This study investigated monthly and annual reference evapotranspiration changes over southwestern China (SWC) from 1960 to 2012, using the Food and Agriculture Organization of the United Nations’ report 56 (FAO-56) Penman–Monteith equation and routine meteorological observations at 269 weather sites. During 1960–2012, the monthly and annual decreased at most sites. Moreover, the SWC regional average trend in annual was significantly negative (p < 0.05); this trend was the same in most months. A new separation method using several numerical experiments was proposed to quantify each driving factor’s contribution to changes and exhibited higher accuracy based on several validation criteria, after which an attribution analysis was performed. Across SWC, the declining annual was mainly due to decreased net radiation (RN). Spatially, the annual changes at most sites in eastern SWC (excluding southeastern West Guangxi) were generally due to RN, whereas wind speed (WND) or vapor pressure deficit (VPD) was the determinant at other sites. Nevertheless, the determinants differed among 12 months. For the whole SWC, increased VPD in February and decreased WND in April, May, and October were the determinant of decreased ; however, decreased RN was the determinant in other months. Overall, the determinant of the monthly changes exhibited a complex spatial pattern. A complete analysis of changes and the related physical mechanisms in SWC is necessary to better understand hydroclimatological extremes (e.g., droughts) and to develop appropriate strategies to sustain regional development (e.g., water resources and agriculture). Importantly, this separation method provides new perspective for quantitative attribution analyses and thus may be implemented in various scientific fields (e.g., climatology and hydrology).
Evapotranspiration (ET), which is one of the most important components of the hydrological cycle, accounts for approximately two-thirds of the precipitation that falls on land (Baumgartner and Reichel 1975). Furthermore, from the perspective of energy balance, ET utilizes approximately 60% of the available annual solar radiation received on Earth’s surface (Wang and Dickinson 2012; Wild et al. 2013). Consequently, global climate patterns (e.g., temperature, precipitation, and heat waves) are influenced by ET-related processes, which control the partitioning of energy and water fluxes. Besides, ET is important for the management of farmland and pasture irrigation, in the maintenance of natural ecosystems (e.g., forest), in ensuring a sustainable water supply to meet domestic and industrial demands, and for estimating environmental and ecological water requirements, which is an indispensable component (McVicar et al. 2007; Zuo et al. 2012; Martí et al. 2015). Therefore, knowledge of the spatiotemporal distribution of ET is necessary for a comprehensive understanding of climate change and its impacts on the hydrological cycle, thus enabling the calculation of the required amount of crop water.
However, direct ET measurements are difficult (Brutsaert 1982), particularly on large spatial and long-term temporal scales, mainly because it is a complicated, physical process that includes both evaporation from soil and from vegetative surfaces and plant transpiration. Usually, ET measurements require specific devices (e.g., a lysimeter, eddy covariance system, and scintillometer; Hargreaves 1989; Allen et al. 2011) or imaging techniques (Allen et al. 2007, 2011); furthermore, accurate measurements of various physical parameters or the soil water balance with these devices are expensive (Allen et al. 1998). Therefore, measured ET records are not available in most cases, and thus the concept of a reference evapotranspiration has been proposed. Commonly, this reference value is defined as the evaporation occurring from a land surface covered by a “reference crop (usually assumed as short, complete and green plant cover)” that is well watered (Allen et al. 1998). Reference evapotranspiration is frequently used as an intermediate variable to estimate ET through various pathways, such as a crop coefficient Kc approach, hydrological models, etc. Most importantly, is easy to directly compute using different mathematical estimation models and routine meteorological observations, such as temperature, wind speed, vapor pressure, and solar radiation (Allen et al. 1998). To date, numerous models (e.g., Hargreaves, Priestley–Taylor, and Penman–Monteith models) have been proposed (Priestley and Taylor 1972; Allen et al. 1998; Hargreaves and Allen 2003), among which the physically based Penman–Monteith model is recommended by the Food and Agriculture Organization of the United Nations (FAO); this model has been widely used in various agricultural, hydrological, environmental, ecological, climatological, and other studies throughout the world (Dinpashoh 2006; Fan and Thomas 2013). In recent decades, several studies have specifically focused on the spatial and temporal changes in induced by global climate change (Wang et al. 2007; Zhang and Shen 2007; Xu et al. 2006, 2015; Zuo et al. 2012; Papaioannou et al. 2011; Irmak et al. 2012; Jhajharia et al. 2012; Hua et al. 2013; Vicente-Serrano et al. 2014; Piticar et al. 2016). For example, Papaioannou et al. (2011) investigated the changes in in Greece and concluded that the annual declined before the early 1980s, and then increased afterward. Irmak et al. (2012) analyzed trends over the past few decades in the Platte River basin of central Nebraska, United States, and found that significantly declined by −0.36 mm yr−1. Xu et al. (2015) noted a decreasing trend in in most of the Jing River basin in China during each season (except for autumn).
