_s, and T_mean in most parts of Iran [17]. However, few studies of humid regions identified VPD as a driving factor of PET change. This study builds on previous research by presenting insights into the influence of increased VPD on increasing PET in the SC-CQ region, a typical humid area in China.

Generally, our sensitivity analysis suggested that annual and seasonal PETs were positively correlated with three climate variables, R_n, VPD, and W_s, and negatively correlated with RH (see Figure 7). This is in agreement with common sense that PET increases with increasing temperature, R_n, VPD, and W_s, but decreases with increasing RH. RH and R_n were the most sensitive factors to PET in this study area. These findings are similar to those of the majority of previous studies that found PET was highly sensitive to both RH and R_n [16,21,29]. However, different sensitivity analysis results existed in some other regions. Ghiami-Shomami et al. (2019) found that PET was significantly most sensitive to air temperature in the forested mountainous watersheds of Japan [9]. Li et al. (2014) found that W_s had the highest sensitivity in northwest China [52]. The sensitivity order of meteorological factors to PET may exist differences due to differences in the study location with respect to topographic and geographic characteristics and differences in the sensitivity analysis procedures.

The effects of climate variables depend not only on the sensitivity of PET to the climate variables, but also on the magnitude of the changes in overall trends of critical climate variables [16,21,25]. Although PET was most sensitive to RH at all time scales, it contributed little to the variation in PET in period II because its variation was small. A similar phenomenon has also been observed in many previous reports. For example, Xu et al. (2006) analyzed the sensitivity of PET to major variables in the Yangtze River catchment and found that RH made little contribution to annual PET, despite regional PET being most sensitive to RH [25]. Zhao et al. (2015) revealed that RH and air temperature were the second-most sensitive factors in the Hai River Basin [35]. However, the distinct weakened W_s was considered as the dominant factor in changing the PET.

4.3. Climate Elements Influencing PET Changes and Possible Impacts from Human Activities

Steeply increasing trends in three temperature variables (T_mean, T_min, and T_max) were found in the study period and area (see Table 2). Previous studies attributed rising air temperature to increases in atmospheric greenhouse gases [2], haze formation [40], and cloud cover [12]. Compared with T_max and T_mean, T_min has a greater influence on PET variation (Figure 6). Turkes and Sumer (2004) and Tao et al. (2017) produced similar results in Turkey and the Xinjiang Autonomous Region, respectively [58,59]. They found that warming trends were more significant for T_min.

However, we do not intend to conclude that PET changes were dominated by increasing temperature based on the foregoing analysis results. A closer look at Figure 6 reveals that W_s is a leading factor in PET changing. Our results illustrate that W_s exhibits a significant decrease in period I, while a statistically significant increasing trend is observed in period II (p < 0.01) (see Table 2). Jiang et al. (2019) found a similar tendency in W_s for Southwest China [16]. However, the weakening of the East Asian monsoon has been demonstrated in many previous studies [60,61]. It is widely accepted that a weakened atmospheric circulation resulting from shrinking temperature differences between the polar and tropical regions is responsible for the decrease in W_s [32,37]. Additionally, recent studies revealed that the increasing surface roughness associated with human activities (e.g., rapid development in high-rise buildings) may be responsible for the decrease in W_s [9,62]. Nevertheless, an increasing trend in W_s was detected from 1997 to 2020 in the study area, while areas of greening continued to expand as a benefit of various ecological restoration projects in Southwest China [63]. The driving factors of W_s in the SC-CQ region therefore require further investigation.

The decreasing R_n appeared to be the major causes for the PET changing in most periods in the study area (Figure 6). This observation is consistent with the results of many previous studies worldwide [7,35,64]. Many researchers have hypothesized that increased cloud cover contributes to decreased solar radiation [65]. However, this hypothesis does not hold in this study area due to the downward trends in precipitation that were observed in most parts of the SC-CQ region [47,48]. A robust alternative explanation is that aggregated aerosols resulting from human activities have a very large negative influence on R_n [64]. Wang et al. (2017) proposed that increasing aerosols were more prevalent in areas with a high population density than in sparsely populated areas [11]. One reason to support this view is that modern buildings obstruct windflow and thus reduce aerosol diffusion [11]. Another reason is that industry in areas with a high population density, such as the megacities Chongqing and Chengdu, is widely acknowledged as a source of serious air pollution [16]. The explanation that intense human activities force changes in solar radiation can be inferred from previous studies.

The impact of changes in RH was found to be considerably less significant than that of W_s, R_n, and T_min in the SC-CQ region (see Figure 6). This finding aligns with the extant literature [66,67]; however, other studies have posited that RH is the dominant variable affecting PET variation in arid and semi-arid regions where water resources are scarce [52,68]. It appears that the response of PET to RH is highly dependent on the specific climatic zone. Notably, a decrease in RH from 1970 to 2020 across the entire study area correlated strongly with decreased precipitation and increased temperature [16,47,48]. Such reductions in RH, along with diminishing atmospheric vapor pressure, are common in this area due to the pronounce the urban dry-island effect (UDI) compounded by significant urban expansion [16]. Decreases in precipitation and RH are fueling the shift toward increasing arid conditions. In the recent past, the study area has witnessed a surge in severe and extreme drought occurrences, attributable to the persistent trend of desiccation [47].

