1 Introduction

The global land surface is temporally and spatially heterogeneous on a range of scales. Many studies have separated the land surface into climatic or eco-geographical regions to allow for specific investigations of regional differences and their significance for agriculture and ecosystems (Bailey 2009; Fu et al. 2001; Holdridge 1967; Huang 1989; Köppen 1931; Lin 1954; Ren and Yang 1961; Thornthwaite 1948; Zheng 1999; Zhu 1930). In recent decades, greenhouse gas and aerosol emissions have had pronounced effects on the global climate, causing an increase in the mean global surface temperature of approximately 0.85 °C (0.65–1.06 °C) between 1880 and 2012 (IPCC 2013). As the global mean temperature continues to increase, land areas are experiencing different climates at an increasingly rapid pace (Mahlstein et al. 2013). While the dynamics of moisture conditions in arid/humid climate regions (AHCR) are known to be particularly complicated, they make an important contribution to the human understanding of the impacts of climate change on, for example, desertification (Huang et al. 2016a; Reynolds et al. 2007), water resources (Crosbie et al. 2012; Scanlon et al. 2006) and crop yields (Mihailovic et al. 2015).

A range of approaches has been previously used to generate valuable information about the dynamics of climatic zones. The impacts of climate change on the spatial distribution of climatic zones on both global (Fraedrich et al. 2001; Huang et al. 2016b; Rohli et al. 2015; Rubel and Kottek 2010) and continental or regional scales (Chan et al. 2016; Engelbrecht and Engelbrecht 2016; Feng et al. 2012; Gerstengarbe and Werner 2009; Wu et al. 2010) are increasingly obvious. For example, Huang et al. (2016b) found that, based on the aridity index, semi-arid regions expanded globally by 7% from 1948–1962 to 1990–2004. Engelbrecht and Engelbrecht (2016) highlighted widespread shifts in climatic regimes using the Köppen–Geiger classification, and reported that there would be a significant expansion of hot desert and steppe zones in southern Africa under future climate change.

The fact that China occupies a vast area and encompasses a wide range of climates and land surfaces, including the unique Asian monsoon climate system, means that changes in factors such as precipitation and dryness have been spatially and temporally heterogeneous. For example, studies have suggested that the climate in the semi-arid northeastern part of China has become warmer and wetter over the past 50 years, whereas it has become warmer and drier in the eastern and central regions (Yang et al. 2005). The AHCR is generally determined from the balance of the atmospheric water supply and the evaporative demand. Reference evapotranspiration (ETo), which represents the atmospheric water demand, is critical for highlighting changes in precipitation (Dai et al. 1997), dryness (Cook et al. 2014; Dai 2011; Greve and Seneviratne 2015) and dryland dynamics (Huang et al. 2016a). As in many other regions, the ETo has decreased significantly across extensive areas of China over the past half-century (Dolman and de Jeu 2010; Roderick and Farquhar 2002; Yin et al. 2010; Zhang et al. 2011). With spatial and temporal variations in the interactions between regional climate factors, increased knowledge of both changes in the AHCR and the causes of these changes could provide an improved understanding of regional responses to climate change. However, the combined effects of the dynamics of both the regional water supply and the evaporative demand on shifts in the AHCR over the recent past remain poorly understood, especially in the context of variations in the extent and degree of dryness, which reflect the interplay between variations in precipitation (P) and ETo.

The major objectives of this study were therefore (1) to quantify spatial and temporal variability in the AHCR by combining the Penman–Monteith ETo model with precipitation and (2) to investigate the influence of regional water supplies and the evaporative demand on shifts in the AHCR for the period from 1961 to 2010. The results have important implications for sustainable planning of agricultural development and eco-construction, and for protecting water resources and ecosystems.

