The Jing-Jin-Ji region (36°02°-42°38°N, 113°25°-119°51°E) is located in North China (Figure 1a). Its altitude ranges from 2 to 2836 m and its general topography consists of mountains, plateaus and plains, with high-relief areas in the northwest and low-relief areas in the southeast. The western part of this region contains the Loess Plateau and the Taihang Mountains, the northern part contains the Mongolian Plateau and the Yanshan Mountains, and the eastern and southeastern parts contain the North China Plain. The Jing-Jin-Ji region is a large, regional, economic community and the fastest-developing region in China. Climatically, it is a semi-humid and semi-arid region, part of the East Asian monsoon region. The annual mean precipitation is about 539 mm, and the annual evaporation is about 470 mm (Liu et al., 2010; Zhao et al., 2014).

Daily meteorological data of 22 national meteorological stations from 1991 to 2015 were downloaded from the National Climate Center of the China Meteorological Administration (http://data.cma.cn). The original variables used to calculate ET₀ were mean air temperature (T_mean), minimum air temperature (T_min), maximum air temperature (T_max), sunshine duration (SD), relative humidity (RH) and wind speed (U₁₀).

We obtained land use data from the Data Center for Resources and Environmental Sciences of the Chinese Academy of Sciences. The resolution of the land use map was 1 km, and land use datasets for the year 2015 were used to identify different land use types (Figure 1b). Depending on the area ratio of a land use type within a 5-km radius around the station site, the stations were classified into three types: natural, farmland and urban stations. Natural stations had the sum of farmland and urban land with a percentage less than 50%, In other words, the percentage of water bodies, forests and grasslands in this region was >50%. The percentage of cultivated land with respect to farmland stations was >50% and that of urban land with respect to urban stations was >50% (Han et al., 2012; Xu et al., 2015).

The FAO56 Penman-Monteith equation, considered to be the most accurate, was applied to calculate daily ET₀. The annual ET₀ value was obtained by summing up the daily values. The equation is expressed as follows (Allen et al., 1998):

(1)

where ET₀ is the daily reference ET rate (mm day^-1), R_n denotes net radiation at the ground surface (MJ m^-2 day^-1), G represents soil heat flux density (MJ m^-2 day^-1), T denotes mean daily air temperature at a 2-m height (℃), U₂ denotes wind speed at a 2-m height (m s^-1), e_s represents saturation vapour pressure (kPa), e_a represents actual vapour pressure (kPa), Δ denotes the slope of the vapour pressure curve (kPa ℃^-1) and γ is the psychometric constant (kPa ℃^-1).

The original wind speed data was obtained at a 10-m height, which then required conversion to wind speed at a 2-m height. The equation for this conversion is expressed as follows:

${{u}_{2}}={{u}_{z}}\frac{4.87}{\ln (67.8z-5.42)}$ (2)

where z is the measurement height above the ground surface (m), u_z is the wind speed at a z-m height (m s^-1).

The Mann-Kendall (MK) trend test is widely used for long-term time series trend analysis in hydrology and meteorology (Jhajharia et al., 2009; Huo et al., 2013; Wang et al., 2015). In this method, the equations for calculating the test statistic S and the standardised test statistic Z are as follows:

$S=\sum\limits_{i=1}^{n-1}{\sum\limits_{j=i+1}^{n}{sgn \left( {{x}_{j}}-{{x}_{i}} \right)}}$ (3)

$sgn ({{x}_{j}}-{{x}_{i}})=\left\{ \begin{matrix} 1\begin{matrix} {} & if({{x}_{j}}-{{x}_{i}})>0 \\\end{matrix} \\ 0\begin{matrix} {} & if({{x}_{j}}-{{x}_{i}})=0 \\\end{matrix} \\ -1\begin{matrix} {} & if({{x}_{j}}-{{x}_{i}})<0 \\\end{matrix} \\\end{matrix} \right.$ (4)

$\sigma _{s}^{2}=\frac{n(n-1)(2n+5)-\sum\limits_{p=1}^{q}{{{t}_{p}}({{t}_{p}}-1)(2{{t}_{p}}+5)}}{18}$ (5)

$Z=\left\{ \begin{matrix} \begin{matrix} \frac{s-1}{\sigma _{s}^{{}}} & if\begin{matrix} s>0 & {} \\\end{matrix} \\\end{matrix} \\ 0\begin{matrix} {} & if\begin{matrix} s=0 & {} \\\end{matrix} \\\end{matrix} \\ \begin{matrix} \frac{s+1}{\sigma _{s}^{{}}} & if\begin{matrix} s<0 & {} \\\end{matrix} \\\end{matrix} \\\end{matrix} \right.$ (6)

where Z is used to estimate the statistical trend. A positive Z value indicates an increasing trend in the time series, while a negative Z value indicates a decreasing trend. There is a significant trend in the time series when |Z| > Z_1-α/2. The tested significance level α was set to 0.05 in this study, and at this significance level, the value of Z_1-α/2 was 1.96.