With continuously increasing temperatures, ET is expected to increase. However, both the observed global pan evaporation and have decreased in recent decades (Liu and Zeng 2004; Burn and Hesch 2007; Roderick and Farquhar 2002, 2004; Roderick et al. 2007; Bandyopadhyay et al. 2009; Zheng et al. 2007, 2009; Sun et al. 2010; Dinpashoh 2006; Liu et al. 2011; D. Zhang et al. 2013), which is known as the “evaporation paradox” (Brutsaert 1982; Roderick and Farquhar 2002; Roderick et al. 2007). Several researchers have widely discussed the possible reasons for the paradoxical decrease in pan evaporation and based on temperature, wind speed, solar radiation, and vapor pressure changes. For example, the decrease in pan evaporation and is mainly due to decreases in wind speed over some regions, such as Australia (Roderick et al. 2007), Iran (Dinpashoh 2006), the Canadian Prairies (Burn and Hesch 2007), the Tibetan Plateau (Zhang et al. 2007; Liu et al. 2011), the Haihe River basin (Zheng et al. 2009), and Jiangxi Province in China (Sun et al. 2010); however, the observed decreases in Ireland (Stanhill and Möller 2008) and the Yangtze River basin (Xu et al. 2006; Wang et al. 2007) have been attributed to decreasing solar radiation. Furthermore, water vapor and temperature exert dominant influences on the decreases in pan evaporation and in some regions (Chattopadhyay and Hulme 1997; Golubev et al. 2001; Tabari et al. 2012; D. Zhang et al. 2013). Notably, the basic reason for the evaporation paradox is that significant global warming has garnered more public attention and is thus expected to have larger impacts on pan evaporation and , whereas other critical climatic variables (e.g., solar radiation, vapor pressure deficits, and wind speed) and their interactions have not been reasonably or sufficiently considered.
A number of previous studies have focused mainly on annual and seasonal changes in , but its monthly variations have received less attention [e.g., in southwestern China (SWC)]; however, the monthly changes are more important and applicable to water resource management and establishing irrigation schemes. Furthermore, the physical mechanisms related to the changes in have mostly been detected in correlation analyses, such as the sensitive coefficient method (D. Zhang et al. 2013) and the differential equation approach (Roderick et al. 2007; Sun et al. 2010, 2012; Liu and Zhang 2013). These methods do not comprehensively consider the impacts from interactions among various driving factors and potentially result in uncertainties and even errors in determining the dominant factors due to inaccurate contributions of irrespective factors (D. Zhang et al. 2013; Liu et al. 2016). On the other hand, the spatial and temporal variabilities of ET or are essential in climatic analyses (Blackie and Simpson 1993), particularly in predicting potential regional climate change scenarios and extreme climate events such as droughts. Recently, based on various drought indices in which was involved (Croitoru et al. 2013), regional and global droughts were revealed to increase with successive and rapid warming (Sheffield et al. 2012; Dai 2013; Trenberth et al. 2014; IPCC 2014). This increase in droughts is also an important reason why researchers have become increasingly concerned about and why numerous research projects related to the implications of changes in have been conducted in many regions (Dalezios et al. 2002).
Using SWC as an example, a number of studies have reported relatively thorough analyses of droughts in this region based on drought indices. There exists a general agreement that the droughts in SWC have been observed to become more frequent and intense during the past five decades, and this trend was projected to continue in future (Wang and Chen 2014; Wang et al. 2015a). Therefore, studies investigating the mechanisms of the changes in and quantifying the magnitudes of these variations in response to ongoing climate change are crucial to better understand the physical mechanisms of drought and its variations. Although some studies have partially addressed seasonal and annual variations and their long-term patterns over SWC or its subregions (Fan and Thomas 2013; Wang et al. 2012; Wen et al. 2012; Liu et al. 2016), a comprehensive spatiotemporal analysis and evaluation of the related physical mechanisms are still needed. In summary, the current study aims to 1) comprehensively evaluate the annual and monthly changes in from 1960 to 2012, which were estimated using observations from 269 weather sites in SWC and the Penman–Monteith model; 2) develop a new separation method for quantifying the impact of each climatic factor on the changes in ; and 3) attribute the changes in the monthly and annual based on the individual contributions of each factor. The present study provides significant insights allowing us to fully understand the spatial and temporal changes in in response to climate and their implications for ongoing and intensifying droughts in SWC (Sun et al. 2016, 2017). Furthermore, the new separation method potentially provides an important reference for quantitative attribution analyses in other scientific fields.
2. Data and methods
a. Study area and data
In the current study, SWC is defined as the area between 21° and 34°N and 97° and 110°E, and it covers a large geographic area, ranging from the high plateaus of western Sichuan and Yunnan to the low-lying Sichuan basin (Fig. 1). SWC is one of the most densely populated regions in China, accounting for approximately one-sixth of the total national population; it is also a main grain-producing area, providing approximately 16% of the national food supply. A typical subtropical monsoon climate prevails across SWC, with a clearly defined dry/wet season and a rainy season that usually begins in April and ends in October.
Routine meteorological observations were required to perform comprehensive analyses. Therefore, the monthly precipitation P (mm month−1), maximum and minimum temperature (°C), relative humidity (%), sunshine duration (h month−1), and wind speed at 10 m (m s−1) measured at 334 weather sites between 1960 and 2012 were collected from the China Meteorological Administration (CMA). With the exception of precipitation, which was reported as a cumulative monthly value, the monthly values for the other four meteorological variables were all computed by averaging the daily measured values within a certain month. Two data quality issues for the historical meteorological observations need to be mentioned here. First, Wijngaard et al. (2003) reported that data inhomogeneity in the long-term meteorological datasets, if not properly controlled, could largely impact the robustness of the study results. Unfortunately, because of a lack of information about station history metadata, we could not adjust the data homogeneity. As an alternative, the Pettitt test, a method used to test time series homogeneity (Wijngaard et al. 2003), was employed to check these variables. Sites with meteorological variables that did not pass the significance test (p < 0.05) were eliminated from the 334 sites. Second, if one site had available data for less than 10 months in every year, it was removed. Then, the missing values in the records of the other sites were completed with data from nearby sites using linear regression equations. Ultimately, data from 269 sites in SWC were utilized (Fig. 1).