2 Materials and methods

2.1 Materials

Quality-controlled monthly meteorological datasets of maximum and minimum air temperatures, precipitation, mean relative humidity, sunshine duration and mean wind speed from 581 meteorological stations (Fig. 1) for the period from 1961 to 2010 were used for this study. The datasets were provided by the Climatic Data Center (CDC) of the National Meteorological Information Center (NMIC) of the Chinese Meteorological Administration (CMA) and were downloaded from the CMA website (http://cdc.cma.gov.cn/index.jsp). Observation stations were deleted from the dataset if (1) the station was opened after 1961, (2) the location of the station changed during the study period, (3) the station was removed before 2010 or (4) more than 5% of the data from the station were missing. For stations with less than 5% of their data missing, the observed values at the same station in other years (before and after the missing year) were averaged to fill gaps in the data. The gap-filling method is based on the assumption that data are missing completely at random. To meet the model input requirements, ground-based point meteorological data were interpolated on a 10 × 10 km grid using a thin plate spline method.

Fig. 1
figure 1

Distribution of the 581 meteorological stations and the six geographical regions of China, namely Northeast China (NE), Inner Mongolia (IM), Northwest China (NW), Northern China (NC), Southern China (SC) and Tibetan Plateau (TP)

Full size image

For the purposes of this study and to provide a better illustration of regional differences, China is divided into six regions, which are as follows: Northeast China (NE), Inner Mongolia (IM), Northwest China (NW), Northern China (NC), Southern China (SC) and the Tibetan Plateau (TP) (Fig. 1).

2.2 Reference evapotranspiration

In general, ETo reflects the maximum water demand needed to maintain the water balance in the environment, while P reflects the water supply over large areas. Since it is difficult to obtain observed data for ETo over large areas, ETo is generally simulated through the use of models. In this study, ETo was simulated with the modified Penman–Monteith model recommended by the Food and Agricultural Organization, hereafter referred to as the FAO56–PM model (Allen et al. 1998), which can provide precise simulations of both arid and humid environments. Radiation in the model is calculated by an Ångström formula and its accuracy is determined by empirical coefficients that have regional limitations. In our previous studies, solar radiation for China was calibrated in the FAO56–PM model. As the model was able to effectively represent the AHCR in China (Yin et al. 2008), it was used to calculate the monthly ETo in this study:

$$E{T_o}=\frac{{0.408\Delta ({R_n} - G)+\gamma \frac{{900}}{{T+273}}{U_2}({e_s} - {e_a})}}{{\Delta +\gamma (1+0.34{U_2})}}$$
(1)

where Δ is the slope of the saturated vapor pressure curve (kPa °C−1), Rn is the net radiation (MJ m−2 day−1), G is the soil heat flux density (MJ m−2 day−1) considered as null for daily estimates, T is the monthly mean air temperature (°C) at 2 m calculated as the average of the maximum and minimum temperatures, U2 is the average wind speed at a height of 2 m (m s−1), es is the saturation vapor pressure (kPa), ea is the actual vapor pressure (kPa), (es–ea) is the saturation vapor pressure deficit (VPD, kPa) at temperature T and γ is the psychrometric constant (0.0677 kPa °C−1).

Rn in the model is calculated by an empirical formula and its accuracy is determined by empirical coefficients that have regional limitations. This study used the following equations to estimate Rn (Yin et al. 2008):

$${R_n}={R_{ns}} - {R_{nl}}$$
(2)
$${R_{ns}}=(1 - 0.23)\left( {0.198+0.787\frac{n}{N}} \right){R_{so}}$$
(3)
$${R_{nl}}=\sigma \left[ {\frac{{T_{{\hbox{max} ,k}}^{4}+T_{{\hbox{min} ,k}}^{4}}}{2}} \right](0.56 - 0.25\sqrt {{e_a}} )\left( {0.1+0.9\frac{n}{N}} \right)$$
(4)

