Printer Friendly

Effects of land use and topography on spatial variety of soil organic carbon density in a hilly, subtropical catchment of China.

Introduction

Soil organic carbon (SOC) plays an important role in the functioning of both natural ecosystems and agricultural systems; it is also an important reservoir for the atmospheric carbon sink, and small changes in SOC may have large effects on atmospheric carbon dioxide (C[O.sub.2]) enrichment (Lai 2003, 2004). Furthermore, SOC is important for determining limits of greenhouse gas production and developing carbon mitigation strategies. The importance of SOC has triggered considerable research performed at global (Jones et al. 2005), national (McGrath and Zhang 2003) and regional (Liu et al. 2006; Wang et al. 2014) scales. However, one shortcoming of these studies is the high degree of uncertainty embedded within the estimates. These uncertainties could be introduced by lack of complete inventory data, the use of different data sources and inherent spatial variability of SOC (Xie et al. 2004). Environmental factors are found to be useful in reducing uncertainties by improving estimations of the spatial distribution of SOC.

First, SOC varies among different land use types (Chuai et al. 2012) because of differences in land cover, soil conditions and micro-organisms (Guo and Gifford 2002; Ayoubi et al. 2012). Second, topographic factors, such as elevation, slope and topographic wetness index (TWI), also affect the spatial distribution of SOC because these topographic factors can affect the processes of soil erosion and sediment transport (Zhong and Xu 2009; Schwanghart and Jarmer 2011) and thus the distribution of SOC attached to soil particles. In addition, topographic factors affect patterns of land cover and decomposition of plant litter by affecting environmental conditions, such as the amount of water available and temperature (Seibert et al. 2007; Schwanghart and Jarmer 2011). Environmental characteristics (e.g. climate, landuse type, elevation, slope) have been widely investigated as ways of improving the accuracy of SOC estimates (Liu et al. 2011). However, shifts in large-scale variables such as climatic conditions likely correspond to an increase SOC homogeneity at smaller spatial scales, and there is a lack of uniformity in the relationship between SOC and environmental factors at different spatial scales (Liu et al. 2011).

Recently, a local, spatial interpolation method of incorporating environmental factors, known as geographically weighted regression (GWR), has received increased attention. GWR is a useful method to deal with the spatial non-stationarity of regression coefficients and to examine spatial variability in data more accurately (Fothcringham et al. 1998). For instance, Jaber and Al-Qinna (2015) demonstrated the superiority of GWR over multiple linear regression (MLR) for the spatial analysis of SOC stocks in the Amman-Zarqa Basin using Landsat Thematic Mapper data. Mishra et al. (2010) used the GWR method to predict spatial variation in SOC in seven states in the Mid western US. In that study, Mishra et al. (2010) compared GWR to regression kriging (RK) and MLR models by including five environmental factors (terrain attributes, climate, land use, bedrock geology and the normalised difference vegetation index), and their results demonstrated that the accuracy of the GWR method greatly exceeded that of the MLR method and was slightly improved compared with the RK method. Zhang et al. (2011) revealed that the GWR method performed better than the ordinary kriging (OK), inverse distance weighted (IDW) and MLR methods in the prediction of SOC in Ireland. In that study, Zhang et al. (2011) used rainfall, land cover and soil type as independent variables and found that GWR could reduce the smoothing effect of spatial interpolation.

Previous studies have focused on the spatial distribution of SOC in multiple regions of China, including the black soil region in the north-east (Liu et al. 2006), the Hexi Corridor in the north-west (Wang et al. 2014), the Loess Plateau region (Liu et al. 2011) and across the landscape of a terrestrial ecosystem (Wu 2011; Xu ef al. 2013). However, few studies on the spatial variation of SOC have been performed in small catchments in subtropical China, and there is little quantitative information about the spatial variability of SOC under different land use types and topography. The lack of environmental data limits our ability to evaluate the carbon budget and to predict ecosystem responses to environmental and land use changes in subtropical China. Furthermore, it should be noted that there is a high positive correlation between SOC and SOC density (SOCD). Generally, the estimation of SOC stock is directly using SOCD rather than SOC. Thus, it is important to analyse SOCD distribution and the factors affecting it, instead of SOC. Therefore, the effect of small-scale variables (e.g. land use and topography) on SOCD should be analysed at the appropriate spatial scale. The aims of the present study were to: (1) analyse the effects of land use and topography on SOCD in a typical catchment of subtropical China; and (2) determine the spatial distribution pattern of SOCD accurately by comparing the performance of multiple geostatistical methods.

Materials and methods

Study site, soil sampling and analyses

The present study was conducted in the Jinjing catchment (134 knr), located in the town of Jinjing, Changsha County, in the Hunan Province of China (Fig. 1a). This area has a humid climate influenced by subtropical monsoons, with an average annual precipitation of 1300 mm and average air temperature of 16.9[degrees]C. In the field, the approximate location of each sampling site was controlled by the spacing distance (50-100 m) between sites. Furthermore, the sampling design was established based on distributions of different topographic factors (elevation, slope and TWI) and land use patterns across the entire catchment. Sampling density was low in regions with a small variability of landscape composition and structure, but high in regions with a large variability of landscape composition and structure. In all, 1033 topsoil (0-20cm) samples were collected during 2010, with 556 soil samples from paddy fields, 429 soil samples from woodland and 48 soil samples from tea fields. For each sample site within a 5-m radius, five to eight replicate samples were randomly collected using a stainless steel auger ([empty set] 3 cm) and homogenised by hand mixing. The centre of the sample's geographical position (longitude and latitude), determined using a global positioning system (GPS; San ding Southern Survey Co.), agricultural land use and management practiccs around the site were recorded. All samples were air-dried at room temperature and passed through a 2-mm sieve. Soil organic matter (SOM) was measured by the [K.sub.2][Cr.sub.2][O.sub.7] oxidation-titration method (Nelson and Sommers 1982) and SOM was converted to SOC by multiplying it by the Van Bemmelen factor of 0.58. Soil bulk density was measured with the core sampling method (Blake and Hartgc 1986).

