Digital Mapping of Soil Organic Carbon Contents and Stocks in Denmark

Kabindra Adhikari; Alfred E. Hartemink; Budiman Minasny; Rania Bou Kheir; Mette B. Greve; Mogens H. Greve

doi:10.1371/journal.pone.0105519

Abstract

Estimation of carbon contents and stocks are important for carbon sequestration, greenhouse gas emissions and national carbon balance inventories. For Denmark, we modeled the vertical distribution of soil organic carbon (SOC) and bulk density, and mapped its spatial distribution at five standard soil depth intervals (0−5, 5−15, 15−30, 30−60 and 60−100 cm) using 18 environmental variables as predictors. SOC distribution was influenced by precipitation, land use, soil type, wetland, elevation, wetness index, and multi-resolution index of valley bottom flatness. The highest average SOC content of 20 g kg⁻¹ was reported for 0−5 cm soil, whereas there was on average 2.2 g SOC kg⁻¹ at 60−100 cm depth. For SOC and bulk density prediction precision decreased with soil depth, and a standard error of 2.8 g kg⁻¹ was found at 60−100 cm soil depth. Average SOC stock for 0−30 cm was 72 t ha⁻¹ and in the top 1 m there was 120 t SOC ha⁻¹. In total, the soils stored approximately 570 Tg C within the top 1 m. The soils under agriculture had the highest amount of carbon (444 Tg) followed by forest and semi-natural vegetation that contributed 11% of the total SOC stock. More than 60% of the total SOC stock was present in Podzols and Luvisols. Compared to previous estimates, our approach is more reliable as we adopted a robust quantification technique and mapped the spatial distribution of SOC stock and prediction uncertainty. The estimation was validated using common statistical indices and the data and high-resolution maps could be used for future soil carbon assessment and inventories.

Citation: Adhikari K, Hartemink AE, Minasny B, Bou Kheir R, Greve MB, Greve MH (2014) Digital Mapping of Soil Organic Carbon Contents and Stocks in Denmark. PLoS ONE 9(8): e105519. https://doi.org/10.1371/journal.pone.0105519

Editor: Dafeng Hui, Tennessee State University, United States of America

Received: May 9, 2014; Accepted: July 21, 2014; Published: August 19, 2014

Copyright: © 2014 Adhikari et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The authors confirm that, for approved reasons, some access restrictions apply to the data underlying the findings. All the soil and environmental data used in this study comes from the National Soil Database of Denmark and are not freely available. However, interested researchers could write to Mogens H. Greve (email: mogensh.greve@agrsci.dk) at the department of Agro-ecology, Aarhus University, Denmark for data access arrangements.

Funding: The study was funded by the Danish Ministry of Climate and Energy through SINKS Project (2009–2012). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

Digital Soil Mapping uses statistical tools to quantify the spatial relationship between soil property values to its environmental covariates [1]. The digital mapping of soil organic carbon (SOC) at fine resolution is a challenging task [2] and the mapping is also a high priority for SOC assessment and monitoring [3]. Spatial models on SOC prediction has a long history (e.g., [4]). A range of techniques have been used to predict and map SOC from landscape to national or continental levels and Minasny et al. [5] provided a comprehensive review. Several researchers applied splines to model the vertical distribution of SOC in the soil profiles and predicted SOC at a landscape scale using data-mining tools and environmental variables as predictors [6]–[8]. Mishra et al. [9] calculated SOC pool in each soil horizon and applied geographic weighted regression to map SOC pool at a regional scale in mid-west USA. Odgers et al. [10] used splines to derive SOC content from soil map units and predicted SOC at six standard soil depths for the entire USA. Arrouays et al. [11], Bou Kheir et al. [12], Chaplot et al. [13], Doblas-Miranda et al. [14], mapped SOC at national level using different statistical tools ranging from statistical aggregation to advanced regression tools such as regression trees.

The distribution of SOC changes across the landscape and it also varies by depth. In most soils, SOC is higher in the surface horizons and it decreases with depth. Such depth-wise variability is mostly continuous [15]–[17] except in soils with a strong human impact like some soils in the Netherlands [18]. Although most SOC studies and inventories are confined to 30 cm soil depth [19]–[21], the amount of SOC stored below 30 cm is of relevant in many ecosystems [22], [23]. For accurate quantification of SOC stocks, a depth function needs to be modeled. Several tools exist: spline function [15], [16], exponential decay function [24] or soil-type specific or profile depth functions [18], [25]. For modeling SOC with depth, the equal-area spline function has been proven to be useful in several studies [7], [8], [10]. Spline predicted SOC values with depth act as a geo-referenced point data to which environmental variables are joined and prediction models are generated using digital soil mapping techniques.

In Denmark, studies on SOC dynamics and its quantification has started after a national wide soil database was established between the years 1975–1985. Based on simple statistical scaling-up techniques, Krogh et al. [26] calculated a total stock of about 579 Tg and reported that 69% of it was stored in the soils under agriculture. Greve et al. [27] estimated topsoil SOC contents (g 100 g⁻¹) for the whole country but have not assessed the SOC stocks. Bou Kheir et al. [12] predicted the spatial extent of organic soils across Denmark. Similarly, Olesen [28] and Taghizadeh-Toosi et al. [29] estimated the stock but did not map its distribution. Most of these studies have estimated the SOC content and stock but they have not explored the spatial distribution of SOC stock nor quantified the uncertainty of the SOC predictions.

We applied digital soil mapping techniques to quantify the SOC content and stocks for Denmark. The main objectives of this study were: to model the vertical distribution of SOC content and bulk density in soil profiles, to predict and map their spatial distribution using environmental variables, to identify major environmental variables responsible for SOC distribution, to estimate the SOC stock for the soils of Denmark, and to assess the uncertainties in the SOC predictions.

Materials and Methods

Study area

Denmark is situated in Northern Europe covering an area of approximately 43,000 km². The temperate climate is characterized by a mild winters with annual mean winter and summer temperatures of about 0°C and 16°C [30]. Precipitation is well distributed throughout the year with an average annual rainfall of 800 mm in the west to 500 mm to the east. The country is relatively flat with a mean elevation above mean sea level is about 31 m. Denmark is divided into 10 physiographic regions−referred in this paper as geo-regions−based on geographical, climatic and soil-formation criteria.

The soils are coarse sandy to clayey to heavy clays as defined in the Danish Soil Classification System [31]. Soil in the western part of the country are developed in non-glaciated sandy parent material along the glacial flood-plains and Saalian moraine, whereas the soils from the central and eastern region have been developed on relatively young basal till high in finer materials [31]. Most of the soils in the north have been formed in sand mixed with uplifted marine sediments. More than 66% of the soils are classified as Podzols (Spodosols) and Luvisols (Alfisols). Podzols occupy a major portion in the west and Luvisols and Cambisols (Inceptisols, Entisols) in the central and eastern part of the country [32]. Peat soils (Histosols) occur in poorly drained basins throughout the country. About 66% of the total land area is used for agriculture with grain and potato as the main crops. Forest areas include spruce, pine and beach and these cover more than 12%.

Point data

Point (pedon) data on soil organic carbon (SOC) (g kg⁻¹) and bulk density (D_b) (Mg m⁻³) were derived from two databases: Danish Soil Classification database (DSC) and Danish Soil Profile database (DSP). DSC consists of about 36,000 point observations from the topsoil (0−20 cm) sampled randomly from agricultural fields in the period 1975−1980. From about 6,000 same pedons, soils from the subsoil (35−55 cm) was also sampled. SOC content was determined by dry combustion using a LECO IR-12 furnace. DSP consists of a grid based data (7×7 km spacing) established during the 1990s for improve fertilizer recommendation in Denmark [33]. At each 850 grid intersection, soil samples were collected based on genetic horizons and were analyzed for SOC by dry combustion. For some selected horizons, samples were taken to determine D_b. In addition, soil data from about 1100 profiles were used, and these were collected during the establishment of main gas pipeline system and other research activities across Denmark [34].

In total 40,250 topsoil point samples and 1,994 soil profiles with horizon based SOC data were used in this study. About 1,133 soil profiles included D_b measurements.