where Rns is the net shortwave radiation (MJ m−2 day−1), Rnl is the net longwave radiation (MJ m−2 day−1), σ is the Stefan–Boltzmann constant (4.903 × 10−9 MJ K−4 m−2 day−1), Tmax,k is the maximum temperature (K), Tmin,k is the minimum temperature (K), Rso is the extraterrestrial radiation (MJ m−2 day−1), n is the actual sunshine hours and N is the potential sunshine hours. N and Rso are calculated by the equations recommended by Allen et al. (1998). Compared with observations from 30 validation stations across China, the calibrated solar radiation was proven to have a relatively high degree of accuracy (correlation coefficient = 0.952, root mean-square error = 1.729 MJ m−2 day−1, mean bias error = 0.030 MJ m−2 day−1) (Yin et al. 2008).

2.3 Classification of the AHCR

Following Budyko (1974), the aridity index (AI) combines first-order climatic drivers and is calculated as the ratio of annual ETo and P. The index is an indicator of regional humid or arid conditions and is, therefore, a suitable criterion for classifying the AHCR. According to Wu et al. (2005), the AI is conventionally used to classify large-scale land surfaces into four zones. These are humid, sub-humid, semi-arid and arid, and are represented by specific natural potential vegetation types, namely forest, forest steppe (including meadow), steppe and desert, as shown in Table 1.

Table 1 Criteria for demarcating AHCR of China according to the aridity index (Wu et al. 2005)
Full size table

2.4 Impact assessment of ET o and P

Detrending of time series is one of the quantitative ways of effectively assessing the influence of climatic variables on ETo and AI (Huo et al. 2013; Li et al. 2017; Liu et al. 2010; Xu et al. 2006). This study has removed the linear trends from ETo and P so that their contributions to the AI can be quantitatively evaluated, which helps to further understand the impact of atmospheric water balance on shifts in the AHCR. The AI was then recalculated and compared with the original AI. The ratio of the difference between the original and recalculated AI was deemed as the change in the AI caused by the trend in ETo or P.

As in Eq. (5), based on the least-squares fit of a straight line to the original series x(t) (t = 1961, 1962, 1963,…, 2010), this study computed and subtracted the predicted series y(t), added the first prediction value y(1961) and generated the detrended series xdetrend(t). When the trend was removed, the time series was then stationary. As the starting points were the same, this study was also able to compare the detrended and the original data. As in Eqs. (6) and (7), AI was recalculated using the detrended ETo and original P, or the original ETo and detrended P, and then compared with its original series. Finally, the percentage ratio of the difference in the AI was calculated.

$${x_{detrend}}\left( t \right)=x\left( t \right) - {\text{y}}\left( t \right)+{\text{y}}\left( {1961} \right)$$
(5)
$$Rati{o_{ETo}}\left( t \right)=\frac{{AI\left( t \right) - \frac{{ET{o_{Detrend}}\left( t \right)}}{{P\left( t \right)}}}}{{AI\left( t \right)}} \times 100\%$$
(6)
$$Rati{o_P}\left( t \right)=\frac{{AI\left( t \right) - \frac{{ETo\left( t \right)}}{{{P_{Detrend}}\left( t \right)}}}}{{AI\left( t \right)}} \times 100\%$$
(7)

3 Results

3.1 Changes in the aridity index

There was considerable spatial variation in the linear trends of the annual AI over the past 50 years from 1961 to 2010, as illustrated in Fig. 2a. The annual values of the AI showed linear decreasing trends at approximately 62.82% of the stations, mainly in NW, TP, the southern part of NE, and the region south of the Huai River. The trend was tested in the AI time series over the last five decades across China with the Mann–Kendal method and it was found that there were statistically significant decreasing and increasing trends at 11.02 and 3.61% of stations, respectively. The significant negative trends were mostly in NW and in the north of the TP, while the significant positive trends were mainly in the west of NC. The spatial distribution of the coefficients of variation (CV) in the annual AI across China is displayed in Fig. 2b. More than 60% of the CV values, mainly at stations in NE, SC, and across most of the TP, were between 0.1 and 0.3. Almost 30% of the stations, mainly in the east of NC, most parts of IM, and in the north of NW, had CV values between 0.3 and 0.5. Approximately 5.34% of the stations, mainly in the south of NW, had CV values greater than 0.5. This analysis shows that there was relatively large inter-annual variability in the annual AI in NC, IM, and, in particular, in NW.

