Application of a combinatorial approach for soil organic carbon mapping in hills
Graphical abstract
Introduction
Accumulation of soil organic carbon (SOC) is essential for mitigating climate change, improving crop production, and sustaining environmental services (McBratney et al., 2003; Lal, 2004; Sanderman and Chappell, 2013). Accurately assessing SOC content is important to regulating soil carbon and adapting to global environmental changes (Florinsky et al., 2002; Zhang et al., 2014; Wiaux et al., 2014; Rovai et al., 2018). The SOC is highly spatial heterogeneous due to the spatial heterogeneities in environmental factors and human activities (Wang et al., 2009; Mishra et al., 2010; Luo et al., 2021). Therefore, accurate representing of the spatial distribution of the SOC in earth system models remains a critical scientific challenge (Zhang et al., 2012; Gelaw et al., 2014; Adhikari et al., 2020).
Many tools based on geographic information systems (GIS) have been developed and tested in analysing soil processes and derive soil properties with spatial continuity (McBratney et al., 2003; Martin et al., 2014; Were et al., 2015). Ordinary Kriging (OK) is a classic method that predicts soil properties in unsampled areas from the neighbour soil observations based on the spatial autocorrelations among measurements of soil sampling points with no environmental factors used as auxiliary variables (McBratney et al., 2003), and thus shows a limited simulation accuracy in some areas with high spatial heterogeneity (Shi et al., 2011, Li et al., 2013; Rial et al., 2017; Keskin et al., 2019). Multiple Linear Regression (MLR) is the most frequently used method for predicting SOC (Grunwald, 2009; Liu et al., 2013; Lamichhane et al., 2019) using both the environmental correlation between soil variables and environmental parameters as well as the spatial autocorrelation in the residuals of the soil variables (McBratney et al., 2003; Minasny et al., 2013; Li et al., 2013). Geographically Weighted Regression (GWR) is developed to analyze spatial heterogeneity. Both MLR and GWR have improved performance by incorporating environmental variables and reducing spatial autocorrelations, respectively. However, relationships between soil properties and environmental variables are not just linear and spatial heterogeneity (McBratney et al., 2003; Li et al., 2017), interior uncertainties such as non-linearity between soil properties and environmental variables are widespread (Drake et al., 2006; Were et al., 2015). Machine learning techniques such as Neural Network (NN) and random forest (RF) can overcome the shortcomings of parametric and non-parametric statistical methods (Sharma et al., 2003; Grimm et al., 2008) including spatial autocorrelation, non-linearity, and overfitting (Behrens et al., 2005; Drake et al., 2006; Vågen et al., 2013; Yang et al., 2016). Although the NN applied to map SOC showed good performance (Li et al., 2013; Were et al., 2015), SOC residuals within local variables still exist. High Accuracy Surface Modelling (HASM) is a new model developed in terms of the fundamental theorem of surfaces (Yue et al. 2007, 2013, 2013) and has been confirmed with advantages in handling spatial residuals (Shi et al., 2011; Li et al. 2013, 2017). These studies have promoted the performance of SOC mapping, but a recent study showed that GWR and RF models with similar performances could not predict SOC accurately under complex topography (Mirchooli et al., 2020). It is thus unlikely that a single model can be developed to be applicable in soil mapping accurately under complex environmental conditions (Mishra et al., 2010; Li et al., 2017; Keskin et al., 2019; Lamichhane et al., 2019).
Soil mapping allows to regionalizing SOC by relating it to environmental variables for improving the prediction accuracy. However, many soil-forming environmental variables are not adequately represented in current models (Lamichhane et al., 2019; Adhikari et al., 2020), and about 80% of studies used less than three types of factors (McBratney et al., 2003; Schillaci et al., 2017; Lamichhane et al., 2019). Moreover, the relationships between environmental variables and soils vary across space (Sumfleth and Duttmann, 2008; Mashimbye et al., 2014; Liu et al., 2016; Lozano-García et al., 2016). Local factors such as topography and vegetation index can lead to additional spatial variations in small areas (Song et al., 2017; Adhikari et al., 2020; Bangroo et al., 2020; Heuvelink et al., 2021; Gibson et al., 2021). The relative importance of different factors is also often region-specific and scale-dependent (Anta et al., 2020; Balkovič et al., 2020; Luo et al., 2020; Abera et al., 2021). Such difference of local factors needs to be considered in SOC mapping. Although a few researchers have developed combined models and local factors for the spatially distributed modelling of SOC (Li et al., 2017; Lamichhane et al., 2019; Heuvelink et al., 2021), how to incorporate these issues within a more generic framework for SOC prediction with high spatial resolution remains insufficient.