Preparation of environmental covariable datasets The digital elevation map (DEM; 5-m spatial resolution) in 2010 and the land use map (5-m spatial resolution; Fig. 1b) for the study area were acquired from the Land Surveying Department in Hunan Province. Elevation (Fig. 1) and slope (Fig. 1d) were directly derived from the DEM. The TWI was calculated as follows (Wilson and Gallant 2000):

TWI = A/tan [beta]

where [beta] is the slope and A is the specific catchment area (based on multiple flow direction [d.sub.[infinity]]; Moore et al. 1993; Fig. 1e). The positions of all the sampling sites in the catchment are shown in Fig. 1d. A 5-m resolution raster map was also generated for the study area, and environmental variables (land use, elevation, slope and TWI) at every soil sampling point were obtained by overlying the soil sampling point layer on the corresponding raster map for each variable. All spatial data were processed, and all maps were produced using ArcGIS 10.1 software (Esri).

Calculation of SOCD

SOCD (measured in kg [m.sup.2]) at each sampling site was calculated using Eqn 1 :

SOCD = [rho]SOC d/100 (1)

where [rho] is the bulk density (g [cm.sup.3]), SOC values are measured in g [kg.sup.-1] and D is soil depth (20 cm; Wang et al. 2014).

Semivariogram model for SOCD

A semivariogram model (Webster and Oliver 2001) was used to quantify spatial variations in SOCD at all locations sampled using geostatistical methods. A semivariogram is equal to half the expected squared difference between any paired values of Z(X) and Z(X+ h), where spatial locations are separated by distance h. For discrete sampling sites, the function used is given in Eqn 2:

[gamma](h) = 1/2N(h) [[summation].sup.N(h).sub.i=1] [[Z([X.sub.i]+h)].sup.2] (2)

where Z([X.sub.i]) is the value at location [X.sub.i], h is the distance between [X.sub.i] and [X.sub.i] + h pairs and N(h) is the number of all data pairs separated by distance h.

Spatial interpolation methods

For mapping the spatial distribution of SOCD in the catchment investigated, five spatial interpolation methods were used, as outlined below.

(1) OK is often used to estimate soil properties. However, the weighting coefficient is calculated from the parameters of an optimally fitted scmivariogram model under unbiased conditions and minimised estimated variance for interpolation (Goovaerts 1999).

(2) 1DW interpolation is a straightforward and non-computationally intensive interpolation method (Tobler 1970). The weights for samples in IDW decrease with increasing distance between the known samples and the point to be estimated. The rate of decrease is proportional to 'inverse distance'.

(3) MLR is a regression model that uses linear relationships between response and explanatory variables. The response and explanatory variables are also known as the dependent and independent variables respectively. This regression model is estimated using the least-squares method to minimise the sum of squares of the differences between dependent and independent values (Chatterjee et al 2000).

(4) Linear mixed-effects models (LMM) do not directly incorporate spatial information. The model allows for local random deviations from overall mean. LMM development, analysis of variance, data exploration and plotting were all performed using R 3.1.3 (http://www.rproject.org/, accessed 25 May 2011) with the lme4 package (Bates et al. 2014). In the present study, LMM was used to represent the effects of topographic factors (elevation, slope and TWI) on SOCD for each of three land use types. Models were fitted using maximum likelihood or restricted maximum likelihood. The best LMM model was selected based on the Akaike information criterion (AIC), the Bayesian information criterion (BIC), root mean square error (RMSE) between predicted and measured values and coefficients of determination ([R.sup.2]), using elevation as a fixed effect and slope and TWI as random effects.

(5) GWR is an improvement of traditional regression models because it allows for parameters to vary locally instead of by a single set of global parameters (Fotheringham et al. 1998). The regression coefficients are estimated at each location using data within a neighbourhood; thus, the model can measure spatial variations in relationships between parameters. The parameters in the GWR model were solved using the weighted least-squares approach. To determine parameters in GWR for a given dataset, a kernel function estimating the weighted influence of neighbouring data points is generally determined.

Validation of model performance

Tenfold cross-validation was used to evaluate the performance of OK, IDW, MLR, LMM and GWR. For this, the total dataset was randomly divided into 10 (approximately) equally sized sub-datasets. For each sub-dataset, the remaining 90% of the data was used as a training set for spatial analyses, and tested by the sub-dataset that was set aside. This procedure was repeated 10 times. In this way, predictions at the test dataset locations were compared with the observed data for each of the 10 test datasets.

The prediction accuracy of each method was evaluated by comparing the predicted values with observed values in the validation subset. The mean absolute error (MAE), RMSE, standardised mean square error (SMSE) and Pearson correlation coefficient (R) were calculated to verify prediction accuracy between the estimated values and observed values using Eqns 3-6 respectively.

MAE = [[summation].sup.n.sub.i=1][absolute value of [[??].sub.i] - [Z.sub.i] / n (3)

RMSE = [square root of ([[summation].sup.n.sub.i=1] [([[??].sub.i] - [Z.sub.i]).sup.2] / n) (4)

SMSE = 1/n [[summation].sup.n.sub.i=1] [([[??].sub.i] - [Z.sub.i]).sup.2] / [s.sup.2] (5)

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)

where n is the number of observations in the validation dataset, [[??].sub.i] and [Z.sub.i] are the observed and predicted values at location i respectively and s is the standard deviation of the estimation error.