Fig. 2
figure 2

Spatial distribution of a linear trends and b CVs in the annual AI across China from 1961 to 2010. Circles with an outline in a denote statistical significance (p < 0.05)

Full size image

The frequency distribution of trends in climatic factors across different regions in China is shown in Fig. 3. The AI decreased at most stations in NW and SC because of decreases in ETo and increases in P. While ETo increased at most stations in TP, P increased at a larger proportion of stations, which resulted in a decrease in AI. The AI most likely decreased across IM because the decreasing trend in ETo was stronger than the decreasing trend in P. The increase in the AI in NC reflects the fact that, while P and ETo both decreased, P decreased at a greater number of stations than ETo.

Fig. 3
figure 3

Frequency distribution of changes in the AI, ETo, and P across different geographical regions in China from 1961 to 2010. The inset panels show the boxplots of anomalies from 1961 to 2010. The box shows the 25th–75th percentile range, while the median value is shown as a solid line within each box, and outliers are represented by whiskers

Full size image

3.2 Spatial and temporal changes in AHCR

The area of the semi-arid region in China increased significantly at an average annual rate of 0.116% (p < 0.001) over the period from 1961 to 2010 (Fig. 4f). While the area of both the humid and sub-humid regions increased slightly, the area of the arid region declined significantly at a rate of − 0.129% per year (p < 0.001). The percentages of the area of AHCR in China over recent decades were obtained by calculating the decadal averages of the annual AI (Table 2). The result of the statistical t test showed that the area of semi-arid region in the 1990s and the 2000s was significantly different (p < 0.05) to the area in the 1960s, as was the area of the arid region. By the 2000s, the semi-arid region had expanded by about 33.53%, and the arid region had decreased by 20.75%, relative to the areas they had covered in the 1960s, which suggests that the semi-arid region in China have expanded as the arid region have contracted over the past five decades.

Fig. 4
figure 4

Spatial distribution of inter-decadal changes in the AHCR across China from a 1961 to 1970, b 1971 to 1980, c 1981 to 1990, d 1991 to 2000, e 2001 to 2010, and f the area percentage (%) from 1961 to 2010. Solid lines with no symbol give the 11-year running means of the data. Dashed lines are the linear fits

Full size image
Table 2 Inter-decadal percentage (%) changes in the area of AHCR in China from the 1960s to 2000s and changes (%) relative to the 1960s
Full size table

The spatial distribution of inter-decadal changes in the AHCR in China over the last 50 years is shown in Fig. 4a–e. The percentages of the area of AHCR in certain geographical regions are as follows. It can be seen that the semi-arid region increased most in TP, NW and NC between the 1960s and the 2000s. Meanwhile, evident decreases were found in the arid region of TP and NW, the sub-humid region of NC and the humid region of NE.

In NW, the area of the semi-arid region grew continuously from 10.96% of NW in the 1960s to 21.80% of NW in the 2000s, and the area of the arid region shrank from 88.79 to 76.98% over the same period (Fig. 4a–e). The shifts in the AHCR were mainly confined to the northern section, as shown by the statistically significant negative trends in the annual AI. Correspondingly, the decadal anomalies in the AI and in ETo and P, the factors that controlled it, meant that the semi-arid region in NW grew continuously. Specifically, the AI was negative from the 1980s, mainly because of increases in precipitation and decreases in ETo (Fig. 3c). However, while there were significant decreasing trends in the annual AI, the original characteristics of most of the NW area did not fundamentally change as it was still mainly characterized by arid conditions with vulnerable ecotopes.