1) Estimation of
Because the primary aim of this study was to attribute variations to climate change (i.e., solar radiation, temperature, vapor pressure deficit, and wind speed), a key criterion for selecting the appropriate estimation method was that a method should involve comprehensive physical considerations of evapotranspiration processes. Therefore, we decided to compute using a modified Penman–Monteith equation (i.e., the FAO-56 type; Allen et al. 1998). The FAO-56 Penman–Monteith equation has been recommended as a standard tool for calculating with climatic data by the International Commission on Irrigation and Drainage (ICID), the FAO, and the American Society of Civil Engineers (ASCE):
where RN [MJ (m2 day)−1] is the difference in incoming net shortwave and outgoing net longwave radiation (detailed information is shown in the supplemental material); G [MJ (m2 day)−1] denotes soil heat flux density, which can be ignored on monthly or longer scales; γ [kPa (°C)−1] is a psychometric constant; Δ [kPa (°C)−1] represents the slope vapor pressure curve; WND (m s−1) is the wind speed at a 2-m height and is converted from the wind speed at a 10-m height; VPD (kPa) denotes the vapor pressure deficit and can be expressed as the difference between saturation and actual vapor pressure; and TAVE (°C) represents the mean temperature. Detailed equations for RN, γ, Δ, and WND are presented in the study by Allen et al. (1998). All the computations were performed on a monthly scale according to Allen et al. (1998), and then the annual was obtained by summing the monthly values within each year.
2) Temporal trends
The trends for each hydroclimatic variable were estimated using the following linear equation:
where yi and xi (i = 1, 2, …, n) represent the hydroclimatic variable and the corresponding year, respectively; a is the temporal trend; b is the intercept; and n indicates the sample size. Here, the regression parameters (i.e., a and b) were estimated using the least squares technique. Generally, a positive (negative) a value suggests an upward (downward) trend. A larger (smaller) value of a denotes a stronger (weaker) trend for y. The statistical variable T (Huang and Chen 2011) was calculated using the following equation to determine whether the linear trend was statistically significant (p < 0.05 in this study):
where R denotes the correlation coefficient between y and x. The significance level of the estimated linear trend was determined by comparing the T values with the p values obtained with a Student’s t test.
3) Method for attributing changes in
In this study, the driving factors of consisted of TAVE, RN, VPD, and WND. Therefore, we designed five experiments with the equation [i.e., Eq. (1)], including one control experiment (Sim_CTR) and one sensitivity experiment for each driving factor to quantify the individual contributions of each factor to between 1960 and 2012. The Sim_CTR experiment was conducted with the four driving factors from 1960 to 2012. The four sensitivity experiments were performed for the same period as the Sim_CTR experiment. For each sensitivity experiment, one driving factor from only 1960 and other factors from 1960 to 2012 were used. For example, the temperature sensitivity experiment (named Sim_TAVE) repeatedly employed temperature data from only 1960, but used data for the other three factors from 1960 to 2012 as the inputs for calculating . The sensitivity experiments for RN, VPD, and WND were named Sim_RN, Sim_VPD, and Sim_WND, respectively. For clarity, we also showed the detailed information about the designations of these experiments in Table S1 in the supplemental material.
Then, we utilized the following equation to distinguish the contributions from each factor:
where represents the ith factor’s contributions to the changes in and is the calculated trend for Sim_CTR, but is for the ith sensitivity experiment. This method is known as approach A.
There are well-known interactions among these driving factors, which potentially introduce some uncertainties into the estimated individual contributions of each factor to the changes in (D. Zhang et al. 2013; Liu et al. 2016). For example, contributions from TAVE alone in approach A might have a larger bias from the real values, mainly because Sim_TAVE includes contributions from the interactions among RN, VPD, and WND. Therefore, we developed an alternative concept for separating the contributions from each driving factor (approach B) to minimize the confounding effects from these interactions. Thus, for a given sensitivity experiment (e.g., Sim_TAVE), its trend could be regarded as the sum of the contributions from the other three factors (e.g., RN, VPD, and WND) and was expressed as follows:
where represents the cumulative contributions of the driving factors (excluding the ith factor) to the changes in ; k denotes the sensitivity experiment for a given driving factor; and n is equal to 4, which is the number of sensitivity experiments. As a result, the individual contributions of each factor to the changes in were obtained by solving Eq. (5):
Importantly, approach B does not involve Sim_CTR [i.e., Eq. (5), excluding ], consequently implying that this new separation method is more independent than approach A. We evaluated the performance of approach B on yearly and monthly scales using several validation criteria [i.e., the correlation coefficient, mean relative error (MRE), and root-mean-square error (RMSE) between or and the Sim_CTR trends] and compared the results with the data obtained using approach A [shown in section 3d(1)] to test the assumption that approach B is the more reasonable approach.
Figure 2 shows the monthly and annual average over five subregions and SWC from 1960 to 2012. As expected, exhibited strong seasonal fluctuations because of the influences from many meteorological variables. For SWC, the annual ranged from 910 to 1022 mm, with a mean value of 958 mm. In the comparison of the annual among these five subregions, Yunnan and West Guangxi had larger values (>1070 mm) than the other three subregions (<900 mm). The maximum monthly in Yunnan and the other four subregions and SWC appeared in May (128 mm) and July (from 117 to 134 mm), respectively, with larger variability among sites in May–September. Nevertheless, from November to February, was less than 75 mm, with smaller variability among sites. Interestingly, in June in Sichuan, Guizhou, and SWC was slightly less than the value calculated in March, which was mainly due to a smaller aerodynamic term of (not shown here). Therefore, the spring (March–May) and summer (June–August) accounted for more than 30% of the annual total in all subregions (excluding Yunnan) and in SWC.