Results and discussion

Descriptive statistics for SOCD

The mean SOCD (3.36 kg [m.sup.-2]) in this catchment was lower than values reported previously for China. For example, SOCD was previously reported to be 6.8 and 21.4 kg [m.sup.-2] in the top 0-20 cm soil layer in eastern tropical and subtropical regions respectively (Zhong et al. 2001), and 4.27 kg [m.sup.-2] for the 0-20 cm soil layer in the paddy fields of north-eastern China (Wang et al. 2007), which is 1.27-fold greater than the value found in the present study (Table 1).

The statistical distribution of the measured SOCD was negatively skewed and had a positive peak. The kurtosis value (3.01 kg [m.sup.-2]; Table 1) indicated that the SOCD data were approximately normally distributed and passed the Kolmogorov Smimov normal distribution test at a significance level higher than 0.05 (McGrath and Zhang 2003). The histogram of SOCD (Fig. 2) demonstrated that it had an approximately normal statistical distribution.

Spatial dependence of SOCD

The fitted model followed an exponential equation with minimum AIC values (Webster and Oliver 2001) for matching the experimental semivariogram (Fig. 3). The nugget ([C.sub.0] = 0.645) was positive, which could be explained by either sampling error, short-range variability, effects of random processes or inherent variability (Liu et al. 2006). The total sill for SOCD ([C.sub.0] + [C.sub.1] = 1.063) represented total variation, with variations in C possibly caused by the properties of the soil parent material, climate and terrain (Chien et al. 1997; Han et al. 2010).

The nugget-to-sill ratio [C.sub.0]/([C.sub.0] + [C.sub.1]) was 60.72%, which was considered to indicate moderate spatial dependence. In general, the nugget-to-sill ratio can be used to classify the spatial dependence of soil properties. Variables are considered to have a strong spatial dependence if the nugget-to-sill ratio is less than 25%, a moderate spatial dependence if the ratio is between 25% and 75% and weak spatial dependence if the ratio is greater than 75% (Cambardella et al. 1994). In the present study, the moderate spatial dependence may be attributed to intrinsic factors, including soil parent material, texture, topography and soil formation processes, as well as extrinsic factors, such as soil fertilisation, human and livestock interference and land use cover (Kilic et al. 2004). The semivariogram model demonstrated the spatial structure of SOCD for a range up to 182 m, is the maximum distance between correlated measurements (Utset et al. 1998). This range was much larger than the sampling interval (50-100 m), indicating that the sampling density adequately revealed spatial structures.

Effects of land use and topography on SOCD

Factors such as land use type, topographic factors of elevation, slope and TWI can affect the spatial variability of SOC. Incorporating the heterogeneity of SOC in predictive models can improve the precision of estimated carbon budgets and assist in the effective implementation of measurements for carbon sequestration (Wang et al. 2009).

In a small-scale region, with the same parent material and a single climate regime, the effects of large-scale factors on SOCD would be masked. Other small-scale factors (e.g. slope, elevation and land use) may be the most influential factors (Ata Rezaei and Gilkes 2005) affecting SOCD. In the present study, we focused on the relationships between SOCD in the 0-20 cm surface layer of soil and four main variables: land use type, elevation, slope and TWI. Climate and soil parent material were not taken into account because the soil was predominantly red soil (humic acrisols), and the climate (temperature and precipitation) was relatively constant in all areas of the present study catchment.

Land use types

The Jinjing catchment includes three main land use types: paddy fields, woodland and tea fields. Land use type significantly affected SOCD (P<0.05; Fig. 4), which was highest in the paddy fields (3.50 kg C [m.sup.-2]), intermediate in the woodland (3.24 kg C [m.sup.-2]) and lowest for the tea fields (2.81 kg C [m.sup.-2]).

It was previously found that farmland has lower SOC concentrations than uncultivated soils because of enhanced rates of decomposition resulting from tillage and removal of organic material during crop harvests (Lal 2004). In contrast, the results of the present study show that paddy fields can sequester more organic carbon than cither woodland or tea fields. Similar results were found in the Jiangsu Province of China, where the SOCD in paddy fields was much higher than in woodlands (Chuai et al. 2012). This may be a result of agricultural practices used in the paddy cultivating system, which include high rates of fertiliser application, increased primary productivity in the system and a high return of crop residues to paddy soils. Moreover, the Hooding water in paddy fields created anaerobic conditions on the soil, which would slow down organic matter decomposition. In addition, agricultural development projects have been shown to improve regional soil resilience (e.g. notill, increased crop density; Xu et al. 2007). SOC concentrations of the woodland are generally high because of a thick forest litter layer, flourishing belowground (root) biomass and reduced rates of litter decomposition, all of which enhance SOC accumulation (Gao et al. 2013). However, the woodland in the present study was dominated by plantations (e.g. for Masson pine and Chinese fir) and included little natural forest. Organic matter accumulation is significantly lower in plantations than in native forest ecosystems (Guo and Gifford 2002). In China, SOCD of total tea fields is smaller than for total woodlands (144 vs 161 MgC [ha.sup.-1] respectively; Zhang and Wang 2010; Li et al. 2011). Thus, the SOCD for tea fields in the present study was lowest, reflecting the high rates of soil organic matter oxidation and the degradation of soil aggregate structure caused by tillage and rotation, and highlighting that tea fields reduce the quantity of organic material returned to the soil.