In general, there were obvious changes in the spatial patterns of the AHCR in NE on the inter-decadal time scale (Fig. 4a–e). The humid region occupied 34.77% of NE in the 1960s, but, by the 1970s, it had shrunk considerably to about 19.19%, while, at the same time, the sub-humid region had expanded from 44.46 to 58.98%. In the 1980s and 1990s, the humid region expanded even further to 46.56 and 41.90%, respectively, and the spatial patterns in the AHCR were quite similar. In the 2000s, the humid region shrank again to 20.69% of NE and covered an area that was similar to what it had occupied in the 1970s. The semi-arid region changed only slightly before the 1990s but it grew strongly thereafter, from 18.33% in the 1990s to 25.07% in the 2000s.

In NC, the sub-humid region has decreased noticeably while the semi-arid region has increased over the past five decades. The humid area also grew and replaced the sub-humid area in the 1980s. The area of the humid region decreased from 20.82% of NC in the 1980s to 8.82% of NC in the 1990s, while the sub-humid and semi-arid regions increased by 3.16% and 8.88%, respectively. It is worth mentioning that a separate patch that was previously sub-humid became semi-arid. By the 2000s, the area of the sub-humid region had decreased to 34.98% of NC, its smallest extent.

In TP, the areas of the humid and sub-humid regions changed only slightly from the 1960s to the 2000s, whereas the area of the semi-arid region grew substantially with considerable corresponding reductions in the arid region. Of the different climatic regions, the arid region accounted for the highest percentage of the total area of TP over the first three decades. However, it shrank considerably from almost 41% to about 34% of TP in the 1990s, which was similar to the area occupied by the semi-arid region. Before the 2000s, the arid region was mainly confined to the northern section and part of the western section, but its area decreased by 9.77% from the 1990s to the 2000s, with most loss occurring in the west. However, the semi-arid region increased by 6.72% and extended towards the west of TP. Comparison of the 1960s with the 2000s shows that the areas of the semi-arid and arid regions changed by 13.24% and 16.45%, respectively.

The spatial distribution of shifts between different AHCRs in China for 1971–1980, 1981–1990, 1991–2000 and 2001–2010 relative to 1961–1970 is shown in Fig. 5. A shift was deemed to have occurred in a grid cell if its AHCR type in the 1970s, 1980s, 1990s, or 2000s was inconsistent with its AHCR type in the 1960s. Generally, 24.14% of the area in China underwent at least one shift, mostly in the northern parts of the country. Based on this study’s classification of geographical regions, the NE experienced most changes, with shifts across 47.96% of the region, followed by the NC, where 47.80% of the area changed. There were shifts across 31.8% of TP, mainly in the central and western parts, while there were shifts across 14.66% of NW, mainly the northern part. Relatively small areas of IM and SC, accounting for 8.88 and 6.26%, respectively, experienced shifts. Shifts that first occurred in the 1970s, 1980s, 1990s, and 2000s relative to the 1960s accounted for 8.33, 5.27, 4.84, and 5.69% of the total area of China, respectively. However, when the spatial patterns in the AHCR across China over the past five decades are examined (Fig. 4a–e), it can be seen that there were complicated changes in the humid and sub-humid regions, with boundaries shifting back and forth, while obviously the semi-arid region saw the most expansion.

Fig. 5
figure 5

Spatial distribution of the first shifts in the AHCR across China from the 1970s–2000s relative to the 1960s

Full size image

3.3 Enlargement of the semi-arid region

In order to confirm and explicate the enlargement of the semi-arid region in China, the abrupt annual change in the area of AHCR from 1961 to 2010 was tested by using the Mann–Kendall method, which detected a change point in the semi-arid region around 1985/1986 (p < 0.05). Moreover, the statistical t test also showed a significant difference in the annual mean area of the semi-arid region between both before and after 1985/1986 (Table 3). For 1986–2010, the semi-arid region was calculated to have increased by 25.13% relative to the 20.81% of China that it had covered in 1961–1985. Meanwhile, the area of the sub-humid region and arid region decreased by 10.02 and 16.05%, respectively.