Environmental covariates

The environmental covariates data used in this study are terrain parameters from the Digital Elevation Model (DEM) of Denmark derived using Light Detection and Ranging (LiDAR) technology. The LiDAR points were interpolated using Delaunay Triangulation and the output Triangular Irregular Network (TIN) surface was converted to a grid DEM with a 1.6×1.6 m spacing. This study resampled the original 1.6 m DEM to 30.4 m using simple aggregation considering the mean value. This 30.4 m grid size was also used in the previous studies in Denmark (e.g., [32], [35]). Before aggregating to a coarser grid, the DEM was corrected by removing the pits and peaks of about 50 cm dimensions in order to ensure a regular flow on the surface. Once the DEM was processed, a number of terrain parameters (e.g., slope aspect, slope gradient, elevation, SAGA wetness index, multi-resolution index of valley bottom flatness (MrVBF), altitude above channel network, slope-length factor, over-land flow distance, and valley depth) were derived. Tools and algorithms incorporated in SAGA GIS [36] and Arc GIS V10.2 [37] were used to process the DEM and to derive these parameters.

Other covariates used were six choropleth maps which were compiled at different cartographic scales including: soil map, landscape types, geo-regions, geology, land use, and wetlands – see Table 1 and for more detail Adhikari et al. [35].

Download:

Table 1. List of environmental variables used to predict the distribution of soil organic carbon and its stock in Denmark.

https://doi.org/10.1371/journal.pone.0105519.t001

Key environmental variables affecting the spatial distribution of SOC and D_b in Denmark were identified. The relative usage of the environmental variables used during SOC and D_b prediction was calculated for each depth and their importance was expressed in percentage. This was done with Cubist software which determines the relative importance of variables based on their usage in the prediction rules. The prediction method adopted in this study is summarized in Figure 1.

Download:

Figure 1. Schematic representation of overall prediction scenario.

https://doi.org/10.1371/journal.pone.0105519.g001

Modeling SOC and D_b distribution

The vertical distribution of SOC and D_b in the soil profiles was modeled with mass preserving equal-area quadratic splines [15] in R [38]. The mathematical derivation of the spline has been described by Malone et al. [8]. As the fitting quality of splines to profile attribute data depends on a smoothing parameter−lambda (λ), we tested seven λ values (0.00001, 0.0001, 0.001, 0.01, 0.1, 1, 10) for SOC and D_b data from all the profiles and selected a λ value that showed the best fit for all the profiles using the root mean square. With increasing λ value, the fit becomes more rough. During the fit, we pegged the spline by introducing a 1 cm thick slice with a same SOC value at the topmost layer to prevent unnecessary extrapolation on the surface horizon.

Once the depth function of SOC and D_b were modeled, a weighted-average value of these properties were derived for five soil depths (0−5, 5−15, 15−30, 30−60, and 60−100 cm) based on the GlobalSoilMap specifications [39]. To these new values for the standard depths from all soil profiles, environmental variables were intersected and used for statistical analysis and spatial prediction.

Mapping to the spatial domain

Spatial prediction of SOC content and D_b at five depths was based on Regression kriging (RK) [40]. RK assumed that the spatial prediction function consists of a deterministic model formed by a regression, and the residuals of the regression (unexplained variation) are spatially correlated. The general principle of RK includes (1) regression, and (2) simple kriging of the residuals from the regression, where outputs from these two steps are added to obtain the final prediction. For the regression step, we adopted Regression-rules (RR) derived using Cubist software [41]. This tool generates a set of condition−based rules where each rule comes with a multiple regression prediction function that only operates once the conditions specified by the rule are met [42].

To build the SOC and D_b prediction model in Cubist, the data set was split randomly into two sets: 75% for calibration and 25% for model validation. Before the data split, SOC content from the topsoil observations (0−20 cm) were joined to the spline predicted SOC content from 0−5 and 5−15 cm soil depths. Similarly, SOC content from subsoil observations (35−55 cm) were attached to the spline predicted SOC at 30−60 cm soil depth. This way, a larger number of point SOC observations were incorporated to the splined SOC data from the 7×7 km grid profiles. The Cubist tool was run for log transformed SOC [log SOC g kg⁻¹] and D_b data from each depth interval and the output was converted to a regular grid map using a program written in FORTRAN. For each calibration location, the difference between measured and RR predicted value was calculated and its spatial distribution over the study area was generated using local variogram and point kriging in VESPER program [43]. This continuous residual surface was added to the corresponding RR output to get a final prediction of SOC and D_b for all five depths. Together with the prediction, a map showing the uncertainty of the prediction was generated. Both SOC prediction and uncertainty maps at each depth were then back-transformed to SOC in g kg⁻¹.

Bulk density in the peat areas

Peat lands are mostly present along the central part of the wetlands across Denmark. Bulk density in those areas were adjusted according to Greve et al. [44]. For the three surface layers (i.e., 0−5, 5−15, and 15−30 cm soil depths), D_b from 0−30 cm was used, and for the 30−60 and 60−100 cm, D_b from 30−60 cm, and 60−90 cm were used (Table 2).

Download:

Table 2. Average soil bulk density (Mg m⁻³) for different soil organic carbon levels (g 100 g⁻¹) within the central wetlands [Source: [44]].

https://doi.org/10.1371/journal.pone.0105519.t002

SOC stocks

SOC stock for each soil depth was calculated according to Eq. (1) using SOC content and D_b data. SOC stocks from the five layers were summed to obtain a SOC stock to 1 m soil depth. D_b in Eq. (1) was corrected for gravel content.(1)Where SOC_stock is the SOC stock (t ha⁻¹), SOC_content the SOC content (g kg⁻¹), D_b the soil bulk density (Mg m⁻³) and D the given soil layer thickness (cm).

Model validation

Model performance in predicting SOC content and D_b was evaluated on 25% of the point data. The following three indices that were calculated:(2)(3)(4)Where obs and pred are observed and predicted SOC and D_b values from n number of observations at i^th locations, ME is mean error, and RMSE is the root mean square error.

Predicted SOC stock was determined for all five depth intervals and then aggregated to 0−30 cm and 0−100 cm soil depth. These two C stock maps were stratified based on soil and land use types. The soil class map of Denmark based on the FAO legend [32] and a land use map (CORINE data) were used. The CORINE database for Denmark has 31 classes [45] but in this study the legend was reduced to five major land uses types: artificial surface (urban areas, industrial areas, road network and ports etc.), agricultural areas, forest and semi-natural areas, wetlands, and others (e.g., coastal lagoon, estuaries etc.) for ease of comparison to other studies (e.g., [11], [22], [26]). We also stratified our stock maps based on Danish geo-regions. We first derived the area of each soil, land use and geo-regions class based on the number of predicted pixels within those classes and then calculated an average and total SOC stock for each class at 0−30 and 0−100 cm soil depth. The 0−30 cm depth represents the plough depth from agriculture areas and estimation of carbon for this depth is of interest to farm management. The top 1 m soil depth mostly represents a rooting depth of many field crops and may act as an important soil depth section for carbon balance and accounting studies.

Results

Summary of SOC and bulk density

SOC content was highly variable and ranged from 0 to 562 g kg⁻¹ for the topsoil (0−20 cm) and from 0 to 569 g kg⁻¹ in the subsoil (Table 3). Mean SOC decreased with soil depth and SOC at 60−100 cm was about four times lower than the SOC in the 0−5 cm layer. With depth, the coefficient of variation (CV) of the SOC content increased. The CV at 0−5 cm was 184% and that for the 60−100 cm was about 466%. The SOC data was positively skewed at all soil depths with a maximum skewness coefficient at 60−100 cm. The equal-area splines modeled the depth-wise distribution and generated a continuous SOC profile to 1 m depth. The best λ value to fit all soil profiles for both SOC and D_b data was 0.1. Also average D_b was found to be increased with soil depth. Up to 30 cm depth the D_b was on average 1.44 Mg m⁻³, whereas it increased to 1.52 Mg m⁻³ below 60 cm depth. Bulk density appeared to be less variable with depth (Table 3).