Human activities have vital impacts on urban weather and climate, eventually leading to regional-scale hydro-climatic changes. With the implementation of national strategies, such as Development of Western Regions and Development of the Yangtze River Economic Belt in recent decades, the SC-CQ region has become one of most rapidly developing and industrializing regions in China. Human activities in the SC-CQ region are mainly concentrated around urban expansion. Zheng et al. (2020) highlighted that changes in land use and land cover greatly influenced the local climate in the Qinhuai River Basin in China [5]. Wang et al. (2017) found that high-rise buildings impeded wind flow, which led to aerosol accumulation and ultimately a decrease in R_n [11]. Huang et al. (2021) implied that an increase in surface roughness caused by human activities was thought to be the main reason for the decrease in W_s [62]. The massive conversion of vegetation cover to urban land use was widely considered to be the leading cause of aggravated UDI and urban heat-island (UHI) effects, which strongly increase evapotranspiration demand [1,60].

4.4. Potential Effects on Agricultural Water Management

The demand for food and the demand for water in agroforestry systems are increasing due to world population rise [5]. Future irrigation systems are supposed to use less water than now, which is beneficial for sustainable water resource management. More efficient water-saving practices are important for meeting the growing demand during acute water shortage [69]. Evapotranspiration contributes to the highest water loss in cultivated areas [12]. Therefore, the accurate calculation of evapotranspiration is essential to irrigation regimes planning, especially in arid and semi-arid areas [68,69]. The investigation of the underlying mechanisms affecting PET variations during the growing season yielded valuable insights into crop water requirements, particularly relevant for the study area—one of the primary rice-producing regions in China. Our data show that the substantial rise in PET during the growing season in period II can be attributed to the upward trend in VPD, which correlates with both atmospheric water demands and surface conductance [4,60]. Consistent with prior research, our findings affirm that PET trends are intrinsically tied to VPD trends, especially during the dry seasons with scarce rainfall in Southwest China [12,70]. Generally speaking, crop yield declines when there is an increasing trend in VPD caused by heightened drought stress [70]. Consequently, the significant positive trend in PET (p < 0.01) during period II may have escalated crop water demands, exacerbating the water shortage crisis [12,70]. Future PET is expected to continue to increase due to ongoing climate change and the likelihood of future droughts, which will increase irrigation water requirements [7,55,69]. These identified trends in PET underscore the imperative need for the development and dissemination of science-based water conservation technologies and sound agronomic management practices. Future work could refine the precision of agricultural water requirements and crop yield estimation by delving into the sudden changes in climate through advanced modeling techniques that utilize high-quality remote sensing products. Future investigations should focus on the applicability of different PET models in this study area to provide more accurate information for advanced studies on PET trends and their impacts on agroforestry systems.

5. Conclusions

This study focused on the abrupt change in PET and its mechanism at annual and seasonal scales in the SC-CQ region in Southwest China for the period of 1970−2020. The detrending method was used to quantify the relative importance of seven climatic drivers to PET variations. The key conclusions of the study are as follows.

(1) Over the SC-CQ region, the results of the Pettitt test exhibited that the abrupt shift point of the long-term PET date appeared around 1996. Annual PET decreased significantly from 1970 to 1996 (−2.3 mm/year, p < 0.01), followed by a significant increasing trend from 1997 to 2020 (2.8 mm/year, p < 0.01). In period I (1970–1996), T_min and RH increased significantly, and W_s, R_n, and VPD decreased significantly. Nonsignificant trends were observed for T_max and T_mean. In period II (1997–2020), all climate variables showed increasing trends. Significant trends were detected in T_max, T_min, T_mean, W_s, and VPD (p < 0.01). Our findings imply that changes in climate variables are closely related to human activities. Urban planning in humid, wet regions must take into account the impact of urban development on climate to reduce the risk of damage to ecosystems and humans.

(2) The “evaporation paradox” existed in 1970–1996, indicting that rising temperature does not necessarily lead to increasing PET due to the variation in other dominant factors. Contribution analysis showed that before 1996, R_n and W_s were the main driving factors of annual PET changes. VPD, R_n, and T_min were responsible for the positive trends of annual PET in 1997–2020. Sensitivity analysis indicated that RH and R_n were considered as the second-most sensitive variables throughout the SC-CQ region. Differences between the results of the contribution method and sensitivity analysis suggested that, in addition to the sensitivity of PET to climate factors, the magnitudes of the changes in climate factors also result in huge impacts on PET dynamics. This present study quantitatively illustrated the temporal variation in climate factors and their sensitivity and contributions to changes in annual and seasonal PETs in 1970–2020. The findings further present a more comprehensive understanding of water cycle process response to climate changes.

(3) Our study concludes that the climate in the SC-CQ region has become drier in recent decades. Higher irrigation water demand for crop growth may well result from increased PET as increasing near-surface atmospheric dryness. VPD governs potential water loss, so accurately modeling VPD has profound significance for local water resource management and agricultural production. Future work must account for climate change and fully consider the implications of VPD variation to better predict crop water requirements and improve agricultural water management in humid, wet regions.