Spatial variability of soil organic carbon in a catchment of the Loess Plateau

Abstract

The issue of soil organic carbon (SOC) is of increasing concern. Because SOC, as an important soil component in farming systems, is essential for improving soil quality, sustaining food production and quality, and maintaining water quality and as a major part of the terrestrial carbon reservoir, it plays an important role in the global carbon cycle. In this paper, a total of 665 soil samples from different depths were collected randomly in the autumn of 2007, and the spatial variability of SOC content at a small catchment of the Loess Plateau was analysed using classical statistics and geo-statistical analysis. In nonsampled areas classical kriging was utilized for interpolation of SOC estimation. The classic statistical analysis revealed moderate spatial variability with all five layers of SOC-content. In addition, the average SOC content decreased with soil depth and the relationship can be modelled by an exponential equation (y=3.1795x ^−1.2015, R ²=0.9866) and all of the SOC-content data in the different depth were normally distributed. The geo-statistical analysis indicated a moderate spatial dependence in 0–60 cm, while in the 60–80 cm depth spatial dependence was strong. The semi-variogram could be fitted by an exponential model for 0–10 cm depth; by a spherical model for 10–20 cm depth and 60–80 cm depth; and by a Gaussian model for 20–60 cm depth. The range increases with increasing depth. In addition, classical kriging could successfully interpolate SOC content in the catchment. In general, the geo-statistics method on a watershed scale could be accurately used to evaluate spatial variability of the SOC content in the Loess Plateau, China.

Keywords:

• Geo-statistical analyses

Introduction

As an important soil component in farming systems, soil organic carbon (SOC) is essential for improving soil quality, sustaining food production and quality, and maintaining water quality (Singh et al., 2007). The SOC, therefore, has been proposed as an indicator of soil quality (DEFRA, 2004). The SOC stock, as a major part of the terrestrial carbon reservoir, plays an important role in the global carbon cycle. In terrestrial ecosystems, the SOC stock is nearly three times as large as carbon storage in vegetation (Schlesinger, 1990), and twice as large as global atmospheric carbon storage (Lal, 2004). Owing to the large quantity stored in terrestrial ecosystems, a slight change in SOC will affect the density of greenhouse gases in the atmosphere. With increasing awareness of the threat posed by global warming that exacerbates environmental problems and endangers human sustainability, the global communities, both international and national, have focused their attention on alleviating global warming through reducing carbon emissions or increasing carbon sequestration (Marland & Schlamadinger, 1999; Wang & Zhou, 1999; Fang & Chen, 2001; Pan et al., 2002; Li et al., 2003; Han et al., 2004; Morisada et al., 2004). As the upper layer of the lithosphere, the pedosphere not only provides the substrate for plants to grow, but also is the interface of the atmosphere, lithosphere, hydrosphere and biosphere (Schindler, 1999). Thus, the soil carbon pool could actively exchange carbon with the atmospheric carbon pool (Wang & Hsieh, 2002). A series of studies on estimating the SOC stock and distribution patterns in the soil on global, continental, country, and regional scales have been reported, such as the work done by Sundquist (1993) on the global carbon dioxide budget, the work by Morisada et al. (2004) on the organic carbon stock in forest soils in Japan, and some of the studies done by Chinese scientists (Li & Wang, 2000; Wang et al., 2000; Li et al., 2001; Sun et al., 2004; Xie et al., 2004) to estimate the soil carbon pool of China and to reveal its distribution patterns.

However, owing to varied data sources, a lack of complete inventory data, and the inherent spatial variability of SOC, estimation of SOC distribution has varied from one study to another (Xie et al., 2004), Especially for the region of the Loess Plateau, where is characterized by extremely complex topography, and significant topographic variability with typical Loess hills and gully slope shapes (Li et al., 2008), which theoretically should result in a large variability in the spatial distribution of SOC. Although some studies have been conducted on spatial variability of SOC (Conant & Paustain, 2002; Schoning et al., 2006; Stavi et al., 2008), Because of the extremely complex topography there is little study about SOC in the Loess Plateau. Only a few experiments have been done on the slope scale (Wang et al., 2001; Fu et al., 2003).