Table 3 Percentage (%) changes in the area of AHCR in China from 1961–1985 to 1986–2010 and changes (%) relative to 1961–1985
Full size table

Based on this time division, the shifts between different AHCRs in China were analysed over the two 25-year periods. Figure 6 shows all the shifts in the semi-arid regions, namely the shifts between semi-arid and other AHCR types. On the one hand, the area of newly formed semi-arid regions that had been arid regions and sub-humid regions occupied 4.82 and 1.34% of China, respectively. On the other hand, the semi-arid region that had transitioned to arid and sub-humid regions accounted for 0.05 and 0.88% of the area of China, respectively. Thus, the net area change in the semi-arid region was 5.23% of China. The expansion of the semi-arid region mainly happened in TP, NW and NC, where it accounted for 11.14, 10.70 and 9.71% of their respective areas. In NC, growth of the semi-arid region was mainly south of the Loess Plateau to the sub-humid region. The semi-arid region in the western TP grew northwards towards the arid region, while the semi-arid region in the northern NW expanded along the basin to the arid region. Generally, the semi-arid region expanded due to shifts from arid regions in the western parts of China (i.e., IM, NW, and TP) and from sub-humid regions in the eastern parts of China (i.e., NE and NC).

Fig. 6
figure 6

Spatial distribution of changed and unchanged semi-arid regions in China for 1986–2010 relative to 1961–1985. The stippled area denotes statistical significance (p < 0.05) with t test in the annual AI between the two periods

Full size image

In addition to this, the statistical t test was also performed at each grid to examine whether the means of the annual AI between the two 25-year periods were different at the 5% significance level (Fig. 6). Results showed that AI had significantly changed in about 76.3% of the area of expanded semi-arid regions. This might imply that the enlargement of the semi-arid region in China from 1961–1985 to 1986–2010 was most significant.

3.4 The role of ET o and P in the expansion of the semi-arid region

To further quantify how climatic factors contributed to the enlargement of the semi-arid region, the contributions of ETo and P to AI were measured after removing their trends (Fig. 7). ETo mainly decreased over the expanded semi-arid region in NW over the past 50 years, with most grids having values between − 4 and − 1 mm/year. P, in turn, showed an increasing trend and ranged from 1 to 2 mm/year. The contrasting trends in ETo and P had consistent effects on the direction of the trend in AI. There was a slight difference between the original and the recalculated AI with detrended ETo (P) early in the study period, with a change ratio of − 1.25% (− 3.50%) in the 1960s. As time went by, it became greater and reached − 13.39% (− 38.05%) in the 2000s. On average, AI decreased by ratios of − 7.29 and − 20.72%, because of the decrease in ETo (− 2.11 mm/year, p < 0.05) and the increase in P (1.27 mm/year, p < 0.05), respectively (Fig. 7c).

Fig. 7
figure 7

Linear trends in the AI, ETo, and P (left column) and comparison between the original AI and the AI recalculated with the detrended ETo or P (right column) in the expanded area of the semi-arid region across different geographical regions in China from 1961 to 2010. The blue numbers in the right column indicate the change ratios of AI with the detrended ETo, and the green numbers indicate the change ratios of AI with the detrended P

Full size image

The expanded area of the semi-arid region in TP and NW experienced similar changes. The trends in ETo and P were mainly between − 2 and 1, and 1 and 5 mm/year, respectively. The difference of the original AI from the recalculated AI gradually amplified during the past five decades. From the 1960s to the 2000s, the change ratio of AI with detrended ETo ranged from − 0.59 to − 6.06%, while that with detrended P ranged from − 10.64 to − 140.91%. Due to the decrease in ETo (− 1.02 mm/year) and the increase in P (3.02 mm/year, p < 0.05), the AI decreased by average ratios of − 3.24 and − 71.61%, respectively (Fig. 7e). This shows that the transition from arid to semi-arid in NW and TP reflected the combined influence of the decreasing trend in ETo and the increasing trend in P. In contrast, the tendency for P to increase was an important factor in the expansion of the semi-arid region.