Figure 3 depicts the spatial patterns of the monthly and annual values averaged from 1960 to 2012. The exhibited an increase from northeast to southwest, southwest to northeast, and northwest to southeast during January–June, August–September, and October–December, respectively, which might be related to the different spatial distributions of the climatic conditions (e.g., temperature, wind speed, solar radiation, and vapor pressure) and the complicated topography in SWC. On the other hand, evident regional differences in the intra-annual fluctuation of (e.g., time of the maximum ) were also observed by comparing the monthly spatial distributions, which were also reflected in the intra-annual fluctuations of the regional average (Fig. 2). Similar to the patterns from January to June, the annual increased from northeast to southwest, with the highest values (>950 mm) observed in Yunnan and West Guangxi.
b. Changes in
The trends in the monthly and annual for each subregion and SWC from 1960 to 2012 were calculated using a regression analysis (Fig. 4). The annual in the five subregions and SWC all decreased. Specifically, with the exception of Yunnan, the significantly (p < 0.05) decreased in each subregion and in SWC, with a maximum of −1.04 mm yr−1 in Chongqing. By combining the data for all the subregions and SWC for each month (12 months × 6 regions), the 50 cases (19 cases with p < 0.05) indicated that the monthly decreased at different magnitudes, and more evident reductions mainly occurred between May and August. Notably, the in Yunnan exhibited an insignificant increase over 7 months. In the other four subregions and SWC, the decreased in most of the 12 months (≥7 months). Table 1 also shows the percentage of the 269 sites in which positive and negative trends were recorded from 1960 to 2012. In broad terms, negative trends were dominant over SWC on both monthly and annual scales. In January, March, June, and July, most sites (>70%) displayed negative trends, which were statistically significant (p < 0.05) for more than 20% of the sites, particularly in January and July (>30%). During February, October, and November, most sites (>50%) exhibited a positive trend, and November had the maximum percentage of sites (21%) with a significant trend (p < 0.05). As a result, during February and November, there was a general increase in across SWC from 1960 to 2012. This pattern was also observed on the annual scale because 70.2% (39.7% with p < 0.05) of the sites exhibited negative trends and the average decreased in the whole region by 0.61 mm yr−1 from 1960 to 2012.
Further analysis did not identify consistent spatial patterns in the absolute magnitude of the monthly and annual changes in (Fig. 5). For most sites in southwestern SWC, with the exception of March–May, July, and September in which decreased, the values increased at different magnitudes in the other months, with a maximum value (>0.15 mm yr−1) in February. In northeastern SWC, the values exhibited decreases at most sites in January–March, June–August, and November–December, with the greatest trends (<−0.2 mm yr−1) occurring from June to August, whereas in September more sites had increased values, and a complex distribution of changes in was observed in other months. For most sites in mideastern SWC, decreased and increased values were recorded in January, March, April, June–August, and October, and in February and November, respectively; however, positive and negative changes in were irregularly distributed over this region during other months. In southeastern SWC, the generally decreased in January, March, May–July, and September and increased in April and November. In general, the annual increased in western SWC but decreased in eastern SWC, both with absolute rates greater than 0.6 mm yr−1 at most sites.
c. Changes in the major driving factors for
Before analyzing the contributions from each driving factor to the changes in , the observed magnitudes of their monthly and annual changes over the 53-yr study period were also estimated with a linear regression analysis. Monthly and annual trends in the major driving factors of over the five subregions and SWC are shown in Fig. 6. Annual RN and WND significantly (p < 0.05) decreased in all subregions and SWC. Among these study regions, the maximum changes in RN [−3.0 MJ (m2 yr2)−1] appeared in Guizhou, and the maximum changes in WND [−0.0062 m (s yr)−1] occurred in West Guangxi. However, in the five subregions and SWC (not including Chongqing), TAVE and VPD significantly (p < 0.05) increased, with the largest rates of 0.018°C yr−1 in Yunnan and 0.002 kPa yr−1 in Yunnan and West Guangxi. Negative trends for the monthly RN were nearly observed in each month over each subregion and SWC, and 17 of 72 cases (12 months × 6 regions) reported significant decreases (p < 0.05), particularly in January, March, and June–September, which exhibited larger decreases [<−0.12 MJ (m2 yr2)−1]. In each subregion and SWC, the monthly TAVE generally showed positive trends, with larger values (>0.02°C yr−1) observed in February and November. Furthermore, 21 of the 72 cases (12 months × 6 regions) displayed significant (p < 0.05) increases in TAVE, particularly in Yunnan during all months except May. For the vast majority (71) of the 72 cases (12 months × 6 regions), the monthly WND exhibited a clear decrease, with 67 cases passing the significance test (p < 0.05), and larger decreases [<−0.003 m (s yr)−1] were observed in February–May in each region. The monthly VPD for most of the 12 months in each study region increased at varying magnitudes. With the exception of Chongqing, which displayed significant (p < 0.05) increases only in November, the VPD significantly increased (p < 0.05) in the other study regions for more than 4 months, mainly during August–December.
Table 2 shows the percentage of the 269 sites in which the trends for each driving factor were recorded from 1960 to 2012. For the annual RN and WND, negative trends prevailed (>75% sites), which were significant (p < 0.05) in approximately 60% of the sites. In contrast, the TAVE and VPD increased at more than 85% of the sites, and around 60% were significant (p < 0.05). Although the RN and WND showed positive trends at some sites and the TAVE and VPD showed negative trends at some sites (<20%), less than 10% of sites exhibited significant (p < 0.05) changes. Generally, negative trends in the monthly RN and WND were dominant over SWC. The RN decreased at more than 60% of the sites in most of the months, except for November, in which RN decreased at more than 40% of the sites. Moreover, January and July exhibited the highest percentages of sites (~50%) with significant reductions in RN (p < 0.05). With the exception of August, the WND decreased at more than 70% of the sites for each month, especially in March and April with more than 80% of the sites. In general, negative trends observed in most months were significant (p < 0.05) at more than 50% of the sites. Positive changes in the monthly TAVE and VPD trends were dominant over SWC. Specifically, TAVE trends at most sites (>70%) were positive in all months (excluding January and March), with the highest percentages (>90%) in February, October, and November. Moreover, more than 20% of sites exhibited significant (p < 0.05) increases in TAVE during each month, particularly in December, which had a maximum of 59.6%. Positive trends in VPD were detected at more than 45% of the sites in all months, particularly in February, August, October, and November, in which positive trends were observed at more than 90% of the sites. Except for January, the increases in VPD at more than 20% of the sites were significant (p < 0.05) in all months, of which August–November displayed the higher percentages (>50%).