To assess the effect of meteorological variables on ET₀, linear regression was used to establish a multiple regression equation, and the relative contribution of each meteorological variable to be ET₀ was noted to be a ratio of each variable’s regression coefficient to the sum of all of the variables’ regression coefficients (Li et al., 2014; Wang et al., 2016; Han et al., 2018).

The meteorological stations in the Jing-Jin-Ji region were classified into three types-urban, farmland and natural stations. If the area of urban land or farmland in the 5-km radius of the station was equal to or greater than 50%, the associated station was referred to as an urban station or a farmland station, respectively. If non-urban and non-farmland areas in the 5-km radius of the stations were >50%, these stations were included under natural stations. Based on the 2015 land use dataset, we noted 10 urban stations, 8 farmland stations and 4 natural stations (Figure 2). The four natural stations, namely Qinglong, Chengde, Fengning and Weichang, were located in the northern mountainous area of the Jing-Jin-Ji region. The proportions of the natural area for each of these stations were 76.16%, 73.27%, 61.96% and 68.30%, respectively. The forests in the northern mountainous area have been planted under the Three-North Shelter Forest Program (Jiang et al., 2015). Furthermore, the western and northern mountainous areas are being considered as ecological conservation areas under the coordinated development of the Jing-Jin-Ji region.

Most of the urban stations are located in the plain area. The urban area proportions for six of the urban stations were >75% and those for four of the stations were between 50% and 75%. As the most urbanised station, the proportion of the urban area around Beijing station was 100%, and those around Zunhua and Qinhuangdao stations were 50.15% and 50.51%, respectively. The urbanised stations were located in four directions: Beijing to Qinhuangdao, Beijing to Tanggu, Beijing to Shijiazhuang and Beijing to Zhangjiakou.

The farmland stations were distributed in the western, southern and eastern parts of the Jing-Jin-Ji region. The proportions of the farmland area for these stations were between 55% and 75% (Figure 2). Three farmland stations (Zhangbei, Huailai and Yuxian) were located in the northwestern part of the plain and four farmland stations (Langfang, Rongyuan, Huanghua and Nangong) were located in the southern part. The land use type for Laoting station was mainly farmland, located in the northeastern part of the plain (Figure 1).

Figure 3a shows the time series of annual ET₀ in the study area. The lowest annual ET₀ was 909.7 mm in 2003, whereas the highest annual ET₀ was 1050.5 mm in 2009. The 5-year moving average annual ET₀ of the 22 selected meteorological stations showed an increasing trend from 1991 to 2015. Furthermore, linear regression analysis indicated that ET₀ significantly increased (p < 0.05) at a rate of 7.4 mm per decade in the Jing-Jin-Ji region.

The change in ET₀ showed a spatial difference with different land use types (Figure 3b). The ET₀ of most stations increased in the northern part (north of 40°N), but decreased in the southern (south of 40°N). The ET₀ of the four natural stations in the northern Yanshan mountainous area showed an increasing trend. The Z value obtained from the MK trend test for the natural stations ranged from 2.58 to 3.373, located above the red dashed line in Figure 4, indicating a significant increase in ET₀. The ET₀ of the six farmland stations in the eastern and southern parts showed a decreasing trend, which was not significant except for that related to Raoyang station located on the central plain (Figures 3b and 4). However, ET₀ increased at Zhangbei and Yuxian stations in the western mountainous area. Furthermore, ET₀ showed a significant decreasing trend at stations that had an urbanisation percentage >80%, while ET₀ showed an increasing trend at Zhangjiakou, Zunhua and Tianjin stations. Figure 4 shows that, among the stations for which a decreasing ET₀ trend was noted, the trend at urban stations was more significant than that at farmland stations. The Z value obtained from the MK trend test for the urban stations ranged from -4.663 to -1.439, while that for farmland stations ranged from -2.927 and -0.248.

Similar findings have been reported for ET₀ trends in northern China (Han and Hu, 2012); an obvious decreasing trend in ET₀ was noted as a result of extensive irrigation in the arid/semi-arid areas. Moreover, Xu et al. (2015) observed a decreasing trend in ET₀ at agricultural and urban stations in the Jinghe River Basin. However, in the case of natural stations, Xu et al. (2015) noted a decreasing ET₀ trend, which is in contrast with our findings.