ETo and P showed declining trends in most grids in the expanded area of the semi-arid region in IM and NC. Trends in ETo ranged from − 3 to 0 mm/year and from − 2 to 1 mm/year in IM and NC, while those in P ranged from − 0.5 to 0 mm/year and from − 3 to 0 mm/year, respectively. Correspondingly, the AI decreased in most grids in the expanded area of the semi-arid region in IM and increased in most grids in the corresponding area in NC. Generally, the significant decrease in ETo (− 1.88 mm/year, p < 0.05) caused the AI to decline by an average of − 5.69% in the expanded area of the semi-arid region in IM (Fig. 7b), and the significant decrease in P (− 2.13 mm/year, p < 0.05) resulted in an average increase of 9.19% in the AI for the expanded area of the semi-arid region in NC (Fig. 7d). In the expanded area of the semi-arid land in NE, the combined influence of the increase in ETo and the decrease in P, while not significant, caused the AI to increase slightly (Fig. 7a).

4 Discussion

This study has assessed changes in the AHCR from the balance between the water demand and supply and found that the semi-arid region expanded over the period from 1961 to 2010. The semi-arid zone, which is mainly at the boundary of the monsoon belt in China, is susceptible to environmental change (Yang et al. 2005). The enlargement of the semi-arid region illustrated in this study is consistent with the results of earlier studies (Li and Ma 2013; Zhang and Yan 2014). In their analysis of combined soil moisture levels in the twentieth century, Li and Ma (2013) found that the semi-arid and semi-humid climate zones were expanding southeastwards, especially in northern China. From cluster analysis using K-means of monthly temperature and precipitation for the period from 1966 to 2005, Zhang and Yan (2014) found that the boundary between an arid temperate zone and a sub-humid temperate zone in China was moving eastward. The results of this study generally agree with those of Huang et al. (2016b), who, from their analysis of the ratio of annual precipitation to annual reference transpiration, recently reported global expansion of semi-arid regions over the period from 1948 to 2008. This study also found that the semi-arid region had enlarged mainly because of transfers from arid regions in western China, including NW, TP, and IM, and from sub-humid regions in eastern China, including NC and NE. Due to limitations with regard to the sparse distribution of meteorological stations in TP, there is relatively high uncertainty associated with the AHCR changes in TP.

There has been a relatively good understanding of changes in the individual climatic factors that are related to moisture conditions. Many studies have reported significant decreasing trends in ETo over a range of scales, from regional to continental, over the last half-century (Dolman and de Jeu 2010; Hobbins et al. 2004; Kukal and Irmak 2016; Peterson et al. 1995; Roderick and Farquhar 2002; Schimel et al. 1997; Yin et al. 2010). From their synthesis of global studies and discussion of trends in measured pan evaporation estimated from reference evapotranspiration, McVicar et al. (2012) reported widespread decreases in the evaporative demand, while Wu et al. (2016) identified distinctive regional patterns in annual total precipitation, with increases in western China and decreases in eastern China over the period from 1960 to 2012. More specifically, almost all the intensity binned precipitation amounts and frequencies increased significantly in northwest China, and the precipitation amount and frequency mostly decreased in northern and southwestern China over the period from 1960 to 2013 (Ma et al. 2015). However, synthetic effects and their relative contributions to changes in aridity and the corresponding AHCR remain less well understood, with considerable spatial heterogeneity in both uncertainty and temporal fluctuations over recent decades.

There has been considerable spatial and temporal variability in trends in aridity over the last few decades. Recently, observed trends in global dryness have been contradictory and very uncertain (Dai 2013; Feng and Fu 2013; Greve et al. 2014; Mueller and Zhang 2016; Sheffield et al. 2012; Trenberth et al. 2014). Regional increases in aridity have been reported in southwestern Spain for the period from 1951 to 2010 (Moral et al. 2016), and for Iran from 1966 to 2005 (Tabari and Aghajanloo 2013). In contrast, increases in wetness in the eastern US from 1895 to 1990 have been attributed to statistically significant increases in precipitation minus reference evaporation (McCabe and Wolock 2002). Consistent with this study’s results, wetness has increased significantly in the northwest of China (Huo et al. 2013), 91.7% of which has been attributed to the increases in precipitation observed from 1960 to 2010 (Liu et al. 2013). The decreasing trend in reference evapotranspiration, together with the significant increasing trend in precipitation, contributed to the change in aridity in NW and the expansion of the semi-arid region. As well as the combined effects of changes in atmospheric water supply and evaporative demand, there have been corrective effects in some other regions such as NC and IM. For example, the expansion of the semi-arid region in NC reflects how the decrease in the evaporative demand was offset by the decrease in precipitation, while the expansion of the semi-arid region in IM demonstrates that the decrease in evaporative demand outpaced the decreases in precipitation.

Semi-arid regions have unique ecosystems and water conditions (Austin and Vivanco 2006; Morgan et al. 2011; Rotenberg and Yakir 2010; Wei et al. 2007). Shifts in the extent of semi-arid regions have great implications for and may have significant impacts on, water resource use, natural vegetation patterns, land degradation and the global carbon cycle. Generally, an eastward shift in the semi-arid zone in Inner Mongolia would exacerbate the development of desertification. The AI has mainly increased in northern China. If the current trend towards decreased humidity continues, it may not be possible to meet the demand for agricultural water meaning that agricultural production may be curtailed in this major grain-producing region. It is thought that semi-arid ecosystems, whose carbon balance is strongly associated with circulation-driven variations in both temperature and precipitation, make a significant contribution to variability in carbon uptake on the global scale (Ahlström et al. 2015; Poulter et al. 2014; Zhang et al. 2016). Enlargement of semi-arid regions would mean an increased contribution to variability in global carbon uptake. Analysis of global trends in aridity shows that severe and widespread droughts are likely to occur in many areas over the next 30–90 years (Dai 2013). Therefore, further analyses of the future shift of AHCR and the relationship with aridity trends and underlying mechanisms would be required in order to mitigate the adverse response of ecosystems to climate change.

5 Conclusions

This study has quantitatively assessed the temporal and spatial changes in AHCR in China using data for atmospheric water demand and supply collected at 581 meteorological stations over the last five decades. It has identified the contributions of changes in the atmospheric water demand and supply to shifts in the region.

There have been inter-decadal variations in the spatial distribution of AHCR across China over the last 50 years. Of the four climatic types, there were significant trends in the arid and semi-arid regions. The semi-arid region increased significantly with a linear upward trend of 0.116%/year (p < 0.001) over the period from 1961 to 2010; in the 1960s, this region accounted for 20.64% of the area of China, but it had increased by 33.53% relative to the area covered in the 1960s by the 2000s. There was a linear decreasing trend of − 0.129%/year (p < 0.001) in the arid region, which shrank by 20.75% relative to the 29.68% of China it covered in the 1960s. The newly expanded semi-arid areas were mainly in TP, NW and NC, while the arid areas mainly decreased in TP and NW.

Changes in precipitation and evaporative demand have either collaborative or counteractive effects on the extension of the semi-arid region. The decreased reference evapotranspiration together with significant increases in precipitation contributed to the expanded semi-arid region in western China, whereas the significantly decreased precipitation accompanied by decreases in evaporative demand led to the expanded semi-arid region in eastern China. For the enlargement of the semi-arid region in NW, TP and NC, the trend in precipitation was the dominant factor.