INTRODUCTION
Soil quality is defined as ‘the capacity of soil to function effectively at present and in the future or as the capacity of a soil to function within ecosystem boundaries to sustain biological productivity, maintain environmental quality and promote plant and animal health’ (Doran and Parkin, Reference Doran and Parkin1994). This definition of soil quality covers a wide range of functions. However, it is unlikely that a particular soil is able to provide all these functions successfully. Some of those functions occur in natural ecosystems whilst the others are the result of human modification. Soil quality depends on the extent to which a soil fulfils the role it is destined for (Singer and Ewing, Reference Singer and Ewing2000). Within the framework of agricultural production, high soil quality equates to the maintenance of high productivity without causing significant soil degradation or environmental consequences.
Soil quality is a combination of soil physical, chemical and biological properties that are able to readily change in response to variations in soil management (Brejda et al., Reference Brejda, Karlen, Smith and Allan2000a). A wide range of indicators are available for the assessment of soil quality, but the interpretation of the indicators is often difficult. Therefore, it is essential to elaborate numerical indices that can be used as synthetic tools to integrate information about soil quality functions deriving from individual parameters. Integrated soil quality indices based on a combination of soil properties provide a better indication of soil quality than individual parameters. These selected properties are grouped into a minimum data set (MDS), and such a collection of selected indicators may have the features of measuring soil state and function from plot to regional scale (Doran and Parkin, Reference Doran and Parkin1994; Karlen et al., Reference Karlen, Mausbach, Doran, Cline, Harris and Schuman1997; Liebig et al., Reference Liebig, Varvel and Doran2001). The concept of MDS of soil quality indicators is widely accepted, but many different methods have been suggested to calculate indices from an MDS (Karlen et al., Reference Karlen, Mausbach, Doran, Cline, Harris and Schuman1997; Liebig et al., Reference Liebig, Varvel and Doran2001; Wienhold et al., Reference Wienhold, Andrews and Karlen2004; Zornoza et al., Reference Zornoza, Mataix-Solera, Guerrero, Arcenegui, Mataix-Beneyto and Go´ mez2008). Generally, the development of a soil quality index (SQI) starts with the establishment of a valid and precise MDS. The different indicators to be included in the MDS are usually expressed by numerical scales which are normalized using the scoring functions of linear and non-linear regressions. The integration of non-dimensional indicators (obtained by normalization) into quality indices is possible through many procedures based on multiplicative (Pierce et al., Reference Pierce, Larson, Dowdy and Graham1983; Singh et al., Reference Singh, Covlin, Erbach and Mughal1992), simple additive (Andrews and Carroll, Reference Andrews and Carroll2001) or weighted additive (Karlen et al., Reference Karlen, Gardner and Rosek1998).
A well-developed SQI can be used to improve soil management especially in those fragile agro-ecosystems such as the semi-arid areas of the western Loess Plateau of China. In many arid and semi-arid areas on the planet, serious soil erosion often occurs, largely due to the use of intensive tillage. In the western Loess Plateau of China, seedbed is typically prepared using three ploughs and two harrows during the period from post-harvest in the fall to the sowing time the following spring. This tillage system is believed to capture and store precipitation in the soil and maximize precipitation use in this ecoregion where precipitation is low and extremely variable. Furthermore, nearly all crop residues are removed from the field at harvest for animal feed or fuel for heating or cooking. The soil surface is left bare for the 7–8 months after harvest in the late summer until early spring (April) the following year. In this region, there is only one crop each year, which coincides with the part of the wet season in July to September. These practices have been shown to exacerbate the degradation of soils, promote erosion and reduce production potential. Thus, management practices must provide protection against the degradation of these soils. Conservation tillage has been shown to play an important role in minimizing soil erosion and improving soil quality. For example, Cai et al. (Reference Cai, Qi and Zhang2008) found that no-till promoted water stability of soil aggregates and stubble retention improved soil organic matter (SOM). The use of conservation tillage improved soil water condition (Enfors et al., Reference Enfors, Barron, Makurira, Rockström and Tumbo2011; Van Wie et al., Reference Van Wie, Adam and Ullman2013). Zhang et al. (Reference Zhang, Luo and Cai2011) and Niu et al. (Reference Niu, Zhang, Luo, Li, Cai, Li and Xie2016) reported that no-till with stubble retention improved soil physical properties compared to conventional tillage in a 7-year study.