Download:

Table 3. Descriptive statistics of soil organic carbon content (g kg⁻¹) and bulk density (D_b) (Mg m⁻³) data used in this study.

https://doi.org/10.1371/journal.pone.0105519.t003

Figure 2 shows a measured and spline predicted SOC and D_b for a coarse sandy soil under agriculture area from the Saalian moraine soilscape in western Denmark (West Jutland). Measured SOC from different horizons decreased with depth except at 53−73 cm where it increased due to podsolization. Although a spline should pass through the mid-point of each measured horizon for this soil profile, in this case the spline slightly extrapolated the SOC value at 35−55 cm due to the selected λ at 0.1.

Download:

Figure 2. Example of a fitted spline for soil organic carbon content (a), and for bulk density (b).

Horizontal bars represent measured soil organic carbon and bulk density at different soil horizons, continuous line through horizons represents a fitted spline, and horizontal olive-green bars give an weighted-average values of these properties at five standard soil depth intervals (i.e., 0−5, 5−15, 15−30, 30−60 and 60−100 cm).

https://doi.org/10.1371/journal.pone.0105519.g002

Prediction rules for SOC and D_b

Depending on the soil depth, 17 to 54 condition-based regression rules were generated while predicting SOC and D_b. This paper only included one of the 54 rules produced during SOC prediction at 0−5 cm soil depth, as an example.

Rule 3: [400 cases, mean 3.1, range 2.2 to 4.5, est err 0.21]
if
Georegions in (1, 2, 3, 6)
Landscape in (5, 6, 7)
MrVBF>6.6
Soil map in (3, 4, 6, 7)
then

log SOC g kg⁻¹ = 1.527+0.124*MrVBF+0.0067*Elevation+0.084*SAGA wetness index–0.0011*precipitation+0.06*slope gradient.

This rule used elevation, SAGA wetness index, MrVBF, precipitation and slope gradient to predict SOC in the areas where the MrVBF index is higher than 6.6 and consisted of fine sandy soils from a Moraine landscapes from the geo-regions (e.g. Himmerland). This rule was only valid for 400 locations where the mean SOC was 3.1 [log SOC g kg⁻¹] and the prediction error was about 0.2 [log SOC g kg⁻¹].

Identifying predictors

Variables were identified for SOC and D_b prediction based on their relative usages in the model (Table 4). In all models, precipitation appeared to be the most dominant variable followed by altitude above channel network and SAGA wetness index to predict SOC. As an example, to predict SOC at 5−15 cm, precipitation had a usage of 98% for both rule setting and developing a linear prediction model followed by SAGA wetness index which had a relative usage of 67% and 94%. For this model, insolation and slope aspect had the lowest contribution. Geology became robust with increasing soil depth, whereas land use was important for SOC prediction of the surface layers. Geology, soil map, wetlands, land use, precipitation, MrVBF, SAGA wetness index, elevation, slope gradient, slope-length factor, and altitude above channel network were among the predictors that had a relative importance of >60%. Similarly, land use, soil map, geology, slope gradient, SAGA wetness index, MrVBF, and elevation appeared to be the most important variables for predicting D_b at all soil depths.

Download:

Table 4. Relative usage (%) of the environmental variables to predict soil organic carbon at different soil depths in Denmark.

https://doi.org/10.1371/journal.pone.0105519.t004

Predicted maps

Predicted maps of SOC content (Figure 3) and D_b (map not shown) at five soil depths were produced at a resolution of 30.4×30.4 m. The highest mean SOC content was in the 0−5 cm layer (mean 20 g kg⁻¹; sd 11 g kg⁻¹). Predicted SOC content decreased with soil depth and at 60−100 cm, it was on average 2.2 g kg⁻¹. The soils of western Denmark have relatively higher SOC content than the rest of the country. The northern part has a moderate amount of SOC with two large raised bogs with high SOC contents. Along the coastline, especially in the west, soils with lower SOC concentration were mapped.

Download:

Figure 3. Predicted soil organic carbon content (a), and standard error maps (b) at five soil depths in Denmark.

https://doi.org/10.1371/journal.pone.0105519.g003

The prediction errors were higher towards the west and along the coastline. The prediction error increased with soil depth. For the 0−5 cm layer, the mean error was 1.1 g kg⁻¹ and it increased to 1.8 g kg⁻¹ at 60−100 cm soil depth.

Predicted SOC content for the FAO−UNESCO soil groups is shown in Table 5. Average SOC content ranged between 11.8 to 52.6 g kg⁻¹ at 0−30 cm and between 1.9 to 37.5 g kg⁻¹ at 30−100 cm. The highest SOC was observed in Histosols and the lowest in Arenosols at most soil depths.

Download:

Table 5. Predicted soil organic carbon content (g kg⁻¹) at five soil depths for each FAO-UNESCO soil groups.

https://doi.org/10.1371/journal.pone.0105519.t005

Model validation

SOC prediction models performance is summarized in Table 6. The best prediction was found at 5−15 cm soil depth for both training and test data sets. The model performance at 60−100 cm was relatively poor compared to the rest of the soil depths. Negative ME values suggested that almost all the prediction models were negatively biased suggesting some under prediction of the mapping of SOC distribution.

Download:

Table 6. Model performance to predict soil organic carbon content [log SOC g kg⁻¹] based on Training and Validation datasets.

https://doi.org/10.1371/journal.pone.0105519.t006

SOC stocks

SOC stock maps were made for two soil depths (Figure 4). For 0−30 cm, average SOC stocks were about 72 t ha⁻¹ and for the top 1 m depth, it was about 120 t SOC ha⁻¹. Most of the western and northern parts of the country have more than 80 t SOC ha⁻¹ in the top 30 cm whereas the average stock in the eastern part of the country was less than 80 t SOC ha⁻¹. Total SOC stock was calculated for each geo-region (Figure 5) and Himmerland and West Jutland had an average stock of about 135 t SOC ha⁻¹ followed by North Jutland and Thy both having a mean stock of >120 t SOC ha⁻¹. The soils of West Jutland and East Denmark contain almost 50% of the total SOC stock in Denmark.

Download:

Figure 4. Predicted soil organic carbon stock maps at 0−30 cm (a), and 0−100 cm (b) soil depths for Denmark.

https://doi.org/10.1371/journal.pone.0105519.g004

Download:

Figure 5. Soil organic carbon stock (1 m depth) for the geo-regions in Denmark.

Percentage values represent the fraction of the total soil organic carbon content stock (570 Tg).

https://doi.org/10.1371/journal.pone.0105519.g005

Luvisols and Podzols contain about 60% of the total SOC stock (Table 7). Other soil groups that contain significant amounts of SOC were Cambisols (6%), Gleysols (9%), and Arenosols (9%). Although Histosols had SOC stock of 176 t ha⁻¹, its total content was 20 Tg SOC. For all soil groups, more than 58% of the total stock was in the top 30 cm. Unmapped areas in the soil map representing major Danish cities that may also contain a substantial amount of carbon [46].

Download:

Table 7. Soil organic carbon stock in the top 1 m soil depth according to FAO−UNESCO soil groups.

https://doi.org/10.1371/journal.pone.0105519.t007

Of the total stock of 570 Tg SOC about 59% is in the upper 30 cm. Soils under agriculture have an average stock of 121 t ha⁻¹ and contain about 444 Tg which is almost 78% of the total estimated stock (Table 8). Another large fraction of SOC stock is found in the soils of the forest and semi-natural areas, and these had a stock of 39 Tg in the top 30 cm and about 67 Tg up to 1 m soil depth. Wetland areas contain large amounts of SOC, and average SOC stock within 1 m soil depth was about 152 t ha⁻¹ which is nearly 2% of the total stock. Almost 90% of the total SOC stock within the top 1 m soil depth is found in the soils under agriculture, forest and natural areas.

Download:

Table 8. Predicted soil organic carbon stock in different land use types derived for two different soil depths.

https://doi.org/10.1371/journal.pone.0105519.t008

Discussion

Here we have predicted the distribution of SOC contents and stocks across Denmark including an assessment of the prediction error. We have also estimated the SOC contents and stocks for different land uses and soil types. This discussion focuses on the prediction model, the importance of the variables, the uncertainty, and the SOC contents and stocks.