Elevation

The SOCD in each land use type (paddy fields, woodlands and tea fields) was grouped according to elevation: <80, 80-110, 110-140, 140-200 and >200 m. These elevations were included in order to ensure that the number of soil samples at each elevation level was as close as possible. When land use type was not considered, the SOCD at elevations of >140 and >200 m were significantly higher than the values for samples at elevations of <80, 80-110 or 110-140m. There were no significant differences between SOCD at elevations of <80, 80-110 and 110-140m (P>0.05; Fig. 5). The overall effect was that SOCD increased with elevation, consistent with the results reported by Wang et al. (2002) and Chuai et al. (2012). This higher SOC at relatively higher elevations could be because of low average soil temperature, fewer human impacts and slower nutrient decomposition (Gao et al. 2013). In addition, elevation also affects the land cover type (Chuai et al. 2012); land at relatively higher elevations in the present study was primarily covered by forests, which acted as carbon reservoirs. SOCD was highest at elevations between 140 and 200 m, primarily because of much less human disturbance than at lower elevations. Above 200 m, SOCD decreased with increasing elevation, which may be due to the removal of soil nutrients and mineral particles by soil erosion processes (Liu et al. 2003).

As for the various land use types, elevation of paddy fields had no significant effect on SOCD (P>0.05), indicating that agriculture activities such as fertilisation may reduce the spatial variability in SOCD in paddy fields. Furthermore, the woodland SOCD was significantly greater in sites with relatively higher (>110m) than lower (<110m) elevation (Fig. 5). In addition, elevation affected SOCD under different land use types at the same elevation level. For elevation at both <80 m and 80-110 m, SOCD was significantly different between paddy fields and woodlands (P<0.05). This suggests that more organic carbon is sequestered by paddy fields than by woodlands (Chuai et al. 2012). At 80-110 m, SOCD in the woodlands and tea fields did differ significantly (P<0.05), which confirms, in part, that human disturbance decreases SOCD in woodlands at this elevation (Gao et al. 2013). Furthermore, for elevation levels of 110-140, 140-200 and >200 m, the no significant differences in SOCD between paddy fields and woodlands (P > 0.05) could be due to less human impacts at relatively higher elevations.

The SOCD demonstrated specific characteristics at different elevation levels. However, the effect of elevation on SOCD is probably indirect and complex. Elevation is related to geomorphology, which could affect soil erosion and geological deposition processes. These processes could cause corresponding changes in moisture, temperature, land use and management. All the factors could possibly affect SOC in topsoil (Liu et al. 2006; Zhong and Xu 2009).

Slope

The slope was divided into 10 levels at intervals of 5[degrees] to ensure that the number of soil samplers at each elevation level was as close as possible. SOCD for each slope level is shown in Fig. 6, with the frequency distribution of land use type determined for each slope level shown in Fig. 7. Slope significantly affected SOCD in the present study (Fig. 6). The SOCD exhibited a sharp peak between slopes of 0[degrees] and 5[degrees] before dropping to a lower value between 5[degrees] and 15[degrees] and then generally increasing with increasing slope. At slopes >45[degrees], SOCD decreased, which could be attributed to slope-dependent differences in land use in the catchment are investigated. Recently, studies have reported strong relationships between slope and SOC accumulation (Liu et al. 2006; Wang et al. 2009). The effects of slope on the SOCD reflected the combined influence of biotic and abiotic factors, such as plant growth, the rate of litter fall and soil moisture content (Wang et al. 2002). Related studies focusing on the effects of slope on SOC have shown that a low slope encourages carbon accumulation (Quinton et al. 2010), possibly because a low slope frequently corresponds to higher water and nutrient availability. Guo et al. (2006) also found that the SOC in flat areas was twice that in sloping fields. In contrast, SOC could be lost from steeper slopes because of soil erosion (Liu et al. 2006); such as the negative correlation (-0.19) between SOCD and slope reported by Wang et al. (2002). A study of the factors affecting SOC of the Jiangsu Province, China, found that steeper slopes tended to result in a decline in SOC (Chuai et al. 2012). The results of the present study are consistent with previous studies reporting high SOCD in low-slope areas.

Fig. 7 shows the frequency distribution of slope according to land use type. Paddy fields located on 0[degrees]-5[degrees] slopes accounted for 60% of total paddy field area, where the value of SOCD peaked. This could explain the ability of paddy fields to sequester organic carbon efficiently. SOCD peaks on tea fields primarily occurred on slopes ranging from 5[degrees] to 15[degrees], which explains why SOCD decreased from slopes of 0[degrees]-5[degrees] to slopes of 5[degrees]-15[degrees] (Fig. 7). SOCD increased at slopes >15[degrees], which could be driven primarily by the presence of mature woodlands or older secondary forests, a lack of human disturbance and an increasing slope.

Topographic wetness index

In this study, TWI was divided into 10 levels to ensure the numbers of soil samples at each TWI level was as close as possible (Fig. 8). There was a significant positive relationship between TWI and SOCD (P = 0.05). The TWI, which was considered a secondary terrain attribute, was related to zones of surface saturation and served as a quantitative variable to evaluate the landscape in terms of water and sediment accumulation (Wilson and Gallant 2000). Previous analysis of the relationship between terrain and soil variables in the Luvisol soil region of Central Bohemia found that TWI was the most appropriate terrain predictor of SOC (Zadorova et al. 2014). Similarly, there was a positive correlation found between SOC and TWI (Zadorova et al. 2014), which corresponds with the findings of the present study.