The aim of the study was to develop a SQI by selecting the best soil quality assessment indicators. The assessment was performed using a long-term field experiment conducted at the semi-arid Loess Plateau of northwest China where different tillage systems were evaluated. For this purpose, soils from the different tillage treatments were sampled and the physical, chemical and biological parameters were determined. The MDS was established by selecting and integrating soil quality indicators together according to the methods described by previous researchers (Doran and Parkin, Reference Doran and Parkin1994; Larson and Pierce, Reference Larson and Pierce1994). Also, to better characterize the soil under investigations, we considered some other parameters and the impact of human activities on the ecosystems in the analysis. Using multivariate statistical analyses and soil quality indices, different soil quality classes were determined for the different tillage/stubble management systems.
MATERIALS AND METHODS
Site description
A long-term conservation tillage experiment was established in 2001 at Dingxi Experimental Station of Gansu Agricultural University (35°28′N, 104°44′E, 1971 m a.s.l.). The station is located in the heart of the semi-arid Loess Plateau of China. Long-term annual precipitation averages 391 mm, with about 54% occurring between July and September. Daily maximum temperatures are up to 38 ℃ in July, whilst minimum temperatures can be −22 ℃ in January. The soil was a Huangmian which is aligning with a Calcaric Cambisols in the FAO soil map of the world. The site had a long history of continuous cropping using conventional tillage system. Field pea (Pisum sativum L.) and spring wheat (Triticum aestivum L.) were grown in alternate years in the experiment since the start of the experiment in 2001 after a previous flax (Linum usitatissimum L.) crop.
Experimental design and treatments
The experiment had a fully phased factorial design, with six tillage treatments (Table 1), two rotation phases and replicated four times (blocks). Spring wheat (cv. Dingxi No. 35) and field pea (cv. Yannong) were sown in rotation with each of the two phases present in each year. There were 48 plots in total (6 tillage treatments x 2 phases x 4 replicates). Further details of the experimental design and plot management are described in Niu et al. (Reference Niu, Zhang, Luo, Li, Cai, Li and Xie2016).
Table 1. Treatments and their description in the long-term conservation tillage experiment.
Soil properties and crop measurements
After 6 years of experiment rotation, soil physical, chemical and biological properties were determined. Soil bulk density and capillary porosity, non-capillary porosity, field capacity and saturation capacity were determined (Nanjing Institute of Soil Science, 1978). Soil aggregates were measured by wet sieved method (Yang and Wander, Reference Yang and Wander1998).
For those measurements, soil samples were taken from the 0–5, 5–10 and 10–30 cm depths after crop was harvested. The three depths were chosen with the consideration that the impacts of tillage systems on the main soil quality indicators are most likely on the surface layers. Five cores (25 mm diameter) were collected from each plot and bulked into one sample per plot at each depth. After mixing using a portable soil mixer, the soil samples were dried and sieved through 2 mm size. Soil organic carbon (Walkley and Black, Reference Walkley and Black1934), total N (Bao, Reference Bao2000), Olsen P (Olsen et al., Reference Olsen, Cole, Watanable and Dean1954) and soil available K (Jackson, Reference Jackson1973) were determined.
Soil biological properties, such as soil catalase activity and urease activity (Yan, Reference Yan1988), alkaline phosphates activity (Zhao and Jiang, Reference Zhao and Jiang1986), invertase activities (Guan, Reference Guan1986) and soil microbial propagules (colony forming units, CFUs) were also measured (Li et al., Reference Li, Yu and He1996). Crop grain yield, straw and chaff weights were obtained when crop was harvest. More details on soil properties and crop yield measurements are given in Niu et al. (Reference Niu, Zhang, Luo, Li, Cai, Li and Xie2016).
Soil quality indicators assessment
In the assessment of soil quality indicators, the three steps were followed: (i) select an MDS of indicators that contributed most to soil quality, (ii) score the MDS indicators based on their performance of soil function and (iii) integrate the indicator scores into a comparative index.