Figure 7 illustrates the spatial distribution of the annual changes in the major driving factors of at 269 observatories from 1960 to 2012. As depicted in Fig. 7a, large decreases [<−4.8 MJ (m2 yr2)−1] in RN were primarily observed in eastern Sichuan, western Chongqing, middle Guizhou, mid-northern Yunnan, and eastern West Guangxi, whereas slight increases [<3.6 MJ (m2 yr2)−1] were sporadically distributed in western SWC. The TAVE increased at the overwhelming majority of the 269 sites (Fig. 7b), with rates higher than 0.012°C yr−1 over the greater part of SWC (not including the northeastern part). Seen from Fig. 7c, WND at most of the sites exhibited decreasing trends, with relatively higher magnitudes [>0.006 m (s yr)−1] in central SWC and West Guangxi. The VPD basically displayed large increases (>0.003 kPa yr−1) in southern Yunnan, West Guangxi, and some parts of Sichuan (Fig. 7d). Furthermore, the spatial distribution of the monthly changes in the major driving factors of at 269 observatories was calculated (see Figs. S1–S4 in the supplemental material). Overall, there was no consistent spatial distribution of the changes in any one major driving factor in each month. For example, maximum decreases in RN were primarily located in observatories in eastern SWC during June–August, whereas the sites with the largest increases were mainly located in southwestern SWC in June (Fig. S1 in the supplemental material). The greatest increase in TAVE in southwestern SWC appeared in January–March and December, whereas higher TAVE trends were observed in other areas (not including the northeastern part) in February (Fig. S2 in the supplemental material). Furthermore, TAVE mostly decreased in the eastern SWC in January and March. Based on the spatial distribution of the monthly WND trends (Fig. S3 in the supplemental material), the large negative changes were mainly detected in eastern Yunnan, West Guangxi, and northwestern Guizhou in all 12 months, especially in February–April, which exhibited the largest changes. The observatories with maximum increases in VPD were mainly located in southwestern SWC in February and West Guangxi in August, and the largest decreases were observed in northeastern SWC in September (Fig. S4 in the supplemental material).
d. Attributing the changes in
1) Selecting the method to quantify the contributions to the changes in
In this study, both approach A and approach B were employed to obtain more accurate results regarding the contribution of each driving factor to the changes in . The cumulative contributions from each method were used in the comparison with the trend obtained from Sim_CTR. Fig. 8 displays scatterplots of and the Sim_CTR trend on annual scale. Based on the annual trends at the 269 sites, the results obtained using approach B were closer to a 1:1 line than the results obtained using approach A (Fig. 8). Combining the annual results of all sites, we derived R of greater than 0.90 between and the Sim_CTR trends for each approach, with higher values (0.996) resulting from approach B (Table 3). The RMSE (MRE) of the trend was 0.397 (37.9%) and 0.099 mm yr−1 (9.5%) for approaches A and B, respectively (Table 3). Furthermore, we performed a statistical analysis of sites with the same sign for the Sim_CTR trend and (Fig. 8). Clearly, compared with approach A, approach B resulted in more sites with the same direction of change (257 sites for approach A vs 266 sites for approach B). Based on these analyses, approach B was able to better quantify the contributions of the driving factors to the annual changes in than approach A. The results from these two methods were compared on a monthly scale to further examine the robustness of approach B (Table 3; Fig. S5 in the supplemental material). Obviously, the results obtained using approach B were closer to a 1:1 line than the results obtained using approach A in all months. According to the statistical analyses of approach A and approach B on the monthly scale (Table 3), approach B (>0.99) had a larger R and a smaller MRE and RMSE in all months than approach A (Table 3), indicating that approach B also achieved better separation of the contributions of each driving factor to the monthly changes in . Therefore, approach B had higher accuracy and efficiency and was chosen to quantify the contributions of climate change to in this study.
2) Causes of changes in
Based on the six experiments using the Penman–Monteith equation and approach B, the contributions of each driving factor to the monthly and annual changes in over the five subregions and SWC during the study period were calculated (shown in Fig. 9). With decreases in RN and WND in each study region, the annual tended to decrease, and the contributions of RN were greater than the contributions of WND for all the study regions except for Yunnan. However, TAVE and VPD increased in each study region, and the VPD contributions were threefold higher than the TAVE contributions; therefore, they offset annual decreases in to some extent. Comparing the contributions of each factor, the major contributor to the decreased annual in Yunnan was the decreased WND, whereas the decreases in the annual were mainly due to the reductions in RN in the other four subregions and SWC. Notably, the impacts of VPD on the annual in Yunnan were greater than the impacts of RN and WND, with respect to the absolute contributions, and the value was approximately equal to the sum of the RN and WND contributions, which could explain why Yunnan displayed a slight decrease in annual (Fig. 4). As shown in the monthly contributions of each driving factor to (Fig. 9), there were evident intra-annual differences, mainly due to the differences in the changes in each driving factor and the responses of to these factors on monthly scale (Fig. 6). We compared the monthly contributions of each factor to the changes in and identified a dominant factor in each month for each subregion and SWC (Table S2 in the supplemental material). In most of the five subregions and SWC, the changes in during January, March, and June–September were attributed to RN, whereas the determinant in April and December was WND. In November, VPD was the dominant factor accounting for the changes in over each subregion and SWC, with the exception of Sichuan and Chongqing, for which WND and TAVE, respectively, were the main determinants. During February, May, and October, the changes in in most subregions and SWC were mainly attributed to changes in WND or VPD.