Figure 5 illustrates the spatial distribution of the changing trend of meteorological parameters in the study area. Wind speed was noted to be significantly increased at all the natural stations, leading to significant ET₀ increase at these stations (Figure 5a). On the other hand, wind speed at most farmland and urban stations decreased significantly, thus causing ET₀ decrease at these stations. With the exception of Zunhua and Hanghua stations, sunshine duration was particularly decreased at 12 of the 20 farmland and urban stations, thus causing a significant ET₀ decreasing trend at these stations (Figure 5b). With respect to relative humidity and air temperature, no obvious consistent changes were noted at the stations. In terms of relative humidity at all the stations, we noted that two natural, three farmland and three urban stations showed increasing trend, causing an increasing ET₀ trend, while two natural, five farmland and seven urban stations showed a downward trend, causing a decreasing ET₀ trend (Figure 5c). The trend of air temperature was roughly opposite to the relative humidity. The air temperature decreased at the stations with upward trend in relative humidity. Therefore, in terms of air temperature, one natural, five farmland and five urban stations showed an increasing trend, whereas three natural, three farmland and five urban stations showed a decreasing trend (Figure 5d).

In addition to human activity, meteorological parameters play a decisive role in spatial variations in ET₀ (Zhao et al., 2014). Li et al. (2017) observed climate change to have a more significant effect on ET than land use change. Wind speed is generally expected to have low values in urban areas than in rural areas because wind is blocked by high buildings in urban areas. Jiang et al. (2009) observed that a decline in average wind speed is closely related to not only weakened winter and summer monsoons but also increasing urbanisation in China. Han et al. (2016) demonstrated that surface wind speed is lowered by farmland and that wind speed values were significantly low at farmland stations. With respect to sunshine duration, Zhang et al. (2004) noted that low values can be attributed to increasing air pollution or high concentration of aerosols in the atmosphere. Furthermore, Liu et al. (2010) found that the aerosol index increased significantly after 1989 in the Haihe River Basin and that this index was higher in areas with higher population density than in those with lower population density. They concluded that an increase in aerosol concentrations in the atmosphere caused by human activities results in low values for sunshine duration and solar radiation. Zhang et al. (2006) also indicated that increasing air pollution may result in reduced sunshine duration in eastern China. Additionally, changes in air temperature can be attributed to factors such as reduction in sunshine duration or irrigation, depending on the land use type. Han and Yang (2013) studied the cooling effect of irrigation in Xinjiang, China, and noted this effect to influence changes in air temperature in the region. In the context of this study, low air temperatures for farmland stations may be attributed to irrigation. Thus, different types of topography and land use largely affect the distribution of wind speed, temperature and other meteorological parameters in a region (Jiang et al., 2009; Zhu et al., 2012). Based on this relationship between land use types and meteorological parameters, we proceeded to analyse the effect of these parameters on variations in ET₀, using the multiple regression equation mentioned in section 2.5.

In general, the dominant meteorological parameters affecting ET₀ at most stations can be ranked as wind speed > sunshine duration > air temperature > relative humidity (Figure 6). However, at each station, the contribution of each parameter to ET₀ was different. Wind speed was the most dominant parameter affecting ET₀ at the four natural stations, with its contribution ranging from 44% to 55% (Figure 6). The second dominant meteorological parameter affecting ET₀ at Weichang and Fengning stations was air temperature, while that at Chengde and Qinglong stations was sunshine duration.

As for the farmland stations, the main parameters controlling ET₀ changes were different. The dominant parameters at Huanghua, Yuxian and Huailai stations were wind speed and air temperature. For Langfang and Laoting, the contribution of sunshine duration and air temperature to ET₀ changes was >65%. Furthermore, sunshine duration and wind speed were the dominant meteorological parameters affecting ET₀ at Raoyang and Nangong stations, and relative humidity and air temperature were the dominant parameters affecting ET₀ at Zhangbei station.

With respect to the urban stations, the dominant parameters affecting ET₀ at most stations were wind speed and sunshine duration. At Tianjin, Zunhua and Zhangjiakou stations, high wind speed values played an important role in the increasing ET₀ trend. The dominant meteorological parameters affecting ET₀ at Beijing and Tangshan stations were sunshine duration and air temperature, while those at Qinhuangdao station were air temperature and wind speed.

Our findings show that different types of land use significantly impact ET₀ in a region; however, quantitatively assessing this impact is not easy. The influence of human activities, such as air pollution and land use change, on ET₀ is ultimately reflected through their influence on meteorological parameters (Zhao et al., 2014). A large number of studies have assessed the attribution of ET₀ to meteorological parameters; for example, wind speed was found to be the main parameter affecting ET₀ in the Haihe River Basin (Wang et al., 2011), the Huanghe River watershed (Ma et al., 2012), southwestern China (Li et al., 2014) and the Jinghe River Basin (Xu et al., 2015). However, few studies have focused on the relationship between different land use types and ET₀. Because of this limitation, it was difficult to compare and assess the accuracy of our findings. Therefore, future studies should focus on verifying these findings on a larger spatial scale.