Representative MDS (Andrews et al., Reference Andrews, Mitchell, Mancinelli, Larlen, Hartz, Horwarth, Pettygrove, Scow and Munk2002b; Doran and Parkin, Reference Doran and Parkin1994) includes only those soil properties that have a significant treatment difference. Significant variables (P < 0.05) were chosen for the next step in MDS formation through principal component analysis (PCA) (Andrews et al., Reference Andrews, Karlen and Mitchell2002a, Reference Andrews, Mitchell, Mancinelli, Larlen, Hartz, Horwarth, Pettygrove, Scow and Munk2002b; Shukla et al., Reference Shukla, Lal and Ebinger2004). Principal components (PC) for a data set are defined as linear combinations of variables that account for maximum variance within the set by describing vectors of closest fit to the n observation in p-dimensional space, subject to being orthogonal to one another. The PC receiving high eigen values and variables with high factor loading were assumed to be variables that best represented system attributes. Therefore, only the PCs with eigen values ≥1 (Brejda et al., Reference Brejda, Moorman, Karlen, Smith and Dao2000b) and those that explained at least 5% of the variation in the data (Wander and Bollero, Reference Wander and Bollero1999) were examined. Within each PC, only highly weighted factors were retained for MDS. Highly weighted factor loadings were defined as having absolute values within 10% of the highest factor loading. When more than one factor was retained under a single PC, multivariate correlation coefficients were employed to determine if the variables could be considered redundant and therefore eliminated from the MDS (Andrews et al., Reference Andrews, Mitchell, Mancinelli, Larlen, Hartz, Horwarth, Pettygrove, Scow and Munk2002b). Well-correlated variables were considered redundant and only one of them was considered for the MDS, with the others being eliminated from the data set. If the highly weighted variables were not correlated, each was considered important and was retained in the MDS.
After the MDS indicators were determined, every observation of each MDS indicator was transformed using a linear scoring method. Indicators were arranged in order depending on whether a higher value was considered ‘good’ or ‘bad’ in terms of soil function, which were assigned applying ‘more is better’ or ‘less is better’. The ‘more is better’ indicators and the ‘less is better’ indicators were calculated, respectively, by the ascending and descending functions. Equations #1 below defines a ‘more is better’ scoring curve for positive slopes, and Equation #2 defines the ‘less is better’ curve for negative slopes. In these equations, numerical values for each soil quality indicator were converted into unit-less scores ranging from 0 to 1.
$$\begin{equation}
F({X_i}) &=& {\raise0.7ex\hbox{${({X_{ij}} - {X_{imim}})}$} \!
/ \!\lower0.7ex\hbox{${({X_{imax}} - {X_{imim}})}$}},\end{equation}$$\\
$$\begin{equation}
F({X_i}) &=& {\raise0.7ex\hbox{${({X_{imax}} - {X_{ij}})}$} \!
/ \!\lower0.7ex\hbox{${({X_{imax}} - {X_{imim}})}$}},
\end{equation}$$
Once observation of each MDS indicator was transformed, the MDS variables for each observation were weighted. There are many ways to assign the weights for each indicator. This includes experience, mathematical statistics or models (Wang, Reference Wang1994). In this study, PCA was used to determine the weights for each indicator (Equation #3).
$$\begin{equation}
{W_i} = {C_i}/\sum\limits_{i = 1}^n {\left( {{C_i}} \right)} ,
\end{equation}$$
where Wi is the weighting factor derived from the PCA; Ci is the communality of i soil quality indicator andnis the number of soil indicators included in the index.
Finally, we summed up the weighted MDS variables scores for each observation using equation #4 below, then the SQI was obtained. Here, the assumption is that higher index scores meant better soil quality or greater performance of soil function, which in this study was to sustain crop yields.
$$\begin{equation}
{\rm{SQI}} = \sum\limits_{i = 1}^n {F\left( {{X_i}} \right)} \times {W_i}.
\end{equation}$$
Data analysis
Analysis of variance (ANOVA) was performed to determine the effects of different tillage systems on soil properties and soil quality. All statistical analyses of data were carried out through the SPSS package (SPSS Software, 13.0, SPSS Institute Ltd, USA). Significances were declared at P < 0.05.