Hilly areas are widely distributed around the world, and provide an array of ecosystem services to humans and animals including logistics supply and ecological buffering. Previous studies have revealed the complex relationships between environmental variables and SOC in hill areas (Chaplot et al., 2010; Li et al., 2013; Gibson et al., 2021; Moura-Bueno et al., 2021). This study aims to (1) identify the relative importance of different soil-forming environmental factors (i.e. land use, soil group, parent material, topographic and vegetation index) on the SOC in a hilly area in southwest China (Li et al., 2016), and (2) develop a combined approach based on HASM and NN to map hill's SOC change more accurately by incorporating environmental variables.
The following sections introduce the materials and methods (Section 2) including study area, data sources and processing, statistical analysis, methods for SOC, and performance assessment. Then Section 3 describes and demonstrates the SOC data statistics, related environmental factors, models for SOC mapping, and models validation. Section 4 discusses drives of SOC spatiotemporal variability and the performance of the methodological framework. Finally, Section 5 provides conclusions.
Section snippets
Study area
The study area is a typical hilly zone of ecological protection for the Yangtze River in the Sichuan basin, China. It covers a 2606 km2, and located between 29°37′ and 30°20′N latitude and 103°54′ to 104°29′E longitude with elevation ranges from 345 to 972 m above sea level. This area is mainly covered by hills (shallow hills, moderate hills, and steep hills) and low mountains (Fig. 1a). The average annual rainfall and temperature are approximately 905 mm and 17.3 °C, respectively
Descriptive statistics
The statistical results of SOC based on different soil groups and land use patterns are listed in Table 2. From 1981 to 2012, the mean value of SOC increased from 6.49 g kg−1 to 13.58 g kg−1, with a declined coefficient of variation (CV). The large differences of mean SOC and CV across soil groups and land use patterns from 1981 to 2012 showed the different contribution of land uses and soil groups to SOC change. The K–S test results (Table S1) revealed that all SOC of 1981 and 2012 had a
Drivers of SOC spatiotemporal variability
Understanding the spatiotemporal variation and factors affecting SOC is a prerequisite to hypothesize future scenarios of C management on yield potential and ecosystem service (Chaplot et al., 2010; Schillaci et al., 2017; Xie et al., 2021). The mean values of topsoil SOC in the study area increased twofold (Table 2), and spatial dependence of SOC was much weaker from 1981 to 2012 (Fig. 2), which suggested dramatic change in environmental factors affecting SOC spatiotemporal variability.
Conclusions
We quantified the impacts of environmental factors on SOC variations from 1981 to 2012 in a typical hilly area of southwest China and developed a combined approach (LS_RBF_HASM) to map SOC spatial variation with improved accuracy. During the period, the mean SOC content increased from 6.49 g kg−1 to 13.58 g kg−1 with moderate spatial variation. The spatial distributions of SOC in most areas were obviously increased by 3–9 g kg−1 with the northwest of low mountain increased by 9–12 g kg−1 owing
Author contributions
QQL and CQW designed the research. QQL and YLL participated in the field campaign. YLL and QQL carried out biogeochemical analyses and data analyses. KW and HXL contributed to the language editing and revision of the paper. All co-authors contributed to the manuscript written by Youlin Luo.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
This work was supported by the National Nature Science Foundation of China (4120124) and Sichuan Province Science and Technology Support Program (2019YJ0433). We thank Aifang Chen for providing language help. We would like to thank the reviewers for their thoughtful comments and efforts towards improving this study.
References (60)
- et al.
Estimating spatially distributed SOC sequestration potentials of sustainable land management practices in Ethiopia
J. Environ. Manag.
(2021) - et al.
Importance and strength of environmental controllers of soil organic carbon changes with scale
Geoderma
(2020) - et al.
Verifiable soil organic carbon modelling to facilitate regional reporting of cropland carbon change: a test case in the Czech Republic
J. Environ. Manag.
(2020) - et al.
Application of predictor variables in spatial quantification of soil organic carbon and total nitrogen using regression kriging in the North Kashmir forest Himalayas
Catena
(2020) - et al.