In topographic depressions, soils are moist because of runoff, sediments (including organic matter) and seepage from surroundings areas, leading to higher SOC values compared with values for drier uphill soils (Yoo et al. 2006). Total storage of SOC in agricultural soils is also primarily linked to reduced mineralisation in soils with a high soil moisture content (Neufeldt 2005). Thus, large TWI values usually indicate higher levels of soil moisture and reduced mineralisation of organic matter, as well as a potentially high SOCD.

Comparison of GWR with other methods

Environmental factors (land use, elevation, slope and TWI) were used as covariables in the interpolation models (MLR, LMM and GWR) in order to reduce interpolation error (Table 2), considering their effects on SOCD. The MAE, RMSE, SMSE and Pearson correlation coefficients indicated that GWR provided better and more reasonable interpolation results of SOCD than the MLR, OK, IDW and LMM methods. The results of the present study are consistent with previously published reports (Zhang et al. 2008, Mishra et al. 2010, Salas et al. 2010, Zhang et al. 2011, Chen et al. 2012, Jaber and Al-Qinna 2015).

Compared with MLR, GWR was more accurate in predicting the spatial distribution of SOC in the Mid western US (Mishra et al. 2010) and provided better spatial estimates of SOC stocks in a typical semi-arid watershed in Jordan (Jaber and Al-Qinna 2015). That GWR provides better results than MLR is due primarily to consideration of spatially varying relationships between the dependent and independent variables. Compared with the OK and IDW methods, GWR usually yields better estimates because it integrates environmental variables into the modelling process. For example, Chen et al. (2012) showed that GWR performed better in estimating forest canopy height than the OK and IDW methods. Zhang et al. (2011) also demonstrated that GWR provided more reasonable results of SOC content in Ireland than the OK and IDW methods. In the present study, GWR had smaller model errors than LMM, similar to results reported by Zhang et al. (2008). However, Salas et al. (2010) found that LMM produced better predictions of tree diameter than GWR. These differences in the performance of GWR and LMM may be due to differences in the spatial data of the dependent and independent variables.

Spatial distribution of SOCD and correlations with covariables

The GWR method was used to estimate SOCD and coefficients between SOCD and environmental variables at points in a 5-m grid system. All estimated values were transformed into grid raster data to create spatial distribution maps of SOCD and the varying coefficients (i.e., changes in the correlation coefficient between SOCD and a particular variable depending on the point where it is measured) using ArcGIS software.

The spatial distribution of SOCD was generally consistent with spatial patterns of elevation, slope, TWI and land use types in the catchment investigated in the present study (Fig. 9). The higher SOCD values in the paddy fields corresponded to areas of lower elevation and slope and larger TWI values. This may be associated with high SOC concentrations under paddy fields (Chuai et al. 2012), which are usually distributed at lower elevations than woodland and have higher levels of soil moisture and larger TWI values. High values of SOCD were also found in the northern mountains, which are characterised by woodlands and a high elevation. This could be explained by the fact that the woodlands fixed more carbon and retained higher concentrations of SOC (Gao et al. 2013), as well as the fact that the high elevation could reduce human disturbance and slow nutrient decomposition (Gao et al. 2013). Lower SOCD values were found in areas with high slope angles, small TWI values or tea fields. These observations are in accordance with findings (sec above) that slope was negatively correlated with SOCD and that TWI was positively correlated with SOCD. Furthermore, the agricultural land use in tea fields reduced SOCD (Zhang and Wang 2010; Li et al. 2011).

The spatial patterns of the effects of environmental covariables on SOCD were further demonstrated by the spatial distribution of correlation coefficients between SOCD and environmental covariables of land use type, elevation, slope and TWI. Generally, a positive relationship was found between SOCD and elevation in mountainous areas with relatively high elevation, and negative relationships were observed at relatively low elevations (Fig. 10a). This is consistent with the positive correlation between SOCD and elevation in the research area in the present study. As shown in Fig. 10b, SOCD was negatively associated with slope in mountainous areas with relatively steeper slope. This is because SOCD can be lost due to soil erosion in areas with relatively steeper slopes. Although the aforementioned results showed a positive correlation between SOCD and TW1, the spatial distribution of the estimated coefficients of TWI (Fig. 10c) were not always consistent with those of SOCD (Fig. 9). This is perhaps due to the presence of confounding factors affecting SOCD in local areas. The spatial distribution of the coefficients of land use was consistent with spatial patterns of SOCD, and this explains the high coefficient values appearing under paddy fields and woodland, and low values under tea fields (Fig. 10d).

In summary, the coefficients could be negative and positive, indicating that the effects of environmental variables 011 SOCD differed at different locations. In addition, there was no single determinant dominating the spatial distribution of SOCD in the research area.

Conclusions

SOCD had a moderate spatial dependence, and its spatial variety was generally shaped by land use and topography in a typical hilly red-soil agricultural catchment in subtropical China. Significant interactions between paddy fields, woodland and tea fields were observed on SOCD (P< 0.05) in the present study. SOCD is relatively higher in soils in the valleys of paddy fields (with low slope and high TWI) and on woodland hills (with high elevation and increased slope). The results of the present study contribute of our understanding of the effects of land use and topography on SOCD, and further demonstrate that GWR performs better in the spatial modelling of SOCD over OK, IDW, MLR. and LMM.

http://dx.doi.org/10.1071 /SR 15038

Acknowledgements

The study was funded by the National Natural Science Foundation of China (Grant no. 41201299) and Strategic Priority Research Program-Climate Change: Carbon Budget and Related Issues (Grant 110. XDA05050505).