Prediction model

The equal-area spline fit the discrete horizon data but it also harmonized the profile by disaggregating the horizon bulk data and generated a continuous function of SOC distribution. Several other researchers have advocated the usefulness of such splines in depth-wise mapping of SOC in different parts of the world applied from local to regional extents (e.g., [7], [8], [10]). Pegging of spline by introducing an artificial horizon on the surface benefitted our splines that restricted biased extrapolation of SOC on the surface horizon.

The spatial prediction method (i.e. rule-based regression using the Cubist software) was capable of exploring the relationship of SOC to its environment predictors. The prediction rules were conditioned to a given environmental settings such that each rule is valid only to that specific boundary within which SOC distribution presumably less heterogeneous compared to the areas outside where other conditions and rules prevail. For example, SOC content in forests or clayey soils might be different than the SOC from agriculture or sandy soils hence different prediction models operated in these two specific areas. Such a beneficial and advanced function has been used by Lacoste et al. [7] who found regression-rules combined with the spline depth function useful for producing a detailed pseudo-3D map of SOC content in heterogeneous agricultural landscape in France. Minasny and McBratney [42] found that regression rules provided a greater accuracy compared to partial least squares while predicting total carbon content. Several other studies have applied this tool in digital soil mapping (e.g., [35], [47], [48]).

Variable importance.

The environmental variables used to predict SOC content showed a varying level of importance in the model. There was a large influence of precipitation, land use, soil type and some terrain parameters such as elevation, slope gradient, SAGA wetness index, and MrVBF on the spatial distribution of SOC content. The influence of topographic parameters on SOC distribution has been documented in other studies [6], [8], [9], [49]–[51]. Similarly, land use, precipitation, soil types, wetlands were found important while mapping SOC distribution [11], [14], [18], [52]–[55]. The influence of terrain parameters on SOC composition and distribution can be linked to its behavior on soil re-distribution through erosion and deposition, in the maintenance of vegetation cover and rooting depths, and in soil drainage that affects SOC decomposition as well as vegetation. In Denmark, the influence of elevation, soil types, geology, and slope gradient was also documented by Bou Kheir et al. [12] when predicting SOC in wet cultivated lands.

The categorical variables such as soil types, geo-regions, and land use were used in defining the rule conditions and continuous variables (terrain parameters) in regression functions. Some terrain parameters like elevation, SAGA wetness index, MrVBF, altitude above channel network etc. were utilized in setting rule conditions. This combined approach in defining the conditions partitioned the study area into several possible strata where SOC distribution was supposed to be more homogeneous and in each stratum several terrain parameters were again used to make sure that the within-stratum SOC variability was well captured by the model. This has made the model robust in predicting SOC content.

Uncertainty.

Based on validation indices, the prediction models showed a higher performance (i.e., higher R² and a lower RMSE) in calibration data (75%) than in validation data (25%). The uncertainty of the SOC prediction increased with depth. The R² value ranged between 0.23–0.63 - the higher values for the surface layers. It suggests that our prediction was able to capture up to 63% of total SOC variability. The range of R² values was comparable to similar SOC mapping studies where internal validation was applied [24], [50], [56], [57]. Values higher than 0.7 are unusual and values <0.5 are quite common in soil attribute predictions [58]. The difference in R² value between the two datasets could probably be attributed to the different data density used in prediction. Minasny et al. [5] reviewed several previous SOC mapping studies, and reported that with increasing data density the R² of prediction was larger. Similarly, SOC maps from the upper soil depths were associated with a low prediction error compared to the maps with depth. This could be linked to the terrain parameters used as predictors because most of these parameters explain soil surface phenomena and the uncertainty increases with depth [24]. Moreover, higher data density from the surface layers (e.g., 0−5 and 5−5 cm) might have a positive influence on prediction performance.

SOC content and stock.

SOC data from two main sources were used in this study: high-density (about 1 observation per km²) topsoil and subsoil samples, and from soil profiles at each 7-km grid-intersections together with the data from profiles along the pipe lines across Denmark. The grid and point data covered the entire country.

The SOC data were all from dry combustion analysis and from the period 1975−85 so we have modeled and predicted here the SOC contents and stocks for that period. Assuming a steady-state condition at that period, i.e. the carbon levels represent the equilibrium with its physical environment and landuse. Likely, SOC contents have changed since that time, thus the map can be used as a baseline to indicate spatio-temporal changes.

A higher SOC content was found in surface soils, and the lowest average SOC content of 3 g kg⁻¹ was found at 60−100 cm soil depth. The western part of the country mostly consists of glacial floodplains and old Saalian moraine landscapes where soils are predominantly sandy. To increase the soil fertility in those areas, farmers has been adding for decades large amounts of manure in the form of pig slurry. This may have led to increased SOC content and gave the highest SOC stock in those areas compared to other geo-regions in Denmark as also reported in Taghizadeh-Toosi et al. [29].

Our results are in line with the previous SOC estimations in Denmark. According to Krogh et al. [26], total stock in Denmark to 1 m soil depth ranged between 563−598 Tg with 579 Tg as the average which was comparable to our prediction of 570 Tg. Similarly, SOC stock from the topsoil (0−28 cm) was about 230 Tg while our prediction showed about 266 Tg. Small differences in stocks estimates between the two studies were noticed. For example, SOC stock in agricultural areas as predicted by Krogh et al. [26] was 404 Tg, whereas the estimate in the current study is 444 Tg. This could be due to a difference in areas estimated for agricultural lands in two studies. Our prediction used CORINE data that suggested an agricultural area of nearly 32,942 km², whereas this area was 28,883 km² in the previous study based on AIS (Area Information System). In a separate study, Olesen [28] reported a total stock of 604 Tg from 0−60 cm soil depth calculated for an area of 34,000 km² based on AIS. Unlike in our study where bulk density from the peat lands were adjusted, Olesen applied a standard bulk density of 1 Mg m⁻³ for all peats or organic soils and that might have increased the stock leading to over estimation.

Previous estimates of SOC stocks had not quantified the spatial distribution of SOC stock nor validated their prediction. Our approach seemed to be more reliable and the data generated could be useful for future SOC content and stock assessments.

A large amount of SOC was present in the soils under agriculture. Our estimation of 121 t ha⁻¹ is in a comparable to the SOC stock of 111 t ha⁻¹ estimated for arable lands in Scotland [59] but it was slightly higher than the arable stock in Southeast Germany [60]. Similarly, our predicted average SOC stock of 121 t ha⁻¹ from the agricultural areas was lower than the findings of Taghizadeh-Toosi et al. [29]. The later study used a soil-type dependent standard D_b established for 7×7 km grid and perhaps overestimated the SOC stock.

The estimation of average SOC stock in different soil groups within 0−30 cm depth was comparable to Arrouays et al. [11], but the total stock was approximately nine times less than reported for France (which is almost 13 times larger than Denmark). Contrary to our study, the soils under forest soils contained more SOC than the arable soils in France. Similar results were also reported by Chaplot et al. [13] from Laos. A slightly lower average stock in the soils under forest in our study might appear due to the inclusion of non-forested areas (semi-natural areas) in the same class and exclusion of litter layer that possibly lowered its value. Decade long intensive soil management practices such as addition of large amount of manure to the agricultural soils might have increased SOC contents and consequently the SOC stock from those areas [29]. Likewise, our estimated average stock for 1 m soil depth from different soil groups, for example, Cambisols (110 t ha⁻¹), and Gleysols (140 t ha⁻¹) were in a agreement with the global estimate of Batjes [22] using the same soil depth. We also observed >50% stock stored within the top 30 cm soil depth for almost all soil groups which also corresponds to the finding of Batjes [22].

Although the SOC stock in Denmark might have changed substantially over the past few decades [29], our estimation based on the available data has provided a baseline SOC. Together with the estimation of uncertainty, the maps are more reliable and could be useful in environmental research in Denmark. It could support national carbon accounting and carbon balance studies and also act as a back ground information for monitoring temporal SOC changes in Denmark.