The objectives of this work were: 1) to estimate the vertical variability of SOC in a small catchment of the Loess Plateau in northern Shaanxi, China; 2) to reveal the spatial variability of SOC in the catchment, and 3) to assess SOC distribution in the catchment.

Materials and methods

The study area

The study catchment, located in Shenmu County, Shaanxi Province, China (38° 46′ to 38° 51′ N, 110° 21′ to110° 23′ E) (Figure 1), is typical of the gullied-hilly area of the Loess Plateau. The watershed has a temperate semiarid climate with a mean annual temperature of 8.4 °C. The mean annual precipitation is 549 mm. Most precipitation occurs in a few large storms, with 77.4% of the annual precipitation falling during the flood season from June to September. The main soil (cultivate loessial soil) is an Ust-Sandiic Entisol which is loess-derived and easily erodible, apart from two small areas of Aeolian sandy soil (Figure 1). As a result, the region is characterized by steep gullies and hills (Tang et al., 1993). The main vegetation is purple medick (Medicago sativa), scattered shrub (Caragana korshinskii), spare poplar (Populus simonii), and little farmland (Figure 1).

Figure 1.  Location of the study area in Shenmu, Shaanxi Province, China (A), the land use (B), and soil types (c) of the catchment.

Soil sampling and chemical analysis

A portable Garmin Summit GPS receiver was used to locate the sampling sites. The sampling sites (Figure 1) were chosen to represent all landform types in the catchment. A depth of 0–80 cm was divided into five layers, 0–10 cm, 10–20 cm, 20–40 cm, 40–60 cm, and 60–80 cm. All of the samples were collected with a 5-cm-diameter hand auger. After being air-dried, samples were passed through a 1-mm sieve for laboratory analysis. In this study soil samples were collected on 169 locations randomly in the watershed and the total number of samples was 665.

SOC content was determined using the dichromate oxidation (external heat applied) method (Nelson & Sommers, 1975)

Statistical analysis

In this study classical and geo-statistical analyses were combined to characterize spatial variability of the soil nutrients. Mean and standard deviation (SD) were computed for each soil depth with the software package SPSS13.0. GS⁺7.0 software was used to perform all geostatistical computations. Software package ArcGIS 9.0 was used to estimate the unsampled locations.

The basic theory of geostatistics has been well established (Journel & Huijbregts, 1978) and reviewed (Trangmar et al., 1985). The experimental variogram estimator is asymptotically unbiased for any intrinsic random function; however, it is very sensitive to outlying values because it is based on squared differences among data. Apart from identifying outliners by exploratory data analysis, a robust variogram [Equation (1)] is a good estimator to reduce the sensitiveness to outliers (Liu et al., 1996):

where γ(h) is the sample semivariance between all observations Z(X _i); N(h) is the number of pairs of Z(X _i) over a separation distance h. The stationary models, i.e. Gaussian [Equation (2)], spherical [Equation (3)], and exponential [Equation (4)] can be written as:

where C ₀ is the nugget, and a is the correlation length.

The degree of spatial dependence(GD) was calculated as in Equation (5):

The simulate precision can be reflected by coefficient R ², calculated as shown in Equation (6):

where are the theoretical data calculated with the model, r(h _i ) are the observation data, and is the mean of the observation data. In order to know if the equation is right, an F-test to the R ²-values was done [Equation (7)], where k is the number of independent variables in the Equation.

Result and discussion

The classical analyses of SOC content

The classical analyses of SOC contents for the five depths are shown in Table I. The mean contents of SOC in the five depths of the catchment are different. SOC decreases with increasing depth and the trend could be modelled by the exponential Equation (8) (Figure 2).