RESULTS
Selection of soil quality indicators
Tillage systems had a different influence on the different soil properties. Taking into account soil and climatic conditions for the specific agro-ecological zone, 16 soil property indicators were used in the index development, namely soil bulk density, total porosity, capillary porosity, non-capillary porosity, aggregates, field capacity, saturation capacity, SOM, total nitrogen, available phosphorus, available potassium, microbial biomass, catalase activities, urease activities, alkaline phosphatase activities and invertase activities. The most sensitive indicators out of the 16 assessment indicators were selected using PCA and these selected indicators were used to evaluate the treatment effect in different soil depths. In the PCA of 16 variables at the 0–5 cm soil depth, five PCs had eigen value >1 and explained 79.2% of the variance in the data (Table 2). Highly weighted variables under PC1 included total nitrogen, available potassium and invertase activity. A correlation matrix for the highly weighted variables under different PCs was run separately for each depth (Table 3). Only variables with the highest correlation sum were included in the MDS. Amongst the three variables in PC1, total nitrogen was chosen for the MDS because of its highest correlation sum. Available potassium (r = 0.746**) and invertase activity (r = 0.635**) were highly correlation with total nitrogen and hence they were dropped. In PC2, bulk density and total porosity were highly weighted. Bulk density was chosen for the MDS and total porosity was dropped because total porosity was calculated by bulk density. Microbial counts were retained under MDS because it was no significantly correlated with bulk density. In PC3, capillary porosity was eliminated because of highly correlated with bulk density (r = 0.417*). Under PC4 and PC5, saturation capacity and field capacity were highly weighted variables and both of them were retained in MDS because of their relative importance in dryland agriculture.
Table 2. Principal component analysis (PCA) of soil quality indicators at 0–5 and 5–10 cm soil depth.
Table 3. Correlation matrix for highly weighted variables under PC's at 0–5, 5–10 and 10–30 cm soil depth.
In the PCA of 16 variable at the 5–10 cm depth, five PCs had eigen value >1 and explained 74.6% of the variance in the data (Table 2). Highly weighted variables under PC1 included bulk density and total porosity. Bulk density was chosen for the MDS and total porosity was dropped. In PC2, SOM, microbial count and urease activity were highly weighted. Microbial count was chosen for the MDS because of its highest correlation sum. SOM plays an important role in maintaining soil quality of erodible environment and was considered under MDS. The urease activity was dropped because it was highly correlated with microbial counts (r = 0.441**). Under PC3, PC4 and PC5, catalase activity, capillary porosity and field capacity were highly weighted variables. Capillary porosity was eliminated because it was highly correlated with bulk density (r = 0.616**). Field capacity and catalase activity was retained in MDS.
In the PCA of 16 variable at the 10–30 cm depth, six PCs had eigen value >1 and explained 78.8% of the variance in the data (Table 4). Highly weighted variables under PC1 included bulk density and total porosity. Bulk density was chosen for the MDS and total porosity was dropped. In PC2, PC3, PC4, PC5 and PC6, no-capillary porosity, microbial counts, catalase, invertase and alkaline phosphatase activity were highly weighted variables. No-capillary porosity, microbial count and catalase activity were retained under MDS. These soil properties have been reported as the early and sensitive indicators of changes in soil quality because they manifest themselves over shorter timescales and are central to the ecological function of a soil (Bandick and Dick, Reference Bandick and Dick1999; Karlen et al., Reference Karlen, Wollenhaupt and Erbach1994). Alkaline phosphatase (r = −0.395*) and invertase activities (r = −0. 453**) were eliminated from the MDS because of its high correlation with microbial counts.
Table 4. Principal component analysis (PCA) of soil quality indicators at 10–30 and 0–30 cm depth.
In the PCA of 16 variable at the 0–30 cm depth, four PCs had eigen value >1 and explained 76.1% of the variance in the data (Table 4). Highly weighted variables under PC1 included no-capillary porosity, aggregates, SOM, total N, available phosphatase, available potassium and invertase activity. Amongst the seven variables in PC1, SOM was chosen for the MDS because of its highest correlation sum. Water-stable soil aggregates play an important role in maintaining soil quality of erodible environment and was considered under MDS. The other variables were highly correlated with SOM and hence they were dropped (Table 5). In PC2, bulk density was chosen for the MDS because of its highest correlation sum, and total porosity was dropped. The microbial count was retained. In PC3 and PC4, capillary porosity and catalase were highly weighted variables. Capillary porosity was eliminated from the MDS because of its high correlation with bulk density (r = −0.641**). Catalase activity was retained as a biochemical soil property.
Table 5. Correlation matrix for highly weighted variables under PC's at 0–30 cm soil depth.