Digital soil mapping of soil organic carbon stocks under different land use and land cover types in montane ecosystems, eastern Himalayas
For. Ecol. Manag.
(2014) - et al.
Prediction of soil properties by digital terrain modelling
Environ. Model. Software
(2002) - et al.
Soil organic carbon and total nitrogen stocks under different land uses in a semi-arid watershed in Tigray, northern Ethiopia
Agric. Ecosyst. Environ.
(2014) - et al.
Assessing digital elevation model resolution for soil organic carbon prediction
Geoderma
(2021) - et al.
Soil organic carbon concentrations and stocks on Barro Colorado Island—digital soil mapping using random forests analysis
Geoderma
(2008) Multi-criteria characterization of recent digital soil mapping and modeling approaches
Geoderma
(2009)
Digital mapping of soil carbon fractions with machine learning
Geoderma
Digital soil mapping algorithms and covariates for soil organic carbon mapping and their implications: a review
Geoderma
Spatially distributed modeling of soil organic matter across China: an application of artificial neural network approach
Catena
Spatiotemporal variations and factors affecting soil nitrogen in the purple hilly area of Southwest China during the 1980s and the 2010s
Sci. Total Environ.
The role of land use, construction and road on terrestrial carbon stocks in a newly urbanized area of western Chengdu, China
Landsc. Urban Plann.
Impacts of agricultural intensity on soil organic carbon pools in a main vegetable cultivation region of China
Soil Tillage Res.
Impact of topographic aspect and vegetation (native and reforested areas) on soil organic carbon and nitrogen budgets in Mediterranean natural areas
Sci. Total Environ.
Loss of organic carbon in suburban soil upon urbanization of Chengdu megacity, China
Sci. Total Environ.
Evaluation of modelling approaches for predicting the spatial distribution of soil organic carbon stocks at the national scale
Geoderma
An evaluation of digital elevation models (DEMs) for delineating land components
Geoderma
On digital soil mapping
Geoderma
Digital mapping of soil carbon
Spatial distribution dependency of soil organic carbon content to important environmental variables
Ecol. Indicat.
Environmental covariates improve the spectral predictions of organic carbon in subtropical soils in southern Brazil
Geoderma
Assessing soil organic carbon stocks under current and potentialforest cover using digital soil mapping and spatial generalisation
Ecol. Indicat.
Understanding the spatial distribution of factors controlling topsoil organic carbon content in European soils
Sci. Total Environ.
Spatio-temporal topsoil organic carbon mapping of a semi-arid Mediterranean region: the role of land use, soil texture, topographic indices and the influence of remote sensing data to modelling
Sci. Total Environ.
Neural networks for predicting nitrate-nitrogen in drainage water
Agric. Water Manag.
Surface modelling of soil pH
Geoderma
Surface modelling of soil properties based on land use information
Geoderma
Cited by (5)
Including soil depth as a predictor variable increases prediction accuracy of SOC stocks
2024, Soil and Tillage ResearchIncorporating agricultural practices in digital mapping improves prediction of cropland soil organic carbon content: The case of the Tuojiang River Basin
2023, Journal of Environmental ManagementApplication of the gleization soil layer to assess centurial drops in shallow groundwater levels across the southeastern coast of China
2022, GeodermaCitation Excerpt :Ordinary Kriging is widely used in spatial interpolation and is based on observations of a target soil variable and its corresponding spatial locations (McBratney et al., 2003). We used it in the Geostatistical Analyst extension in ArcGIS (Luo et al., 2021b) for depicting the spatial distributions of different shallow groundwater levels. All graphs were made by Origin 9.0 and ArcGIS 10.6.
Divergent responses of cropland soil organic carbon to warming across the Sichuan Basin of China
2022, Science of the Total EnvironmentCitation Excerpt :Then, in such coarse-textured soil, the balance between carbon input and SOC decomposition may have not been additionally affected by the warming-induced soil drying. The significant increase in cropland SOC suggested cropland soils played an important role in carbon sequestration over past decades in the Sichuan Basin, which is consistent with the previous studies conducted in the subareas such as the Chengdu Plain in the west of the basin (Li et al., 2020a) and the purple hilly areas in the basin (Luo et al., 2021). However, the parabolical relationship between warming and SOC increment implies the carbon sequestration capability of cropland soil may be unsustainable under climate warming.
Impact of Climate Warming on Paddy Soil Organic Carbon Change in the Sichuan Basin of China
2023, Huanjing Kexue/Environmental Science