The difference between the minimum and maximum SOC contents is large for the five depths, especially for 0–10 cm and 10–20 cm, where the values are relatively large (0.6 and 0.66), as topsoils are more easily affected by disturbance. Because the SOC variance is different at different depths, according to the coefficient of variance (CV) at different depths, we can obtain a cursory classification. The variable is considered to have a strong spatial dependence if the CV is less than 10%, and has a moderate spatial dependence if the ratio is between 10 and 100%; otherwise, the variable has a weak spatial dependence (Warrick & Nielsen, 1980). The CV was different at different depths, from 32.91 to 50.84% and all values indicated a moderate spatial dependence.

Figure 2.  Mean SOC content at different depths.

Table I. Classical analyses of the SOC parameters.

Depth	10 cm	20 cm	40 cm	60 cm	80 cm
Minimum	0.14	0.04	0.02	0.01	0.01
Maximum	0.74	0.70	0.27	0.28	0.23
Mean	0.40	0.21	0.14	0.12	0.11
Median	0.39	0.20	0.15	0.13	0.11
Std. Deviation	0.13	0.11	0.06	0.05	0.05
Kolmogorov–Smirnov test	0.57	0.66	1.07	0.83	0.80
Variance coefficient (CV) (%)	32.91	50.84	45.30	43.53	46.23

A previous study of the spatial variability of soil properties has shown that the soil properties are not independent but instead are stochastic variables (Nielsen et al., 1983). This is caused by the continuity of the soil-forming processes and the gradual change of the climate. So it is necessary to traverse the spatial variability of the SOC using the geo-statistical analyses method. However, if the data are not normally distributed, the scale effect will affect the variance function. Before geo-statistical analyses, we tested the data with a one-sample Kolmogorov–Smirnov test (Meulman & Heiser, 2001). In this study the data of SOC content at different depths were normally distributed (Table I).

Table II. The variance function model and parameters for SOC.

Depth (cm)	C 0	C 0+C	GD (%)	Range h (m)	Model	R ^2	F	Significance
10	0.0054	0.019	28.7	79.8	Exponential	0.755	200.31	0.00
20	0.0034	0.013	25.7	143.9	Spherical	0.879	232.46	0.00
40	0.0013	0.0048	27.1	147.2	Gaussian	0.895	272.76	0.00
60	0.0011	0.0035	37.8	153.1	Gaussian	0.854	187.12	0.00
80	0.00052	0.0031	16.8	155.1	Spherical	0.805	132.1026	0.00

The geo-statistical analyses of SOC distribution

Semi-variances were computed at different lags (multiples of the sampling interval of 12.5 m) using Equation (1). The semi-variance charts of SOC distribution are shown in Figure 3 and the variance function model and parameters for SOC are shown in Table II. The semi-variogram was fitted by exponential model in the 0–10 cm depth; by spherical model in the 10–20 cm and 60–80 cm depths; and by Gaussian model in 20–60 cm depth. All of the regression coefficients’R ²-values were higher than 0.755. From this we know that the semi-variogram can be effectively modelled. Through the F-test the R ²-values of all nutrients in the catchment were prominent. C ₀ is the nugget that is caused by random factors like fertilizer application, crop planting, and soil management. C ₀+C is the sill that is the value when the semi-variogram doesn't increase with increasing distance. The sill indicates the total variance. So C is the variance that is caused by the property of soil parent materials, climate, and terrain (Chien et al., 1997). The nugget effect (GD) is the ratio of C ₀ to C ₀+C. The GD relative degrees of <25%, 25–75%, and >75% could be used to describe the proportion of the spatial structure that showed strong, moderate, and weak spatial autocorrelation, respectively (Cambardella et al., 1994). In this catchment C ₀ decreased with increasing depth. A possible reason is that the SOC for the deeper layer is less affected by stochastic factors. As for the values for GD, they were lower in the 0–60 cm depth, showing a moderate spatial dependence, while in the 60–80 cm depth they showed a high magnitude of spatial dependence. The range (h) of the semi-variograms was the distance at which semi-variance attained the maximum value (sill), and the sill approximately equalled the sample variance (Journel & Huijbregts, 1978). The range expressed as a distance could be interpreted as the diameter of the zone of influence that represents the average maximum distance over which a soil property of two samples was related. At distances less than the range, measured properties of two samples became similar with decreasing distance between the two points. Thus, the range provides an estimate of areas of similarity. In the catchment, the range increased with increasing depth. The lowest value (only 79.8 m) was in surface soil but this value was still larger than the distance of sampling locations (15 m).

Figure 3.  Semi-variance charts of SOC distribution.

The SOC spatial distribution

The maps of the SOC content distribution can clearly characterize the concrete distribution in the catchment. In order to obtain good spatial patterns, the spatial changes of SOC content for unsampled locations can be directly interpolated with classical kriging in the ArcGIS software based on the observed semi-variogram values (Smith et al., 1993; Goovaerts & Journel, 1995). According to the correlation length, 12 neighboring samples were in the correlation length. To obtain the overall patterns of the SOC distribution, the spatial interpolation method of kriging was applied with 12 neighboring samples used for estimation of each grid point. The distribution of SOC content for five depths in the study area is shown in Figure 4.

Figure 4.  Spatial distribution of SOC content in the catchment.

The topsoil layer (0–10 cm) contained more SOC than did the other layers, and the horizontal variability of the topsoil layer was strong. The SOC content in the middle of the catchment, where the gully terrain was, was lower than in other places. The lowest SOC content was in the northwest part of the catchment, with the sandy soil. In the layer at 10–20 cm depth there were two obvious spots. One was the highest-content spot in the southwest and the other was the lowest-content spot in the northwest of the catchment and the range is 2.63 g kg⁻¹. The horizontal variability of the other three layers from 20–80 cm was similar and the SOC content was lower. A previous study had shown that structural factors, such as parent material, climate, terrain, and water table, as well as random factors, such as fertilizer application, crop planting, soil management, and co-determined soil properties, can effect the SOC distribution (Goovaerts, 1999). In this catchment, the area is small, only 0.3 km². So the climate and parent materials are similar. The SOC variable is caused mainly by soil type, landscape, and other human activity, especially the soil type. The lower SOC content spots (white spots) of the SOC distribution maps at different depth were covered by Aeolian sandy soil, and most of the higher SOC content spots (black spots) were covered by wood or shrub (Figure 1). The human activity affected mainly the topsoil. The classical analyses of SOC content for five depths in the study area show moderate spatial dependence, because the CV of the different layers is from 32.91 to 50.84%, between 10 and 100%. But through the geo-statistical analyses the values of GD at 0–60 cm depth were lower, showing a moderate spatial dependence, while at the 60–80 cm depth the value was different from that from the classical analyses, showing a high magnitude of spatial dependence. So if we want to know the true variability of the soil properties, the geo-statistical analyses is a useful method.

The classic statistical analysis showed that the mean SOC content at different depths of the catchment decreased with increasing depth, and the trend could be modelled by exponential Equation (9).

The SOC content at different depths showed moderate spatial variability, and all of the SOC content data at different depths were normally distributed. However, the geo-statistical analysis indicated that the values of GD in the 0–60 cm range were lower, showing a moderate spatial dependence, while at 60–80 cm depth GD showed a high magnitude of spatial dependence. The semi-variogram could be fitted by an exponential model for 0–10 cm depth; by a spherical model for 10–20 cm depth and 60–80 cm depth; and by a Gaussian model for 20–60 cm depth. The range increased with increasing depth. In addition, kriging could successfully interpolate SOC content in the catchment. In general, the geo-statistics method could be well used to evaluate spatial variability of the SOC content at the watershed scale in the Loess Plateau, China.

Acknowledgements

Financial support of this research was jointly provided by National Key Basic Research Support Foundation of China (No.2007CB106803), “Innovation Team” Grant of Northwest Sci-Tech University of Agriculture and Forestry, National Research Support Foundation of China (No.2006BAD09B06), West Light Foundation of The Chinese Academy of Sciences. (2006YB04), IRP0749.