Calculation of indicators weights
Having finalized the MDS indicators, numerical values for each soil quality indicator were converted into unit-less scores ranging from 0 to 1. The indicators retained in the MDS were considered ‘good’ in an increasing order except bulk density, and they were scored with Equation #1, as ‘more is better’. Excessively high soil bulk density was considered ‘poor’, and it was scored with Equation #2, as ‘less is better’. Once transformed, the MDS variables for each observation were weighted using PCA results (Supplementary Table S1, available online at http://dx.doi.org/10.1017/S0014479716000521).
Assessment of soil quality
The SQI calculated for the different tillage systems in the 0–5 cm depth decreased in the following order: 0.647 (NTS) > 0.558 (TS) > 0.516 (NTP) > 0.462 (NT) > 0.440 (TP) > 0.369 (T) (Supplementary Table S2). The SQI calculated for the different tillage systems in the 5–10 cm depth decreased in the following order: 0.493 (NTS) > 0.484 (TS) > 0.471 (NTP) > 0.414 (NT) > 0.377 (T) >0.374 (TP). The SQI calculated for the different tillage systems in the 10–30 cm depth decreased in the following order: 0.301 (NTS) > 0.278 (NT) > 0.265 (NTP) > 0 .252 (TS) > 0.215 (T) > 0.194 (TP). The SQI calculated for the different tillage systems in the 0–30 cm depth decreased in the following order: 0.527 (NTS) > 0.432 (TS) > 0 .419 (NTP) > 0.396 (NT) > 0.307 (T) > 0.303 (TP). These results clearly showed that the tillage systems had a significant impact on soil quality as shown by the SQI. The residue retention treatment significantly increased the soil quality in the top (0–5 and 5–10 cm) soil, whilst minimum tillage significantly increased the soil quality in the deeper soil layers (10–30 cm).
The soil quality indices were correlated with crop productivity under different tillage and stubble management treatments. Correlation analysis amongst SQI of different soil depths and grain yields showed that grain yield had a significant (P < 0.01) positive correlation with SQI in the 0–5 cm depth (Supplementary Table S3). Also, grain yield was significantly (P < 0.05) correlated with SQI in the 0–30 cm depth. However, there was no correlation amongst grain yield and SQI in the 5–10 or 10–30 cm depths.
DISCUSSION
Selection of soil depth for soil quality assessment
Tillage systems have a different effect on soil properties in different soil depths. Previous studies on conservation tillage have concentrated on single soil property evaluations related to different soil depths, such as changes in bulk density (Ferreras et al., Reference Ferreras, Costa, Garcia and Pecorari2000; Unger and Jones, Reference Unger and Jones1998), SOM (Chan and Heenan, Reference Chan and Heenan2005) and nutrients available to plants (Tracy et al., Reference Tracy, Westfall, Elliott, Peterson and Cole1990). Less attention has been paid to a comprehensive assessment of soil quality changes in different soil layers. The present study showed that grain yield had a positive correlation with SQI in the 0–5 cm (P < 0.01) and 0–30 cm (P < 0.05) soil depths, indicating that soil quality assessment in relation to cropping treatments should be conducted in the soil (0–5 cm) or tilth (0–30 cm) soil layers. Soil properties in the 0–5 cm depth can be considered as the early and sensitive indicators of management-induced changes in soil, whilst those in the 5–30 cm depth may have a strong effect on crop growth and grain yield. Therefore, under the experimental conditions, an effective soil quality assessment is in the depth of 0–5 and 0–30 cm for the typical soil in the Loess Plateau.
Selection of soil quality indicators
Selection of an MDS for soil quality evaluation took into account general soil and climatic conditions for the specific agro-ecological zone and their interaction. Most soil quality indicators suggested by previous researchers (Doran and Parkin, Reference Doran and Parkin1994; Karlen and Stott, Reference Karlen, Stott, Doran, Coleman, Bezdicek and Stewart1994; Larson and Pierce, Reference Larson and Pierce1994; Singer and Ewing, Reference Singer and Ewing2000) were included in the list of variables assessed in the present study. Also, our list also included some other properties like saturation capacity and soil enzyme activities, but excluded some others like earthworm (Glover et al., Reference Glover, Reganold and Andrews2000), pH and soil texture (Doran and Parkin, Reference Doran and Parkin1994). These excluded ones are not applicable for the present agro-ecological zone, or are of no relevance to our soil quality comparison on a small regional scale. Inclusion of more biological soil properties such as soil enzyme activity helped improve the understanding of the specific soil systems. Andrews et al. (Reference Andrews, Mitchell, Mancinelli, Larlen, Hartz, Horwarth, Pettygrove, Scow and Munk2002b) reported that choice amongst well-correlated variables could be based on the practicability of the variables. Options to retain or drop a variable from the final MDS may depend on various factors, such as the ease of sampling, cost of estimation, and logic and interpretability. Following these principles, we determined the final MDS for the various soil depths. In the 0–5 cm depth, the MDS consisted of total nitrogen, bulk density, microbial counts, capillary porosity, field capacity and saturation capacity. The final MDS in the 5–10 cm consisted of bulk density, field capacity, SOM, microbial counts and catalase activity. The final MDS in the 10–30 cm consisted of bulk density, no-capillary porosity, microbial counts and catalase activity. Two common indicators appeared in the MDSs of the three soil layers are bulk density and microbial counts, suggesting that these two properties play a key role in affecting the quality of the particular soil.