Figure 10 depicts the contributions of the major driving factors to the changes in the annual from 1960 to 2012 at 269 sites. The RN negatively contributed to the changes in the annual throughout SWC, and most of the sites had contributions less than −1.20 mm yr−1, whereas RN positively contributed to some sites in southwestern Yunnan (Fig. 10a). Although no obvious spatial distribution of TAVE contributions was observed, TAVE tended to increase the at the overwhelming majority of the 269 sites, with contributions less than 0.4 mm yr−1 (Fig. 10b). In contrast, WND decreased for almost all sites, and higher contributions (<−1.2 mm yr−1) were observed in eastern Yunnan and central West Guangxi (Fig. 10c). Generally, the annual tended to increase because the VPD increased at most of the 269 sites, with particularly higher contributions in northern Yunnan and eastern West Guangxi (>0.8 mm yr−1), as shown in Fig. 10d.
Annual and monthly dominant factors were identified by comparing the contributions of each driving factor to further analyze the main causes of the changes in observed at each site (Fig. 9, Fig. S6 in the supplemental material), as shown in Table 4, Fig. 11, and Fig. S7 in the supplemental material. Table 4 shows the site percentage for each dominant factor attributed to the changes in monthly and annual in SWC. The changes in the annual at 50% of the sites in east SWC (excluding southeastern West Guangxi) were attributed to RN (Fig. 11, Table 4); however, WND (VPD; Table 4) was the main determinant at approximately 26.8% (22.1%) of the sites, which were generally located in western SWC (southeastern West Guangxi; Fig. 11). By comparing the site percentage of the four determinants to the changes in the monthly (Table 4), the dominant factor RN corresponded to the largest percentage of sites (>30%) in most months, except for February, April, November, and December. Among the 12 months, the sites with RN as the determinant accounted for a higher (>50%; from 56.3% in September to 79.0% in July) percentage during June–September and were mostly distributed in eastern SWC, with the exception of September (Table 4; Figs. S7f–h in the supplemental material). In February, VPD was the dominant factor for 51.8% of the sites, predominantly in the middle Sichuan, Yunnan, and Guizhou, whereas RN and WND were the determinants for most sites in northeastern Sichuan and West Guangxi, respectively (Table 4; Fig. S7b in the supplemental material). In April, RN, VPD and WND were the dominant factors in approximately 30% of sites, indicating a complex spatial distribution of these determinants in SWC (Table 4; Fig. S7d in the supplemental material). In May, RN was the dominant factor in 35.3% sites, followed by WND in 33.1% of sites and VPD in 31.3% of sites (Table 4); however, the determinant of RN was irregularly distributed over SWC (Fig. S7e in the supplemental material). During October, RN, WND, and VPD corresponded to site percentages of 30.9%, 30.5%, and 38.6%, respectively (Table 4); furthermore, the determinants in most of the sites were RN in Guizhou, WND in West Guangxi, and VPD in Yunnan (Fig. S7j in the supplemental material). In November, WND and VPD were the dominant factors in approximately 40% of sites, whereas RN was the dominant factor in only 15.8% of sites (Table 4). Moreover, most sites with the determinants of VPD were located in the middle Yunnan and western Guizhou, whereas WND was the dominant factor for most sites in Sichuan and West Guangxi (Fig. S7k in the supplemental material). In December, WND and VPD were the dominant factors for more than 35% of the sites (Table 4), which were broadly spread over SWC (not including eastern Sichuan); however, less than 20% of sites for which RN was the determinant were located in eastern Sichuan (Fig. S7l in the supplemental material). Notably, TAVE was the dominant factor in less than 4% of sites in SWC on both annual and monthly scales, indicating that the impacts of TAVE on the changes in were very limited compared with the other three factors.
a. Comparison of the results
Like most parts of China (Zhang et al. 2009; Sun et al. 2010, 2012, 2014; Fan and Thomas 2013; Huo et al. 2013; Xing et al. 2014; Zheng and Wang 2014; Shan et al. 2015), the annual exhibited a significant decreasing trend in SWC and the five subregions (with the exception of Yunnan) from 1960 to 2012, and the change rate ranged from −1.04 to −0.06 mm yr−1. Moreover, negative trends mainly occurred at 70.2% (39.7% with p < 0.05) of the 269 sites, which were mainly located in eastern SWC. For the regional mean, the decreasing annual values calculated in Sichuan, Chongqing, Guizhou, and SWC were all attributed to a decrease in RN, which was the dominant factor at approximately 50% of the sites. This result agreed with the observations of general decreases in pan evaporation over the Northern Hemisphere, which was possibly related to the measured widespread decreases in sunlight due to increasing cloud coverage and aerosol concentrations over the past five decades (Roderick and Farquhar 2002; Yin et al. 2010). Yang et al. (2012) reported that the annual and seasonal sunshine duration in SWC have decreased from 1969 to 2009, which coincided with our findings (seen in Fig. S8 in the supplemental material), and further pointed out that weakening WND and increasing relative humidity were the principal reason of declined sunshine duration. Additionally, by comparing the magnitudes of the decreases in sunshine duration between urban and rural weather sites, Li et al. (2012) noted that the effects of increased air pollution and increased aerosol loading induced by rapid urbanization were the other causes of declining sunshine duration in SWC. However, in Yunnan and West Guangxi, the major factor influencing the decreased annual was WND, which was consistent with the results in other study regions (Roderick et al. 2007; Zhang et al. 2009; Sun et al. 2010, 2012, 2014; Huo et al. 2013). In addition, the dominant factor influencing the changes in differed from site to site. In eastern SWC (excluding southern West Guangxi), RN was the dominant factor in approximately 50% of sites, whereas WND and VPD were the dominant factors in most sites in the western SWC and southern West Guangxi, respectively. The different dominant factors attributed to the changes in at the site and regional scales were presumably related to differences in the responses of to climate (Huo et al. 2013; D. Zhang et al. 2013; Zheng and Wang 2014), the magnitudes of climate change, and the basic climate conditions (e.g., multiyear mean RN, TAVE, WND, and VPD), which might also be the cause of the different dominant factors at monthly scale.