References

Ata Rezaci S, Gilkes RJ (2005) The effects of landscape attributes and plant community on soil chemical properties in rangelands. Geoderma 125, 167-176. doi: 10.1016/j.geoderma.2004.07.010

Ayoubi S, Karchegani PM, Mosaddeghi MR, Honarjoo N (2012) Soil aggregation and organic carbon as affected by topography and land use change in western Iran. Soil & Tillage Research 121, 18-26. doi: 10.1016/j.still.2012.01.011

Bates D, Maechler M, Bolker B. Walker S (2014) Ime4: linear mixed-effects models using Ligen and S4. R package version 1.1-7. Available at http:// keziamanlove.com/wp-eontent/uploads/2015/04/StatsInRTutorial.pdf [accessed 1 April 2014].

Blake GR, Hartge KH (1986) Bulk density. In 'Methods of soil analysis. Part I: physical and mineralogical methods. Vol. 9'. 2nd edn. (F.d. A Klute) pp. 363-375. (American Society of Agronomy: Madison, WI)

Cambardella CA, Moorman TB, Parkin TB, Karlen DL, Novak JM, Turco RF, Konopka AH (1994) Field-scale variability of soil properties in central Iowa soils. Soil Science Society of America Journal 58, 1501-1511. doi: 10.2136/sssaj1994.03615995005800050033x

Chatterjce S, Hadi A, Price B (2000) 'The use of regression analysis by example.' (John Wiley & Sons: Chichester, UK)

Chen Ci, Zhao K, McDermid GJ, Hay GJ (2012) The influence of sampling density on geographically weighted regression: a case study using forest canopy height and optical data. International Journal of Remote Sensing 33, 2909-2924. doi: 10.1080/01431161.201-1.624130

Chien YJ, Lee DY, Guo HY, Houng KH (1997) Geostatistical analysis of soil properties of mid-west Taiwan soils. Soil Science 162, 291-298. doi: 10.1097/00010694-199704000-00007

Chuai XW, Huang X.I, Wang WJ, Zhang M, Lai L, Liao QL (2012) Spatial variability of soil organic carbon and related factors in Jiangsu Province, China. Pedosphere 22, 404-414. doi: 10.1016/ S1002-0160(12)60026-5

Fotheringham AS, Charlton ML, Brunsdon C (1998) Geographically weighted regression: a natural evolution of the expansion method for spatial data analysis. Environment & Planning A 30, 1905-1927. doi:10.1068/a301905

Gao P, Wang B, Geng GP, Zhang GC (2013) Spatial distribution of soil organic carbon and total nitrogen based on GIS and geostatistics in a small watershed in a hilly area of northern China. PLoS One 8. e83592. doi: 10.1371/journal.pone.0083592

Goovaerts P (1999) Geostatistics in soil science: state-of-the-art and perspectives. Geoderma 89, 1-45. doi: 10.1016/S0016-7061 (98) 00078-0

Guo LB, Gifford RM (2002) Soil carbon stocks and land use change: a meta analysis. Global Change Biology 8, 345-360. doi: 10.1046/j.13541013.2002.00486.x

Guo Y, Gong P, Amundson R, Yu Q (2006) Analysis of factors controlling soil carbon in the conterminous United States. Soil Science Society of America Journal 70. 601-612. doi: 10.2136/sssaj2005.0163

Han F, Hu W, Zheng J, Du F, Zhang X (2010) Spatial variability of soil organic carbon in a catchment of the Loess Plateau. Acta Agriculturae Scandinavica Section B Soil and Plant Science 60, 136-143. doi: 10.1080/09064710902721354

Jaber SM, Al-Qinna MI (2015) Global and local modeling of soil organic carbon using Thematic Mapper data in a semi-arid environment. Arabian Journal of Geosciences 8. 3159-3169. doi: 10.1007/s12517014-1370-6

Jones C, McConnell C, Coleman K, Cox P, Falloon P, Jenkinson D, Powlson D (2005) Global climate change and soil carbon stocks; predictions from two contrasting models for the turnover of organic carbon in soil. Global Change Biology II, 154-166. doi: 10.1111/j.1365-2486.2004. 00885.x

Kilic K, Ozgoz F, Akbac F (2004) Assessment of spatial variability in penetration resistance as related to some soil physical properties of two Auvents in Turkey. Soil & Tillage Research lb, 1-11. doi: 10.1016/j.still.2003.08.009

Lai R (2003) Global potential of soil carbon sequestration to mitigate the greenhouse effect. Critical Reviews in Plant Sciences 22, 151-184. doi: 10.1080/713610854

Lai R (2004) Soil carbon sequestration impacts on global climate change and food security. Science 304, 1623-1627. doi: 10.1126/science. 1097396

Li SY. Wu X, Xue H, Gu BJ, Cheng H, Zeng JM, Peng CH, Ge Y, Chang J (2011) Quantifying carbon storage for tea plantations in China. Agriculture, Ecosystems & Environment 141, 390-398. doi: 10.1016/j.agee.2011.04.003

Liu S, Bliss N, Sundquist F, Huntington TG (2003) Modeling carbon dynamics in vegetation and soil under the impact of soil erosion and deposition. Global Biogeochemical Cycles 17. 1074-1098. doi: 10.1029/ 2002GB002010

Liu D, Wang Z, Zhang B, Song K, Li X, Li J, Li F, Duan H (2006) Spatial distribution of soil organic carbon and analysis of related factors in croplands of the black soil region. Northeast China. Agriculture. Ecosystems & Environment 1-13, 73-81. doi: 10.1016/ j.agcc.2005.09.006

Liu Z, Shao M, Wang Y (2011) Effect of environmental factors on regional soil organic carbon stocks across the Loess Plateau region, China. Agriculture, Ecosystems & Environment 142. 184-194. doi: 10.1016/j. agce.2011.05.002

McGrath D, Zhang C (2003) Spatial distribution of soil organic carbon concentrations in grassland of Ireland. Applied Geochemistry 18, 1629-1639. doi: 10.1016/S0883-2927(03(00045-3

Mishra U, Lai R, Liu DS, Van Meirvenne M (2010) Predicting the spatial variation of the soil organic carbon pool at a regional scale. Soil Science Society of America Journal 74, 906-914. doi: 10.2136/ sssaj2009.0158

Moore ID, Gessler PH, Nielsen OA, Peterson GA (1993) Soil attribute prediction using terrain analysis. Soil Science Society of America Journal 57, 443-452.