Conclusions

This study predicted the spatial distribution of SOC content (g kg⁻¹) at five soil depths intervals (0−5, 5−15, 15−30, 30−60, and 60−100 cm) and quantified its stock (t ha⁻¹) to 1 m soil depth for Denmark. DEM derivatives and soil maps were used as predictors where condition-based regression rules were applied to quantify the spatial relationship between measured SOC and D_b with the predictors. The following can be concluded from this study:

Equal area spline modeled the continuous depth function of SOC and D_b data from discrete soil horizons in soil profiles from Denmark.
The most important variables that influenced SOC distribution across Denmark were precipitation, wetlands, land use, soil types, elevation, and saga wetness index.
Model performance was better for surface soil layers and almost all prediction models suffered from underpredictions.
The total estimated SOC stock at 0−30 cm soil depth was about 340 Tg and that for 0−100 cm was 570 Tg.
Almost 60% of the total SOC stock was found in Luvisols and Podzols.
About 90% of SOC was held in soils under agriculture, forest and semi-natural vegetation. For the soils under agriculture, 60% of the SOC was found in the top 30 cm.
West Jutland and east Denmark contained almost 50% of the total SOC stock.
This article is an example for a national level SOC mapping based on GlobalSoilMap procedures and the methods applied can be tested and used in other part of the world.

Acknowledgments

The authors thank all soil surveyors in Denmark for compiling a national Danish soil database used in this study. We are grateful to the GlobalSoilMap consortium for providing the specifications that were used in this paper to predict soil carbon. We also thank the academic editor and two reviewers for their comments on this paper. During the course of writing, we were extremely distressed as our friend and co-author Prof. Rania Bou Kheir suddenly passed away. We dedicate this paper to her.

Author Contributions

Conceived and designed the experiments: KA MHG. Performed the experiments: KA MHG. Analyzed the data: KA AEH BM MHG. Contributed reagents/materials/analysis tools: KA AEH BM RBK MBG MHG. Contributed to the writing of the manuscript: KA.

References

1. McBratney AB, Mendonça-Santos ML, Minasny B (2003) On digital soil mapping. Geoderma 117: 3–52.
- View Article
- Google Scholar
2. Hartemink AE, McSweeney K (2014) Soil Carbon. Dordrecht: Springer.
3. Hartemink AE, Lal R, Gerzabek MH, Jama B, McBratney AB, et al. (2014) Soil carbon research and global environmental challenges. PeerJ PrePrints 2: e366v361.
- View Article
- Google Scholar
4. Jenny H, Salem AE, Wallis JR (1968) Interplay of soil organic matter and soil fertility with state factors and soil properties. “Study Week on Organic Matter and Soil Fertility”. , Pontificiae Academiae Scientiarvm Scripta Varia 32: 5–36.
- View Article
- Google Scholar
5. Minasny B, McBratney AB, Malone BP, Wheeler I (2013) Digital mapping of soil carbon. Advances in Agronomy 118: 1–47.
- View Article
- Google Scholar
6. Dorji T, Odeh IOA, Field DJ, Baillie IC (2014) Digital soil mapping of soil organic carbon stocks under different land use and land cover types in montane ecosystems, Eastern Himalayas. Forest Ecology and Management 318: 91–102.
- View Article
- Google Scholar
7. Lacoste M, Minasny B, McBratney AB, Michot D, Viaud V, et al. (2014) High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape. Geoderma 213: 296–311.
- View Article
- Google Scholar
8. Malone BP, McBratney AB, Minasny B, Laslett GM (2009) Mapping continuous depth functions of soil carbon storage and available water capacity. Geoderma 154: 138–152.
- View Article
- Google Scholar
9. Mishra U, Lal R, Liu D, 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.
- View Article
- Google Scholar
10. Odgers NP, Libohova Z, Thompson JA (2012) Equal-area spline functions applied to a legacy soil database to create weighted-means maps of soil organic carbon at a continental scale. Geoderma 189–190: 153–163.
- View Article
- Google Scholar
11. Arrouays D, Deslais W, Badeau V (2001) The carbon content of topsoil and its geographical distribution in France. Soil use and Management 17: 7–11.
- View Article
- Google Scholar
12. Bou Kheir R, Greve MH, Bøcher PK, Greve MB, Larsen R, et al. (2010) Predictive mapping of soil organic carbon in wet cultivated lands using classification-tree based models: The case study of Denmark. Journal of Environmental Management 91: 1150–1160.
- View Article
- Google Scholar
13. Chaplot V, Bouahom B, Valentin C (2010) Soil organic carbon stocks in Laos: spatial variations and controlling factors. Global Change Biology 16: 1380–1393.
- View Article
- Google Scholar
14. Doblas-Miranda E, Rovira P, Brotons L, Martínez-Vilalta J, Retana J, et al. (2013) Soil carbon stocks and their variability across the forests, shrublands and grasslands of peninsular Spain. Biogeosciences 10: 8353–8361.
- View Article
- Google Scholar
15. Bishop TFA, McBratney AB, Laslett GM (1999) Modelling soil attribute depth functions with equal-area quadratic smoothing splines. Geoderma 91: 27–45.
- View Article
- Google Scholar
16. Poncehernandez R, Marriott FHC, Beckett PHT (1986) An improved method for reconstructing a soil-profile from analyses of a small number of samples. Journal of Soil Science 37: 455–467.
- View Article
- Google Scholar
17. Hartemink AE, Minasny B (2014) Towards digital soil morphometrics. Geoderma 230−231: 305–317.
- View Article
- Google Scholar
18. Kempen B, Brus DJ, Stoorvogel JJ (2011) Three-dimensional mapping of soil organic matter content using soil type-specific depth functions. Geoderma 162: 107–123.
- View Article
- Google Scholar
19. IPCC (2003) Good practice guidance for land use, land-use change and forestry. Institute for Global Environmental Strategies.
20. Janssens IA, Freibauer A, Schlamadinger B, Ceulemans R, Ciais P, et al. (2005) The carbon budget of terrestrial ecosystems at country-scale–a European case study. Biogeosciences 2: 15–26.
- View Article
- Google Scholar
21. Smith J, Smith P, Wattenbach M, Zaehle S, Hiederer R, et al. (2005) Projected changes in mineral soil carbon of European croplands and grasslands, 1990–2080. Global Change Biology 11: 2141–2152.
- View Article
- Google Scholar
22. Batjes NH (1996) Total carbon and nitrogen in the soils of the world. European Journal of Soil Science 47: 151–163.
- View Article
- Google Scholar
23. Jobbágy EG, Jackson RB (2000) The vertical distribution of soil organic carbon and its relation to climate and vegetation. Ecological applications 10: 423–436.
- View Article
- Google Scholar
24. Minasny B, McBratney AB, Mendonça-Santos ML, Odeh IOA, Guyon B (2006) Prediction and digital mapping of soil carbon storage in the Lower Namoi Valley. Australian Journal of Soil Research 44: 233–244.
- View Article
- Google Scholar
25. Mishra U, Lal R, Slater B, Calhoun F, Liu DS, et al. (2009) Predicting soil organic carbon stock using profile depth distribution functions and ordinary kriging. Soil Science Society of America Journal 73: 614–621.
- View Article
- Google Scholar
26. Krogh L, Noergaard A, Hermansen M, Greve MH, Balstroem T, et al. (2003) Preliminary estimates of contemporary soil organic carbon stocks in Denmark using multiple datasets and four scaling-up methods. Agriculture, Ecosystems & Environment 96: 19–28.
- View Article
- Google Scholar
27. Greve MH, Greve MB, Bøcher PK, Balstrom T, Madsen HB, et al. (2007) Generating a Danish raster-based topsoil property map combining choropleth maps and point information. Geografisk Tidsskrift-Danish Journal of Geography 107: 1–12.
- View Article
- Google Scholar
28. Olesen JE (1991) Forelobig beregning af CO₂ emission fra danske landbrugsjord. AJMET Notat no 24 Afdeling for Jordbrugsmeteorologi. Tjele, Denmark: Research Center Foulum, Danish Institute of Agricultural Sciences.
29. Taghizadeh-Toosi A, Olesen JE, Kristensen K, Elsgaard L, Østergaard HS, et al. (2014) Changes in carbon stocks of Danish agricultural mineral soils during 1986−2009. European Journal of Soil Science.
30. Danmarks Meteorologiske Institut (1998) Danmarks Klima 1997. Copenhagen: Danmarks Meteorologiske Institut.
31. Madsen HB, Nørr AH, Holst KA (1992) The Danish soil classification: Atlas over Denmark Copenhagen, Denmark: The Royal Danish Geographical Society.
32. Adhikari K, Minasny B, Greve MB, Greve MH (2014) Constructing a soil class map of Denmark based on the FAO legend using digital techniques. Geoderma 214–215: 101–113.
- View Article
- Google Scholar
33. Østergaard HS (1990) Kvadratnettet for nitratundersøgelser i Denmark 1986−89. Skejby, Århus: Landbrugets Rådgivningscenter, Landskontoret for Planteavl.
34. Madsen HB, Jensen NH (1985) The establishment of pedological soil databases in Denmark. Geografisk Tidsskrift-Danish Journal of Geography 85: 1–8.
- View Article
- Google Scholar
35. Adhikari K, Kheir RB, Greve MB, Bøcher PK, Malone BP, et al. (2013) High-resolution 3-D mapping of soil texture in Denmark. Soil Science Society of America Journal 77: 860–876.
- View Article
- Google Scholar
36. SAGA GIS System for Automated Geoscientific Analyses. Available: http://www.saga-gis.org.
37. ESRI (2012) ArcGIS Desktop: Release 10.1. Redlands, CA: Environmental Systems Research Institute.
38. R Development Core Team (2008) R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.
39. Arrouays D, Grundy MG, Hartemink AE, Hempel JW, Heuvelink GBM, et al. (2014) GlobalSoilMap: Toward a fine-resolution global grid of soil properties. Advances in Agronomy 125: 93–134.
- View Article
- Google Scholar
40. Odeh IOA, McBratney AB, Chittleborough DJ (1995) Further results on prediction of soil properties from terrain attributes - Heterotopic cokriging and Regression-kriging. Geoderma 67: 215–226.
- View Article
- Google Scholar
41. Quinlan JR (1993) C4. 5: Programs for machine learning. San Mateo, CA, USA: Morgan Kaufmann.
42. Minasny B, McBratney AB (2008) Regression rules as a tool for predicting soil properties from infrared reflectance spectroscopy. Chemometrics and intelligent laboratory systems 94: 72–79.
- View Article
- Google Scholar
43. Minasny B, McBratney AB, Whelan BM (2005) VESPER version 1.62. Australian Centre for Precision Agriculture, McMillan Building A05, The University of Sydney. NSW.
44. Greve MH, Christiansen OF, Greve MB, Bou Kheir R (2014) Change in peat coverage in Danish cultivated soils during the past 35 Years. Soil Science (Accepted).
45. Stjernholm M, Kjeldgaard A (2004) CORINE landcover update in Denmark-Final report. Denmark: National Environment Research Institute (NERI).
46. Vasenev VI, Stoorvogel JJ, Vasenev II, Valentini R (2014) How to map soil organic carbon stocks in highly urbanized regions? Geoderma 226–227: 103–115.
- View Article
- Google Scholar
47. Lacoste M, Lemercier B, Walter C (2011) Regional mapping of soil parent material by machine learning based on point data. Geomorphology 133: 90–99.
- View Article
- Google Scholar
48. Bui EN, Henderson BL, Viergever K (2006) Knowledge discovery from models of soil properties developed through data mining. Ecological Modelling 191: 431–446.
- View Article
- Google Scholar
49. Razakamanarivo RH, Grinand C, Razafindrakoto MA, Bernoux M, Albrecht A (2011) Mapping organic carbon stocks in eucalyptus plantations of the central highlands of Madagascar: A multiple regression approach. Geoderma 162: 335–346.
- View Article
- Google Scholar
50. Mueller TG, Pierce FJ (2003) Soil carbon maps: Enhancing spatial estimates with simple terrain attributes at multiple scales. Soil Science Society of America Journal 67: 258–267.
- View Article
- Google Scholar
51. Mendonça-Santos ML, Dart RO, Santos HG, Coelho MR, Berbara RLL, et al. (2010) Digital soil mapping of topsoil organic carbon content of Rio de Janeiro State, Brazil. In: J. L Boettinger, D. W Howell, A. C Moore, A. E Hartemink, S Kienast-Brown, editors. Digital Soil Mapping. Springer Netherlands. 255−266.
52. Ross CW, Grunwald S, Myers DB (2013) Spatiotemporal modeling of soil organic carbon stocks across a subtropical region. Science of The Total Environment 461–462: 149–157.
- View Article
- Google Scholar
53. Martin MP, Wattenbach M, Smith P, Meersmans J, Jolivet C, et al. (2011) Spatial distribution of soil organic carbon stocks in France. Biogeosciences 8: 1053–1065.
- View Article
- Google Scholar
54. Wiesmeier M, Barthold F, Blank B, Kögel-Knabner I (2011) Digital mapping of soil organic matter stocks using Random Forest modeling in a semi-arid steppe ecosystem. Plant Soil 340: 7–24.
- View Article
- Google Scholar
55. Meersmans J, van Wesemael B, Goidts E, van Molle M, De Baets S, et al. (2011) Spatial analysis of soil organic carbon evolution in Belgian croplands and grasslands, 1960–2006. Global Change Biology 17: 466–479.
- View Article
- Google Scholar
56. Bui EN, Henderson BL, Viergever K (2009) Using knowledge discovery with data mining from the Australian Soil Resource Information System database to inform soil carbon mapping in Australia. Global biogeochemical cycles 23: 1–15.
- View Article
- Google Scholar
57. Zhao Y-C, Shi X-Z (2010) Spatial prediction and uncertainty assessment of soil organic carbon in Hebei province, China. Digital Soil Mapping. Springer. 227−239.
58. Beckett P, Webster R (1971) Soil variability: a review. Soils and Fertilizers 34: 1–15.
- View Article
- Google Scholar
59. Chapman SJ, Bell JS, Campbell CD, Hudson G, Lilly A, et al. (2013) Comparison of soil carbon stocks in Scottish soils between 1978 and 2009. European Journal of Soil Science 64: 455–465.
- View Article
- Google Scholar
60. Wiesmeier M, Spörlein P, Geuβ U, Hangen E, Haug S, et al. (2012) Soil organic carbon stocks in southeast Germany (Bavaria) as affected by land use, soil type and sampling depth. Global Change Biology 18: 2233–2245.
- View Article
- Google Scholar
61. Böhner J, Antonić O (2009) Land-surface parameters specific to topo-climatology. In: T Hengl and H. I Reuter, editors. Geomorphometry- Concepts, Software, Applications. New York: Elsevier. 195−226.
62. Bendix J (2004) Geländeklimatologie – Gebrüder Borntraeger. Berlin, Stuttgart.
63. Zevenbergen LW, Thorne CR (1987) Quantitative analysis of land surface topography. Earth Surface Processes and Landforms 12: 47–56.
- View Article
- Google Scholar
64. Moore ID, Gessler PE, Nielsen GA, Peterson GA (1993) Soil attribute prediction using terrain analysis. Soil Science Society of America Journal 57: 443–452.
- View Article
- Google Scholar
65. Gallant JC, Dowling TI (2003) A multiresolution index of valley bottom flatness for mapping depositional areas. Water Resources Research 39: 1347–1359.
- View Article
- Google Scholar
66. Desmet PJJ, Govers G (1996) A GIS procedure for automatically calculating the USLE LS factor on topographically complex landscape units. Journal of Soil and Water Conservation 51: 427–433.
- View Article
- Google Scholar

[ref1] 1. McBratney AB, Mendonça-Santos ML, Minasny B (2003) On digital soil mapping. Geoderma 117: 3–52.
View Article
Google Scholar

[2] View Article

[3] Google Scholar

[ref2] 2. Hartemink AE, McSweeney K (2014) Soil Carbon. Dordrecht: Springer.