To explore the possible relationship between geographical location (i.e., latitude, longitude, and altitude) and the contributions of each climate factor to the changes in , their spatial correlations were examined. Generally, climate-induced changes in were significantly (p < 0.05) correlated with the geographic parameters. However, the variance of the climate-induced changes in explained by each geographic parameter was less than 10%, indicating that the impacts of the geographical location on the contributions of different climate factors to the changes in were limited.
b. Implications of the changes in on droughts and terrestrial ecosystems
The changes in have more practical importance because they exert a substantial influence on terrestrial ecosystems and agriculture, which is largely determined by the concomitant variations in precipitation. In SWC, water resource shortages are becoming an urgent problem because of frequent and intense droughts, especially since the start of twenty-first century (Duan et al. 2000; Hu et al. 2009; Huang 2011; Barriopedro et al. 2012; Zhang and Jia 2013; Feng et al. 2014; W. Zhang et al. 2013, 2014; Wang et al. 2015a,b, 2016). For instance, droughts in SWC have become more frequent and intense during the past 50 years and this trend is expected to continue during the twenty-first century (Wang et al. 2015a). By analyzing the average annual precipitation over the five subregions and SWC (Table 5), they all exhibited decreases during the past 53 years, with a range from −0.61 to −1.75 mm yr−1. Despite that the regional mean annual decreased by different magnitudes among these study regions (Fig. 4, right), the differences between precipitation and trends were also negative (with the exception of West Guangxi), particularly in Sichuan and Guizhou, which had a large change rate (>−0.90 mm yr−1; Table 5). Furthermore, monthly differences in precipitation and trends were calculated and are displayed in Table 5. In West Guangxi and SWC, the negative differences mainly appeared in April and August–December, of which West Guangxi exhibited the largest differences in April and August (<−0.70 mm yr−1) and SWC exhibited the largest differences in August and September (<−0.40 mm yr−1). In Sichuan, Guizhou, and Yunnan, the differences between monthly precipitation and decreased in more than half of the 12 months; in particular, decreases occurred in September in Sichuan (−0.75 mm yr−1), April and August in Guizhou (<−0.50 mm yr−1), and June and August in Yunnan (<−0.55 mm yr−1). In Chongqing, negative differences between monthly precipitation and trends were detected in 5 months, with a maximum in September (−1.16 mm yr−1). Notably, there were spatial differences in the combination of the precipitation and trends from region to region. In SWC, more than 75% of the 269 sites showed negative trends in annual precipitation characterized by larger decreases that were generally located in the central region. Correspondingly, the negative differences in their trends were concentrated in southern (with the exception of West Guangxi) and mideastern SWC (>65% sites). With the exception of January, March, and May–July, the decreased differences were observed between the monthly precipitation and trends in more than half of the 269 sites, particularly between August and December, which had a maximum percentage (>65%). Based on the results for the differences between the precipitation and trends, annual and monthly droughts intensified over SWC, which generally agreed well with previous results (Duan et al. 2000; Hu et al. 2009; Huang 2011; Barriopedro et al. 2012; Zhang and Jia 2013; Feng et al. 2014; W. Zhang et al. 2013, 2014; Wang et al. 2015a,b, 2016; Sun et al. 2016, 2017). Consequently, the coupling of the precipitation and changes has restricted regional vegetation growth. Some scholars have noted that droughts that occurred over the past decades have exerted significant and adverse effects on ecosystem productivity in SWC (J. Zhang et al. 2012; L. Zhang et al. 2012). For example, in the study by J. Zhang et al. (2012), more than 50% of vegetation suffered from the drought event from September 2009 to March 2010, and significant spatiotemporal variability was observed in the range and intensity of the adverse impacts. Based on the results reported by L. Zhang et al. (2012), the spring (March–May) drought in 2010 substantially decreased the enhanced vegetation index (EVI) and hence reduced the regional annual gross primary productivity (GPP) and net primary productivity (NPP) in this year by 65 and 46 TgC yr−1, respectively. Notably, was used as a proxy of ET in our analyses. Because ET is always impacted by climate conditions and soil water storage, changes in the water deficit reflected by differences between the precipitation and trends might be underestimated. Therefore, the drought risk over SWC might be higher than the risk determined in our analyses.
Although the FAO-56 Penman–Monteith equation has been widely utilized in various scientific fields (e.g., hydrology, climatology, agriculture, and ecology) throughout the world and is mainly dependent on various climate variables, several factors (e.g., vegetation responses to elevated atmospheric CO2 levels and land surface albedo) that have been ignored by this method have the potential to introduce more or less uncertainty into our results. Based on the assumptions of the FAO-56 Penman–Monteith equation that the stomatal resistance of a single leaf is 100 s m−1 under well-watered conditions and the active (sunlit) leaf area index is 1.44 m2 m−2, the bulk surface resistance is fixed at a constant (~70 s m−1; Allen et al. 1998). However, in practice, the elevated atmospheric CO2 levels are expected to directly influence plant physiology through declining stomatal and canopy conductance, thereby increasing water use efficiency. The expectations have been extensively detected by various vegetation datasets and further validated by numerous global modeling studies during the past few decades (Field et al. 1995; Betts et al. 2007; Cramer et al. 2001; Medlyn et al. 2001; Cao et al. 2010; de Boer et al. 2011; Lammertsma et al. 2011; Miglietta et al. 2011; Wiltshire et al. 2013; Hao et al. 2017; Rigden and Salvucci 2017). Therefore, the absence of data for vegetation responses to elevated atmospheric CO2 levels tends to induce bias in the FAO-56 Penman–Monteith from its real value (involving CO2 effects), potentially impacting the confidence level of some related studies (e.g., Roderick et al. 2015; Milly and Dunne 2016). For example, by comparing the obtained using several variants of the Penman–Monteith equation (including the FAO-56 type) with the value obtained using different climate models, Milly and Dunne (2016) reported that all these variants had a higher estimated than climate models, and thus continental drying might be overestimated by drought metrics with these estimates. Furthermore, the discrepancy between the Penman–Monteith variants and climate models was largely explained by neglecting the bulk stomatal conductance responses to increased CO2 levels within the Penman–Monteith variants. However, empirical evidences of the magnitudes of the vegetation responses to increased CO2 levels are still highly uncertain. For instance, Domec et al. (2009) investigated loblolly pine acclimation to long-term elevated CO2 levels and showed that the effects of elevated CO2 levels on stomatal conductance were manifested only during specified times (e.g., high soil moisture). Moreover, uncertainties may also be introduced in the current study because of the constant ratio [i.e., land surface albedo = 0.23; see Eq. (S10) in the supplemental material] of solar radiation reflected by the land surface. The land surface albedo is a known variable in different climate conditions or changes in soil moisture and vegetation (X. Zhang et al. 2012); therefore, the fixed land surface albedo potentially biases the estimated RN and even .
In summary, the lack of consideration of the impacts of the elevated CO2 levels and the variable land surface albedo may lead to some uncertainties in the current study. However, the magnitude of the impacts of these factors on should be investigated further in the future because this information will provide an important reference for accurate predictions of future hydroclimatological conditions (e.g., drying/drought).
Analyses of the spatiotemporal variations in , an important, integrated variable of climate change, and the related physical mechanisms will help us comprehensively understand climate change and its potential impacts on hydrology, agriculture, and ecology. In the current study, we analyzed the evolution of the spatial patterns of and its four major driving factors in SWC from 1960 to 2012, which were estimated using the FAO-56 Penman–Monteith model. An increase in the annual from northeast to southwest was detected, with an SWC regional mean of 958 mm. Because of the impacts of many driving factors, showed strong seasonal fluctuations and peaked in July in SWC. Based on the trends in the driving factors, the regional average RN and WND both significantly (p < 0.05) decreased in SWC, with change rates of −2.45 MJ (m2 yr2)−1 and −0.005 m (s yr)−1, respectively; however, there were significant (p < 0.05) increases in TAVE (0.01°C yr−1) and VPD (0.0014 kPa yr−1). The monthly RN (excluding in November) and WND, and TAVE and VPD values all decreased and increased to different extents, separately. Because of their combined impacts, the annual trend averaged over SWC was significantly (p < 0.05) negative (−0.61 mm yr−1) and accompanied by negative changes in in most months.
We proposed a new separation approach based on one control and four sensitivity experiments using the Penman–Monteith model to quantify and attribute the changes in . By comparing this method with a classical approach (named approach A in this study), this new method had higher efficiency and accuracy, based on the monthly and annual criteria of R, MRE, and RMSE between the cumulative contributions of the four driving factors to the changes in and the trends observed in the control experiment. Overall, a decreasing RN was the main contributor to declining annual throughout SWC. In eastern SWC (with the exception of southeastern West Guangxi), changes in the annual at most sites were due to RN, whereas changes in other areas were attributed to changes in WND or VPD. Notably, the dominant factors differed in the 12 months, and RN was the main contributor in January, March, and June–September, whereas decreased was mainly attributed to the increased VPD observed in February and the decreased WND observed in April, May, and October.
Based on a full analysis of the variations in in SWC on different spatial (i.e., regional and site) and temporal (i.e., monthly and annual) scales, we confirmed that exhibited significant changes in this study region due to climate change from 1960 to 2012; moreover, the driving factors attributed to the changes in were identified for each site. This detailed information will be useful for obtaining a better understanding of hydroclimatological extreme events (e.g., ongoing and intensifying droughts in SWC) and developing specific measures to sustain regional development (e.g., water resources, agriculture, and ecology). Besides, the new separation method, which uses relatively reasonable concepts (e.g., experiments designations and algorithms) and achieves better performance, provided a new perspective for quantitative attribution analyses. This method has great potential for use in various scientific fields (e.g., climatology, hydrology, and ecology), for example, quantitatively assessing the impacts of changes in land use/cover, climate, and vegetation on the hydrological cycle.
This work was jointly supported by the National Natural Science Foundation of China (Grants 41605042, 41401016, 41375099, 91337108, and 41625019); the Natural Science Foundation of Jiangsu Province, China (Grants BK20151525, BK20140998, and BK20160948); the Special Public Sector Research Program of Ministry of Water Resources (Grant 201301040); the Natural Science Foundation for Higher Education Institutions in Jiangsu Province, China (Grant 16KJB170007); and the Priority Academic Program Development (PAPD) of Jiangsu Higher Education Institutions.
Supplemental information related to this paper is available at the Journals Online website: http://dx.doi.org/10.1175/JHM-D-16-0118.s1.