Nelson DW, Sommers LE (1982) Total carbon, organic carbon, and organic matter. In 'Methods of soil analysis', (lids AL Page, RH Miller, DR Keeney) pp. 539-579. (American Society of Agronomy Inc., Soil Science Society of America Inc.: Madison, WI)

Neufeldt H (2005) Carbon stocks and sequestration potentials of agricultural soils in the federal state of Baden-Wiirttemberg, SW Germany. Journal of Plant Nutrition and Soil Science 168, 202-211. doi: 10.1002/jpln. 200421441

Quinton .IN, Covers G, Oost KV, Bardgett RD (2010) The impact of agricultural soil erosion on biogeochemical cycling. Nature Geoscience 3, 311-314. doi: 10.1038/ngeo838

Salas C, line L, Gregoire TG, Naesset Ii, Gobakken T (2010) Modelling tree diameter from airborne laser scanning derived variables: a comparison of spatial statistical models. Remote Sensing of Environment 114, 1277-1285. doi: 10.1016/j.rsc.2010.01.020

Schwanghart W, Jarmer T (201 I) Linking spatial patterns of soil organic carbon to topography: a case study from south-eastern Spain. Geomorphology 126, 252-263. doi:10.1016/j.geomorph.2010.11.008

Seibert J. Stendahl J, Serensen R (2007) Topographical influences on soil properties in boreal forests. Geoderma 141, 139-148. doi : 10.1016/ j.geoderma.2007.05.013

Tobler W (1970) A computer movie simulating urban growth in the Detroit region. Economic Geography 46, 234-240. doi: 10.2307/143141

Utset A, Ruiz ML, Herrera J, de Leon DP (1998) A geostatistical method for soil salinity sample site spacing. Geoderma 86, 143-151. doi: 10.1016/ S0016-7061(98)00037-8

Wang HQ. Hall CAS, Cornell JD, Hall MHP (2002) Spatial dependence and the relationship of soil organic carbon and soil moisture in the Luquillo experimental forest, Puerto Rico. Landscape Ecology 17, 671-684. doi: 10.1023/A: 1022962116313

Wang H, Liu Q, Shi X, Yu D, Zhao Y, Sun W, Darilek JL (2007) Carbon storage and spatial distribution patterns of paddy soils in China. Frontiers of Agriculture in China 1, 149-154. doi: 10.1007/s11703007-0026-7

Wang Y, Zhang X, Zhang J, Li S (2009) Spatial variability of soil organic carbon in a watershed on the Loess Plateau. Pedosphere 19, 486-495. doi: 10.1016/S1002-0160(09)60141-7

Wang M, Su Y, Yang X (2014) Spatial distribution of soil organic carbon and its influencing factors in desert grasslands of the Hexi Corridor, northwest China. PLoS One 9, e94652. doi:10.1371/journal.pone.009 4652

Webster R, Oliver MA (2001) 'Geostatistics for environmental scientists.' (John Wiley & Sons: Chichester, UK)

Wilson JP, Gallant JC (2000) Secondary topographic attributes. In 'Terrain analysis: principles and applications'. (Rds JP Wilson, JC Gallant) pp. 87-131. (John Willey & Sons: New York, NY)

Wu J (2011) Carbon accumulation in paddy ecosystems in subtropical China: evidence from landscape studies. European Journal of Soil Science 62, 29-34. doi:10.1-111/j. 1365-2389.2010.01325.x

Xie XL, Sun B, Zhou HZ, Li AB (2004) Soil organic carbon storage in China. Pedosphere 14, 491-500.

Xu Q, Rui WY, Bian XM, Zhang W (2007) Regional differences and characteristics of soil organic carbon density between dry land and paddy field in china. Agricultural Sciences in China 6, 981-987. doi: 10.1016/S1671-2927(07)60137-0

Xu GC, Li ZB, Li P, Lu K.X, Wang Y (2013) Spatial variability of soil organic carbon in a typical watershed in the source area of the middle Dan River, China. Soil Research 51, 41-49. doi: 10.1071/SR12327

Yoo K, Amundson R, Heimsath AM, Dietrich WE (2006) Spatial patterns of soil organic carbon on hillslopes: integrating geomorphic processes and the biological C cycle. Geoderma 130, 47-65. doi: 10.1016/ j.geoderma.2005.01.008

Zadorova T, Zizala D, Penizek V, C ejkova S (2014) Relating extent of colluvial soils to topographic derivatives and soil variables in a luvisol sub-catchment. Central Bohemia, Czech Republic. Soil and Water Research 9, 47-57.