[ref3] 3. Hartemink AE, Lal R, Gerzabek MH, Jama B, McBratney AB, et al. (2014) Soil carbon research and global environmental challenges. PeerJ PrePrints 2: e366v361.
View Article
Google Scholar

[6] View Article

[7] Google Scholar

[ref4] 4. Jenny H, Salem AE, Wallis JR (1968) Interplay of soil organic matter and soil fertility with state factors and soil properties. “Study Week on Organic Matter and Soil Fertility”. , Pontificiae Academiae Scientiarvm Scripta Varia 32: 5–36.
View Article
Google Scholar

[9] View Article

[10] Google Scholar

[ref5] 5. Minasny B, McBratney AB, Malone BP, Wheeler I (2013) Digital mapping of soil carbon. Advances in Agronomy 118: 1–47.
View Article
Google Scholar

[12] View Article

[13] Google Scholar

[ref6] 6. Dorji T, Odeh IOA, Field DJ, Baillie IC (2014) Digital soil mapping of soil organic carbon stocks under different land use and land cover types in montane ecosystems, Eastern Himalayas. Forest Ecology and Management 318: 91–102.
View Article
Google Scholar

[15] View Article

[16] Google Scholar

[ref7] 7. Lacoste M, Minasny B, McBratney AB, Michot D, Viaud V, et al. (2014) High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape. Geoderma 213: 296–311.
View Article
Google Scholar

[18] View Article

[19] Google Scholar

[ref8] 8. Malone BP, McBratney AB, Minasny B, Laslett GM (2009) Mapping continuous depth functions of soil carbon storage and available water capacity. Geoderma 154: 138–152.
View Article
Google Scholar

[21] View Article

[22] Google Scholar

[ref9] 9. Mishra U, Lal R, Liu D, 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.
View Article
Google Scholar

[24] View Article

[25] Google Scholar

[ref10] 10. Odgers NP, Libohova Z, Thompson JA (2012) Equal-area spline functions applied to a legacy soil database to create weighted-means maps of soil organic carbon at a continental scale. Geoderma 189–190: 153–163.
View Article
Google Scholar

[27] View Article

[28] Google Scholar

[ref11] 11. Arrouays D, Deslais W, Badeau V (2001) The carbon content of topsoil and its geographical distribution in France. Soil use and Management 17: 7–11.
View Article
Google Scholar

[30] View Article

[31] Google Scholar

[ref12] 12. Bou Kheir R, Greve MH, Bøcher PK, Greve MB, Larsen R, et al. (2010) Predictive mapping of soil organic carbon in wet cultivated lands using classification-tree based models: The case study of Denmark. Journal of Environmental Management 91: 1150–1160.
View Article
Google Scholar

[33] View Article

[34] Google Scholar

[ref13] 13. Chaplot V, Bouahom B, Valentin C (2010) Soil organic carbon stocks in Laos: spatial variations and controlling factors. Global Change Biology 16: 1380–1393.
View Article
Google Scholar

[36] View Article

[37] Google Scholar

[ref14] 14. Doblas-Miranda E, Rovira P, Brotons L, Martínez-Vilalta J, Retana J, et al. (2013) Soil carbon stocks and their variability across the forests, shrublands and grasslands of peninsular Spain. Biogeosciences 10: 8353–8361.
View Article
Google Scholar

[39] View Article

[40] Google Scholar

[ref15] 15. Bishop TFA, McBratney AB, Laslett GM (1999) Modelling soil attribute depth functions with equal-area quadratic smoothing splines. Geoderma 91: 27–45.
View Article
Google Scholar

[42] View Article

[43] Google Scholar

[ref16] 16. Poncehernandez R, Marriott FHC, Beckett PHT (1986) An improved method for reconstructing a soil-profile from analyses of a small number of samples. Journal of Soil Science 37: 455–467.
View Article
Google Scholar

[45] View Article

[46] Google Scholar

[ref17] 17. Hartemink AE, Minasny B (2014) Towards digital soil morphometrics. Geoderma 230−231: 305–317.
View Article
Google Scholar

[48] View Article

[49] Google Scholar

[ref18] 18. Kempen B, Brus DJ, Stoorvogel JJ (2011) Three-dimensional mapping of soil organic matter content using soil type-specific depth functions. Geoderma 162: 107–123.
View Article
Google Scholar

[51] View Article

[52] Google Scholar

[ref19] 19. IPCC (2003) Good practice guidance for land use, land-use change and forestry. Institute for Global Environmental Strategies.

[ref20] 20. Janssens IA, Freibauer A, Schlamadinger B, Ceulemans R, Ciais P, et al. (2005) The carbon budget of terrestrial ecosystems at country-scale–a European case study. Biogeosciences 2: 15–26.
View Article
Google Scholar

[55] View Article

[56] Google Scholar

[ref21] 21. Smith J, Smith P, Wattenbach M, Zaehle S, Hiederer R, et al. (2005) Projected changes in mineral soil carbon of European croplands and grasslands, 1990–2080. Global Change Biology 11: 2141–2152.
View Article
Google Scholar

[58] View Article

[59] Google Scholar

[ref22] 22. Batjes NH (1996) Total carbon and nitrogen in the soils of the world. European Journal of Soil Science 47: 151–163.
View Article
Google Scholar

[61] View Article

[62] Google Scholar

[ref23] 23. Jobbágy EG, Jackson RB (2000) The vertical distribution of soil organic carbon and its relation to climate and vegetation. Ecological applications 10: 423–436.
View Article
Google Scholar

[64] View Article

[65] Google Scholar

[ref24] 24. Minasny B, McBratney AB, Mendonça-Santos ML, Odeh IOA, Guyon B (2006) Prediction and digital mapping of soil carbon storage in the Lower Namoi Valley. Australian Journal of Soil Research 44: 233–244.
View Article
Google Scholar

[67] View Article

[68] Google Scholar

[ref25] 25. Mishra U, Lal R, Slater B, Calhoun F, Liu DS, et al. (2009) Predicting soil organic carbon stock using profile depth distribution functions and ordinary kriging. Soil Science Society of America Journal 73: 614–621.
View Article
Google Scholar

[70] View Article

[71] Google Scholar

[ref26] 26. Krogh L, Noergaard A, Hermansen M, Greve MH, Balstroem T, et al. (2003) Preliminary estimates of contemporary soil organic carbon stocks in Denmark using multiple datasets and four scaling-up methods. Agriculture, Ecosystems & Environment 96: 19–28.
View Article
Google Scholar

[73] View Article

[74] Google Scholar

[ref27] 27. Greve MH, Greve MB, Bøcher PK, Balstrom T, Madsen HB, et al. (2007) Generating a Danish raster-based topsoil property map combining choropleth maps and point information. Geografisk Tidsskrift-Danish Journal of Geography 107: 1–12.
View Article
Google Scholar

[76] View Article

[77] Google Scholar

[ref28] 28. Olesen JE (1991) Forelobig beregning af CO₂ emission fra danske landbrugsjord. AJMET Notat no 24 Afdeling for Jordbrugsmeteorologi. Tjele, Denmark: Research Center Foulum, Danish Institute of Agricultural Sciences.

[ref29] 29. Taghizadeh-Toosi A, Olesen JE, Kristensen K, Elsgaard L, Østergaard HS, et al. (2014) Changes in carbon stocks of Danish agricultural mineral soils during 1986−2009. European Journal of Soil Science.

[ref30] 30. Danmarks Meteorologiske Institut (1998) Danmarks Klima 1997. Copenhagen: Danmarks Meteorologiske Institut.

[ref31] 31. Madsen HB, Nørr AH, Holst KA (1992) The Danish soil classification: Atlas over Denmark Copenhagen, Denmark: The Royal Danish Geographical Society.

[ref32] 32. Adhikari K, Minasny B, Greve MB, Greve MH (2014) Constructing a soil class map of Denmark based on the FAO legend using digital techniques. Geoderma 214–215: 101–113.
View Article
Google Scholar

[83] View Article

[84] Google Scholar

[ref33] 33. Østergaard HS (1990) Kvadratnettet for nitratundersøgelser i Denmark 1986−89. Skejby, Århus: Landbrugets Rådgivningscenter, Landskontoret for Planteavl.

[ref34] 34. Madsen HB, Jensen NH (1985) The establishment of pedological soil databases in Denmark. Geografisk Tidsskrift-Danish Journal of Geography 85: 1–8.
View Article
Google Scholar

[87] View Article

[88] Google Scholar

[ref35] 35. Adhikari K, Kheir RB, Greve MB, Bøcher PK, Malone BP, et al. (2013) High-resolution 3-D mapping of soil texture in Denmark. Soil Science Society of America Journal 77: 860–876.
View Article
Google Scholar

[90] View Article

[91] Google Scholar

[ref36] 36. SAGA GIS System for Automated Geoscientific Analyses. Available: http://www.saga-gis.org.

[ref37] 37. ESRI (2012) ArcGIS Desktop: Release 10.1. Redlands, CA: Environmental Systems Research Institute.

[ref38] 38. R Development Core Team (2008) R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.

[ref39] 39. Arrouays D, Grundy MG, Hartemink AE, Hempel JW, Heuvelink GBM, et al. (2014) GlobalSoilMap: Toward a fine-resolution global grid of soil properties. Advances in Agronomy 125: 93–134.
View Article
Google Scholar

[96] View Article

[97] Google Scholar

[ref40] 40. Odeh IOA, McBratney AB, Chittleborough DJ (1995) Further results on prediction of soil properties from terrain attributes - Heterotopic cokriging and Regression-kriging. Geoderma 67: 215–226.
View Article
Google Scholar

[99] View Article

[100] Google Scholar

[ref41] 41. Quinlan JR (1993) C4. 5: Programs for machine learning. San Mateo, CA, USA: Morgan Kaufmann.

[ref42] 42. Minasny B, McBratney AB (2008) Regression rules as a tool for predicting soil properties from infrared reflectance spectroscopy. Chemometrics and intelligent laboratory systems 94: 72–79.
View Article
Google Scholar

[103] View Article

[104] Google Scholar

[ref43] 43. Minasny B, McBratney AB, Whelan BM (2005) VESPER version 1.62. Australian Centre for Precision Agriculture, McMillan Building A05, The University of Sydney. NSW.

[ref44] 44. Greve MH, Christiansen OF, Greve MB, Bou Kheir R (2014) Change in peat coverage in Danish cultivated soils during the past 35 Years. Soil Science (Accepted).

[ref45] 45. Stjernholm M, Kjeldgaard A (2004) CORINE landcover update in Denmark-Final report. Denmark: National Environment Research Institute (NERI).

[ref46] 46. Vasenev VI, Stoorvogel JJ, Vasenev II, Valentini R (2014) How to map soil organic carbon stocks in highly urbanized regions? Geoderma 226–227: 103–115.
View Article
Google Scholar

[109] View Article

[110] Google Scholar

[ref47] 47. Lacoste M, Lemercier B, Walter C (2011) Regional mapping of soil parent material by machine learning based on point data. Geomorphology 133: 90–99.
View Article
Google Scholar

[112] View Article

[113] Google Scholar

[ref48] 48. Bui EN, Henderson BL, Viergever K (2006) Knowledge discovery from models of soil properties developed through data mining. Ecological Modelling 191: 431–446.
View Article
Google Scholar

[115] View Article

[116] Google Scholar

[ref49] 49. Razakamanarivo RH, Grinand C, Razafindrakoto MA, Bernoux M, Albrecht A (2011) Mapping organic carbon stocks in eucalyptus plantations of the central highlands of Madagascar: A multiple regression approach. Geoderma 162: 335–346.
View Article
Google Scholar

[118] View Article

[119] Google Scholar

[ref50] 50. Mueller TG, Pierce FJ (2003) Soil carbon maps: Enhancing spatial estimates with simple terrain attributes at multiple scales. Soil Science Society of America Journal 67: 258–267.
View Article
Google Scholar

[121] View Article

[122] Google Scholar

[ref51] 51. Mendonça-Santos ML, Dart RO, Santos HG, Coelho MR, Berbara RLL, et al. (2010) Digital soil mapping of topsoil organic carbon content of Rio de Janeiro State, Brazil. In: J. L Boettinger, D. W Howell, A. C Moore, A. E Hartemink, S Kienast-Brown, editors. Digital Soil Mapping. Springer Netherlands. 255−266.

[ref52] 52. Ross CW, Grunwald S, Myers DB (2013) Spatiotemporal modeling of soil organic carbon stocks across a subtropical region. Science of The Total Environment 461–462: 149–157.
View Article
Google Scholar

[125] View Article

[126] Google Scholar

[ref53] 53. Martin MP, Wattenbach M, Smith P, Meersmans J, Jolivet C, et al. (2011) Spatial distribution of soil organic carbon stocks in France. Biogeosciences 8: 1053–1065.
View Article
Google Scholar

[128] View Article

[129] Google Scholar

[ref54] 54. Wiesmeier M, Barthold F, Blank B, Kögel-Knabner I (2011) Digital mapping of soil organic matter stocks using Random Forest modeling in a semi-arid steppe ecosystem. Plant Soil 340: 7–24.
View Article
Google Scholar

[131] View Article

[132] Google Scholar

[ref55] 55. Meersmans J, van Wesemael B, Goidts E, van Molle M, De Baets S, et al. (2011) Spatial analysis of soil organic carbon evolution in Belgian croplands and grasslands, 1960–2006. Global Change Biology 17: 466–479.
View Article
Google Scholar

[134] View Article

[135] Google Scholar

[ref56] 56. Bui EN, Henderson BL, Viergever K (2009) Using knowledge discovery with data mining from the Australian Soil Resource Information System database to inform soil carbon mapping in Australia. Global biogeochemical cycles 23: 1–15.
View Article
Google Scholar

[137] View Article

[138] Google Scholar

[ref57] 57. Zhao Y-C, Shi X-Z (2010) Spatial prediction and uncertainty assessment of soil organic carbon in Hebei province, China. Digital Soil Mapping. Springer. 227−239.

[ref58] 58. Beckett P, Webster R (1971) Soil variability: a review. Soils and Fertilizers 34: 1–15.
View Article
Google Scholar

[141] View Article

[142] Google Scholar

[ref59] 59. Chapman SJ, Bell JS, Campbell CD, Hudson G, Lilly A, et al. (2013) Comparison of soil carbon stocks in Scottish soils between 1978 and 2009. European Journal of Soil Science 64: 455–465.
View Article
Google Scholar

[144] View Article

[145] Google Scholar

[ref60] 60. Wiesmeier M, Spörlein P, Geuβ U, Hangen E, Haug S, et al. (2012) Soil organic carbon stocks in southeast Germany (Bavaria) as affected by land use, soil type and sampling depth. Global Change Biology 18: 2233–2245.
View Article
Google Scholar

[147] View Article

[148] Google Scholar

[ref61] 61. Böhner J, Antonić O (2009) Land-surface parameters specific to topo-climatology. In: T Hengl and H. I Reuter, editors. Geomorphometry- Concepts, Software, Applications. New York: Elsevier. 195−226.

[ref62] 62. Bendix J (2004) Geländeklimatologie – Gebrüder Borntraeger. Berlin, Stuttgart.

[ref63] 63. Zevenbergen LW, Thorne CR (1987) Quantitative analysis of land surface topography. Earth Surface Processes and Landforms 12: 47–56.
View Article
Google Scholar

[152] View Article

[153] Google Scholar

[ref64] 64. Moore ID, Gessler PE, Nielsen GA, Peterson GA (1993) Soil attribute prediction using terrain analysis. Soil Science Society of America Journal 57: 443–452.
View Article
Google Scholar

[155] View Article

[156] Google Scholar

[ref65] 65. Gallant JC, Dowling TI (2003) A multiresolution index of valley bottom flatness for mapping depositional areas. Water Resources Research 39: 1347–1359.
View Article
Google Scholar

[158] View Article

[159] Google Scholar

[ref66] 66. Desmet PJJ, Govers G (1996) A GIS procedure for automatically calculating the USLE LS factor on topographically complex landscape units. Journal of Soil and Water Conservation 51: 427–433.
View Article
Google Scholar

[161] View Article

[162] Google Scholar

Figures

Abstract

Introduction

Materials and Methods

Study area

Point data

Environmental covariates

Modeling SOC and Db distribution

Mapping to the spatial domain

Bulk density in the peat areas

SOC stocks

Model validation

Results

Summary of SOC and bulk density

Prediction rules for SOC and Db

Identifying predictors

Predicted maps

Model validation

SOC stocks

Discussion

Prediction model

Variable importance.

Uncertainty.

SOC content and stock.

Conclusions

Acknowledgments

Author Contributions

References

Modeling SOC and D_b distribution

Prediction rules for SOC and D_b