EN中文

Journal of Geographical Sciences >

2016 , Vol. 26 >Issue 1: 102 - 124

DOI: https://doi.org/10.1007/s11442-016-1257-4

Orginal Article

Spatially distributed modelling and mapping of soil organic carbon and total nitrogen stocks in the Eastern Mau Forest Reserve, Kenya

  • Kennedy WERE , 1, 2 ,
  • Bal Ram SINGH 3 ,
  • Øystein Bjarne DICK 1
Expand
  • 1. Department of Mathematical Sciences and Technology, Norwegian University of Life Sciences, P.O. Box 5003, NO-1432, Ås, Norway
  • 2. Kenya Agricultural and Livestock Research Organisation, Kenya Soil Survey, P.O. Box 14733-00800, Nairobi, Kenya
  • 3. Department of Environmental Sciences, Norwegian University of Life Sciences, P.O. Box 5003, NO-1432, Ås, Norway

Author: Kennedy Were, PhD, specialized in application of GIS and remote sensing techniques in environmental research. E-mail:

Received date: 2015-02-18

  Accepted date: 2015-06-12

  Online published: 2016-01-25

Copyright

Journal of Geographical Sciences, All Rights Reserved
Fold

Abstract

Detailed knowledge about the estimates and spatial patterns of soil organic carbon (SOC) and total nitrogen (TN) stocks is fundamental for sustainable land management and climate change mitigation. This study aimed at: (1) mapping the spatial patterns, and (2) quantifying SOC and TN stocks to 30 cm depth in the Eastern Mau Forest Reserve using field, remote sensing, geographical information systems (GIS), and statistical modelling approaches. This is a critical ecosystem offering essential services, but its sustainability is threatened by deforestation and degradation. Results revealed that elevation, silt content, TN concentration, and Landsat 8 Operational Land Imager band 11 explained 72% of the variability in SOC stocks, while the same factors (except silt content) explained 71% of the variability in TN stocks. The results further showed that soil properties, particularly TN and SOC concentrations, were more important than that other environmental factors in controlling the observed patterns of SOC and TN stocks, respectively. Forests stored the highest amounts of SOC and TN (3.78 Tg C and 0.38 Tg N) followed by croplands (2.46 Tg C and 0.25 Tg N) and grasslands (0.57 Tg C and 0.06 Tg N). Overall, the Eastern Mau Forest Reserve stored approximately 6.81 Tg C and 0.69 Tg N. The highest estimates of SOC and TN stocks (hotspots) occurred on the western and northwestern parts where forests dominated, while the lowest estimates (coldspots) occurred on the eastern side where croplands had been established. Therefore, the hotspots need policies that promote conservation, while the coldspots need those that support accumulation of SOC and TN stocks.

Key words： soil organic carbon; total nitrogen; carbon sequestration; climate change; digital soil mapping; Eastern Mau

Cite this article

Kennedy WERE , Bal Ram SINGH , Øystein Bjarne DICK . Spatially distributed modelling and mapping of soil organic carbon and total nitrogen stocks in the Eastern Mau Forest Reserve, Kenya[J]. Journal of Geographical Sciences, 2016 , 26(1) : 102 -124 . DOI: 10.1007/s11442-016-1257-4

1 Introduction

Soil organic carbon (SOC) and total nitrogen (TN) are key determinants of biogeochemical cycling, as well as soil quality and properties (Obade and Lal, 2013; Wang et al., 2013; Yang et al., 2014). They vary spatially and temporally in response to a mix of climatic, edaphic, biotic, topographical, and lithological factors. Such dynamics also affect the contributions of SOC and TN to atmospheric greenhouse gases, particularly carbon dioxide (CO2) and nitrous oxide (N2O). The world soils contain about 1500 petagrams of carbon (Pg C) to 1 m depth (1 Pg = 1015 g), which is twice the amount of C in the atmospheric pool and three times the amount in the biotic pool (Lal, 2004; Smith, 2004, 2008). This implies that even slight changes in SOC pool can significantly affect global C cycle and climate. Therefore, current research is geared towards quantifying and mapping SOC and TN stocks, in space and over time, with a view to understanding climate change and land degradation processes. Unfortunately, the traditional soil mapping techniques are expensive, time-consuming, and yield coarse qualitative information (Mora-Vallejo et al., 2008; Mehrjardi et al., 2014). Consequently, there is increasing effort in the emerging field of digital soil mapping (DSM) to develop, evaluate, and apply new techniques for spatial prediction and mapping of soil properties.
The existing DSM techniques fall into two categories, namely (1) measure and multiply (MM) and (2) soil-landscape modelling (SLM) techniques (Mishra et al., 2010; Cambule et al., 2014). In MM approach, the study area is stratified and then the point estimates of a target soil property within a stratum are averaged and multiplied by the stratum’s area. In contrast, in SLM approach, the spatial variability of a target soil property is explained by its relationships with soil-forming factors, such as topography, climate, land use, vegetation, parent material, and soil type. In particular, field observations and ancillary environmental data are used to calibrate an empirical model, which is then applied to generate a prediction surface of the target soil variable (Mishra et al., 2010; Li et al., 2013b; Cambule et al., 2014). SLM approach has been boosted by the improvements in computer technology and accessibility to inexpensive environmental data from remote sensors and existing spatial databases. Although MM approach is simple, it ignores the complex interactions of environmental factors with the target soil variables, which account for the spatial variability. Thus, MM approach yields predictions with lesser accuracy than SLM approach. McKenzie and Ryan (1999), McBratney et al. (2003), and Scull et al. (2003) have provided detailed reviews of DSM.
Literature is replete with examples of SLM techniques that have been applied so far to model and map the spatial patterns of SOC and TN stocks. The techniques range from multiple linear regression (Lesch and Corwin, 2008; Meersmans et al., 2008) and partial least square regression (Selige et al., 2006; Amare et al., 2013) to generalized linear models (Yang et al., 2008), classification and regression trees (Kheir et al., 2010; Martin et al., 2011; Razakamanarivo et al., 2011), kriging (Wu et al., 2009; Zhang et al., 2010; Liu et al., 2011; Li et al., 2013a; Cambule et al., 2014), regression-kriging (Hengl et al., 2004, 2007; Lamsal et al., 2006; Mora-Vallejo et al., 2008; Sumfleth and Duttmann, 2008; Li, 2010; Vasques et al., 2010a, 2010b; Dorji et al., 2014; Martin et al., 2014), geographically weighted regression (Mishra et al., 2010; Zhang et al., 2011; Mishra and Riley, 2012; Kumar et al., 2013; Wang et al., 2013), geographically weighted regression-kriging (Kumar et al., 2012), neural networks (Malone et al., 2009; Jaber and Al-Qinna, 2011; Li et al., 2013b), random forests (Grimm et al., 2008; Vågen and Winowiecki, 2013a, 2013b), rule-based models (Lacoste et al., 2014), and linear mixed models (Doetterl et al., 2013; Karunaratne et al., 2014). Of these techniques, multiple linear regression (MLR) is most popular because of its simplicity, computational efficiency, and straightforward interpretation (Li et al., 2013b). However, its assumptions of spatial stationarity in the effects of environmental variables and spatial independence in the target soil properties are mostly violated leading to misspecification of prediction models. Hybrid methods, particularly regression-kriging (MLRK), which combines ordinary kriging with MLR are also gaining currency in digital soil mapping because of their detailed results and lower prediction errors compared to pure geostatistical, or statistical methods (Hengl et al., 2004). Geographically weighted regression (GWR) is the most recent technique, which has drawn the attention of environmental scientists. GWR was designed to deal with the spatially varying relationships between the target and environmental variables (i.e., spatial non-stationarity); hence, the estimated parameters also vary spatially (Wang et al., 2013). Even though some comparative studies have shown that it outperforms MLRK in spatial prediction of SOC stocks (Mishra et al., 2010), the application of GWR is still limited. Few studies have also attempted to couple GWR with kriging (geographically weighted regression-kriging; GWRK) to predict the spatial distribution of environmental phenomena; for example, urban heat island in Wrocław, Poland (Syzomanoski and Kryza, 2012) and SOC stocks in Pennsylvania State, USA (Kumar and Lal, 2011; Kumar et al., 2012).
The objective of this study was to estimate and map the spatial distribution of SOC and TN stocks to 30 cm depth in the Eastern Mau Forest Reserve by integrating field sampling, remote sensing, geographical information systems (GIS), and statistical modelling. The 30 cm depth is consistent with the Intergovernmental Panel on Climate Change (IPCC) guidelines (IPCC, 2006). The Eastern Mau Forest Reserve was selected because it had undergone wanton deforestation and degradation since the mid-1990s owing to ill-advised forest excisions and illegal loggings, encroachments, and charcoal burning (Government of Kenya 2009; UNEP 2009). Despite this, no complete studies had been undertaken to quantify the storage and map the spatial patterns of SOC and TN. This study aimed to bridge this gap and contribute information for designing spatially-targeted, effective, and sustainable strategies for ecosystem restoration and management.

2 Materials and methods

2.1 Study area

The Eastern Mau Forest Reserve, which covers approximately 650 km2, is part of East Africa’s largest closed-canopy indigenous montane forest, and Kenya’s key water catchment area. It is bounded by the latitudes 0º15´-0º40´S and the longitudes 35º40´-36º10´E (Figure 1) with the altitudes ranging from 2210 to 3070 m above sea level. The climate is cool and humid; that is, the average annual temperatures vary from 9.8°C to 17.5°C, while the average annual rainfall ranges between 935 and 1287 mm (Jaetzold et al., 2010). The rainfall distribution pattern is tri-modal with peaks in April, August, and November. The Njoro, Naishi, and Larmudiac Rivers drain the eastern slopes into Lake Nakuru, while the Nessuiet and Rongai flow northwards into Lake Bogoria, and Baringo, respectively. The area’s physiography is characterized by major scarps and uplands comprising pyroclastic rocks (e.g., pumice tuffs) of Tertiary-Quaternary volcanic age. These soft light brown rocks have insets of yellow pumice and angular trachyte, which decompose into deep to very deep, and dark reddish brown clayey soil aggregates (McCall, 1967). The soils, classified as Mollic Andosols by FAO, are friable and smeary with humic topsoils (Jaetzold et al., 2010). The dominant land cover types are forests, grasslands, and croplands (Figure 1). The floristic composition of forests and grasslands comprise indigenous tree species, such as Prunus africana, Arundinaria alpina, Juniperus procera, Olea europaea ssp. africana, Olea capensis ssp. hochstetteri, Podocarpus latifolius, Nuxia congesta, Clematis hirsuta, Schefflera volkensii, and Dombeya torrida, exotic tree species like Pinus patula and Cupressus lusitanica, and grass species, such as Pennisetum clandestinum. The major crops grown are maize (Zea mays), beans (Phaseolus vulgaris), wheat (Triticum aestivum), and potatoes (Solanum tuberosum) (Were et al., 2015).
Figure 1 Geographical location of the study area

2.2 Data sources and pre-processing

Figure 2 summarizes the data sources and spatial modelling framework of this study. The overall methodology involved seven major steps: (1) soil sampling and analysis, (2) preparation of the environmental predictors and target soil variables, (3) calibration of the regression-based models, (4) application of the models, (5) interpolation of the regression-based residuals and their addition to the fitted trend surfaces, (6) validation, and (7) production of the thematic maps for SOC and TN stocks.
Figure 2 Illustration of the data sources and modelling framework
2.2.1 Soil sampling and analysis
The soil sampling campaign was conducted between June and August 2012. Before the campaign, sampling points were generated in a completely randomized design using agro-ecological zones map as the base in a GIS environment. A map showing the distribution of these sampling points was created and used in the field. At each sampling point, an auger was used to collect samples at 0-15 cm and 15-30 cm depths from the centre and corners of a plot measuring 30 m × 30 m. The samples taken from corresponding depths in a plot were mixed thoroughly and bulked into one composite sample of about 500 g. To determine bulk density (BD), a core sampler (5 cm in diameter and 5 cm in height) was used to collect one undisturbed sample at the centre of each plot and at each depth. The geographical coordinates, elevation, vegetation, and land management practices were also recorded. A total of 320 soil samples were collected from 160 sampling plots to analyze the chemical and physical properties, and a similar number to determine BD at the National Agricultural Research Laboratories. Supplementary soil data that had been collected similarly from 60 other sampling plots to assess the effects of land cover changes on SOC and TN stocks (Were et al., 2015) were also used. Overall, soil data from 220 sampling plots (Figure 1) were used to model the spatial distribution of SOC and TN stocks.
The soil samples were air-dried, ground, and sieved through a 2 mm mesh. SOC concentrations, TN concentrations, and BD were then determined using the Walkley-Black wet oxidation method (Nelson and Sommers, 1982), Kjeldahl digestion method (Bremner and Mulvaney, 1982), and core method (Blake, 1965), respectively. These three properties were used to calculate SOC and TN stocks (i.e., the target variables) at each depth. Additional soil properties were also analyzed. The hydrometer method (Day, 1965) was used to determine particle size distribution, while the Mehlich method (Okalebo et al., 2002) was used to estimate phosphorous (P) content. A flame-photometer was used to measure potassium (K) content, an atomic absorption spectrophotometer to measure the contents of calcium (Ca) and magnesium (Mg), and a pH meter to measure pH (1:2.5 soil-water) (Okalebo et al., 2002).
Eq. (1) was used to calculate SOC stocks (Mg C ha-1) for each depth (Aynekulu et al., 2011):
where SOCst is the soil organic carbon stock (Mg C ha-1), SOC is the soil organic carbon concentration (%, which is then converted to g C g-1 soil), BD is the bulk density (g cm-3), D is the depth (cm), and 100 is the multiplication factor to convert the SOC per unit area from g C cm-2 to Mg C ha-1. Stone contents were negligible due to the softness of the volcanic rocks; hence, are not accounted for in Eq. (1). Similarly, TN stocks (TNst; Mg N ha-1) for each depth were computed by substituting TN for SOC in Eq. (1). The SOC and TN stocks in the surface (0-15 cm) and subsurface soils (15-30 cm) were summed up to obtain the total stocks to 30 cm depth.
2.2.2 Remote sensing and GIS analysis
Twenty candidate environmental predictors that had been selected a priori based on the scorpan conceptual model (McBratney et al., 2003) were obtained from existing spatial databases and fieldwork. The scorpan model captures six key soil-forming factors; namely, soil properties (s), climate (c), organisms (o), topography (r), parent material (p), age (a), and space (n). Table 1 provides the sources of temperature, rainfall, land cover, elevation (digital elevation model; DEM), and Landsat 8 Operational Land Imager (OLI) data. Although Landsat 8 OLI imagery was captured in 2013, whereas fieldwork was conducted in 2012 (i.e., when Landsat 5 TM operational imaging had ended, and before the launch of Landsat 8 OLI), it was assumed that soil properties had not changed significantly within such a short time to affect their spectral response. Slope, curvature, aspect, and compound topographic index (CTI) were also extracted from the DEM. Eq. (2) was used to extract CTI:
where As is the upslope area and β is the slope (McKenzie and Ryan, 1999).
Table 1 Properties of the environmental predictors for spatial modelling
Variables Data format Date Source Scale Soil-forming factor
Target variables
1. SOC stocks Points 2012 Field work
2. TN stocks Points 2012 Field work
Predictor variables
1. SOC concentration Raster 2012 Interpolated field data 30 m S
2. TN concentration Raster 2012 Interpolated field data 30 m S
3. Magnesium Raster 2012 Interpolated field data 30 m S
4. Potassium Raster 2012 Interpolated field data 30 m S
5. Calcium Raster 2012 Interpolated field data 30 m S
6. Clay content Raster 2012 Interpolated field data 30 m S
7. Silt content Raster 2012 Interpolated field data 30 m S
8. Sand content Raster 2012 Interpolated field data 30 m S
9. pH Raster 2012 Interpolated field data 30 m S
10. Elevation Raster - ASTER GDEM
http://gdem.ersdac.jspacesystems.or.jp/		 30 m R
11. Slope Raster - ASTER GDEM 30 m R
12. Aspect Raster - ASTER GDEM 30 m R
13. Curvature Raster - ASTER GDEM 30 m R
14. CTI Raster - ASTER GDEM 30 m S
15. Temperature Raster 1950-2000 www.worldclim.org 1 km C
16. Rainfall Raster 1950-2000 www.worldclim.org 1 km C
17. Surface reflectance & thermal emission Raster 30.05.2013 Landsat 8 OLI (bands 2, 3, 4, 5, 6, 7, 10 & 11) http://earthexplorer.usgs.gov/ 30 m C, S
18. NDVI Raster 30.05.2013 Landsat 8 OLI (bands 4 & 5) 30 m O
19. PC bands Raster 30.05.2013 Landsat 8 OLI (bands 2, 3, 4, 5, 6 & 7) 30 m S
20. Land cover Raster 17.01.2011 Landsat 5 TM; Were et al. (2013) 30 m O

Note: SOC=soil organic carbon; TN=total nitrogen; CTI=compound topographic index; NDVI=normalized difference vegetation index; PC= principal component; S=soil properties; C=climate; O=organisms; and, R=topography

The Normalized Difference Vegetation Index (NDVI) (Eq. (3)) was derived after the digital numbers of OLI band 4 (red; R) and 5 (near infra-red; NIR) were converted to top-of-atmosphere reflectance (ρ) (http://landsat.usgs.gov/Landsat8_Using_Product.php).
Moreover, principal component analysis was performed to reduce dimensionality, while capturing as much variability as possible from OLI bands 2, 3, 4, 5, 6, and 7. The first principal component (PC1), which explained 98% of the variability, was chosen for spatial modelling. All raster grids were transformed to Universal Transverse Mercator coordinate system (UTM WGS84 Zone 36S) prior to extracting the area of interest from each. The 1 km climatic grids were resampled to 30 m to synchronize them with the rest. Soil data from the laboratory, including sand content, silt content, clay content, TN concentrations, C concentrations, pH, Mg, Ca, P, and K were also integrated into the GIS database both as points in vector format and raster grids after interpolation by ordinary kriging. Ordinary kriging method was chosen because it has been widely used to optimize the prediction of soil properties at unvisited locations in pedological studies (Chaplot et al., 2010; Pachomphon et al., 2010; Kumar and Lal, 2011; Tesfahunegn et al., 2011; Marchetti et al., 2012; Elbasiouny et al., 2014). Finally, the attribute values of all the other raster grids (e.g., slope, rainfall, temperature) were extracted to the points to allow the analysis of relationships between the target variables and environmental predictors.

2.3 Spatial modelling

2.3.1 Exploratory data analysis
Firstly, descriptive statistics of the target variables were estimated. This was followed by pairwise Pearson’s product-moment correlation analysis to detect collinearity between the predictor variables, as well as their correlation with the target variables. Predictors entered the model only if their correlation with the target variable was, or exceeded 0.2. Again, two highly correlated predictors (r≥0.8) were retained in a model only if their variance inflation factors (VIFs) did not exceed 10; otherwise, one was removed (Montgomery et al., 2006).
2.3.2 Model development
The processed point dataset from the GIS database (n=220) was randomly split into two: (i) training data (n=176) to calibrate the models of SOC and TN stocks and, subsequently, create prediction surfaces, and (ii) test data (n=44) to validate the surfaces. Multiple linear regression (MLR), multiple linear regression-kriging (MLRK), geographically weighted regression (GWR), and geographically weighted regression-kriging (GWRK) techniques were used to calibrate the models.
2.3.2.1 MLR and MLRK
Eq. 4 gives the form of MLR model used to define the relationship between the target variables and predictors at the sampled locations (Montgomery et al., 2006):
where yi is the value of the target variable at ith location, β0 is the regression coefficients, xi is the value of the predictor variable at ith location, k is the number of predictors, and εi is the error term.
Full MLR models were fitted by ordinary least square estimator, after which the best subset models were ranked based on Mallow’s Cp using all possible regressions variable selection method. The final reduced model for each target variable was selected from the three best subset models after scrutiny for physical correctness. T-tests were used to determine significance of the model parameters, while analysis of variance F-tests were used to determine significance of the regression at a level of 5%. The adequacy of the models was checked using residual plots, normal probability plots, measures of influence and leverage (e.g., Cook’s D), VIFs, and coefficients of determination (R2). Finally, the models were applied to create prediction surfaces of the target variables.
To develop MLRK models and prediction surfaces, the deterministic component of the target variable modelled by MLR (Eq. 4) and the spatially correlated stochastic component modelled by kriging the MLR residuals were summed up. Eq. 5 summarizes the MLRK model (Hengl et al., 2004; Vasques et al., 2010a):
where y(uivi) is the target variable at location (uivi), (uivi) is the coordinates of the ith location, mmlr(uivi) is the deterministic component, ε’ok(uivi) is the spatially correlated random component, and ε’’(uivi) is the spatially independent residuals error (noise).
2.3.2.2 GWR and GWRK
Similar predictors were used to build GWR models to allow comparison with MLR models. Unlike MLR that assumes spatial stationarity and locational independence, GWR takes into account the spatial location of samples. This allows the estimated parameters to vary locally; hence, representing the spatially varying relationships between the target and predictor variables much better (Zhang et al., 2011). Eq. 6 expresses the form of GWR model (Fotheringham et al., 2002):
where yi is the value of the target variable at ith location, (uivi) is the coordinates of the ith location, β(uivi) are the regression coefficients, xi is the value of the predictor variable at ith location, k is the number of predictors, and εi is the error term.
The GWR parameters were estimated using adaptive (bi-square) spatial kernel functions, where the bandwidth of the samples included for estimation varied with sample density (Fotheringham et al., 2002; Wang et al., 2013). The optimal bandwidth was determined by Akaike Information Criterion (AICc). Lastly, the estimated parameters were applied to create spatially distributed maps of the target variables.
To develop GWRK models, the deterministic component of the target variable modelled by GWR (Eq. 6) and the spatially correlated stochastic component represented by kriged GWR residuals were added. Eq. 7 provides the form of GWRK model (Kumar et al., 2012):
where y(uivi) is the target variable at location (uivi), mgwr(uivi) is the deterministic component, ε’ok(uivi) is the spatially correlated random component, and ε’’(uivi) is the spatially independent residuals error.
Additionally, Moran’s I was calculated to measure spatial autocorrelation in the residuals. The Moran’s I values range from -1 to +1, with 0 indicating absence of spatial autocorrelation, positive values indicating positive autocorrelation, and negative values indicating the opposite (Overmars et al., 2003).
2.3.3 Model evaluation
A ten-fold validation procedure was employed to evaluate the prediction surfaces produced by the fitted models. In this procedure, the original dataset (n=220) was randomly split into training (n=176) and testing (n=44) datasets ten times. The training data were used to calibrate models and generate prediction surfaces, while the testing data were used to validate them. Root mean squared error (RMSE) and mean error (ME) were computed from the differences between the predicted and measured values to determine the precision and bias of the predictions, respectively (Eqs. 8 and 9):
where is the estimated value, yi is the measured value, and n is the number of measured values in the testing data. The ME should be close to zero, while RMSE should be as small as possible. Average ME and RMSE values of the ten-fold validation are reported in this paper. Statistical validation was supplemented by visual inspection of the spatial patterns of the target variables.
The method with the lowest prediction error indices was chosen to provide the final estimates of SOC and TN stocks for the Eastern Mau Forest Reserve. To estimate the stocks under different land cover types, the prediction surfaces were overlaid with the land cover map of the area and zonal statistics extracted. All data management, analyses, and geovisualization functions were performed using ArcGIS® 10.1, ERDAS IMAGINE® 2013, GWR4, Microsoft Excel® 2010, and R version 3.0.1 (R Core Team, 2013) with its add-in packages “sp” (Pebesma et al., 2013) and “automap” (Hiemstra, 2013).

3 Results

3.1 Exploratory data analysis

Table 2 presents the numerical summaries of SOC and TN stocks at the sampled locations. Soil organic carbon stocks range from 42.0 to 193.4 Mg C ha-1 with a mean of 102.7 Mg C ha-1. The standard deviation is 24.6 Mg C ha-1 and coefficient of variation is 23.9%, which suggests moderate variability. The skewness is 0.39 indicating an approximately normal distribution of the data, whereas kurtosis is 0.97 implying less peaked values in the distribution of the data. Similarly, TN stocks vary from 4.2 to 19.1 Mg N ha-1 with a mean of 10.3 Mg C ha-1. The standard deviation is 2.4 Mg C ha-1 and coefficient of variation is 23.8%, while skewness and kurtosis are 0.28 and 0.76, respectively. Again, this shows moderate variability and minimal departure from normality. Hence, spatial modelling of both SOC and TN stocks was performed using the raw, non-transformed data. Pearson’s correlation analysis shows that some of the predictors were highly correlated (r≥0.80), and that only 13 met the threshold correlation (r≥0.20) with the target variables (Table 3). Thus, the candidate predictors for developing full models reduced from 20 to 13; namely, elevation, aspect, rainfall, temperature, TN, Mg, silt, clay, land cover, PC1, NDVI, and OLI band 10 and 11 (Table 3). The predictors that were highly correlated include: temperature and elevation, temperature and land cover, elevation and land cover, elevation and OLI band 11, land cover and OLI band 11, and PC1 and OLI band 11. Therefore, VIFs of the predictors in the reduced models were also checked for multi-collinearity.
Table 2 Descriptive statistics of SOC and TN stocks (0-30 cm)
Variable n Mean Median SD CV (%) Min. Max. Range Skewness Kurtosis
SOCst 220 102.7 103.2 24.6 23.9 42.0 193.4 151.4 0.39 0.97
TNst 220 10.3 10.3 2.4 23.8 4.2 19.1 14.9 0.28 0.76

SD=standard deviation; CV=coefficient of variation; n=number of observations

Table 3 Pearson’s correlation coefficient between the predictors and target variables selected for spatial modelling
Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1. SOC stock 1.00
2. TN stock 0.99 1.00
3. TN content 0.84 0.85 1.00
4. SOC content 0.85 0.84 0.99 1.00
5. Silt -0.41 -0.42 -0.56 -0.55 1.00
6. Magnesium 0.35 0.35 0.44 0.44 -0.36 1.00
7. Clay 0.28 0.29 0.40 0.39 -0.61 0.08 1.00
8. Temperature -0.50 -0.50 -0.63 -0.63 0.28 -0.04 -0.34 1.00
9. Rainfall 0.44 0.45 0.56 0.55 -0.23 0.25 0.10 -0.61 1.00
10. Elevation 0.51 0.51 0.65 0.65 -0.30 0.06 0.35 -0.99 0.65 1.00
11. Aspect 0.22 0.23 0.18 0.17 -0.08 0.01 0.02 -0.16 0.16 0.16 1.00
12. NDVI 0.30 0.30 0.39 0.39 -0.25 0.07 0.24 -0.50 0.25 0.50 0.11 1.00
13. Land cover -0.48 -0.48 -0.54 -0.53 0.31 0.00 -0.41 0.83 -0.46 -0.84 -0.16 -0.56 1.00
14. PC1 -0.48 -0.48 -0.52 -0.52 0.15 -0.03 -0.23 0.71 -0.50 -0.73 -0.28 -0.32 0.74 1.00
15. Landsat 8 OLI band 11 -0.58 -0.58 -0.65 -0.65 0.35 -0.09 -0.37 0.81 -0.56 -0.84 -0.29 -0.63 0.89 0.82 1.00

Note: SOC=soil organic carbon; TN=total nitrogen; NDVI=normalized difference vegetation index; PC= principal component. Bold form shows that the correlation coefficient between the predictors and target variables exceeded the threshold value (r > 0.2).

3.2 Spatial models

3.2.1 MLR
The subset SOC and TN stock models selected by all possible regressions method had Mallow’s Cp values (3.8 and 10.3, respectively) that were lower than the number of model parameters. Table 4 provides the summaries of the models. Elevation, silt content, TN concentration, and OLI band 11 have significant effects on SOC stocks explaining 72% of its variability (adjusted R2=0.72), whereas OLI band 11, elevation, and SOC concentration have significant effects on TN stocks explaining 71% of its variability (adjusted R2=0.71). Total nitrogen concentrations have the largest magnitude of effect on SOC stocks, while SOC concentrations have the largest magnitude of effect on TN stocks. OLI band 11 has the smallest magnitude of effect on both TN and SOC stocks. Visual analysis of the residual and normal probability plots indicated equality of variance and normality in the distribution of error terms, as well as linearity in the model parameters. The few outliers that were evident on these plots were not sufficiently influential to warrant their removal from the data because Cook’s D indices were less than 1. Despite the high correlation between elevation and OLI band 11 (r=0.84), the associated VIFs do not exceed 10 in the models. Moran’s indices are very low, but statistically significant; that is, 0.11 (p=0.0141) and 0.08 (p=0.0550) for SOC and TN stocks models, respectively. This shows very weak tendency for clustering of similar residuals. The high nugget-to-sill ratios (NSRs) of 78.6% for the residuals of SOC stock model, and 73.7% for the residuals of TN stock model to a range of 4 km (Table 6 and Figure 3) also demonstrate this weak spatial structure. However, the spatial dependency of SOC and TN stocks data are moderate (NSRs of 58.1% and 45.6%, respectively) to a range of 4.8 km. Total sills for the residuals are 182 Mg C ha-1 and 1.9 Mg N ha-1, which are close to the variance (σ2) estimates of the respective MLR models (170.6 Mg C ha-1 and 1.7 Mg N ha-1).
Table 4 Parameter estimates of the MLR models
Parameter SOC stocks model TN stocks model
Estimate SE t value Pr (>|t|) VIF Estimate SE t value Pr (>|t|) VIF
Intercept 143.502 50.757 2.827 0.0053** - 16.741 5.131 3.263 0.0013** -
Silt 0.443 0.202 2.191 0.0298* 1.531 - - - - -
Band 11 -0.003 0.001 -2.360 0.0194* 3.511 -0.000 0.000 -2.475 0.0143* 3.489
Elevation -0.022 0.009 -2.503 0.0133* 3.613 -0.002 0.001 -2.305 0.0223* 3.558
TN 178.200 12.269 14.524 0.0000*** 2.471 - - - - -
SOC - - - - - 1.597 0.106 15.103 0.0000*** 1.807
Adjusted R2 0.72 0.71
RMSE 13.07 1.33
Moran’s I 0.11 0.08

Significance codes: 0 '***' 0.001 '**' 0.01 '*'

Figure 3 (a) Experimental variograms (points) and fitted models (lines) of SOC stocks (b) MLRsoc residuals (c) GWRsoc residuals (d) TN stocks (e) MLRtn residuals, and (f) GWRtn residuals
3.2.2 GWR
Table 5 shows the summaries of parameter estimates of the GWR models for SOC and TN stocks. Once again, TN concentrations have the highest magnitude of effect on SOC stocks, while SOC concentrations have the highest magnitude of effect on TN stocks. OLI band 11 has the lowest magnitude of effect on both TN and SOC stocks. Unlike the MLR models, the GWR models show that the magnitude of the effects of predictors varies with sampling location. This means that the interactions between the target variables and environmental factors are spatially non-stationary. Hence, the summaries of GWR estimates are given in ranges instead of mean values. Although the magnitudes vary spatially, the directions of the effects are constant. The Moran’s indices are lower than for MLR models; that is, 0.06 (p=0.1798) and 0.02 (p=0.1620) for SOC and TN stocks models, respectively. This indicates that the GWR residuals are approximately uncorrelated, and that the models are better specified than the MLR models. The high NSRs of 84.4% for the GWR residuals of SOC stocks, and 87.5% for the GWR residuals of TN stocks to a range of 6 km and 3 km, respectively (Table 6 and Figure 3), also reveal this weak spatial dependency. Total sills for the residuals are 167 Mg C ha-1 and 1.6 Mg N ha-1, which are lower than for the MLR residuals. However, the range is shorter for the GWR residuals than for the MLR residuals only in the case of TN stocks.
Table 5 Parameter estimates of the GWR models
Parameter SOC stocks model TN stocks model
Mean SD Min. Max. Range Mean SD Min. Max. Range
Intercept 129.525 31.412 50.031 199.881 149.851 14.790 4.771 8.037 26.423 18.387
Silt 0.436 0.115 0.203 0.614 0.411 - - - - -
Band 11 -0.003 0.001 -0.004 -0.001 0.004 -0.000 0.000 -0.000 -0.000 0.000
Elevation -0.021 0.007 -0.041 -0.004 0.037 -0.002 0.001 -0.005 -0.001 0.004
TN 177.230 23.072 142.790 238.558 95.768 - - - - -
SOC - - - - - 1.576 0.197 1.087 1.899 0.812
Global adjusted R2 0.73 0.72
Global RMSE 12.86 1.29
Moran’s I 0.06 0.02
Table 6 Parameters of the fitted variogram models for SOC and TN stocks, and the residuals of the respective GWR and MLR models
Variable Model Nugget Mg ha-1 Partial sill Mg ha-1 Total sill Mg ha-1 Range (m) Nugget- to- sill ratio (%) Spatial
dependence
SOC stocks Gaussian 386 278 664 4845 58.1 Moderate
MLRsoc residuals Gaussian 143 39 182 4434 78.6 Weak
GWRsoc residuals Gaussian 141 26 167 5760 84.4 Weak
TN stocks Exponential 3.1 3.7 6.8 4493 45.6 Moderate
MLRtn residuals Exponential 1.4 0.5 1.9 4132 73.7 Weak
GWRtn residuals Exponential 1.4 0.2 1.6 2944 87.5 Weak

3.3 Model evaluation

Table 7 presents the results of ten-fold validation procedure used to quantify the errors attached to the prediction maps of SOC and TN stocks. The average MEs for all prediction models are close to 0, which indicate a small tendency for over- or underestimation. In addition, the average RMSEs range from 16.7 to 19.9 Mg C ha-1 and from 1.5 to 1.9 Mg N ha-1, which are slightly higher than the RMSEs of the fitted models (13.07 Mg C ha-1 and 1.33 Mg N ha-1 for MLR models, and 12.86 Mg C ha-1 and 1.29 Mg N ha-1 for GWR models). This suggests that the models do not predict new data as precise as they fit the original ones. However, the differences in RMSEs are slight, and it should also be noted that the proportions of observations used for calibration and validation were not equal. The GWR models show better performance in predicting SOC and TN stocks at new locations than MLR models given their lower average RMSEs and MEs. The average RMSEs are also slightly lower than the standard deviations of the measured values (Table 3), which means that incorporation of the predictors and spatial correlation gives better estimations than what would be achieved by just using the measured values for predictions. However, addition of the stochastic part (kriged residuals) to the GWR and MLR outputs does not reduce the prediction errors. The RMSEs of the GWRK and GWR models are similar, and so are the RMSEs of the MLRK and MLR models.
Table 7 Summary statistics of the spatial prediction errors
SOCst (Mg C ha-1) TNst (Mg N ha-1)
ME RMSE ME RMSE
GWRK -0.48 16.74 0.04 1.53
GWR -0.86 16.66 -0.03 1.51
MLRK 0.39 19.42 -0.31 1.93
MLR 0.30 19.89 -0.33 1.93

ME= mean error; RMSE=root mean squared error

3.4 Spatial distribution and estimates of SOC and TN stocks

Figures 4 and 5 display the different prediction surfaces of SOC and TN stocks produced by MLR, MLRK, GWR, and GWRK models. The maps reveal similar spatial patterns of SOC and TN stocks meaning that SOC and TN stocks respond similarly to the environmental factors. There is a general decrease of SOC and TN stocks from west to east. The highest estimates of SOC and TN stocks occur in the western and northwestern parts, which according to the environmental data, have higher forest cover, elevations, and SOC and TN concentrations, but lower silt contents and surface temperatures. These hotspots are parts of the Logoman, Nessuiet, Kiptunga, and Baraget forests that are undisturbed. The lowest estimates, on the other hand, occur on the eastern side where croplands have been established, including Teret, Nessuiet, Kapkembu, Tuiyotich, and Sururu locations. These coldspots are areas with higher crop cover, silt contents, and surface temperatures, but lower elevations, and SOC and TN concentrations. In the northern and southeastern parts where crop cover is also high, the SOC and TN stocks are moderate to high. The GWR and GWRK prediction surfaces give more realistic pictures of the moderate to high SOC and TN stocks at Sururu forest in the southeasternmost part, which is more degraded than the forests in the western and northwestern parts.
The models generated minimum and maximum values that approximate the measured values (cf. Table 3). The MLR and MLRK estimates of TN stocks range from 5.8 to 15.1 Mg N ha-1, whereas the GWR and GWRK estimates vary from 5.3 to 15.8 Mg N ha-1. Similarly, the MLR and MLRK estimates of SOC stocks range from 56.7 to 146 Mg C ha-1, while the GWR and GWRK estimates vary from 55.6 to 146 Mg C ha-1.
Table 8 gives the magnitude of SOC and TN stocks under different land cover categories based on GWR method, which has lower prediction error indices compared to other methods. Forests stores the highest amounts of SOC and TN (3.78 Tg C and 0.38 Tg N) followed by croplands (2.46 Tg C and 0.25 Tg N), and grasslands (0.57 Tg C and 0.06 Tg N) (1 Tg = 1012 g = 1 million tons). This is because forests cover the largest area (32,228 ha), while grasslands cover the smallest area (5509 ha). In total, the Eastern Mau Forest Reserve stores about 6.81 Tg and 0.69 Tg of SOC and TN, respectively.
Table 8 Soil organic carbon and nitrogen stocks under different land cover types
Land cover Area SOC stocks TN stocks
Min. Max. Mean Total Min. Max. Mean Total
(Ha) (Mg ha-1) (Tg) (Mg ha-1) (Tg)
Forests 32228.4 75.5 142.9 110.4 3.78 7.5 15.3 11.1 0.38
Grasslands 5509.4 66.7 129.8 103.5 0.57 6.7 12.6 10.4 0.06
Croplands 25828.1 62.9 126.9 95.2 2.46 6.5 12.2 9.6 0.25
Total 65565.9 6.81 0.69
Figure 4 Maps showing the spatial patterns of the predicted SOC stocks using MLR, MLRK, GWR and GWRK
Figure 5 Maps showing the spatial patterns of the predicted TN stocks using MLR, MLRK, GWR and GWRK

4 Discussion

4.1 Spatial models

The significant effects of elevation, silt content, TN concentration, and OLI band 11 in the SOC models, and elevation, SOC concentration, and OLI band 11 in the TN models implies that topographical, edaphic, and climatic factors control the spatial patterns of SOC and TN stocks in the Eastern Mau Forest Reserve. Pachomphon et al. (2010) and Li and Shao (2014) reported similar combination of controlling factors in Laos and north-western China, respectively. The magnitudes of the effects of these predictors indicates that soil properties, particularly TN and SOC concentrations, are more important than the other factors in determining the observed variability of SOC and TN stocks, respectively. This was expected because of the high statistical correlation between them and the tight coupling of C and N cycles. Nitrogen supply increases the net uptake of C in terrestrial ecosystems, which in turn leads to higher inputs of C and N to the soils (Zaehle et al., 2011). The high coefficient of determination (i.e., R2>0.70) obtained for the fitted MLR and GWR models further confirms the explanatory power of these soil properties. In a similar study at four contrasting East African landscapes, Vågen and Winowiecki (2013a) also concluded that intrinsic soil properties determined more the SOC dynamics than other environmental factors alone. The significant effects of OLI band 11 (proxy for surface temperature) and elevation on SOC and TN stocks suggests that: (i) some solutions to the problem of up-scaling soil survey data to landscape level in the region exist in the freely available remotely sensed and topographical data, and (ii) computationally intensive remote sensing- and DEM-derived parameters (e.g., NDVI, CTI) do not always improve the spatial prediction of soil properties. Generally, poor prediction performance (R2<0.50) has been the norm in the region. For instance, Mora-Vallejo et al. (2008) developed MLR and MLRK models using topographical and geomorphological variables that explained less than 25% of SOC variability in south-eastern Kenya. But recently, Vågen et al. (2013b) and Winowiecki et al. (2015) achieved better performances (i.e., R2>0.70) when they predicted SOC stocks using remotely-sensed imagery and random forest models in Ethiopia and Tanzania, respectively.
In terms of spatial structure, the NSRs of raw SOC and TN stocks data revealed moderate spatial dependency (Table 6), which compare with the findings of Sumfleth and Duttmann (2008). This suggests that in the short-range, random and structural processes are equally influential in explaining the spatial variability of SOC and TN stocks. The structural processes that determine the variability of SOC and TN stocks in the Eastern Mau Forest Reserve are the natural soil-forming factors, including topography, soil properties, and climate, while the random processes that explain the remaining variability are human activities, such as illegal loggings, encroachments, and charcoal burning, as well as land management practices. In contrast, the residuals obtained from the GWR and MLR models exhibited weaker patterning as evidenced by the low Moran’s indices and high NSRs. This means that the global trend models partly explained the variability and spatial correlation of SOC and TN stocks leaving only a small, less structured, short-range variation unexplained (Vasques et al., 2010a, 2010b). The unexplained short-range spatial variation reflects the inherent data errors and spatial sources of variations at distances smaller than the shortest sampling interval. Theoretically, this can be resolved by increasing the sampling intensity, but practically, this may be difficult to implement due to resource constraints. The NSRs also hint at the proportion of variation that can be explained by the spatial models. As expected, the NSRs for the MLR models of SOC and TN stocks (79% and 74%) were close to the proportions of variation that the models explained (adjusted R2=72% and 71%).

4.2 Model evaluation

The GWR-based models were better than MLR-based models in predicting new data (Table 7); thus, GWR models were chosen to quantify the total stocks of SOC and TN in the area and under different land cover types (Table 8). Mishra et al. (2010), Zhang et al. (2011), and Syzomanoski and Kryza (2012) also obtained similar results. Basically, the MLR approach assumed that the environmental factors, which affected the variability of SOC and TN stocks, were spatially stationary. Hence, it represented their relationships using a global statistic. However, such global values can lead to large errors and be misleading since most of the variability in SOC and TN stocks stem from local interaction of processes (Kumar et al., 2012). In contrast, the GWR approach applied regressions locally, which accounted for both the spatial trends and local variations resulting in superior estimations of SOC and TN stocks. The major weakness of GWR approach was that even though the variation of regression coefficients locally implied the selection of different predictors at different locations (Zhang et al., 2011; Kumar et al., 2013), this did not happen; hence, some predictors may have been redundant at some locations. Unlike other studies (Mishra et al., 2010; Kumar et al., 2012; Syzomanoski and Kryza, 2012; Zhang et al., 2012), the addition of stochastic component (kriged residuals) to the MLR and GWR outputs did not yield lower prediction errors in this case. The small proportion of spatially correlated random component in the residuals as indicated by the low Moran’s indices and NSRs (Table 6) explains this. Mora-Vallejo et al. (2008) and Li et al. (2013b) also reported that MLRK did not outperform MLR in their study.

4.3 Spatial distribution and estimates of SOC and TN stocks

The prediction maps revealed spatial patterns of SOC and TN stocks that were similar and reflected the environmental predictors. The given characteristics of the hotspots of SOC and TN stocks in the western and northwestern parts, as well as the highly fertile Andosols of the area favour accumulation of SOC and TN stocks. For instance, the high rainfall and low temperatures associated with higher altitudes increase net primary productivity of the forests and decrease SOC turnover. The lower silt content relative to clay content in the forest soils also indicates the presence of organo-complexes, or allophane, imogolite, and ferrihydrite clay minerals, which stabilize organic matter and plant nutrients (Lemenih et al., 2005; Chaplot et al., 2010). The smaller pore spaces of clay particles also promote aggregation and physical protection of SOC. In contrast, the characteristics of the coldspots of SOC and TN stocks on the eastern side are unfavourable for accumulation of SOC and TN stocks. For example, the higher crop cover is attributed to the conversion of forests to croplands, which began in the mid-1990s. In these croplands, biomass removal after harvesting, erosion, and frequent tillage, which breaks up the soil aggregates and alters aeration, can explain the lower SOC and TN stocks (Murty et al., 2002; Smith 2008; Eclesia et al., 2012; Wiesmeier et al., 2012). Thus, the coldspots of SOC and TN stocks also highlight human-induced soil degradation and sources of C and N emissions. The altitudinal gradient in SOC stocks mentioned above corresponds with previous studies in the Tana River basin, Kenya (Tamooh et al., 2012), while the highest SOC and TN stocks under forests coincide with other studies in the region (Bewketa and Stroosnidjer, 2003; Lemenih et al., 2005; Girmay and Singh, 2012; Demessie et al., 2013). The hotspots and coldspots on the prediction maps of SOC and TN stocks are the sites to target the best management practices (BMPs) for climate change mitigation and sustainable land management. For example, the western and northwestern parts need practices that promote retention, whereas the eastern part requires those that enhance accumulation of SOC and TN stocks.
The SOC and TN stocks in the Eastern Mau Forest Reserve to 30 cm depth were estimated at 6.81 Tg and 0.69 Tg, respectively. This accounts for 0.36% of the total SOC stock to 30 cm depth reported for Kenya (Batjes, 2004). Batjes (2004) further reported that Andosols of the humid and semi-humid regions in Kenya stored an average of 9.1 Kg C m-2 (91 Mg C ha-1) to 30 cm depth, which slightly differs from the present findings (i.e., 10.3 Kg C m-2 or 102.7 Mg C ha-1). This difference can be ascribed to the properties of the data used in the two studies. Batjes (2004) used coarse resolution legacy data from SOil and TERrain (SOTER) database and Africa Land Cover Characteristics (ALCC) database, while the present study used newly collected, fine resolution field data to estimate SOC stocks.

4.4 Limitations of the study

We acknowledge the limitations of this study. Firstly, the soil properties used as predictors were themselves products of interpolation by ordinary kriging. Thus, interpolation errors may have been propagated to the subsequent prediction of SOC and TN stocks. These predictors would have enhanced the prediction accuracy more had they been sampled more intensely than the target variables. Similarly, estimation of SOC and TN stocks under different land cover types was based on a land cover map that had been produced through classification of Landsat 5 TM satellite imagery. Thus, the inherent classification errors may have influenced the estimates of SOC and TN stocks under the different land cover classes. Additionally, the auxiliary spatial data (e.g., DEM, Landsat imagery, and climate) were sourced from various databases; hence, their quality was different. Poor coverage of samples in the southeastern most and central parts, which were dominated by thick impenetrable bamboo forests may have also affected prediction accuracy in these areas. Lastly, some soil-forming factors (e.g., parent material and age) were omitted owing to lack of suitable data. Their inclusion, if significant, may improve the predictive power of future models. The foregoing factors introduced uncertainties, the quantification of which was beyond the scope of this study. Future work will assess the implications of error propagation through sensitivity analysis of model parameters estimated using multi-source auxiliary spatial data with varied accuracy.

5 Conclusions and recommendations

This study has demonstrated an integrated approach of field sampling, GIS, remote sensing, and statistical analysis to quantify and map SOC and TN stocks to 30 cm depth in the Eastern Mau Forest Reserve, Kenya. Based on the results, the conclusions drawn are: (1) Forests have the largest SOC and TN pools followed by croplands and grasslands. Altogether, the Eastern Mau Forest Reserve stores about 6.81 Tg of C and 0.69 Tg of N. (2) The hotspots of SOC and TN stocks are the native systems in the western and northwestern parts, namely Logoman, Nessuiet, Kiptunga, and Baraget forests, while the coldspots are the human-dominated landscapes in the eastern part, including Teret, Nessuiet, Kapkembu, Tuiyotich, and Sururu locations. Thus, conversion of forests to croplands is a driver of soil degradation in this area. (3) Climatic, edaphic, and topographic factors control the observed spatial patterns of SOC and TN stocks; however, soil properties, particularly TN and SOC concentrations are the most important determinants. Despite the limitations, this study provides the first spatially exhaustive soil information for Eastern Mau forest reserve at a finer scale. The resultant outputs will assist to monitor SOC and TN stocks, as well as to formulate spatially targeted climate change mitigation and sustainable land management policies. Also, the approach used offers a cost-effective framework to derive knowledge of soil processes and multi-purpose soil information in other data-poor environments in Eastern Africa.

The authors have declared that no competing interests exist.

References

Publishing order | Descend order by publishing year | Descend order by cited within
1
Amare T, Hergarten C, Hurni H et al., 2013. Prediction of soil organic carbon for Ethiopian highlands using soil spectroscopy. ISRN Soil Science, 720589 (11 pp), .

2
Aynekulu E, Vågen T-G, Shepherd Ket al., 2011. A protocol for measurement and monitoring soil carbon stocks in agricultural landscapes. Version 1.1. World Agroforestry Centre, Nairobi.Introduction The determinants that differentiate men and women from the biological, sexual, cultural and social viewpoints have a bearing on what kind of health problems affect them and how as well as on the required health interactions. Objectives To identify those differences in mortality indexes between Cuban men and women in 2006. Methods A cross-sectional descriptive study using mortality index databases from the National Division of Statistics. Age-specific, sex-specific and potential years of life lost-specific mortality rates along with rate rations for the first causes of death were estimated to gather proper information. Results It was found that men had an overmortality rate in most of the main causes of death, and died earlier than women from a group of chronic diseases, accidents, self-inflicted lesions; additionally, malignant tumors negatively affected both sexes. Conclusions There are differences between men and women that favor the latter with respect to potential years of life lost, although the gap between men and women shows negative tendency towards the women.

DOI

3
Batjes N H, 2004. Soil carbon stocks and projected changes according to land use and management: A case study for Kenya.Soil Use and Management, 20: 350-356.Abstract. Soil organic (SOC) and inorganic carbon (SIC) stocks of Kenya were determined using four different methods to provide baseline data. The assessments used an updated version of the 1:1 M soil and terrain database for the country. Estimates for national SOC stocks to 1 m depth ranged from 3452 to 3797 Tg C. The findings highlight the need for comprehensive databases of soil and terrain data of good quality that consider more than one representative profile per soil component. The 95% confidence limits for the median, area-weighted SOC content were largest in the humid highlands (15.4–15.7 kg C m) and smallest in the hot arid zone (4.4–4.5 kg C m). Conversely, for SIC these values were largest in the arid zone (4.3–4.5 kg C m) and smallest in high rainfall areas (<0.1 kg C m). Many croplands in Kenya have been over-exploited, resulting in nutrient depletion and loss of organic matter. The SOC gains considered ecologically and technically feasible upon improved management of croplands were estimated at 5.8–9.7 Tg C over the next 25 years. This corresponds to an estimated annual mitigation potential of 5–9% of Kenya's CO-C emissions from fossil fuels, cement manufacturing and land use change for 1990.

DOI

4
Bewketa W, Stroosnijder L, 2003. Effects of agro-ecological land use succession on soil properties in Chemoga watershed, Blue Nile basin, Ethiopia.Geoderma, 111: 85-98.

5
Blake G R, 1965. Bulk density. In: Black C A (ed.). Methods of Soil Analysis, Part 1. Physical and Mineralogical Properties, Including Statistics of Measurement and Sampling, American Society of Agronomy, Inc., Madison, Wisconsin, USA.

6
Bremner J M, Mulvaney C S, 1982. Nitrogen - total. In: Page A L (ed.). Methods of Soil Analysis, Part 2. Chemical and Microbiological Properties. 2nd ed. American Society of Agronomy, Inc., Madison, Wisconsin, USA.

7
Cambule A H, Rossiter D G, Stoorvogel J Jet al., 2014. Soil organic carbon stocks in the Limpopo National Park, Mozambique: Amount, spatial distribution and uncertainty.Geoderma, 213: 46-56.Many areas in sub-Saharan African are data-poor and poorly accessible. The estimation of soil organic carbon (SOC) stocks in these areas will have to rely on the limited available secondary data coupled with restricted field sampling. We assessed the total SOC stock, its spatial variation and the causes of this variation in Limpopo National Park (LNP), a data-poor and poorly accessible area in southwestern Mozambique. During a field survey, A-horizon thickness was measured and soil samples were taken for the determination of SOC concentrations. SOC concentrations were multiplied by soil bulk density and A-horizon thickness to estimate SOC stocks. Spatial distribution was assessed through: i) a measure-and-multiply approach to assess average SOC stocks by landscape unit, and ii) a soil-landscape model that used soil forming factors to interpolate SOC stocks from observations to a grid covering the area by ordinary (OK) and universal (UK) kriging. Predictions were validated by both independent and leave-one-out cross validations. The total SOC stock of the LNP was obtained by i) calculating an area-weighted average from the means of the landscape units and by ii) summing the cells of the interpolated grid. Uncertainty was evaluated by the mean standard error for the measure-and-multiply approach and by the mean kriging prediction standard deviation for the soil-landscape model approach. The reliability of the estimates of total stocks was assessed by the uncertainty of the input data and its effect on estimates. The mean SOC stock from all sample points is 1.5902kg02m 61022 ; landscape unit averages are 1.13–2.4602kg02m 61022 . Covariables explained 45% (soil) and 17% (coordinates) of SOC stock variation. Predictions from spatial models averaged 1.6502kg02m 61022 and are within the ranges reported for similar soils in southern Africa. The validation root mean square error of prediction (RMSEP) was about 30% of the mean predictions for both OK and UK. Uncertainty is high (coefficient of variation of about 40%) due to short-range spatial structure combined with sparse sampling. The range of total SOC stock of the 10,41002km 61022 study area was estimated at 15,579–17,90802Gg. However, 90% confidence limits of the total stocks estimated are narrower (5–15%) for the measure-and-multiply model and wider (66–70%) for the soil-landscape model. The spatial distribution is rather homogenous, suggesting levels are mainly determined by regional climate.

DOI

8
Chaplot V, Bouahom B, Valentin C, 2010. Soil organic carbon stocks in Laos: Spatial variations and controlling factors.Global Change Biology, 16: 1380-1393.Surface soils, which contain the largest pool of terrestrial organic carbon (C), may be able to sequester atmospheric C and thus mitigate climate change. However, this remains controversial, largely due to insufficient data and knowledge gaps in respect of organic C contents and stocks in soils and the main factors of their control. Up to now and despite numerous evaluations of soil organic carbon (SOC) stocks worldwide, the sloping lands of southeast Asia, one of the most biogeochemically active regions of the world, remain uninvestigated. Our main objective was to quantify SOC stocks and to evaluate the impact of various environmental factors. We, therefore, selected Laos with 2306556665kmof mostly forested steep slopes, and where cultivation is still mainly traditional, i.e. a system of shifting cultivation without fertilization or mechanical tillage. Analytical data from 3471 soil profiles demonstrated that the top 165m of soil depth holds an estimated 4.64 billion tons of SOC, 65% of which is in the first 0.365m. SOC stocks to 0.365m exhibit a high coefficient of variation (CV=62%) with values from 1.8 to 77165Mg65C65haand a mean at 12965Mg65C65ha. Furthermore, these stocks are significantly (at <0.05 level) affected by land use as shown by principal components analysis and -tests with the largest amount being found under forest, less under shifting cultivation and the smallest under continuous cultivation. Moreover, SOC stocks correlated regionally to total annual rainfalls and latitude, and locally at the hill-slope level to the distance to the stream network and the slope angle. It is hypothesized that this correlation is through actions on mineral weathering, soil clay content, soil fertility and SOC redistributions in landscapes. These relationships between SOC stocks and environmental factors may be of further use in (1) predicting the impact of global changes on future SOC stocks; and (2) identifying optimal strategies for land use planning so as to minimize soil C emissions to the atmosphere while maximizing carbon sequestration in soils.

DOI

9
Day P R, 1965. Particle fractionation and particle size analysis. In: Black C A (ed.). Methods of Soil Analysis, Part 1. Physical and Mineralogical Properties, Including Statistics of Measurement and Sampling. American Society of Agronomy, Inc., Madison, Wisconsin, USA.

10
Demessie A, Singh B R, Lal R, 2013. Soil carbon and nitrogen stocks under chronosequence of farm and traditional agro-forestry uses in Gambo district, southern Ethiopia.Nutr. Cycl. Agroecosys., 95: 365-375.

11
Doetterl S, Stevens A, van Oost Ket al., 2013. Spatially explicit regional scale prediction of soil organic carbon stocks in cropland using environmental variables and mixed model approaches.Geoderma, 204/205: 31-42.

12
Dorji T, Odeh I O A, Field D Jet al., 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.

13
Eclesia R P, Jobbagy E G, Jackson R Bet al., 2012. Shifts in soil organic carbon for plantation and pasture establishment in native forests and grasslands of South America.Global Change Biology, 18: 3237-3251.Abstract The replacement of native vegetation by pastures or tree plantations is increasing worldwide. Contradictory effects of these land use transitions on the direction of changes in soil organic carbon (SOC) stocks, quality, and vertical distribution have been reported, which could be explained by the characteristics of the new or prior vegetation, time since vegetation replacement, and environmental conditions. We used a series of paired-field experiments and a literature synthesis to evaluate how these factors affect SOC contents in transitions between tree- and grass-dominated (grazed) ecosystems in South America. Both our field and literature approaches showed that SOC changes (0&ndash;20cm of depth) were independent of the initial native vegetation (forest, grassland, or savanna) but strongly dependent on the characteristics of the new vegetation (tree plantations or pastures), its age, and precipitation. Pasture establishment increased SOC contents across all our precipitation gradient and C gains were greater as pastures aged. In contrast, tree plantations increased SOC stocks in arid sites but decreased them in humid ones. However, SOC losses in humid sites were counterbalanced by the effect of plantation age, as plantations increased their SOC stocks as plantations aged. A multiple regression model including age and precipitation explained more than 50% ( p <0.01) of SOC changes observed after sowing pastures or planting trees. The only clear shift observed in the vertical distribution of SOC occurred when pastures replaced native forests, with SOC gains in the surface soil but losses at greater depths. The changes in SOC stocks occurred mainly in the silt+clay soil size fraction (MAOM), while SOC stocks in labile (POM) fraction remained relatively constant. Our results can be considered in designing strategies to increase SOC storage and soil fertility and highlight the importance of precipitation, soil depth, and age in determining SOC changes across a range of environments and land-use transitions.

DOI

14
Elbasiouny H, Abowaly M, Abu_Alkheir Aet al., 2014. Spatial variation of soil carbon and nitrogen pools by using ordinary kriging method in an area of north Nile delta, Egypt.Catena, 113: 70-78.Nile Delta includes part of the most fertile and populated lands in the world. However, there is no accurate and reliable database about C and N pools of this region; in addition there are no published data in this regard. Spatial variation of soil C and N pools was studied based on Ordinary Kriging (OK) as a geostatistical method. This method was used for converting sampled soil C and N data to continuous maps of C and N pools in Northern part on Nile Delta, Egypt. The data revealed different levels of variability of C and N pools in the study area. The total C pool (TCP) ranged between 1.6 and 122.7Mg/ha; while total N pool (TNP) ranged between 0.3 and 7.6Mg/ha. Soil organic carbon pool (SOCP) ranged between 0.3 and 76.4Mg/ha, whereas soil inorganic carbon pool (SICP) ranged between 1.2 and 90.5Mg/ha. Soil C and N pools are the lowest in the Northern part in the study area which is close to Mediterranean Sea coast because of low organic matter inputs in addition to salinity and alkalinity.

DOI

15
Fotheringham A S, Brunsdon C, Charlton M E, 2002. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. England: John Wiley & Sons Inc.

16
Girmay G, Singh B R, 2012. Changes in soil organic carbon stocks and soil quality: Land use system effects in northern Ethiopia. Acta Agriculturae Scandinavica, Section B &#x2013,Soil & Plant Science, 62(6): 519-530.In Tigray, Ethiopia, land degradation is a dominant environmental problem and hence the regional government has undertaken restoration measures on degraded soils since 1991. The present study was aimed to assess the impact of land uses and soil management practices on soil properties, and consequently on soil quality of degraded soils. The catchments selected were Maileba and Gum Selassa, and land uses included cultivated (CL), grazing (GL), plantation (PA) and area exclosure (AE). Replicated soil samples were collected from topsoil and profiles of four land-use types in both catchments. Soils in area exclosure showed higher soil organic carbon (SOC), total N and extractable K than grazing land, cultivated land and plantation area mainly at 0–40 cm soil depth. Estimated soil organic carbon stock at Maileba in 0–40 cm depth varied between 54 to 74 Mg C ha, being lowest in cultivated land and highest in area exclosure, and the soil organic carbon stock in area exclosure represents 63% of total carbon stock stored in the profile. Soil organic carbon stock (0–40 cm) at Gum Selassa ranged between 33 to 38 Mg C ha, being higher in cultivated land and lower in plantation area. Soil quality index (SQI) of area exclosure (0.794) at Maileba and cultivated land (0.721) at Gum Selassa scored highest among all land uses, and the order was area exclosure>grazing land>plantation area>cultivated land at Maileba and cultivated land>grazing land>plantation area at Gum Selassa, highlighting the effectiveness of area exclosure in restoring soil quality of degraded soils.

DOI

17
[Accessed 2014, January 19].

18
Grimm R, Behrens T, Märker Met al., 2008. Soil organic carbon concentrations and stocks on Barro Colorado Island: Digital soil mapping using Random Forests analysis.Geoderma, 146: 102-113.Spatial estimates of tropical soil organic carbon (SOC) concentrations and stocks are crucial to understanding the role of tropical SOC in the global carbon cycle. They also allow for spatial variation of SOC in environmental process models. SOC is spatially highly variable. In traditional approaches, SOC concentrations and stocks have been derived from estimates for single or very few profiles and spatially linked to existing units of soil or vegetation maps. However, many existing soil profile data are incomplete and untested as to whether they are representative or unbiased. Also single means for soil or vegetation map units cannot characterize SOC spatial variability within these units. We here use the digital soil mapping approach to predict the spatial distribution of SOC. This relies on a soil inference model based on spatially referenced environmental layers of topographic attributes, soil units, parent material, and forest history. We sampled soils at 165 sites, stratified according to topography and lithology, on Barro Colorado Island (BCI), Panama, at depths of 0–10cm, 10–20cm, 20–30cm, and 30–50cm, and analyzed them for SOC by dry combustion. We applied Random Forest (RF) analysis as a modeling tool to the SOC data for each depth interval in order to compare vertical and lateral distribution patterns. RF has several advantages compared to other modeling approaches, for instance, the fact that it is neither sensitive to overfitting nor to noise features. The RF-based digital SOC mapping approach provided SOC estimates of high spatial resolution and estimates of error and predictor importance. The environmental variables that explained most of the variation in the topsoil (0–10cm) were topographic attributes. In the subsoil (10–50cm), SOC distribution was best explained by soil texture classes as derived from soil mapping units. The estimates for SOC stocks in the upper 30cm ranged between 38 and 116Mg ha 61021 , with lowest stocks on midslope and highest on toeslope positions. This digital soil mapping approach can be applied to similar landscapes to refine the spatial resolution of SOC estimates.

DOI

19
Hengl T, Heuvelink G B M, Rossiter D G, 2007. About regression-kriging: From equations to case studies.Computers & Geosciences, 33: 1301-1315.This paper discusses the characteristics of regression-kriging (RK), its strengths and limitations, and illustrates these with a simple example and three case studies. RK is a spatial interpolation technique that combines a regression of the dependent variable on auxiliary variables (such as land surface parameters, remote sensing imagery and thematic maps) with simple kriging of the regression residuals. It is mathematically equivalent to the interpolation method variously called “Universal Kriging” (UK) and “Kriging with External Drift” (KED), where auxiliary predictors are used directly to solve the kriging weights. The advantage of RK is the ability to extend the method to a broader range of regression techniques and to allow separate interpretation of the two interpolated components. Data processing and interpretation of results are illustrated with three case studies covering the national territory of Croatia. The case studies use land surface parameters derived from combined Shuttle Radar Topography Mission and contour-based digital elevation models and multitemporal-enhanced vegetation indices derived from the MODIS imagery as auxiliary predictors. These are used to improve mapping of two continuous variables (soil organic matter content and mean annual land surface temperature) and one binary variable (presence of yew). In the case of mapping temperature, a physical model is used to estimate values of temperature at unvisited locations and RK is then used to calibrate the model with ground observations. The discussion addresses pragmatic issues: implementation of RK in existing software packages, comparison of RK with alternative interpolation techniques, and practical limitations to using RK. The most serious constraint to wider use of RK is that the analyst must carry out various steps in different software environments, both statistical and GIS.

DOI

20
Hengl T, Heuvelink G B M, Stein A, 2004. A generic framework for spatial prediction of soil variables based on regression-kriging.Geoderma, 120: 75-93.A methodological framework for spatial prediction based on regression-kriging is described and compared with ordinary kriging and plain regression. The data are first transformed using logit transformation for target variables and factor analysis for continuous predictors (auxiliary maps). The target variables are then fitted using step-wise regression and residuals interpolated using kriging. A generic visualisation method is used to simultaneously display predictions and associated uncertainty. The framework was tested using 135 profile observations from the national survey in Croatia, divided into interpolation (100) and validation sets (35). Three target variables: organic matter, pH in topsoil and topsoil thickness were predicted from six relief parameters and nine soil mapping units. Prediction efficiency was evaluated using the mean error and root mean square error (RMSE) of prediction at validation points. The results show that the proposed framework improves efficiency of predictions. Moreover, it ensured normality of residuals and enforced prediction values to be within the physical range of a variable. For organic matter, it achieved lower relative RMSE than ordinary kriging (53.3% versus 66.5%). For topsoil thickness, it achieved a lower relative RMSE (66.5% versus 83.3%) and a lower bias than ordinary kriging (0.15 versus 0.69 cm). The prediction of pH in topsoil was difficult with all three methods. This framework can adopt both continuous and categorical soil variables in a semi-automated or automated manner. It opens a possibility to develop a bundle algorithm that can be implemented in a GIS to interpolate soil profile data from existing datasets.

DOI

21
Hiemstra P, [Accessed 2013, December 15].

22
IPCC, 2006. IPCC Guidelines for national greenhouse gas inventories, prepared by the national greenhouse gas inventories programme, Eggleston H S, Buendia L, Miwa K et al. (eds.). Published: IGES, Japan.

23
Jaber S M, Al-Qinna M I, 2011. Soil organic carbon modelling and mapping in a semi-arid environment using thematic mapper data.Photogrammetric Engineering & Remote Sensing, 77(7): 709-719.This study evaluated the effectiveness of using Thematic Mapper (TM) data for estimating soil organic carbon (SOC) content in the Zarqa Basin in Jordan, a typical semi-arid environment, under natural surface conditions by testing a variety of statistical modeling techniques. This evaluation is essential for implementing carbon crediting programs for ameliorating the effects of global warming. Although none of the developed models was powerful in predicting SOC, a stepwise regression model was selected since it provided the lowest validation root mean square error (RMSE) of 10.4 metric tons per hectare (ton/ha). Using this model, a SOC map for the basin was constructed by applying map algebra. The total SOC content to 0.2 m depth of the basin was calculated to be 9,423,986.4 metric tons with SOC density of 26.3 ton/ha. This study suggested that, in semi-arid environments and using the statistical modeling techniques that were tested, TM-based SOC models cannot be used for implementing carbon crediting programs; however they can estimate total surface SOC pools in large areas to with a few percent error.

DOI

24
Jaetzold R, Schmidt H, Hornetz Bet al., 2010. Farm management handbook of Kenya, Vol. II. Natural conditions and farm management information. 2nd ed., Part B Central Kenya, Subpart B1a Southern Rift Valley Province. Ministry of Agriculture, Kenya and German Agency for Technical Cooperation (GTZ), Nairobi.

25
Karunaratne S B, Bishop T F A, Baldock J Aet al., 2014. Catchment scale mapping of measureable soil organic carbon fractions.Geoderma, 219/220: 14-23.This study aims to map the measurable fractions of soil organic carbon related to the RothC carbon model at the catchment scale and to assess the model and prediction quality. It also discusses how the outputs can be used to provide initial pool estimates for process modelling of soil carbon in a spatial context. The study was carried out in Cox's Creek catchment in northern New South Wales, Australia. Samples were collected in 2010 using a design-based sampling scheme. The measurable fractions of the RothC soil carbon model considered in this study were resistant organic carbon, humus organic carbon and particulate organic carbon. It has been reported that these measurable fractions of soil organic carbon can successfully substitute for the conceptual pools of carbon in the RothC soil carbon model. All the samples were scanned to create MIR spectra and recently developed spectroscopic models by Commonwealth Scientific and Industrial Research Organisation (CSIRO) under the national soil carbon research programme (2009-2012) were used to carry out the prediction of respective fractions. We used linear mixed models to create a model for mapping the measurable fractions of soil organic carbon across the catchment. The cross validation results revealed that the highest Lin's concordance correlation between measured and predicted values was recorded for resistant organic carbon (0.78), followed by humus organic carbon (0.74) and particulate organic carbon (0.58). Finally, to assess the uncertainty of the predictions we carried out conditional sequential Gaussian simulations. We demonstrated that measurable fractions of carbon related to the RothC model can be mapped at catchment scale with reasonable accuracy. The derived maps could be used in future studies to initialize the RothC model at any location across the landscape with quantified uncertainties.

DOI

26
Kheir R B, Greve M H, BØcher P Ket 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.<h2 class="secHeading" id="section_abstract">Abstract</h2><p id="">Soil organic carbon (SOC) is one of the most important carbon stocks globally and has large potential to affect global climate. Distribution patterns of SOC in Denmark constitute a nation-wide baseline for studies on soil carbon changes (with respect to Kyoto protocol). This paper predicts and maps the geographic distribution of SOC across Denmark using remote sensing (RS), geographic information systems (GISs) and decision-tree modeling (un-pruned and pruned classification trees). Seventeen parameters, i.e. parent material, soil type, landscape type, elevation, slope gradient, slope aspect, mean curvature, plan curvature, profile curvature, flow accumulation, specific catchment area, tangent slope, tangent curvature, steady-state wetness index, Normalized Difference Vegetation Index (NDVI), Normalized Difference Wetness Index (NDWI) and Soil Color Index (SCI) were generated to statistically explain SOC field measurements in the area of interest (Denmark). A large number of tree-based classification models (588) were developed using (i) all of the parameters, (ii) all Digital Elevation Model (DEM) parameters only, (iii) the primary DEM parameters only, (iv), the remote sensing (RS) indices only, (v) selected pairs of parameters, (vi) soil type, parent material and landscape type only, and (vii) the parameters having a high impact on SOC distribution in built pruned trees. The best constructed classification tree models (in the number of three) with the lowest misclassification error (ME) and the lowest number of nodes (<em>N</em>) as well are: (i) the tree (T1) combining all of the parameters (ME&#xA0;=&#xA0;29.5%; <em>N</em>&#xA0;=&#xA0;54); (ii) the tree (T2) based on the parent material, soil type and landscape type (ME&#xA0;=&#xA0;31.5%; <em>N</em>&#xA0;=&#xA0;14); and (iii) the tree (T3) constructed using parent material, soil type, landscape type, elevation, tangent slope and SCI (ME&#xA0;=&#xA0;30%; <em>N</em>&#xA0;=&#xA0;39). The produced SOC maps at 1:50,000 cartographic scale using these trees are highly matching with coincidence values equal to 90.5% (Map T1/Map T2), 95% (Map T1/Map T3) and 91% (Map T2/Map T3). The overall accuracies of these maps once compared with field observations were estimated to be 69.54% (Map T1), 68.87% (Map T2) and 69.41% (Map T3). The proposed tree models are relatively simple, and may be also applied to other areas.</p>

DOI

27
Kumar S, Lal R, 2011. Mapping the organic carbon stocks of surface soils using local spatial interpolator.Journal of Environmental Monitoring, 13: 3128-3135.The largest uncertainties are associated with estimating the soil organic carbon (SOC) stock because of natural soil variability and data scarcity. Thus, a local spatial geostatistical hybrid approach, the geographically weighted regression kriging (GWRK), was used in the present study to overcome some of these uncertainties. This study was designed to estimate the SOC stock (kg C m(-2)) for the surface 0 to 15 cm depth using the state of Pennsylvania as the study region. A total of 920 soil profiles were extracted from the National Soil Survey Center database and were divided into calibration (80%) and validation (20%) periods. Some soil parameters that include clay content, bulk density ((b)), total nitrogen (TN) content, pH, Ca(2+), Na(+), extractable acidity (EXACID), and cation exchange capacity (CEC) were used as covariates for estimating the SOC stock. These covariates exhibited spatial autocorrelation (Moran's Index, I = 0.62 to 0.89). Further, residuals of geographically weighted regression were spatially autocorrelated, and hence support the use of the GWRK approach. Validation results concluded that the performance of the GWRK approach was the best with the lowest values of root mean square error, mean estimation error and mean absolute estimation error. The estimated SOC stock for the surface 0 to 15 cm depth ranged from 1.41 to 3.94 kg m(-2). Results from this study show that the GWRK captures spatial dependent relationships, and addresses spatial non-stationarity issues, hence this approach improves the estimations of SOC stock.

DOI PMID

28
Kumar S, Lal R, Liu D, 2012. A geographically weighted regression kriging approach for mapping soil organic carbon stock.Geoderma, 189/190: 627-634.Local variations in the model parameters can play an important explanatory role in the spatial modeling of soil organic carbon (SOC) stock. Linear regression models assume parameters to be spatially invariant and are unable to account for the spatially varying relationships in the variables. A recently developed approach, geographically weighted regression kriging (GWRK), was used in this study to examine the relationships between environmental variables and SOC stock for the state of Pennsylvania, USA. The specific objectives were to (i) estimate the SOC stock (kg02C02m 61022 ) to 1.0-m depth, and (ii) compare the GWRK results with those obtained from regression kriging (RK). Data for 878 georeferenced soil profiles, extracted from National Soil Survey Center database, were divided into calibration ( n 02=02702) and validation ( n 02=02176) datasets. Environmental variables including temperature, precipitation, elevation, slope, geology, land use, and normalized difference vegetation index were explored and included as independent variables to establish the model for estimating the SOC stock. Results using Pennsylvania as a case study conclude that GWRK was the least biased and more accurate compared to RK for estimating the SOC stock based on the lowest root mean square error (2.61 vs. 4.6102kg02m 61022 ), and high R 2 (0.36 vs. 0.23) values. Higher stock was consistent with higher precipitation and cooler temperature of the region. Total SOC stock ranged from 1.12 to 1.1802Pg for the soils of Pennsylvania. Forests store the highest SOC stock (64% of the total), followed by croplands (22%), wetlands (2.3%), and shrubs (2%). Results show that GWRK enhances the precision for estimating the SOC stock compared to the RK since the former takes into account the spatial non-stationarity coupled with spatial autocorrelation of the residuals.

DOI

29
Kumar S, Lal R, Liu D, 2013. Estimating the spatial distribution of organic carbon density for the soils of Ohio, USA.Journal of Geographical Sciences, 23(2): 280-296.Abstract<br/><p class="a-plus-plus">Historical database of National Soil Survey Center containing 1424 geo-referenced soil profiles was used in this study for estimating the organic carbon (SOC) for the soils of Ohio, USA. Specific objective of the study was to estimate the spatial distribution of SOC density (C stock per unit area) to 1.0-m depth for soils of Ohio using geographically weighted regression (GWR), and compare the results with that obtained from multiple linear regression (MLR). About 80% of the analytical data were used for calibration and 20% for validation. A total of 20 variables including terrain attributes, climate data, bedrock geology, and land use data were used for mapping the SOC density. Results showed that the GWR provided better estimations with the lowest (3.81 kg m<sup class="a-plus-plus">−2</sup>) root mean square error (RMSE) than MLR approach. Total estimated SOC pool for soils in Ohio ranged from 727 to 742 Tg. This study demonstrates that, the local spatial statistical technique, the GWR can perform better in capturing the spatial distribution of SOC across the study region as compared to other global spatial statistical techniques such as MLR. Thus, GWR enhances the accuracy for mapping SOC density.</p><br/>

DOI

30
Lacoste M, Minasny B, McBratney Aet al., 2014. High resolution 3D mapping of soil organic carbon in a heterogeneous agricultural landscape.Geoderma, 213: 296-311.Soil organic carbon (SOC) is a key element of agroecosystems functioning and has a crucial impact on global carbon storage. At the landscape scale, SOC spatial variability is strongly affected by natural and anthropogenic processes and linear anthropogenic elements (such hedges or ditches). This study aims at mapping SOC stocks distribution in the A-horizons for a depth up to 10502cm, at a high spatial resolution, for an area of 1002km 2 in a heterogeneous agricultural landscape (North-Western France). We used a data mining tool, Cubist, to build rule-based predictive models and predict SOC content and soil bulk density (BD) from a calibration dataset at 8 standard layers (0 to 7.502cm, 7.5 to 1502cm, 15 to 3002cm, 30 to 4502cm, 45 to 6002cm, 60 to 7502cm, 75 to 9002cm and 90 to 10502cm). For the models calibration, 70 sampling locations were selected within the whole study area using the conditioned Latin hypercube sampling method. Two independent validation datasets were used to assess the performance of the predictive models: (i) at landscape scale, 49 sampling locations were selected using stratified random sampling based on a 300-m square grid; (ii) at hedge vicinity, 112 sampling locations were selected along transects perpendicular to 14 purposively chosen hedges. Undisturbed samples were collected at fixed depths and analysed for BD and SOC content at each sampling location and continuous soil profiles were reconstructed using equal-area splines. Predictive environmental data consisted in attributes derived from a light detection and ranging digital elevation model (LiDAR DEM), geological variables, land use data and a predictive map of A-horizon thickness. Considering the two validation datasets (at landscape scale and hedge vicinity), root mean square errors (RMSE) of the predictions, computed for all the standard soil layers (up to a depth of 10502cm), were respectively 7.74 and 5.0202g02kg 61021 for SOC content, and 0.15 and 0.2102g02cm 61023 for BD. Best predictions were obtained for layers between 15 and 6002cm of depth. The SOC stocks were calculated over a depth of 10502cm by combining the prediction of SOC content and BD. The final maps show that the carbon stocks in the soil below 3002cm accounted for 33% of the total SOC stocks on average. The whole method produced consistent results between the two predicted soil properties. The final SOC stocks maps provide continuous data along soil profile up to 10502cm, which may be critical information for supporting carbon policy and management decisions.

DOI

31
Lal R, 2004. Soil carbon sequestration to mitigate climate change.Geoderma, 123: 1-22.The increase in atmospheric concentration of CO 2 by 31% since 1750 from fossil fuel combustion and land use change necessitates identification of strategies for mitigating the threat of the attendant global warming. Since the industrial revolution, global emissions of carbon (C) are estimated at 270±30 Pg (Pg=petagram=10 15 g=1 billion ton) due to fossil fuel combustion and 136±55 Pg due to land use change and soil cultivation. Emissions due to land use change include those by deforestation, biomass burning, conversion of natural to agricultural ecosystems, drainage of wetlands and soil cultivation. Depletion of soil organic C (SOC) pool have contributed 78±12 Pg of C to the atmosphere. Some cultivated soils have lost one-half to two-thirds of the original SOC pool with a cumulative loss of 30–40 Mg C/ha (Mg=megagram=10 6 g=1 ton). The depletion of soil C is accentuated by soil degradation and exacerbated by land misuse and soil mismanagement. Thus, adoption of a restorative land use and recommended management practices (RMPs) on agricultural soils can reduce the rate of enrichment of atmospheric CO 2 while having positive impacts on food security, agro-industries, water quality and the environment. A considerable part of the depleted SOC pool can be restored through conversion of marginal lands into restorative land uses, adoption of conservation tillage with cover crops and crop residue mulch, nutrient cycling including the use of compost and manure, and other systems of sustainable management of soil and water resources. Measured rates of soil C sequestration through adoption of RMPs range from 50 to 1000 kg/ha/year. The global potential of SOC sequestration through these practices is 0.9±0.3 Pg C/year, which may offset one-fourth to one-third of the annual increase in atmospheric CO 2 estimated at 3.3 Pg C/year. The cumulative potential of soil C sequestration over 25–50 years is 30–60 Pg. The soil C sequestration is a truly win–win strategy. It restores degraded soils, enhances biomass production, purifies surface and ground waters, and reduces the rate of enrichment of atmospheric CO 2 by offsetting emissions due to fossil fuel.

DOI

32
Lamsal S, Grunwald S, Bruland G Let al., 2006. Regional hybrid geospatial modeling of soil nitrate-nitrogen in the Santa Fe River watershed.Geoderma, 135: 233-247.Typically, regional assessment of the spatial variability and distribution of environmental properties are constrained by sparse field observations that are costly and labor intensive. We adopted a hybrid geospatial modeling approach that combined sparsely measured soil NO 3 –N observations collected in three seasons (Sept. 2003, Jan. and May 2004) with dense auxiliary environmental datasets to predict NO 3 –N within the Santa Fe River Watershed (3585km 2 ) in north-east Florida. Elevated nitrate–nitrogen concentrations have been found in this watershed in spring, surface and ground water. We collected soil samples at four depths (0–30, 30–60, 60–120, 120–180cm) based on a random-stratified sampling design. Classification and regression trees were used to develop trend models for soil NO 3 –N predictions based on environmental correlation and predict values at the watershed scale. Residuals were spatially autocorrelated only for the Jan. 2004 sampling and regression kriging was used to combine the kriged residuals with tree-based trend estimates for this event. At each step of the upscaling process, error assessment provided important information about the uncertainty of predictions, which was lowest for the Jan. sampling event. Sites that showed consistently high NO 3 –N values throughout the cropping season (Jan–May 2004) with values ≤5μg g 611 covered 95.7% (3363.9km 2 ) of the watershed. Values in the 5–10μg g 611 range covered 4.3% (150.7km 2 ), while values exceeding 10μg g 611 covered only 0.59% (20.7km 2 ) of the watershed. Elevated soil NO 3 –N on karst, unconfined areas with sand-rich soils, or in close proximity to streams and water bodies pose the greatest risk for accelerated nitrate leaching contributing to elevated nitrogen found in spring, surface and ground water in the watershed. This approach is transferable to other land resource problems that require the upscaling of sparse site-specific data to large watersheds.

DOI

33
Lemenih M, Karltun E, Olsson M, 2005. Assessing soil chemical and physical property responses to deforestation and subsequent cultivation in smallholders farming system in Ethiopia.Agriculture, Ecosystems and Environment, 105: 373-386.

34
Lesch S M, Corwin D L, 2008. Prediction of spatial soil property information from ancillary sensor data using ordinary linear regression: Model derivations, residual assumptions and model validation tests.Geoderma, 148: 130-140.

35
Li D, Shao M, 2014. Soil organic carbon and influencing factors in different landscapes in an arid region of north-western China.Catena, 116: 95-104.Knowledge of the spatial pattern of soil organic carbon (SOC) and the factors influencing it in various landscapes is essential for understanding carbon cycles. An arid region with an area of 10002km 2 in northwestern China consisted of desert, cropland and wetland was investigated. The vertical patterns of SOC density in the three different landscapes and the horizontal distribution of SOC density in the study area were evaluated. The differences in SOC density among different landscapes and soil layers were analyzed, and the primary factors influencing SOC density were determined. The density of SOC was low and remained homogeneous in the profiles of desert soil. The vertical distributions of SOC density in cropland and wetland were well described by logarithmic functions ( R 2 02=020.97 and 0.92, respectively, P 02<020.001). Geostatistical analysis showed that SOC density presented moderate spatial variability and strong spatial dependence across all depths. Wetland and desert were easily recognized by the highest and lowest SOC densities in the study area, respectively. The densities of SOC in the 3-m profiles were 59.35, 149.6 and 174.402Mg02ha 61021 for desert, cropland and wetland, respectively. The SOC in the 1–302m layer represented 67.0, 52.7 and 58.0% of the total SOC stored in the 0–302m profiles of desert, cropland and wetland, respectively. Clay and silt particles were the major determinant of SOC in the study area. The variability in SOC density explained by clay02+02silt content increased with depth ranging from 46.0 to 82.2% in desert and from 45.3 to 76.7% in cropland. The variability in SOC density accounted for by clay02+02silt content decreased from 52.2% in the 0–0.302m layer to 43.3% in the 0–102m layer of wetland. The remaining SOC density variability could be attributed to factors not included in this study, such as geography, vegetation and the degree of erosion. Errors in the measurement of SOC concentration and the distribution of soil-particle size, however, may introduce uncertainty in the determination of soil bulk density and thus the estimation of SOC density. The concentration of SOC in the 0–0.302m layer increased by 196.3% after the reclamation of native desert less than 4002years ago and decreased by 5.3% after the cultivation of wetland as cropland for less than 3002years. Short-term cultivation is insufficient to significantly alter SOC concentration in the deeper layers of desert and wetland soils. The results of this study may be of further use in optimizing strategies for the protection of wetland, ecological restoration of desertified land and the sustainable management of cropland in arid regions of northwestern China.

DOI

36
Li M, Zhang X, Pang Get al., 2013b. The estimation of soil organic carbon distribution and storage in a small catchment area of the Loess Plateau.Catena, 101: 11-16.The distribution and storage of soil organic carbon serve as basic data for the study of soil productivity, soil hydrological properties, and the balance among carbon-based greenhouse gases. In this study, the organic carbon storage and density distribution characteristics of the soil in the Zhifanggou catchment on the Loess Plateau were studied based on field investigations, laboratory measurement, and geostatistics analysis. A total of 1,282 soil samples were collected from 215 sites in addition to 10 profiles from the catchment. The landuse within the catchment was divided into 4 types: farmland, grassland, shrubland, and woodland. The following results were obtained. 1. In the Zhifanggou catchment, the average organic carbon content of the soil at a depth of 0-100 cm is between 2.81 and 3.50 g. kg(-1). The soil organic carbon content (SOCC) for different landuse types follows the trend shrubland > woodland > grassland > farmland, whereas the soil bulk density follows. The relationship between soil organic carbon content and bulk density follows a power law function. 2. The soil organic carbon density (SOCD) at a depth of 0-100 cm is 1.24-8.34 kg m(-2). The coefficient of variation is 0.40, indicating a moderate variation in the average carbon density of 2.63 kg m(-2). The soil organic carbon density for different landuse types follows the trend shrubland > woodland > grassland > farmland. In the entire catchment, the proportion of average soil organic carbon density at a depth of 0-20 cm is 50.24%, which decreases with soil depth. The spatial distribution of the soil organic carbon density is closely related to the landuse types and topography. 3. The total soil organic carbon storage (SOCS) at a depth of 0-100 cm in the Zhifanggou catchment is 21.84x 10(6) kg. The relationship between soil depth and total organic carbon storage is linear. (c) 2012 Elsevier B.V. All rights reserved.

DOI

37
Li Q, Yue T, Wang Cet al., 2013a. Spatially distributed modeling of soil organic matter across China: An application of artificial neural network approach.Catena, 104: 210-218.Accurate prediction of spatial distribution of soil organic matter (SOM) at different scales is important for various applications related to land use and environmental problems. This study proposed a radial basis function neural networks model (RBFNN), combined with principal component analysis (PCA), to predict the spatial distribution of SOM content across China. To assess its feasibility, 6241 soil samples collected during the second national soil survey period were used. To predict the SOM at such scale, the entire study area was firstly divided into 22 different soil-landscape units according to soil types and vegetation types; then 11 quantitative environmental factors derived from climate, topography, and vegetation were converted into principal components (PC) and the first five PCs which explain 92.97% of the total data variance were selected as predictors for the purpose of eliminating the mulicollinearity of these actual variables and reducing the number of predictors; finally, a specific artificial neural network model was trained for each soil-landscape unit to capture the relationships between SOM and PCs and then used to predict the distribution of SOM content within the corresponding soil-landscape unit. The performance of this approach was evaluated by several validation indices and compared with multiple linear regression (MLR) and regression kriging (RK). The results have shown that RBFNN performs much better than both MLR and RK with much higher ratio of performance to deviation (RPD) and lower prediction errors (mean absolute error (MAE), mean relative error (MRE) and root mean squared error (RMSE)). The RPD obtained by RBFNN was 1.94, which resulted in relative improvement of 29.33% compared with RK and MLR. The three prediction errors of RBFNN were smaller than that of MLR and RK by 3.10 g.kg(-1), 17.25%, 6.25 g.kg(-1), and by 234 g.kg(-1), 5.93%, 6.24 g.kg(-1) respectively. Also, RBFNN presented a more realistic spatial pattern of SOM than RK and MLR. The good performance of this method can be attributed to the division of the study area and the capability of RBFNN to capture the nonlinear relationships between SOM and environmental factors within different soil-landscape units. The result suggests that the proposed method can play a vital role in improving prediction accuracy of SOM within a large area. (C) 2012 Elsevier B.V. All rights reserved.

DOI

38
Li Y, 2010. Can the spatial prediction of soil organic matter contents at various sampling scales be improved by using regression kriging with auxiliary information?Geoderma, 159: 63-75.

39
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 and Environment, 142: 184-194.

40
Malone B P, McBratney A B, Minasny Bet al., 2009. Mapping continuous depth functions of soil carbon storage and available water capacity.Geoderma, 154: 138-152.There is a need for accurate, quantitative soil information for natural resource planning and management. This information shapes the way decisions are made as to how soil resources are assessed and managed. This paper proposes a novel method for whole-soil profile predictions (to 1 m) across user-defined study areas where limited soil information exists. Using the Edgeroi district in north-wes...

DOI

41
Marchetti A, Piccini C, Francaviglia Ret al., 2012. Spatial distribution of soil organic matter using geostatistics: A key indicator to assess soil degradation status in central Italy.Pedosphere, 22(2): 230-242.Soil organic matter (SOM) content is one of the main factors to be considered in the evaluation of soil health and fertility. As timing, human and monetary resources often limit the amount of available data, geostatistical techniques provide a valid scientific approach to cope with spatial variability, to interpolate existing data and to predict values at unsampled locations for accurate SOM status survey. Using geostatistical and geographic information system (GIS) approaches, the spatial variability of some physical and chemical soil parameters was investigated under Mediterranean climatic condition in the Abruzzo region of central Italy, where soil erosion processes accelerated by human induced factors are the main causes of soil degradation associated with low SOM content. Experimental semivariograms were established to determine the spatial dependence of the soil variables under investigation. The results of 250 soil sampling point data were interpolated by means of ordinary kriging coupled with a GIS to produce contour maps distribution of soil texture, SOM content related to texture, and C/N ratio. The resulting spatial interpolation of the dataset highlighted a low content of SOM in relation with soil texture in most of the surveyed area (87%) and an optimal C/N ratio for only half of the investigated surface area. Spatial location of degraded area and the assessment of its magnitude can provide decision makers with an accurate support to design appropriate soil conservation strategies and then facilitate a regional planning of agri-environmental measures in the framework of the European Common Agricultural Policy.

42
Martin M P, Orton T G et al., 2014. Evaluation of modelling approaches for predicting the spatial distribution of soil organic carbon stocks at the national scale. Geoderma, .

43
Martin M P, Wattenbach M, Smith Pet al., 2011. Spatial distribution of soil organic carbon stocks in France.Biogeosciences, 8: 1053-1065.Soil organic carbon plays a major role in the global carbon budget, and can act as a source or a sink of atmospheric carbon, thereby possibly influencing the course of climate change. Changes in soil organic carbon (SOC) stocks are now taken into account in international negotiations regarding climate change. Consequently, developing sampling schemes and models for estimating the spatial distribution of SOC stocks is a priority. The French soil monitoring network has been established on a 16 km x 16 km grid and the first sampling campaign has recently been completed, providing around 2200 measurements of stocks of soil organic carbon, obtained through an in situ composite sampling, uniformly distributed over the French territory.<br/>We calibrated a boosted regression tree model on the observed stocks, modelling SOC stocks as a function of other variables such as climatic parameters, vegetation net primary productivity, soil properties and land use. The calibrated model was evaluated through cross-validation and eventually used for estimating SOC stocks for mainland France. Two other models were calibrated on forest and agricultural soils separately, in order to assess more precisely the influence of pedo-climatic variables on SOC for such soils.<br/>The boosted regression tree model showed good predictive ability, and enabled quantification of relationships between SOC stocks and pedo-climatic variables (plus their interactions) over the French territory. These relationships strongly depended on the land use, and more specifically, differed between forest soils and cultivated soil. The total estimate of SOC stocks in France was 3.260 +/- 0.872 PgC for the first 30 cm. It was compared to another estimate, based on the previously published European soil organic carbon and bulk density maps, of 5.303 PgC. We demonstrate that the present estimate might better represent the actual SOC stock distributions of France, and consequently that the previously published approach at the European level greatly overestimates SOC stocks.

DOI

44
McBratney A B, Santos M L M, Minasny B, 2003. On digital soil mapping.Geoderma, 117: 3-52.

45
McCall G J H, 1967. Geology of the Nakuru-Thomson’s falls-Lake Hannington area: Degree sheet No. 35, S.W. Quarter and 43 N.W. Quarter, Report No. 78. Government Printer, Nairobi.

46
McKenzie N J, Ryan P J, 1999. Spatial prediction of soil properties using environmental correlation.Geoderma, 89: 67-94.Conventional survey methods have efficiencies in medium to low intensity survey because they use relationships between soil properties and more readily observable environmental features as a basis for mapping. However, the implicit predictive models are qualitative, complex and rarely communicated in a clear manner. The possibility of developing an explicit analogue of conventional survey practice suited to medium to low intensity surveys is considered. A key feature is the use of quantitative environmental variables from digital terrain analysis and airborne gamma radiometric remote sensing to predict the spatial distribution of soil properties. The use of these technologies for quantitative soil survey is illustrated using an example from the Bago and Maragle State Forests in southeastern Australia. A design-based, stratified, two-stage sampling scheme was adopted for the 50,000 ha area using digital geology, landform and climate as stratifying variables. The landform and climate variables were generated using a high resolution digital elevation model with a grid size of 25 m. Site and soil data were obtained from 165 sites. Regression trees and generalised linear models were then used to generate spatial predictions of soil properties using digital terrain and gamma radiometric survey data as explanatory variables. The resulting environmental correlation models generate spatial predictions with a fine grain unmatched by comparable conventional survey methods. Example models and spatial predictions are presented for soil profile depth, total phosphorus and total carbon. The models account for 42%, 78% and 54% of the variance present in the sample respectively. The role of spatial dependence, issues of scale and landscape complexity are discussed along with the capture of expert knowledge. It is suggested that environmental correlation models may form a useful trend model for various forms of kriging if spatial dependence is evident in the residuals of the model.

DOI

47
Meersmans J, de Ridder F, Canters Fet al., 2008. A multiple regression approach to assess the spatial distribution of Soil Organic Carbon (SOC) at the regional scale (Flanders, Belgium).Geoderma, 143: 1-13.

48
Mehrjardi R T, Minasny B, Sarmadian Fet al., 2014. Digital mapping of soil salinity in Ardakan region, central Iran.Geoderma, 213: 115-128.Salinization and alkalinization are the most important land degradation processes in central Iran. In this study we modelled the vertical and lateral variation of soil salinity (measured as electrical conductivity in saturation paste, ECe) using a combination of regression tree analysis and equal-area smoothing splines in a 72,00002ha area located in central Iran. Using the conditioned Latin hypercube sampling method, 173 soil profiles were sampled from the study area, and then analysed for ECe and other soil properties. Auxiliary data used in this study to represent predictive soil forming factors were terrain attributes (derived from a digital elevation model), Landsat 7 ETM + data, apparent electrical conductivity (ECa)—measured using an electromagnetic induction instrument (EMI), and a geomorphologic surfaces map. To derive the relationships between ECe (from soil surface to 102m) and the auxiliary data, regression tree analysis was applied. In general, results showed that the ECa surfaces are the most powerful predictors for ECe at three depth intervals (i.e. 0–15, 15–30 and 30–6002cm). In the 60–10002cm depth interval, topographic wetness index was the most important parameter used in regression tree model. Validation of the predictive models at each depth interval resulted in R 2 values ranging from 78% (0–1502cm) to 11% (60–10002cm). Thus we can recommend similar applications of this technique could be used for mapping soil salinity in other parts in Iran.

DOI

49
Mishra U, Lal R, Liu Det al., 2010. Predicting the spatial variation of the soil organic carbon pool at a regional scale.Soil Science Society of America Journal, 74: 906-914.Estimates of soil organic C (SOC) storage and their variability at various spatial scales are essential to better understand the global C cycle, estimate C sink capacity, identify effective C sequestration strategies, and quantify the amount of SOC sequestered during a specific period of time. This study used a geographically weighted regression (GWR) approach to predict the SOC pool at a regional scale. The GWR considers varying relationships between the SOC pool and environmental variables across the study area. The range of the variogram of SOC observations was used to define a search radius in the GWR. Terrain attributes, climate data, land use data, bedrock geology, and normalized difference vegetation index data were used to predict the SOC pool for seven states in the midwestern United States. The prediction accuracy of this SOC pool map was compared with the multiple linear regression (MLR) and regression kriging (RK) approaches. Higher contrast and wider variability (1.73-39.3 kg m(-2)) of the SOC pool were predicted with lower global prediction errors (mean estimation error = -0.11 kg m(-2), RMSE = 6.40 kg m(-2)) in GWR compared with the other approaches. A relative improvement of 22% over MLR and 2% over RK was observed in SOC prediction. The total SOC pool to the 0.5-m depth was estimated to be 6.22 Pg. The results suggest that the GWR approach is a promising tool for regional-scale SOC prediction.

DOI

50
Mishra U, Riley W J, 2012. Alaskan soil carbon stocks: Spatial variability and dependence on environmental factors.Biogeosciences, 9: 3637-3645.The direction and magnitude of soil organic carbon (SOC) changes in response to climate change depend on the spatial and vertical distributions of SOC. We estimated spatially-resolved SOC stocks from surface to C horizon, distinguishing active-layer and permafrost-layer stocks, based on geospatial analysis of 472 soil profiles and spatially referenced environmental variables for Alaska. Total Alaska state-wide SOC stock was estimated to be 77 Pg, with 61% in the active-layer, 27% in permafrost, and 12% in non-permafrost soils. Prediction accuracy was highest for the active-layer as demonstrated by highest ratio of performance to deviation (1.5). Large spatial variability was predicted, with whole-profile, active-layer, and permafrost-layer stocks ranging from 1-296 kg C m-2, 2-166 kg m-2, and 0-232 kg m-2, respectively. Temperature and soil wetness were found to be primary controllers of whole-profile, active-layer, and permafrost-layer SOC stocks. Secondary controllers, in order of importance, were: land cover type, topographic attributes, and bedrock geology. The observed importance of soil wetness rather than precipitation on SOC stocks implies that the poor representation of high-latitude soil wetness in Earth System Models may lead to large uncertainty in predicted SOC stocks under future climate change scenarios. Under strict caveats described in the text and assuming temperature changes from the A1B Intergovernmental Panel on Climate Change emissions scenario, our geospatial model indicates that the equilibrium average 2100 Alaska active-layer depth could deepen by 11 cm, resulting in a thawing of 13 Pg C currently in permafrost. The equilibrium SOC loss associated with this warming would be highest under continuous permafrost (31 %), followed by discontinuous (28 %), isolated (24.3 %), and sporadic (23.6 %) permafrost areas. Our high resolution mapping of soil carbon stock reveals the potential vulnerability of high-latitude soil carbon and can be used as a basis for future studies of anthropogenic and climatic perturbations.

DOI

51
Montgomery D C, Peck E A, Vining G G, 2006. Introduction to Linear Regression Analysis. John Wiley & Sons, Inc., New Jersey.

52
Mora-Vallejo A, Claessens L, Stoorvogel Jet al., 2008. Small-scale digital soil mapping in southeastern Kenya.Catena, 76: 44-53.Digital soil mapping techniques appear to be an interesting alternative for traditional soil survey techniques. However, most applications deal with (semi-)detailed soil surveys where soil variability is determined by a limited number of soil forming factors. The question that remains is whether digital soil mapping techniques are equally suitable for exploratory or reconnaissance soil surveys in more extensive areas with limited data availability. We applied digital soil mapping in a 13,500 km2 study area in Kenya with the main aim to create a reconnaissance soil map to assess clay and soil organic carbon contents in terraced maize fields. Soil spatial variability prediction was based on environmental correlation using the concepts of the soil forming factors equation. During field work, 95 composite soil samples were collected. Auxiliary spatially exhaustive data provided insight on the spatial variation of climate, land cover, topography and parent material. The final digital soil maps were elaborated using regression kriging. The variance explained by the regression kriging models was estimated as 13% and 37% for soil organic carbon and clay respectively. These results were confirmed by cross-validation and provide a significant improvement compared to the existing soil survey.

DOI

53
Murty D, Kirschbaum M F, McMurtrie R Eet al., 2002. Does conversion of forest to agricultural land change soil carbon and nitrogen? A review of the literature.Global Change Biology, 8: 105-123.Soil carbon is a large component of the global carbon cycle and its management can significantly affect the atmospheric COconcentration. An important management issue is the extent of soil carbon (C) release when forest is converted to agricultural land. We reviewed the literature to assess changes in soil C upon conversion of forests to agricultural land. Analyses are confounded by changes in soil bulk density upon land-use change, with agricultural soils on average having 13% higher bulk density. Consistent with earlier reviews, we found that conversion of forest to cultivated land led to an average loss of approximately 30% of soil C. When we restricted our analysis to studies that had used appropriate corrections for changes in bulk density, soil C loss was 22%. When, from all the studies compiled, we considered only studies reporting both soil C and nitrogen (N), average losses of C and N were 24% and 15%, respectively, hence showing a decrease in the average C : N ratio. The magnitude of these changes in the C : N ratio did not correlate with either C or N changes. When considering the transition from forest to pasture, there was no significant change in either soil C or N, even though reported changes in soil C ranged from -50% to +160%. Among studies that reported changes in soil N as well as soil C, C : N ratios both increased and decreased, with trends depending on changes in system N. Systems with increasing soil N generally had decreased C : N ratios, whereas systems with decreasing soil N had increased C : N ratios. Our survey confirmed earlier findings that conversion of forest to cropland generally leads to a loss of soil carbon, although the magnitude of change might have been inflated in many studies by the confounding influence of bulk-density changes. In contrast, conversion of forest to uncultivated grazing land did not, on average, lead to loss of soil carbon, although individual sites may lose or gain soil C, depending on specific circumstances, such as application of fertiliser or retention or removal of plant residues.

DOI

54
Nelson D W, Sommers L E, 1982. Total carbon, organic carbon and organic matter. In: Page A L (ed.) Methods of Soil Analysis, Part 2, Chemical and Microbiological Properties. 2nd ed. American society of agronomy, Inc., Madison, Wisconsin, USA.Analytical methods for determining total carbon in soils are reviewed. Details are given of methods for determining total carbon by dry combustion (medium-temperature resistance furnace method, high-temperature induction furnace method and instrumental methods) and wet combustion. Procedures are given for the determination of organic carbon in calcareous and non-calcareous soils based on the di...

DOI

55
Obade V P, Lal R, 2013. Assessing land cover and soil quality by remote sensing and geographical information systems (GIS).Catena, 104: 77-92.Precise soil quality assessment is critical for designing sustainable agriculture policies, restoring degraded soils, carbon (C) modeling, and improving environmental quality. Although the consequences of soil quality reduction are generally recognized, the spatial extent of soil degradation is difficult to determine, because no universal equation or soil quality prediction model exists that fits all ecoregions. Furthermore, existing soil organic C (SOC) models generate estimates with uncertainties that may exceed 50%. Therefore it is possible that drastic changes in soil quality may be occurring in sites which are not identifiable on existing maps. Soil quality can either be directly inferred from SOC concentration, or through the assessment of the soil physical, chemical and biologic properties. Assessing the spatial distribution of SOC over large areas requires the calibration and development of models derived from laboratory or field based techniques. However, mapping SOC concentration in all soils is logistically challenging by using normal standard survey techniques. The availability of new generations of remotely sensed datasets and geographical information system (GIS) models (i.e. GEMS, RothC, and CENTURY) provides new opportunities for predicting soil properties and quality at different spatial scales. This article discusses the current approaches, identifies gaps and proposes improvements in techniques for measuring soil quality within agricultural fields. (C) 2012 Elsevier B.V. All rights reserved.

DOI

56
Okalebo J R, Gathna K W, Woomer P L, 2002. Laboratory methods for soil and plant analysis: A working manual. 2nd ed. Tropical Soil Biology and Fertility Programme, Nairobi.

57
Overmars K P, Verburg P H, 2005. Analysis of land use drivers at the watershed and household level: Linking two paradigms at the Philippine forest fringe.International Journal of Geographical Information Science, 19(2): 125-152.Land use and land cover change (LUCC) is the result of the complex interactions between behavioural and structural factors (drivers) associated with the demand, technological capacity, social relations and the nature of the environment in question. Although no general theory of land use change exists, different disciplinary theories can help us to analyse aspects of LUCC in specific situations. However, paradigms and theories applied by the different disciplines are often difficult to integrate and their specific research results do not easily combine into an integrated understanding of LUCC. Geographical approaches often aim to identify the location of LUCC in a spatially explicit way, while socio-economic studies aim to understand the processes of LUCC, but often lack spatial context and interactions. The objective of this study is to integrate process information from a socio-economic study into a geographical approach. First, a logistic regression analysis is performed on household survey data from interviews. In this approach the occurrence of the land use types corn, wet rice and banana is explained by a set of variables that are hypothesised to be explanatory for those land use types, with fields as the unit of analysis. The independent variables consist of household characteristics, like ethnicity and age, and plot and field information, like tenure, slope and travel time. The results of these analyses are used to identify key variables explaining land use choice, which subsequently are also collected at watershed level, using maps, census data and remote sensing imagery. Logistic regression analysis of this spatial dataset, where a ten percent sample of a 50 by 50 m grid was analysed, shows that the key variables identified in the household analysis are also important at the watershed level. Important drivers in the study area are, among others, slope, ethnicity, accessibility and place of birth. The differences in the contribution of the variables to the models at household and watershed level can be attributed to differences in spatial extent and data representation. Comparing the model with a mainstream geographical approach indicates that the spatial model informed by the household analysis gives a better insight in the actual processes determining land use than the mainstream geographic approach.

DOI

58
Pachomphon K, Dlamini P, Chaplot V, 2010. Estimating carbon stocks at regional level using soil information and easily accessible auxiliary variables.Geoderma, 155: 372-380.One of the most important challenges of digital soil mapping is the development of methods that allow the characterisation of large areas with a high-resolution. Surface soils, forming the largest pool of terrestrial organic carbon, may be able to sequester atmospheric carbon and thus mitigate climate change. However, this remains controversial, largely due to insufficient information on SOC st...

DOI

59
Pebesma E, Bivand R S, Rowlingson B et al., 2013. Classes and methods for spatial data in R. Available: [Accessed 2013, December 15].

60
R Core Team, .

61
Razakamanarivo R H, Grinand C, Razafindrakoto M Aet al., 2011. Mapping organic carbon stocks in eucalyptus plantations of the central highlands of Madagascar: A multiple regression approach.Geoderma, 162: 335-346.Recent concerns about global warming have resulted in more concerted studies on quantification and modeling of carbon (C) storage in different ecosystems. The aim of this study was to assess and map the carbon stocks in above (ABG), below-ground (BLG) biomass and soil organic carbon contained in the 30 centimeter top-layer (SOC) in coppices of eucalyptus plantations in the central highlands of Madagascar in an area of 1590 ha. Relationships between C stock and various biophysical (stool or shoot stockings and ages, circumferences) and spatial (elevation, slope, and soil type) factors that may affect C storage within each pool were investigated. Three different modeling techniques were tested and compared for various factor sets: (i) simple linear regression (SLM), (ii) multiple linear (MLM) models and, (iii) boosted regression tree (BRT) models. Weights of the factors in the respective model were analyzed for the three pool-specific models that produced the highest accuracy measurement. A regional spatial prediction of carbon stocks was performed using spatial layers derived from a digital elevation model, remote sensing imagery and expert knowledge. Results showed that BRT had the best predictive capacity for C stocks compared with the linear regression models. Elevation and slope were found to be the most relevant predictors for modeling C stock in each pool, and mainly for the SOC. A factor representing circumferences of stools and their stocking (stools.ha(-1)) largely influenced BLG. Shoot circumference at breast height and shoot age were the best factors for ABG fitting. Accuracy assessment carried out using coefficient of determination (R-2) and ratio of standard deviation to prediction error (RPD) showed satisfactory results, with 0.74 and 1.95 for AGB, 0.85 and 2.59 for BLG, and 0.61 and 1.6 for SOC respectively. Application of the best fitted models with spatial explanatory factors allowed to map and estimate C contained within each pool : 32 +/- 13 Gg C for ABG, 67 +/- 15 Gg C for BLG and, 139 +/- 36 Gg C for SOC (1 Gg=10(9) g). A total of 238 +/- 40 Gg C was obtained for the entire study area by combining the three C maps. Despite their relatively low predictive quality, models and C maps produced herein provided relevant reference values of C storage under plantation ecosystems in Madagascar. This study contributed to the reducing of uncertainty related to C monitoring and baseline definition in managed terrestrial ecosystem. (C) 2011 Elsevier B.V. All rights reserved.

DOI

62
Scull P, Franklin J, Chadwick O Aet al., 2003. Predictive soil mapping: A review. Progress in Physical Geography, 27(2): 171-197.Predictive soil mapping (PSM) can be defined as the development of a numerical or statistical model of the relationship among environmental variables and soil properties, which is then applied to a geographic data base to create a predictive map. PSM is made possible by geocomputational technologies developed over the past few decades. For example, advances in geographic information science, digital terrain modeling, remote sensing, fuzzy logic has created a tremendous potential for improvement in the way that soil maps are produced. The State Factor soil-forming model, which was introduced to the western world by one of the early Presidents of the American Association of Geographers (C.F.Marbut), forms the theoretical basis of PSM. PSM research is being driven by a need to understand the role soil plays in the biophysical and biogeochemical functioning of the planet. Much research has been published on the subject in the last 20 years (mostly outside of geographic journals) and methods have varied widely from statistical approaches (including geostatistics) to more complex methods, such as decision tree analysis, and expert systems. A geographic perspective is needed because of the inherently geographic nature of PSM.

DOI

63
Selige T, Böhner J, Schmidhalter U, 2006. High resolution topsoil mapping using hyperspectral image and field data in multivariate regression modeling procedures.Geoderma, 136: 235-244.The spatial variability of within field topsoil texture and organic matter was studied using airborne hyperspectral imagery so as to develop improved fine-scale soil mapping procedures. Two important topsoil variables for precision farming applications, soil organic matter and soil texture, were found to be correlated with spectral properties of the airborne HyMap scanner. The percentage sand, clay, organic carbon and total nitrogen content could be predicted quantitatively and simultaneously by a multivariate calibration approach using either partial least-square regression (PLSR) or multiple linear regression (MLR). The different topsoil parameters are determined simultaneously from the spectral signature contained in the single hyperspectral image, since the various variables were represented by varying combinations of wavebands across the spectra. The methodology proposed provides a means of simultaneously estimating topsoil organic matter and texture in a rapid and non-destructive manner, whilst avoiding the spatial accuracy problems associated with spatial interpolation. The use of high spatial resolution and hyperspectral remotely sensed data in the manner proposed in this paper can also be used to monitor and better understand the influence of management and land use practices on soil organic matter composition and content.

DOI

64
Smith P, 2004. Soils as carbon sinks: The global context.Soil Use and Management, 20: 212-218.Abstract. Soil carbon sequestration could meet at most about one-third of the current yearly increase in atmospheric CO-carbon, but the duration of the effect would be limited, with significant impacts lasting only 20–50 years. Coupled with this limited duration, increases in population and per-capita energy demand mean that soil carbon sequestration could play only a minor role in closing the difference between predicted and target carbon emissions by 2100. However, if atmospheric COconcentrations are to be stabilized at reasonable levels (450–650 ppm), drastic reductions in carbon emissions will be required over the next 20–30 years. Given this, carbon sequestration should form a central role in any portfolio of measures to reduce atmospheric COconcentrations over this crucial period, while new energy technologies are developed and implemented. International agreements, such as the Kyoto Protocol, encourage soil carbon sequestration and could be used to formulate soil carbon sequestration polices. Such policies need to take account of other environmental impacts as well as political, economic and societal needs, so that they form part of a raft of measures encouraging sustainable development. Of the carbon sequestration options available, those of a ‘win–win’ nature, that is, those that increase carbon stocks at the same time as improving other aspects of the environment, and those that protect or enhance existing stocks (‘no regrets’ implementation) show the greatest promise in meeting these goals.

DOI

65
Smith P, 2008. Land use change and soil organic carbon dynamics.Nutr. Cycl. Agroecosyst., 81: 169-178.Historically, soils have lost 40-90 Pg (C) globally through cultivation and disturbance with current rates of C loss due to land use change of about 1.6 ± 0.8 Pg C y63641; mainly in the tropics. Since soils contain more than twice the C found in the atmosphere, loss of C from soils can have a significant effect of atmospheric CO60 concentration, and thereby on climate. Halting land-use conversion would be an effective mechanism to reduce soil C losses, but with a growing population and changing dietary preferences in the developing world, more land is likely to be required for agriculture. Maximizing the productivity of existing agricultural land and applying best management practices to that land would slow the loss of, or is some cases restore, soil C. There are, however, many barriers to implementing best management practices, the most significant of which in developing countries are driven by poverty. Management practices that also improve food security and profitability are most likely to be adopted. Soil C management needs to considered within a broader framework of sustainable development. Policies to encourage fair trade, reduced subsidies for agriculture in developed countries and less onerous interest on loans and foreign debt would encourage sustainable development, which in turn would encourage the adoption of successful soil C management in developing countries. If soil management is to be used to help address the problem of global warming, priority needs to be given to implementing such policies.

DOI

66
Sumfleth K, Duttmann R, 2008. Prediction of soil property distribution in paddy soil landscapes using terrain data and satellite information as indicators.Ecological Indicators, 485-501.Sustainable land management and land use planning require reliable information about the spatial distribution of the physical and chemical soil properties affecting both landscape processes and services. Although many studies have been conducted to identify the spatial patterns of soil property distribution on various scales and in various landscapes, only little is known about the relationships underlying the spatial distribution of soil properties in intensively used, finely structured paddy soil landscapes in the southeastern part of China. In order to provide adequate soil information for the modelling of landscape processes, such as soil water movement, nutrient leaching, soil erosion and plant growth, this study investigates to what extent cheap and readily available ancillary information derived from digital elevation models and remote sensing data can be used to support soil mapping and to indicate soil characteristics on the landscape scale. This article focuses on the spatial prediction of the total carbon and nitrogen content and of physical soil properties such as topsoil silt, sand and clay content, topsoil depth and plough pan thickness. Correlation analyses indicate that, on the one side, the distribution of C, N and silt contents is quite closely related to the NDVI of vegetated surfaces and that, on the other side, it corresponds significantly to terrain attributes such as relative elevation, elevation above nearest drainage channel and topographical wetness index. Geostatistical analyses furthermore reflect a moderately structured spatial correlation of these soil variables. The combined use of the above mentioned terrain variables and the NDVI in a multiple linear regression accounted for 29% (silt) to 41% (total C) of the variance of these soil properties. In order to select the best prediction method to accurately map soil property distribution, we compared the performance of different regionalization techniques, such as multi-linear regression, simple kriging, inverse distance to a power, ordinary kriging and regression kriging. Except for the prediction of topsoil clay content, in all cases regression kriging model performed best. Compared to simple kriging, the spatial prediction was improved by up to 14% (total C), 13% (total N) and 10% (silt). Since the autocorrelation lengths of these spatially well correlated soil variables range between three and five times the soil sampling density, we consider regression kriging model to be a suitable method for reducing the soil sampling density. It should help to save time and costs when doing soil mapping on the landscape scale, even in intensively used paddy soil landscapes.

DOI

67
Szymanowski M, Kryza M, 2012. Local regression models for spatial interpolation of urban heat island: An example from Wrocław, SW Poland.Theor. Appl. Climatol., 108: 53-71.

68
Tamooh F, van den Meersche K, Meysman Fet al., 2012. Distribution and origin of suspended matter and organic carbon pools in the Tana River basin, Kenya.Biogeosciences, 9: 2905-2920.

69
Tesfahunegn G B, Tamene L, Vlek P L G, 2011. Catchment scale spatial variability of soil properties and implications on site-specific soil management in northern Ethiopia.Soil & Tillage Research, 117: 124-139.Scientific information on the spatial variability and distribution of soil properties is critical for understanding ecosystem processes and designing sustainable soil-crop and environmental management decisions. However, little is known on spatial distribution and variability of soil properties at catchment-scale in many tropical developing regions including Ethiopia. This study aims to examine catchment-scale spatial dependence and variability of soil properties using classical and geostatistical methods to indicate for site-specific soil management in the Mai-Negus catchment, northern Ethiopia. Soil samples were collected based on sampling zones identified by the knowledge of local farmers and field observation and analyzed following standard laboratory procedures for selected soil properties. The coefficient of variation of the soil properties ranged from 8.6% (pH) to 73.4% (clay) at catchment-scale. The mean soil organic carbon (OC) (1.21%), total nitrogen (TN) (0.12%), and available phosphorus (Pay) (7.8 mg kg(-1)) of the soils in the catchment were low, whereas high in exchangeable potassium (Ex K) (0.77 cmol(c) kg(-1)), and medium in cation exchange capacity (CEC) (23.4 cmol(c) kg(-1)) compared to the rate for African soils reported in literature. The results of semivariograms indicated a strong (8%) to moderate (63%) degree of spatial dependence for the soil properties. In addition, the goodness-of-prediction criterium (G) are higher than zero indicating that spatial soil properties mapped based on kriging interpolation are more accurate than the catchment average value (classical statistics) for site-specific management decisions. This study indicates a wide range of variability in the soil properties as the kriged maps of the soil properties at catchment-scale showed for sand (15-70%), silt (18-77%), clay (3-51%), bulk density (1.00-2.00 Mg m(-3)), OC (0.20-4.5%), TN (0.05-1.0%), Pay (1-26 mg kg(-1)), Ex K (0.10-1.30 cmol(c) kg(-1)), exchangeable calcium, Ex Ca (5-28 cmol(c) kg(-1)), exchangeable magnesium, Ex Mg (2-15 cmol(c) kg(-1)), CEC (8-51 cmol(c) kg(-1)), and iron (3-45 mg kg(-1)). The lowest soil nutrients and fine soil particles were measured on the sub-sampling zones such as low soil quality, eroded sites, and marginal land soils. Introducing appropriate interventions such as conservation tillage, fertilizer rates, agro-forestry practices, crop rotation, exclosure degraded lands, and conservation measures based on the kriged soil properties maps produced is crucial for sustainable production and environmental services. (C) 2011 Elsevier B.V. All rights reserved.

DOI

70
UNEP, 2009[Accessed 2013, August 28].

71
Vågen T G, Winowiecki L A, 2013a. Mapping of soil organic carbon stocks for spatially explicit assessments of climate change mitigation potential. Environmental Research Letters 8, 015011 (9pp). doi: 10.1088/1748- 9326/8/1/015011.Current methods for assessing soil organic carbon (SOC) stocks are generally not well suited for understanding variations in SOC stocks in landscapes. This is due to the tedious and time-consuming nature of the sampling methods most commonly used to collect bulk density cores, which limits repeatability across large areas, particularly where information is needed on the spatial dynamics of SOC stocks at scales relevant to management and for spatially explicit targeting of climate change mitigation options. In the current study, approaches were explored for (i) field-based estimates of SOC stocks and (ii) mapping of SOC stocks at moderate to high resolution on the basis of data from four widely contrasting ecosystems in East Africa. Estimated SOC stocks for 0-30 cm depth varied both within and between sites, with site averages ranging from 2 to 8 kg m. The differences in SOC stocks were determined in part by rainfall, but more importantly by sand content. Results also indicate that managing soil erosion is a key strategy for reducing SOC loss and hence in mitigation of climate change in these landscapes. Further, maps were developed on the basis of satellite image reflectance data with multiple R-squared values of 0.65 for the independent validation data set, showing variations in SOC stocks across these landscapes. These maps allow for spatially explicit targeting of potential climate change mitigation efforts through soil carbon sequestration, which is one option for climate change mitigation and adaptation. Further, the maps can be used to monitor the impacts of such mitigation efforts over time.

DOI

72
Vågen T G, Winowiecki L A, Abegaz Aet al., 2013b. Landsat-based approaches for mapping of land degradation prevalence and soil functional properties in Ethiopia.Remote Sensing of Environment, 134: 266-275.Agriculture is the basis of the Ethiopian economy, accounting for the majority of its employment and export earnings. Land degradation is, however, widespread and improved targeting of land management interventions is needed, taking into account the variability of soil properties that affect agricultural productivity and land degradation risk across landscapes. In the current study we demonstrate the utility of Landsat ETM + imagery for landscape-level assessments of land degradation risk and soil condition through a combination of systematic field methodologies, infrared (IR) spectroscopy and ensemble modeling techniques. The approaches presented allow for the development of maps at spatial scales that are appropriate for making spatially explicit management recommendations. Field data and soil samples collected from 38 sites, each 100 km(2), were used to develop predictive models that were applied as part of a case study to an independent dataset from four sites in Ethiopia. The predictions based on Landsat reflectance were robust, with R-squared values of 0.86 for pH and 0.79 for soil organic carbon (SOC), and were used to create predicted surfaces (maps) for these soil properties. Further, models were developed for the mapping of the occurrence of soil erosion and root depth restrictions within 50 cm of the soil surface (RDR50), with an accuracy of about 80% for both variables. The maps generated from these models were used to assess the spatial distribution of soil pH and SOC, which are important indicators of soil condition, and land degradation risk factors in order to target relevant management options. (C) 2013 Elsevier Inc. All rights reserved.

DOI

73
Vasques G M, Grunwald S, Comerford N Bet al., 2010a. Regional modelling of soil carbon at multiple depths within a subtropical watershed.Geoderma, 156: 326-336.Environmental factors that exert control over fine-scale spatial patterns of soil organic carbon (SOC) within profiles and across large regions differ by geographic location and landscape setting. Regions with large SOC storage and high variability can serve as natural laboratories to investigate how environmental factors generate vertical and horizontal SOC patterns across the landscape. This was investigated in the Santa Fe River watershed (SFRW), Florida, where we modelled the spatial distribution of total C (TC) at four depths to 18002cm (0–30, 30–60, 60–120, and 120–18002cm) and at an aggregated depth of 0–10002cm. A total of 554 samples from 141 sampling sites distributed along land use and soil order combinations were analyzed for TC by high-temperature combustion. A vertical trend of TC stocks decreasing with depth was identified. Horizontal trends of TC were modelled to identify the environmental determinants of TC in the SFRW. We used analysis of variance (ANOVA) and compared regression block kriging with lognormal block kriging to scale up TC across the SFRW. Total soil C was influenced by soil depth, land use, soil type, soil drainage class, and geologic unit. Regression kriging performed better than block kriging to scale up TC at three out of five depth intervals. This indicates that in the majority of cases environmental factors were the major determinants of the spatial distribution of TC relative to its spatial autocorrelation. At 60–120 and 120–18002cm, the local spatial dependence of TC was more important than environmental factors to explain its variation across the watershed. Our models show that 54.002Tg (teragrams) of C is held in the upper 102m of soils in the SFRW, and significant amounts are stored in deeper layers. They also identified the major factors responsible for regional spatial patterns of TC in this subtropical region, providing information to support current efforts of conservation of soil resources in Florida, and under similar environmental conditions in the southeastern U.S. and elsewhere.

DOI

74
Vasques G M, Grunwald S, Sickman J Oet al., 2010b. Up-scaling of dynamic soil organic carbon pools in a north-central Florida watershed.Soil Science Society of America Journal, 74. doi: 10.2136/sssaj2009.0242.Regional-scale assessment of soil C pools is essential to provide information for C cycling models, land management, and policy decisions, and elucidate the relative contribution of different C pools to total C (TC). We estimated TC and four soil C fractions, namely recalcitrant C (RC), hydrolyzable C (HC), hot-water-soluble C (SC), and mineralizable C (MC), at 0 to 30 cm across a 3585-km(2) mixed-use watershed in north-central Florida. We used lognormal block kriging (BK) and regression block kriging (RK) to upscale soil C using 102 training samples and compared the models using 39 validation samples. Regression kriging produced the most accurate models for TC and RC, whereas the labile C fractions (HC, SC, and MC) were best modeled by BK. Maps produced by BK showed similar spatial patterns due to the strong correlation between the labile C fractions and the similarity of their spatial dependence structure. Estimates of TC and RC were similar due to their high correlation and the similarity of their global trend models. Total soil C amounted to 27.40 Tg across the watershed, indicating the potential of these soils to store C. Recalcitrant C totaled 22.49 Tg (82% of TC), suggesting that a large amount of TC could be potentially stored for centuries to millennia. Our estimates of soil C and fractions within a mixed-use watershed in Florida highlight the importance of appropriately characterizing the inherent spatial dependence structure of soil C, as well as relevant regional environmental patterns (e.g., hydrology), to better explain the variability of soil C.

DOI

75
Wang K, Zhang C, Li W, 2013. Predictive mapping of soil total nitrogen at a regional scale: A comparison between geographically weighted regression and co-kriging.Applied Geography, 42: 73-85.

76
Were K O, Dick Ø B, Singh B R, 2013. Remotely sensing the spatial and temporal land cover changes in Eastern Mau forest reserve and Lake Nakuru drainage basin, Kenya.Applied Geography, 41: 75-86.This study aimed at characterizing land cover dynamics for four decades in Eastern Mau forest and lake Nakuru basin, Kenya. The specific objectives were to: (i) identify and map the major land cover types in 1973, 1985, 2000 and 2011; (ii) detect and determine the magnitude, rates and nature of the land cover changes that had occurred between these dates, and; (iii) establish the spatial and temporal distribution of these changes. Land cover types were discriminated through partitioning, hybrid classification and spatial reclassification of multi-temporal Landsat imagery. The land cover products were then validated and overlaid in post-classification comparison to detect the changes between 1973 and 2011. The accuracies of the land cover maps for 1973, 1985, 2000 and 2011 were 88%, 95%, 80% and 89% respectively. Six land cover classes, namely forests-shrublands, grasslands, croplands, built-up lands, bare lands and water bodies, were mapped. Forests-shrublands dominated in 1973, 1985 and 2000 covering about 1067 km(2), 893 km(2) and 797 km(2) respectively, but were surpassed by croplands (953 km(2)) in 2011. Bare lands occupied the least area that varied between 2 km(2) and 7 km(2) during this period. Overall, forests-shrublands and grasslands decreased by 428 km(2) and 258 km(2) at the annual average rates of 1% each, whereas croplands and built-up lands expanded by 660 km(2) and 24 km(2) at the annual rates of 6% and 16% respectively. The key hotspots of these changes were distributed in all directions of the study area, but at different times. Therefore, policies that integrate restoration and conservation of natural ecosystems with enhancement of agricultural productivity are strongly recommended. This will ensure environmental sustainability and socio-economic well-being in the area. Future research needs to assess the impacts of the land cover changes on ecosystem services and to project the future patterns of land cover changes. (C) 2013 Elsevier Ltd. All rights reserved.

DOI

77
Were K O, Singh B R, Dick Ø B, 2015. Effects of land cover changes on soil organic carbon and nitrogen stocks in the Eastern Mau Forest Reserve, Kenya. In: Lal R, Singh B R, Mwaseba D L et al., (eds.). Sustainable Intensification to Advance Food Security and Enhance Climate Resilience in Africa. Springer International Publishing, Switzerland, 113-133.

78
Wiesmeier M, Spörlein P, Geuß Uet 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.Precise estimations of soil organic carbon () stocks are of decided importance for the detection of C sequestration or emission potential induced by land use changes. For Germany, a comprehensive, land use–specific data set has not yet been compiled. We evaluated a unique data set of 1460 soil profiles in southeast Germany in order to calculate representative stocks to a depth of 102m for the main land use types. The results showed that grassland soils stored the highest amount of , with a median value of 11.802kg02m, whereas considerably lower stocks of 9.8 and 9.002kg02mwere found for forest and cropland soils, respectively. However, the differences between extensively used land (grassland, forest) and cropland were much lower compared with results from other studies in central European countries. The depth distribution of showed that despite low concentrations in A horizons of cropland soils, their stocks were not considerably lower compared with other land uses. This was due to a deepening of the topsoil compared with grassland soils. Higher grassland stocks were caused by an accumulation of in the B horizon which was attributable to a high proportion of C‐rich Gleysols within grassland soils. This demonstrates the relevance of pedogenetic inventories instead of solely land use–based approaches. Our study indicated that cultivation‐induced depletion was probably often overestimated since most studies use fixed depth increments. Moreover, the application of modelled parameters in inventories is questioned because a calculation of stocks using different pedotransfer functions revealed considerably biased results. We recommend stocks be determined by horizon for the entire soil profile in order to estimate the impact of land use changes precisely and to evaluate C sequestration potentials more accurately.

DOI

79
Winowiecki L, Vågen T G, Huising J, .

80
Wu C, Wu J, Luo Yet al., 2009. Spatial prediction of soil organic matter content using co-kriging with remotely sensed data.Soil Science Society of America Journal, 73: 1202-1208.Accurately measuring soil organic matter content (SOM) in paddy fields is important because SOM is one of the key soil properties controlling nutrient budgets in agricultural production systems. Estimation of this soil property at an acceptable level of accuracy is important; especially in the case when SOM exhibits strong spatial dependence and its measurement is a time- and labor-consuming pr...

DOI

81
Yang R, Su Y Z, Wang Met al., 2014. Spatial pattern of soil organic carbon in desert grasslands of the diluvial-alluvial plains of northern Qilian Mountains.Journal of Arid Land, 6(2): 136-144.The soil properties in arid ecosystems are important determinants of vegetation distribution patterns. Soil organic carbon (SOC) content, which is closely related to soil types and the holding capacities of soil water and nutrients, exhibits complex variability in arid desert grasslands; thus, it is essentially an impact factor for the distri-bution pattern of desert grasslands. In the present study, an investigation was conducted to estimate the spatial pattern of SOC content in desert grasslands and the association with environmental factors in the diluvial-alluvial plains of northern Qilian Mountains. The results showed that the mean values of SOC ranged from 2.76 to 5.80 g/kg in the soil profiles, and decreased with soil depths. The coefficients of variation (CV) of the SOC were high (ranging from 48.83% to 94.67%), which indicated a strong spatial variability. SOC in the desert grasslands of the study re-gion presented a regular spatial distribution, which increased gradually from the northwest to the southeast. The SOC distribution had a pattern linked to elevation, which may be related to the gradient of climate conditions. Soil type and plant community significantly affected the SOC. The SOC had a significant positive relationship with soil moisture (<em>P</em>&lt;0.05); whereas, it had a more significant negative relationship with the soil bulk density (BD) (<em>P</em>&lt;0.01). However, a number of the variations in the SOC could be explained not by the environmental factors involved in this analysis, but rather other factors (such as grazing activity and landscape). The results provide important references for soil carbon storage estimation in this study region. In addition, the SOC association with environmental variables also provides a basis for a sustainable use of the limited grassland resources in the diluvial-alluvial plains of north-ern Qilian Mountains.

DOI

82
Yang Y, Fang J, Tang Yet al., 2008. Storage, patterns and controls of soil organic carbon in the Tibetan grasslands.Global Change Biology, 14: 1592-1599.The soils of the Qinghai-Tibetan Plateau store a large amount of organic carbon, but the magnitude, spatial patterns and environmental controls of the storage are little investigated. In this study, using data of soil organic carbon (SOC) in 405 profiles collected from 135 sites across the plateau and a satellite-based dataset of enhanced vegetation index (EVI) during 2001-2004, we estimated storage and spatial patterns of SOC in the alpine grasslands. We also explored the relationships between SOC density (soil carbon storage per area) and climatic variables and soil texture. Our results indicated that SOC storage in the top 165m in the alpine grasslands was estimated at 7.465Pg C (165Pg=1065g), with an average density of 6.565kg65m. The density of SOC decreased from the southeastern to the northwestern areas, corresponding to the precipitation gradient. The SOC density increased significantly with soil moisture, clay and silt content, but weakly with mean annual temperature. These variables could together explain about 72% of total variation in SOC density, of which 54% was attributed to soil moisture, suggesting a key role of soil moisture in shaping spatial patterns of SOC density in the alpine grasslands.

DOI

83
Zaehle S, Ciais P, Friend A Det al., 2011. Carbon benefits of anthropogenic reactive nitrogen offset by nitrous oxide emissions.Nature Geoscience, 4: 601-605.Additions of reactive nitrogen to terrestrial ecosystems--primarily through fertilizer application and atmospheric deposition--have more than doubled since 1860 owing to human activities . Nitrogen additions tend to increase the net uptake of carbon by the terrestrial biosphere, but they also stimulate nitrous oxide release from soils. However, given that the magnitude of these effects is uncertain, and that the carbon and nitrogen cycles are tightly coupled, the net climatic impact of anthropogenic nitrogen inputs is unknown . Here we use a process-based model of the terrestrial biosphere to evaluate the overall impact of anthropogenic nitrogen inputs on terrestrial ecosystem carbon and nitrous oxide fluxes between 1700 and 2005. We show that anthropogenic nitrogen inputs account for about a fifth of the carbon sequestered by terrestrial ecosystems between 1996 and 2005, and for most of the increase in global nitrous oxide emissions in recent decades; the latter is largely due to agricultural intensification. We estimate that carbon sequestration due to nitrogen deposition has reduced current carbon dioxide radiative forcing by 96+/-14mWm. However, this effect has been offset by the increase in radiative forcing resulting from nitrous oxide emissions, which amounts to 125+/-20mWm.

DOI

84
Zhang C, Tang Y, Xu Xet al., 2011. Towards spatial geochemical modelling: Use of geographically weighted regression for mapping soil organic carbon contents in Ireland.Applied Geochemistry, 26: 1239-1248.It is challenging to perform spatial geochemical modelling due to the spatial heterogeneity features of geochemical variables. Meanwhile, high quality geochemical maps are needed for better environmental management. Soil organic C (SOC) distribution maps are required for improvements in soil management and for the estimation of C stocks at regional scales. This study investigates the use of a geographically weighted regression (GWR) method for the spatial modelling of SOC in Ireland. A total of 1310 samples of SOC data were extracted from the National Soil Database of Ireland. Environmental factors of rainfall, land cover and soil type were investigated and included as the independent variables to establish the GWR model. The GWR provided comparable and reasonable results with the other chosen methods of ordinary kriging (OK), inverse distance weighted (IDW) and multiple linear regression (MLR). The SOC map produced using the GWR model showed clear spatial patterns influenced by environmental factors and the smoothing effect of spatial interpolation was reduced. This study has demonstrated that GWR provides a promising method for spatial geochemical modelling of SOC and potentially other geochemical parameters. (C) 2011 Elsevier Ltd. All rights reserved.

DOI

85
Zhang S, Huang Y, Shen Cet al., 2012. Spatial prediction of soil organic matter using terrain indices and categorical variables as auxiliary information.Geoderma, 171/172: 35-43.ABSTRACT Soil organic matter (SOM) is one of the most important indicators of the soil quality. Accurate information about the spatial variation of SOM is critical to sustainable soil utilization and management. Although utilizing spatially correlated auxiliary information to improve the prediction accuracy of soil properties has been widely recognized in pedometrics, not all studies have taken account of the influence of categorical variables (e.g., land use types, soil texture and soil genetic types) and did not systematically analyze the relationship between auxiliary variables and soil properties to be predicted. This paper aimed to examine whether inclusion of categorical variables can improve the accuracy of SOM prediction based on systematical analyses of variability. The least-significant difference (LSD) method and Pearson correlation analysis were used to systematically and quantitatively analyze the relationship between SOM and other environment variables (terrain indices, land use types, soil texture and soil genetic types). Spatial distribution of SOM was predicted by multiple linear stepwise regressions, ordinary Kriging and regression Kriging. Results indicated that spatial distribution of SOM was mainly affected by terrain indices, soil texture and soil genetic types. The root mean squared error of predictions based on elevation, which is used frequently as an auxiliary variable, was reduced when categorical variables were added as predictors. Our study suggested that introduction of categorical variables, such as soil genetic types, improved the prediction accuracy for a given prediction method. At the same time, systematic and exploratory analyses of the relationship between variables to be predicted and auxiliary was also important to ensure good predictions.

DOI

86
Zhang Z, Yu C, Shi Xet al., 2010. Application of categorical information in the spatial prediction of soil organic carbon in the red soil area of China.Soil Science and Plant Nutrition, 56: 307-318.Abstract Predicting soil organic carbon (SOC) content distribution accurately from limited soil samples has received a great deal of attention recently in an effort to support soil fertility mapping and to improve our understanding of carbon sequestration variability. Kriging methods combined with auxiliary variables are frequently used at present. However, studies using categorical information, such as soil type and land use, which are closely related to local trends in SOC spatial variation, as auxiliary variables are seldom conducted. In the present investigation, a total of 254 surficial soil samples were collected in the study area, Yujiang county in the hilly red soil region of South China, and a comparison of performance of four kriging approaches was conducted, ordinary kriging (OK), kriging combined with soil-type information (KST), land use (KLU) and combined land use–soil type information (KLUST). Results of the assessment were based on 85 validation samples. The results indicate that the best correlation between the measured and predicted values for validation location was obtained with KLUST (65=650.854), whereas the lowest was obtained using OK (65=650.383). Furthermore, the root mean square error (RMSE) from KLUST (3.4765g65kg) is the lowest, whereas the one obtained using OK (6.4965g65kg) is the highest. The correlation coefficient and RMSE from KST (65=650.784, RMSE65=654.1565g65kg) and KLU (65=650.795, RMSE65=653.9565g65kg) are the second and third most correlated, respectively. Comparing the SOC distribution maps generated by the four prediction approaches, the KLUST rendering best reflects the local change associated with soil types and land uses, whereas the map from the OK is the least representative. The results demonstrate that soil type and land use have an important impact on SOC spatial distribution, and KLUST, which reduces their influence as a local trend, is an efficient and practical prediction approach for the hilly red soil region of South China.

DOI PMID

Options
Outlines

/

〈 〉