Zhang Q, Wang C (2010) Carbon density and distribution of six Chinese temperate forests. Science China. Life Sciences 53, 831-840. doi: 10.1007/s 11427-010-4026-0

Zhang L, Ma Z, Guo L (2008) Spatially assessing model errors of four regression techniques for three types of forest stands. Forestry 81, 209-225. doi : 10.1093/forcstry/cpn014

Zhang CS, Tang Y, Xu XL. Kiely G (2011) Towards spatial geochemical modelling: use of geographically weighted regression for mapping soil organic carbon contents in Ireland. Applied Geochemistry 26, 1239-1248. doi: 10.1016/j.apgeochem,2011.04.014

Zhong B, Xu YJ (2009) Topographic effects on soil organic carbon in Louisiana watersheds. Environmental Management 43, 662-672. doi: 10.1007/S00267-008-9182-7

Zhong L, Xiao J, Xianzhang P, Qiguo Z (2001) Organic carbon storage in soils of tropical and subtropical China. Water, Air, and Soil Pollution 129, 45-60. doi: 10.1023/A: 1010356828990 http://dx.doi.org/10.1071/SR15322

Huanyao Liu (A,B), jiaogen Zhou (A,D), Qingyu Feng (C), Yuyuan Li (A), Yong Li (A), and Jinshui Wu (A)

(A) Changsha Research Station for Agricultural & Environmental Monitoring, Institute of Subtropical Agriculture, Chinese Academy of Sciences, No. 644, Yunda 2nd Rd, Changsha 410125, China.

(B) University of Chinese Academy of Sciences, Yuquan Rd, Shijiangshan District, Beijing 100049, China.

(C) Agricultural and Biological Engineering, Purdue University, 225 South University Street, West Lafayette, IN 47907, USA.

(D) Corresponding author. Email: zhoujg@isa.ac.cn

Caption: Fig. 1. (a) Location of the study area in the Jinjing catchment, Hunan Province, China. (b) Land use types, (c) elevation map, (d) slope, (e) topographic wetness index (TWI) and (f) soil organic carbon density (SOCD) distribution.

Caption: Fig. 2. Histogram showing the relative frequency of soil organic carbon density (SOCD) in the catchment investigated in the present study.

Caption: Fig. 3. Parameters used for the semivariogram model to determine soil organic carbon density (SOCD). The exponential model was [C.sub.o] = 0.645, [C.sub.0] + [C.sub.1] = 1.063, range = 182 m.

Caption: Fig. 4. Box plots of soil organic carbon density (SOCD) for woodland (WL), paddy field (PF) and tea fields (TF). The boxes indicate the interquartile intervals (25th and 75th percentile), the line across the box indicates the median, the open square symbols indicate the mean value, and the bars represent 90% intervals (5th and 95th percentiles). Different lowercase letters denote significant differences determined by Duncan's multiple range test (P < 0.05).

Caption: Fig. 5. Mean ([+ or -] s.e.m.) soil organic carbon density (SOCD) measured in the 0-20cm soil layer at different elevations. WL, woodland; PF, paddy field; TF, tea fields. Different uppercase letters denote significant differences among elevation classes for a given land use type (superscripts PF represents paddy field and WL represents woodland; no superscripts denotes all land use types combined). The uppercase letters denote comparison among groups for a particular land use type i.e. comparison of values obtained for the same land use type at different elevation classes. The lowercase italic letters denote comparison within groups for a particular group. Different lowercase italic letters denote significant differences among land use types and the mean SOCD value for all land use types combined, within an individual elevation (P<0.05, Duncan's multiple-range test).

Caption: Fig. 6. Mean ([+ or -] s.e.m.) soil organic carbon density (SOCD) in the top 0-20 cm soil layer for different slope levels. Columns with different superscript letters differ significantly (P < 0.05).

Caption: Fig. 7. Frequency distribution of the slope according to land use types. WL, woodland; PF, paddy field; TF, tea fields.

Caption: Fig. 8. Mean ([+ or -] s.e.m.) soil organic carbon density (SOCD) in the top 0-20 cm soil layer for different topographic wetness index levels. Columns with different superscript letters differ significantly (P<0.05).

Caption: Fig. 9. Spatial distribution of soil organic carbon density (SOCD) in the catchment investigated using geographically weighted regression.

Caption: Fig. 10. Spatially varying coefficients for (a) elevation, (b) slope, (c) topographic wetness index, (d) land use types and (e) intercept in the geographically weighted regression model.
Table 1. Descriptive statistics for measured soil organic carbon
density (SOCD) in the study area (n = 1033)
K, kurtosis; S, skewness

                  Mean   Median   Minimum   Maximum   s.d.   cv

SOCD              3.36    3.53     0.31      5.67     1.07   32%
(kg [m.sup.-2])

                   K       S

SOCD              3.01   -0.52
(kg [m.sup.-2])

Table 2. Comparative performance of ordinary kriging (OK), inverse
distance weighted (IDW) interpolation, multiple linear regression
(MLR), linear mixed-effects model (LMM) and geographically weighted
regression (GWR) models on soil organic carbon density MAE, mean
absolute error; RMSF, root mean square error; SMSE, standardised mean
square error; r, Pearson correlation coefficient

Method   MAE    RMSD    SMSE     r     P-value

OK       0.86   I.I 1   4.22    0.57    0.01
IDW      0.81   0.99    14.91   0.58    0.01
MLR      0.88   1.09    14.82   0.56    0.05
LMM      0.84   1.04    9.12    0.64    0.01
GWR      0.80   1.01    3.92    0.66    0.01
COPYRIGHT 2017 CSIRO Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2017 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Liu, Huanyao; Zhou, Jiaogen; Feng, Qingyu; Li, Yuyuan; Li, Yong; Wu, Jinshui
Publication:Soil Research
Article Type:Report
Geographic Code:9CHIN
Date:Mar 1, 2017
Words:7524
Previous Article:Increases in organic carbon concentration and stock after clay addition to sands: validation of sampling methodology and effects of modification...
Next Article:Soil organic and organomineral fractions as indicators of the effects of land management in conventional and organic sugar cane systems.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters