The CLP is the largest region of loess deposits in the world. It is characterized by temperate semiarid and subhumid climate, modified by latitude, longitude, and terrain. Both the MAAT and the MAP show a clear decrease northwestward (Fig. 1). Dominated by the strength of the East Asian summer monsoon (EASM) system, the precipitation occurs mostly in the summer months (from May to September), which accounts for approximately 68–87 % of the total annual precipitation (Ding, 1994; Liu et al., 2005a). The present-day CLP is mainly covered by shrub (e.g. Sophora viciifolia and Vitex chinensis) and grasses (e.g. Artemisia and Gramineae) (Liu, 1985). In general, the vegetation progressively becomes sparser and less dense from southeast to northwest, resulting from increasing dryness (Liu et al., 2005c).

A total of 97 samples were collected in late September, 2012, from 33 sites in the CLP and the surrounding areas (Fig. 1, Supplement Table 1). For each site, 2–5 samples were collected at locations tens to hundreds of metres apart, except for site DengkouB with only one sample being collected. Samples from the CLP (WLPS-1–WLPS-18 and WLPS-79–WLPS-97, totaling 37 samples) were collected from natural grassland or grassland that has been in restoration for > 10 a. At each sampling location, three randomly collected samples were pooled and mixed to make one composite sample representing that location. Most samples were collected from the uppermost layer of soil with a depth of less than 5 cm and they were transported to the laboratory immediately after collection and stored at -20 ∘C. The details of the samples are provided in the Supplement Table 1.

The SWC (soil water content) of the soils was obtained by weighting the sample before and after freeze-drying it.

Soil pH was measured following Wang et al. (2012): ca. 4 g of freeze-dried sample was added in 10 mL of distilled water; the mixture was stirred for 1 min, left to stand for 30 min, and pH of the supernatant was measured with a Sartorius PB-10 pH meter. The standard deviation for triplicate measurements was ±0.03.

The meteorological data for our sampling sites were obtained from the China Meteorological Data Sharing Service System (http://cdc.cma.gov.cn/). For each sampling site, MAP, MAAT, mean annual ground surface temperature (MAGST), mean annual relative humidity (MARH) and mean annual evaporation (MAE) were generally estimated from the nearest weather station. However, if there were 2–3 stations nearby, the meteorological data from all of them were averaged for the sampling site. A further correction to the meteorological MAGST and MAAT values for each sample was made according to a lapse rate of -0.6 ∘C/100 m.

The freeze-dried and homogenized samples (30–50 g) were extracted (4 × 10 min) with dichloromethane (DCM) : methanol (9:1, v/v) using an accelerated solvent extractor (ASE 350, Dionex) at 100 ∘C and 1500 psi. The total extract was dried under N2 in a water bath. A known amount of C46 GDGT internal standard (IS; Huguet et al., 2006) was then added to the extract, which was redissolved in DCM : methanol (9:1, v/v) and divided into two halves. One half was dried under N2, redissolved in hexane/isopropanol (99:1, v/v) and filtered over a 0.45 µm PTFE (polytetrafluoroethylene) filter.

An aliquot (10 µL) of sample was injected for HPLC-atmospheric pressure chemical ionization (APCI)-MS analysis according to a procedure slightly modified from Schouten et al. (2007) and Zhang et al. (2012). Separation of bGDGTs was achieved on an Alltech Prevail Cyano column (150 mm × 2.1 mm, 3 µm). The elution gradient was isocratic (5 min) at 99 % hexane / 1 % isopropanol followed by a linear gradient to 1.8 % propanol in 45 min. After each analysis, the column was cleaned by back-flushing using 90 % hexane / 10 % propanol. The flow rate was set at 0.2 mL min-1. In order to increase sensitivity and reproducibility, selective ion monitoring (SIM) mode of the [M+H]+ (protonated molecular ion) of the different bGDGTs (Fig. 2) was used to detect and quantify them. Quantification of each bGDGT was achieved by calculating the peak area in the chromatogram and comparing it with that of the IS. Ionization efficiency for bGDGTs and the IS was assumed identical.

The CBT index was calculated following Weijers et al. (2007a): CBT=-log((Ib+IIb)/(Ia+IIa)).

The MBT' index was calculated according to the following equation (Peterse et al., 2012): MBT′=(Ia+Ib+Ic)/(Ia+Ib+Ic+IIa+IIb+IIc+IIIa).

The MBT' / CBT-derived MAAT and MBT' / pH-derived MAAT was calculated as follows (Peterse et al., 2012): MAAT=0.81-5.67×CBT+31.0×MBT',MAAT=-23.20+2.86×pH+33.71×MBT'.

MAAT was also estimated using the Chinese SSM calibration model based on the fractional abundance of bGDGTs (Yang et al., 2014a): MAAT=20.9-13.4×IIa-17.2×IIIa-17.5×IIb+11.2×Ib.

In addition to the soils newly collected in the CLP and the surrounding areas in this study, we included other alkaline (pH > 7) soil samples in China with published bGDGT data available to generate the largest possible data set for determining the environmental controls on the distribution of bGDGTs in Chinese alkaline soils (Supplement Table 1). This combined data set consisted of 97 soils analysed in this study, 37 soils in Xie et al. (2012) and Yang et al. (2012), and 41 soils in Yang et al. (2014a). The MAP, pH and MAAT ranges for this data set are 7.2–9.2, 140–1138 mm, and 3.8–16.8 ∘C, respectively. Two soils (XJBC-3 and XJBC-5) of Yang et al. (2014a) collected at the northwestern margin of the Taklimakan Desert were not included in the data set because they were likely from the wet sites with surface/groundwater supply, and may not represent typical soils with a MAP of ca. 60 mm.

We performed a redundancy analysis (RDA) on the combined data set to visualize the environmental controls on the variation in fractional abundances of bGDGTs in Chinese alkaline soils employing the software Canoco for Windows 4.5 (ter Braak and Smilauer, 2002). According to previous studies (Peterse et al., 2012; Yang et al., 2014a), MAP, soil pH and MAAT are set as environment data while species data are the fractional abundances of individual bGDGTs. Detrended correspondence analysis (DCA) on the fractional abundances of bGDGTs showed that the gradient length was 1.4, much lower than 3 (ter Braak, 1988; ter Braak and Prentice, 1988; Yang et al., 2014a). Therefore, variability in the bGDGTs was linear and RDA was appropriate in this case.

The Pearson correlation coefficients between environmental variables and bGDGT distributions were determined with the Windows SPSS 16.0 software program. We accepted a p value of < 0.05 as being significant.

The results of the RDA show that the first two RDA axes cumulatively explained 51.9 % of the bGDGT distribution data and 99.9 % of the relationships between bGDGTs and environmental factors. The first axis (axis 1) alone explains 50.3 % of the variance of bGDGT distribution and captures the gradients in MAP and MAAT. However, the correlation coefficient of axis 1 and MAP is 0.81, higher than that for axis 1 and MAAT (0.74). Soil pH correlates more weakly (r = -0.24) with axis 1. In addition, based on the angles between the fractional abundances of bGDGTs and environmental factors, the relative abundances of bGDGTs with a high degree of cyclization (Ib, Ic, IIb, IIc, IIIb and IIIc) correlate positively with MAP (Fig. 3). For acyclic bGDGTs, Ia varied positively with MAAT and MAP, whereas IIa and IIIa varied negatively with MAAT and MAP.

The relationships between bGDGT distributions and environmental variables (i.e. pH, MAP and MAAT) can be further demonstrated by the correlation coefficients calculated by SPSS (Table 1). As expected, each individual bGDGT correlates significantly (p < 0.01) with MAP and MAAT, with the correlation coefficients much higher than those with soil pH (except for IIIb exhibiting the weakest correlation with MAAT). For the proportions of cyclic bGDGTs and acyclic bGDGT IIa, they are most closely related to MAP (r = 0.81, 0.70, 0.69, 0.56, 0.50, 0.48 and -0.78 for Ib, Ic, IIb, IIc, IIIb, IIIc and IIa, respectively). For the proportions of acyclic Ia and IIIa, they are most closely related to MAAT (r = 0.60 for Ia and -0.61 for IIIa). These are in agreement with the visualized RDA results (Fig. 3). Moreover, the MBT index representing the degree of methylation for bGDGTs correlates significantly (p < 0.01) with both MAAT (r = 0.76) and MAP (r = 0.76), while the CBT index expressing the relative extent of cyclopentane moieties in bGDGTs is most related to MAP (r = -0.74, p < 0.01) and to a lesser extent to MAAT (r = -0.59, p < 0.01).

For the Chinese alkaline soil data set, pH seems unaccountable for the variance in the distribution of bGDGTs (Fig. 3) and there is only a slightly positive relationship between CBT and pH (Table 1 and Fig. 4; r = 0.22, p=0.01, n = 165). This seems different from previous regional or global soil investigations at a large pH range, which have showed a strong positive control of pH on the relative abundance of bGDGTs containing cyclopentyl rings and thus CBT correlated negatively with pH (Weijers et al., 2007a; Peterse et al., 2009b, 2012; Menges et al., 2014). However, we note that the proportion of alkaline soils is relatively small in each soil sample set of the aforementioned studies. In fact, by extending the pH to > 9, Xie et al. (2012) and Yang et al. (2014a) have pointed out that the CBT index appears to be unable to distinguish pH variation, due to the weak positive CBT–pH relationship or a flattening off of CBT when the pH is > 7, despite that the CBT values still exhibit a significant (p < 0.01) negative correlation with soil pH when pH is < 7 (Fig. 4). The flattening off of CBT at higher soil pH values is also observed by Peterse et al. (2010), who analysed bGDGTs in long-term soil pH manipulation plots in Scotland. Moreover, in lacustrine systems, a seemingly similar pattern also exists in the surface sediments of 23 lakes in China and Nepal (Sun et al., 2011), as well as in suspended particulate matter (SPM) from 23 lakes in the USA (Schoon et al., 2013). The weak correlation between CBT and pH under alkaline conditions suggests that variation in CBT might be problematic for tracing pH variation in alkaline soils.

The insensitivity of CBT to soil pH variation in alkaline soils might be caused by the inaccurate identification and subsequent quantification of 5-methyl bGDGTs. It has been observed that higher pH may decrease the fractional abundances of 5-methyl IIa but increase the fractional abundances of 6-methyl IIa in globally distributed soils (De Jonge et al., 2014). Since the two isomers are regarded as “normal” IIa (which is positively correlated with pH) in the calculation of CBT (which is negatively correlated with pH) using the traditional HPLC method (Weijers et al., 2007a), the CBT values would be overestimated if 6-methyl IIa exists. The exclusion of 6-methyl bGDGTs in the CBT index (defined as CBT5ME), however, exhibits only slight improvement in relation to soil pH when pH is > 7 for the global soil data set (De Jonge et al., 2014). Therefore, De Jonge et al. (2014) further defined a new proxy for soil pH reconstruction, the so-called CBT' index, which comprises the pH-dependent 6-methyl bGDGTs and separates the opposite relations of 6-methyl IIa and 5-methyl IIa with pH. It correlates strongly with pH in globally distributed soils, with the RMSE (root mean square error) of the subset of arid soils decreasing from 2.9 to 0.3 pH units (De Jonge et al., 2014). The newly proposed CBT' index seems more promising for pH reconstruction in palaeoclimate studies, particularly in arid and alkaline soils.

Alternatively, the absence of a clear CBT–pH relationship under alkaline conditions possibly reflects that a certain threshold is reached for bGDGT producers in adapting their membrane lipids to pH. For the global soil data set, the correlations between the relative abundance of cyclic 5-methyl bGDGTs (e.g. Ib and IIb) and pH are indeed very weak under alkaline conditions (De Jonge et al., 2014). Previously, Weijers et al. (2007a) proposed two likely mechanisms for the negative CBT–pH relationship when pH is < 7. Firstly, it is important for the microbes to keep the internal pH constant within the cell, and therefore the bGDGT-producing bacteria tend to reduce the production of cyclopentyl moieties (causing higher CBT) at lower pH for better membrane packing and lower membrane permeability, which helps to protect cells against low pH (Weijers et al., 2007a). When pH is > 7, however, there seems to be no need for soil bacteria to overcome the inhibition of acidity. Secondly, the proton permeability (i.e. the pH gradient) of the cell membrane plays a crucial role in driving energy reactions over the cell membrane (e.g. Booth, 1985). Introduction of cyclopentyl moieties would allow more water molecules to get trapped in the membrane and consequently increase the membrane proton permeability, whereas a steeper proton gradient (thus lower ambient pH) is counteracted by a more impermeable membrane resulting from fewer cyclopentyl moieties and thus higher CBT (Weijers et al., 2007a). Under alkaline conditions, however, further adaptation of the cell membrane would not be needed to overcome H+ leakage (Schoon et al., 2013) if the bGDGT producers turn to use Na+ for energy transduction as neutrophiles and alkaliphiles, which are able to perform energy transduction by establishing a Na+ gradient in combination with an H+ gradient (Speelmans et al., 1995; Schoon et al., 2013). These mechanisms could possibly explain the apparent insensitivity of CBT to pH in alkaline conditions theoretically. Hence, variation in CBT values should be attributed as a response to other factor(s) for the alkaline soils.

Soil moisture can potentially affect the distribution of bGDGTs in modern soils (Loomis et al., 2011; Peterse et al., 2012; Dirghangi et al., 2013; Menges et al., 2014). SWC is a direct measurement of soil moisture. Notably for our soils collected in the CLP and its adjacent arid/semiarid areas, the CBT index correlates more strongly with SWC (r = -0.67, p < 0.01, n = 97; Fig. 5a) than with soil pH (r = 0.50, p < 0.01, n = 97; Fig. 5b), pointing to a likely negative control of soil moisture on the CBT index in alkaline soils in the water-limited regions. Some of the scatters in the SWC–CBT relationship might be due to that our measured SWC is an instantaneous value that is liable to differences in local rainfall for different sites and may not accurately represent the average soil moisture condition during the growth of bGDGT producers. For instance, soils at site Yongdeng (MAP: 284 mm) should be drier than those at the adjacent site Gulang (MAP: 352 mm). However, the measured SWC values for soils in Yongdeng (13, 13, and 12 %) collected immediately after a rainfall (according to our field note) were much higher than those for Gulang (3, 6, and 3 %). In such a case, the use of SWC at this sampling time could lead to an overestimation of the effect of soil moisture availability on bGDGT producers in Yongdeng (Fig. 5a). Omitting the Yongdeng samples, the CBT–MAP relationship was improved (r = -0.75). For extensive surface soil investigations, therefore, variation in SWC at the time of sampling can only roughly reflect differences in mean soil moisture conditions.

In contrast, MAP data would be more representative of the mean soil moisture than the instantaneous SWC value when continuous local soil moisture observation data were not available. Therefore, the MAP that is generally available in the data sets of previous studies is used to represent soil moisture in further discussions. As is expected, amongst the three environmental variables (MAAT, MAP, and soil pH) MAP is the dominant one that affects the relative abundance of cyclic bGDGTs with cyclopentyl moieties and thus the CBT index for the Chinese alkaline soils (Fig. 3, Table 1), further supporting soil moisture as the dominant environmental control on the CBT index in alkaline soils.

It might be argued that the negative MAP–CBT relationship (Fig. 6a) is possibly an artifact of CBT and MAAT being negatively correlated (Fig. 6b) and MAP and MAAT being intercorrelated (Fig. 6c). However, the correlation of MAP–CBT is stronger than that of MAAT–CBT (r=-0.74 for MAP–CBT and -0.59 for MAAT–CBT; Table 1), particularly when only considering the soils newly collected in this study (r=-0.72 for MAP–CBT and -0.28 for MAAT–CBT, n = 97; Fig. 6a, b). Moreover, we observed that the negative relationship between MAP and CBT is also significant (p < 0.01) in alkaline soils of two other regional data sets that show no MAAT–CBT correlation (in the USA: r = -0.80 for MAP–CBT and 0.26 for MAAT–CBT; Fig. 6d, e; De Jonge et al., 2014) or inverse MAAT–CBT correlation (across the Iberian Peninsula: r = -0.65 for MAP–CBT and 0.59 for MAAT–CBT; Fig. 6g, h; Menges et al., 2014). The results of these studies in different regions collectively suggest that the negative correlation between CBT and MAP might be valid for alkaline soils in water-limited regions, despite that the influence of temperature on CBT cannot be excluded at present.

The reason why CBT correlates with soil moisture in alkaline soils in arid–subhumid regions remains speculative as the exact biological source of bGDGTs is unknown. A likely explanation for the positive correlation between MAP and the relative abundance of cyclic bGDGTs (and thus the negative MAP–CBT relationship) might be that, under drier conditions, a more dense packing of membrane lipids is needed for bGDGT-producing organism(s) to avoid overevaporation of intracellular water. As a result, the formation of cyclopentyl moieties, which could result in loosening of the packing of the membrane lipids (Weijers et al., 2007a), is inhibited in soils at lower MAP values. Another possible reason for the MAP–CBT relationship in alkaline soils in water-limited regions might be a response of the community change of bGDGT producers with different soil moisture conditions; however, we currently know nothing about what organism(s) would produce more acyclic bGDGTs in drier soils under alkaline conditions. Further studies are needed to explore the exact underlying mechanism responsible for the observed moisture dependence of the CBT index in alkaline soils from arid–subhumid regions.

It should be noted that we only suggest the relation between CBT and soil moisture valid in alkaline soils from arid–subhumid regions. There is no evidence for a strong relation between soil moisture and CBT over the whole pH range of the global calibration set (De Jonge et al., 2014) or other regional data sets (e.g. Menges et al., 2014; Yang et al., 2014a). In fact, both pH (Weijers et al., 2007) and soil moisture (as discussed previously) might be important in controlling the membrane lipid composition of soil bGDGT producers. In humid regions where water availability is not a limiting factor, changes in soil pH might dominate the variations in CBT values (Weijers et al., 2007; Peterse et al., 2012). In contrast, water availability is generally a limiting factor for the growth of microorganisms in soils from arid–subhumid regions (e.g. Wang et al., 2013; Menges et al., 2014). Since it seems that CBT is insensitive to soil pH in alkaline soils (Xie et al., 2012; Yang et al., 2014a and Sect. 3.2), variations in CBT might be predominantly controlled by changes in soil moisture. In acid soils in arid–subhumid regions, however, both pH and soil moisture can influence the CBT values. Therefore, we tentatively restrict the CBT–soil moisture relationship to alkaline soils from arid–subhumid regions.

The significant (p < 0.01) correlation between CBT and MAP in Chinese alkaline soils implies that the CBT index may be a useful tool for palaeorainfall reconstruction in LPS. When only CLP soils collected in this study were considered, the positive CBT–MAP correlation exhibits an obvious improvement (Fig. 7a): CBT=-0.0021×MAP+1.7(r=-0.93,p<0.01,n=37).

We apply this initial calibration on three LPSs with published bGDGT data covering the past 70 ka, in order to validate the CBT index as a palaeorainfall proxy on the CLP. The residual errors of the estimated MAP are < 127 mm, with a RMSE of 50 mm and without following a clear trend with MAP (Fig. 7b). Analytically, the standard deviation for CBT is 0.01, equivalent to 6 mm of MAP, indicating that palaeorainfall estimates can be obtained with high analytical reproducibility. Located from west to east on the southern CLP, the three LPSs are the Yuanbao (Fig. 8a; Jia et al., 2013), Lantian (Fig. 8b; Gao et al., 2012), and Mangshan (Fig. 8c; Peterse et al., 2014) sections, respectively. The CBT-derived MAP records exhibit pronounced precession-driven variation in rainfall amount at the three sites. Within dating uncertainties, the fluctuations of each record resemble the speleothem δ18O record from the Hulu (Wang et al., 2001) and Sanbao (Wang et al., 2008) caves in southeastern China (Fig. 8d), which are widely regarded as a robust record for the EASM intensity, predominantly for monsoon precipitation (Peterse et al., 2014). All these records indicate that rainfall amount was highest during marine isotope stage (MIS) 1, relatively higher during MIS 3, and lowest during MIS 2 and probably MIS 4. Overall, the consistency of the CBT-inferred MAP records with the independent speleothem δ18O monsoon rainfall record (Wang et al., 2001, 2008) strongly supports the CBT index as a palaeorainfall proxy on the CLP.

According to the preliminary modern CBT–MAP calibration, the minimum MAP values for the three sites (559 mm, 726 mm, and 616 mm for Yuanbao, Lantian, and Mangshan, respectively) since the last glacial occurred at the Last Glacial Maximum (LGM), while the maximum values (780 mm, 834 mm, and 858 mm for Yuanbao, Lantian, and Mangshan, respectively) were reached during the Holocene, showing an overall ca. 200 mm enhancement in MAP on average during the transitional deglacial period in the southern CLP. Moreover, the maximum and minimum values of reconstructed MAP for the Yuanbao LPS are much lower than those of Lantian and Mangshan LPSs. This is in agreement with the modern isohyet's pattern showing a decreasing MAP from east to west CLP (Fig. 1). However, these quantitative results based on CBT data measured by different laboratories should be interpreted with caution, since recent round-robin studies with dozens of laboratories have shown that GDGT indices may vary due to differences in interlaboratory instrumental characteristics (Schouten et al., 2013b).

The CBT–MAP relationship in alkaline soils provides an empirical basis for the direct reconstruction of palaeorainfall based on the CBT index on the CLP instead of linking them via pH. Previously, the CBT index has been used successfully as a qualitative indicator of past precipitation in tropical Africa, given that large-scale changes in soil pH reflect changes in regional precipitation (Weijers et al., 2007b, and references therein). Recently, Peterse et al. (2014) proposed that the comparison of CBT-derived pH (likely precipitation-induced) and bGDGT-derived temperature in the LPS might enable us to further assess the relative timing and magnitude of hydrological and thermal changes in continental East Asia, independently of potential biases associated with age model uncertainties. However, the reconstructed pH is unexpectedly higher in palaeosols than in loess (Jia et al., 2013; Peterse et al., 2014), resulting in an inference of lower precipitation when palaeosols formed according to the assumption of Weijers et al. (2007b) that higher palaeosoil pH is related to drier conditions. In this study, we show that the CBT index possibly correlates directly with precipitation intensity, providing an explanation for the perplexing results of previous studies (Jia et al., 2013; Peterse et al., 2014). In view of the great potential of bGDGTs in reconstructing continental air temperature (Weijers et al., 2007a; Peterse et al., 2012; De Jonge et al., 2014; Yang et al., 2014a), the application of the CBT index derived from the same suite of lipids as a palaeorainfall proxy might be particularly promising on the CLP.

The MBT' / CBT-derived MAATs are consistently lower than the meteorological MAAT for the Chinese alkaline soils (Fig. 9a). This is in agreement with a number of recent investigations which observed that the MBT' / CBT proxy underestimates MAAT in arid and alkaline conditions (Peterse et al., 2012; Dirghangi et al., 2013; Menges et al., 2014; Yang et al., 2014a). Such a “cold bias” of MBT' / CBT-derived MAAT in arid soils can be attributed in part to an enhanced hydrological (MAP) influence on MBT' (Peterse et al., 2012; Dirghangi et al., 2013; Menges et al., 2014). For this data set, the MBT' values are indeed generally lower than those for the globally distributed soils (Peterse et al., 2012) with the same MAAT (Fig. 9b). Recently, De Jonge et al. (2014) further suggest the underestimation of MAAT for arid soils may be forthcoming from the fact that higher pH (induced by more aridity; Xie et al., 2012) could increase the production of 6-methyl bGDGTs. When using the traditional cyano HPLC column, the co-eluting 6-methyl bGDGTs increase the peak areas of the “normal” 5-methyl bGDGTs for IIa, IIb, IIc and IIIa. This would decrease the calculated MBT' and reconstructed temperature, whereas it is actually caused by a higher pH rather than lower temperature. Moreover, the insensitivity of CBT to soil pH variation in alkaline soils might also be responsible for the “cold bias” of MBT' / CBT-derived MAAT. Actually, when applying the global MBT' / pH MAAT calibration (Peterse et al., 2012), in which CBT is replaced by measured soil pH, we observed that the reconstructed temperature for the Chinese alkaline soil data set is improved (Fig. 9a). Overall, therefore, due to the problem in both MBT' and CBT indices, the MBT' / CBT temperature proxy should be applied with caution for quantitative palaeotemperature reconstructions in soils from arid–subhumid regions.

Given the uncertainty in the MBT' / CBT proxy in arid/semiarid regions, Yang et al. (2014a) proposed an alternative transfer function (i.e. the Chinese SSM calibration) for temperature reconstruction based on the fractional abundances of bGDGTs. According to this calibration, the reconstructed MAAT compares favourably to meteorological MAAT for the Chinese alkaline soil data set (Fig. 9a), with a mean difference of 1.8 ∘C and no obvious bias. Therefore, the Chinese SSM calibration might be generally more applicable for quantitative palaeotemperature reconstructions in alkaline soils from arid–subhumid China, in particular on the CLP.