Soil quality assessment
Accurate assessment of soil quality requires a systematic method for measuring and interpreting soil properties that adequately serve as soil quality indicators. It is well known that individual soil properties may not be an adequate measure of soil quality. Integrated soil quality indicators based on a combination of soil properties can better reflect the status of soil quality than individual parameters. To select a representative MDS only those soil properties that showed significant treatment effect were chosen, which was determined by PCA. Based on the soil quality evaluation, the SQI was developed using the relative value of each of the selected soil properties and their weights.
In all of soil depths, the highest SQI occurred under NTS, which was also the treatment with highest and most stable yield. The results, in line with numerous studies in temperate regions, demonstrated that decreasing tillage intensity or increasing amount of crop residues retained on the soil surface leads to improved soil quality (Halvorson et al., Reference Halvorson, Peterson and Reule2002; Lal et al., Reference Lal, Mahboubi and Fausey1994; Soon et al., Reference Soon, Clayton and Rice2001). No-till or minimum tillage with crop residues covering soil surface offer the best opportunity to increase C sequestration, soil microbial biomass and nutrient cycling (Salinas-Garcia, Reference Salinas-Garcia2001). In contrast, the T and TP treatments had the lowest SQI value mainly because of the human disturbance and the absence of residue retention. Soil inversion and pulverization by repeated tillage accelerates decomposition of organic matter thus affecting soil physical, chemical and biological properties (Cannell and Hawes, Reference Cannell and Hawes1994). Care must be taken when looking at the SQI values of TS and NT, which were higher than the other three treatments but lower than NTS treatment. Improvement in soil quality depends mainly on the modification of soil properties through various means such as tillage reduction, residue retention or the combination of both. In rainfed agro-ecosystems of the loess plateau, these two management practices have been proven to be better than either practice used alone to elevate soil nutrient levels and improve soil quality. The lower SQI of the NTP treatment, as compared with NTS, indicated that soil degradation is associated with the removal of crop residues and the mulch of plastic film. The latter practice has been considered an innovative technique in boosting crop yield in arid and semi-arid areas (Gan et al., Reference Gan, Siddique, Turner, Li, Niu, Yang, Liu, Chai and Donald2013), but it may disturb soil eco-environment, leading to decreased soil aggregates and reduced microbial biomass. Accelerated decomposition of SOM due to plastic film leads to the reduction of SOM (Li et al., Reference Li, Li and Li2003).
CONCLUSION
Considering the SQI for soil depths, the present study indicated that soil quality assessment should be done in the top soil (0–5 cm) layer. Different sets of indicators were found for MDSs of soil layers, but bulk density and microbial counts appeared in all MDSs, suggesting that these two indicators are closely related to the tillage systems tested. The greatest SQI score and the highest crop yield occurred with the no-till and stubble covering treatments, indicating that no-till with stubble retention is the most effective option for improving soil quality and increasing crop productivity in the semi-arid Loess Plateau of northwest China.
Acknowledgments
The project was supported by the National Natural Science Foundation of China (No. 31171513, No. 41461067 and No. 31560379), Natural Science Foundation of Gansu Province (145RJZA208) and the Fundamental Research Funds for the Gansu Provincial Universities (037041014, 035041049). Special thanks to all postgraduate and undergraduates students from Gansu Agricultural University who were involved in the project for different periods of time.
SUPPLEMENTARY MATERIAL
For supplementary material for this article, please visit http://dx.doi.org/10.1017/S0014479716000521.
