# 7 Spatial prediction and assessment of Soil Organic Carbon

Edited by: Hengl T. & Sanderman J.

## 7.1 Introduction

This chapter was prepared as supplementary material for the Sanderman, Hengl, and Fiske (2018) article. It explains how to map Soil Organic Carbon Stocks (OCS) using soil samples (point data). It also demonstrates derivation of values at both the site level (per profile) and by using raster calculus (per pixel). We then illustrate how to estimate total OCS for any area of interest (which can be a field plot, farm and/or administrative region). For an introduction to soil mapping using Machine Learning Algorithms refer to section 6. To access ISRIC’s global compilation of soil profiles please refer to: http://www.isric.org/explore/wosis.

## 7.2 Measurement and derivation of soil organic carbon

Carbon below ground can be organic or non-organic/ mineral (usually carbonates and bicarbonates) i.e. CaCO$$_3$$ in the rocks. Organic carbon stocks below ground (0–2 m) in terrestrial ecosystems consist of two major components:

1. Living organism biomass i.e. mainly:
• Plant roots,
• Microbial biomass (Xu, Thornton, and Post 2013),
2. Plant and animal residues at various stages of decomposition (organic matter).

Xu, Thornton, and Post (2013) estimated that total global microbial biomass is about 17 Pg C, which is only about 2% of the total for organic matter. Therefore, the amounts of C in microbial biomass can reasonably be considered negligible, in comparison to the total C stock. However, if one includes all living organisms, and especially tree roots, then the portion of total C fouind in living organisms could be more significant, especially in areas under dense forests.

Soil Organic Carbon Stock (OCS) is the mass of soil organic carbon per standard area and for a specific depth interval, usually expressed in kg/m$$^2$$ or t/ha. It can be derived using (laboratory and/or field) measurement of soil organic carbon content (ORC; expressed in % or g/kg of <2mm mineral earth), then taking into account bulk density (BLD), thickness of the soil layer, and volume percentage of coarse fragments (CRF) (Nelson and Sommers 1982; Poeplau, Vos, and Axel 2017):

$$${\rm OCS} [{\rm kg/m^2}] = {\rm ORC} [\%] / 100 \cdot {\rm BLD} [{\rm kg/m^3}] \cdot (1- {\rm CRF} [\%]/100) \cdot {\rm HOT} [m] \tag{7.1}$$$

Note that if one has soil organic carbon content measured in g/kg then one should divide by 1000 instead of 100. A correction for gravel content is necessary because only material less than 2 mm is analyzed for ORC concentration. Ignoring the gravel content can result in an overestimation of the organic carbon stock. Note also that OCS always refers to a specific depth interval or horizon thickness (HOT), e.g.:

• kg/m$$^2$$ for depth 0–30 cm (Berhongaray and Alvarez 2013),

Values of OCS in kg/m$$^2$$ can also be expressed in tons/ha units, in which case a simple conversion formula can be applied:

$$$1 \cdot {\rm kg/m^2} = 10 \cdot {\rm tons/ha} \tag{7.2}$$$

Total OCS for an area of interest can be derived by multiplying OCS by total area e.g.:

$$$120 {\rm tons/ha} \cdot 1 {\rm km^2} = 120 \cdot 100 = 12,000 {\rm tons} \tag{7.3}$$$

Another way to express soil organic carbon is through soil organic carbon density (OCD in kg/m$$^3$$), which is in fact equivalent to OCS divided by the horizon thickness:

$$${\rm OCD} [{\rm kg/m^3}] = {\rm ORC} [\%]/100 \cdot {\rm BLD} [{\rm kg/m^3}] \cdot (1- {\rm CRF} [\%]/100) = {\rm OCS} / {\rm HOT} \tag{7.4}$$$

While OCS is a summary measure of SOC, always associated with a specific depth interval, OCD is a relative measure of soil organic carbon distribution and can be associated with any support size i.e. to an arbitrary depth. In principle, OCD (kg/m$$^3$$) is strongly correlated with ORC (g/kg) as indicated in the figure below. However, depending on soil mineralogy and coarse fragment content, OCD can be lower or higher than what the smoothed line indicates (notice the range of values around the smoothed line is relatively wide). It is important to understand, however, that, as long as ORC, BLD and CRF are known, one can convert the values from ORC to OCD and OCS and vice versa, without loosing any information about the soil organic carbon stock.

In summary, there are four main variables used to represent soil organic carbon:

1. Soil Organic Carbon fraction or content (ORC) in g/kg (permille) or dg/kg (percent),
2. Soil Organic Carbon Density (OCD) in kg/m$$^3$$,
3. Soil Organic Carbon Stock (OCS) in kg/m$$^2$$ or in tons/ha and for the given soil depth interval,
4. Total Soil Organic Carbon Stock (TOCS) in million tonnes or Pg i.e. OCS multiplied by surface area,

Global estimates of the total soil organic carbon stock are highly variable (Scharlemann et al. 2014). Current estimates of the present total soil organic carbon stock range between 800–2100 Pg C (for 0–100 cm), with a median estimate of about 1500 Pg C (for 0–100 cm). This means that the average OCS for the 0–100 cm depth interval for the global land mask (148,940,000 km$$^2$$) is about 11 kg/m$$^2$$ or 110 tons/ha, and that average soil organic carbon density (OCD) is about 11 kg/m$$^3$$ (compare to the standard bulk density of fine earth of 1250 kg/m$$^3$$); standard OCS for 0–30 cm depth interval is 7 kg/m$$^2$$ i.e. the average OCD is about 13 kg/m$$^3$$.

The average Organic Carbon Stock for tha 0–100 cm depth interval for the land mask (148,940,000 km$$^2$$) is about 11 kg/m$$^2$$ or 110 tons/ha. The average soil Organic Carbon Density (OCD) is about 11 kg/m$$^3$$ (compared to the standard bulk density of fine earth of 1250 kg/m$$^3$$). Standard Organic Carbon Stock for 0–30 cm depth interval is 7 kg/m$$^2$$ i.e. the average OCD is about 13 kg/m$$^3$$.

The distribution of soil organic carbon in the world is, however, highly patchy with large areas with OCS $$\ll 100$$ tons/ha, and then some pockets of accumulated organic material i.e. organic soil types (histosols) with OCS up to 850tons/ha (for 0–30 cm depth interval). The world’s soil organic matter accumulation areas are usually found in the following biomes / land cover classes: wetlands and peatlands, mangroves, tundras and taigas.

Land use and agriculture, in particular, have led to dramatic decreases in soil carbon stocks in last 200+ years (agricultural and industrial revolutions). Lal (2004) estimated that approximately 54 Pg C have been added to the atmosphere due to agricultural activities with another 26 Pg C being lost from soils due to erosion. Wei et al. (2014) estimated that, on average, conversion from forests to various agricultural land results in a 30–50% decrease in SOCS. Modelling and monitoring of soil organic carbon dynamics is therefore of increasing importance (see e.g. FAO report “Unlocking the Potential of Soil Organic Carbon”).

## 7.3 Derivation of OCS and OCD using soil profile data

As mentioned previously, OCS stock is most commonly derived from measurements of the organic carbon (ORC) content, soil bulk density (BLD) and the volume fraction of gravel (CRF). These are usually sampled either per soil layers or soil horizons (a sequence of horizons makes a soil profile), which can refer to variable soil depth intervals i.e. are non-standard. That means that, before one can determine OCS for standard fixed depth intervals (e.g. 0–30 cm or 0–100 cm), values of ORC, BLD and CRF need to be standardized so they refer to common depth intervals.

Consider, for example, the following two real life examples of soil profile data for a standard agricultural soil and an organic soil. In the first example, profile from Australia, the soil profile data shows:

Table 7.1: Laboratory data for a profile 399 EDGEROI ed079 from Australia (Karssies 2011).
upper_limit lower_limit carbon_content bulk_density CF SOCS
0 10 8.2 1340 6 1.1
10 20 7.5 1367 6 1.0
20 55 6.1 1382 7 3.0
55 90 3.3 1433 8 1.7
90 116 1.6 1465 8 0.6

Note that BLD data were not given for the described horizons (the original soil profile description / laboratory data indicates that no BLD were recorded for this profile). In the absence of measured field BLD we can substitute BLD estimated using SoilGrids250m data. It (unfortunately) commonly happens that soil profile observations lack BLD measurements, and consequently BLD needs to be generated using a Pedo-Transfer function or estimated from soil maps.

To determine OCS for standard depth intervals 0–30, 0–100 and 0–200 cm, we first fit a mass-preserving spline (Malone et al. 2009):

library(GSIF)
library(aqp)
library(sp)
library(plyr)
lon = 149.73; lat = -30.09;
id = "399_EDGEROI_ed079"; TIMESTRR = "1987-01-05"
top = c(0, 10, 20, 55, 90)
bottom = c(10, 20, 55, 90, 116)
ORC = c(8.2, 7.5, 6.1, 3.3, 1.6)
BLD = c(1340, 1367, 1382, 1433, 1465)
CRF = c(6, 6, 7, 8, 8)
#OCS = OCSKGM(ORC, BLD, CRF, HSIZE=bottom-top)
prof1 <- join(data.frame(id, top, bottom, ORC, BLD, CRF),
data.frame(id, lon, lat, TIMESTRR), type='inner')
#> Joining by: id
depths(prof1) <- id ~ top + bottom
#> Warning: converting IDs from factor to character
site(prof1) <- ~ lon + lat + TIMESTRR
coordinates(prof1) <- ~ lon + lat
proj4string(prof1) <- CRS("+proj=longlat +datum=WGS84")
ORC.s <- mpspline(prof1, var.name="ORC", d=t(c(0,30,100,200)), vhigh = 2200)
#> Fitting mass preserving splines per profile...
#>
|
|                                                                 |   0%
|
|=================================================================| 100%
BLD.s <- mpspline(prof1, var.name="BLD", d=t(c(0,30,100,200)), vhigh = 2200)
#> Fitting mass preserving splines per profile...
#>
|
|                                                                 |   0%
|
|=================================================================| 100%
CRF.s <- mpspline(prof1, var.name="CRF", d=t(c(0,30,100,200)), vhigh = 2200)
#> Fitting mass preserving splines per profile...
#>
|
|                                                                 |   0%
|
|=================================================================| 100%

now we can derive OCS for top-soil by using:

OCSKGM(ORC.s$var.std$0-30 cm,
BLD.s$var.std$0-30 cm,
CRF.s$var.std$0-30 cm, HSIZE=30)
#> [1] 2.88
#> attr(,"measurementError")
#> [1] 3.84
#> attr(,"units")
#> [1] "kilograms per square-meter"

and for sub-soil using:

OCSKGM(ORC.s$var.std$30-100 cm,
BLD.s$var.std$30-100 cm,
CRF.s$var.std$30-100 cm, HSIZE=70)
#> [1] 3.62
#> attr(,"measurementError")
#> [1] 9.18
#> attr(,"units")
#> [1] "kilograms per square-meter"

Note that the OCSKGM function requires soil organic carbon content in g/kg. If one has contents measured in % then first multiply the values by 10. Bulk density data should be provided in kg/m3, gravel content in %, and layer depth in cm. Running the OCSKGM function for the Edgeroi profile gives the following estimates of OCS for standard depth intervals (Fig. 7.1):

• 0–30 cm: 2.9 kg / m-square

• 0–100 cm: 6.5 kg / m-square

• 0–200 cm: 8.5 kg / m-square (85 tonnes / ha)

Values of OCS between 5–35 kg/m$$^2$$ for 0–100 cm are commonly reported for a variety of mineral soils with e.g. 1–3% of soil organic carbon.

Organic Carbon Stock for standard depths can be determined from legacy soil profile data either by fitting a spline function to organic carbon, bulk density values, or by aggregating data using simple conversion formulas. A standard mineral soil with 1–3% soil organic carbon for the 0–100 cm depth interval should have about 5–35 kg/m$$^2$$ or 50–350 tonnes/ha. An organic soil with >30% soil organic carbon may have as much as 60–90 kg/m$$^2$$ for the 0–100 cm depth interval.

Note that the measurement error is computed from default uncertainty values (expressed in standard deviations) for organic carbon (10 g/kg), bulk density (100 kg/m3) and coarse fraction content (5%). When these are not provided by the user, the outcome should be interpreted with care.

In the second example we look at a profile from Canada (a histosol with >40% of organic carbon):

Table 7.2: Laboratory data for an organic soil profile from Canada (Shaw, Bhatti, and Sabourin 2005).
upper_limit lower_limit carbon_content bulk_density CF SOCS
0 31 472 185 5 25.7
31 61 492 172 6 23.9
61 91 487 175 6 24.1
91 122 502 166 6 24.3
122 130 59 830 6 3.7

Here also BLD values were not provided and so had to be estimated. To estimate BLD, we use a simple Pedo-Transfer rule e.g. from Köchy, Hiederer, and Freibauer (2015):

$$$BLD.f = (-0.31 \cdot log(ORC/10) + 1.38) \cdot 1000 \tag{7.5}$$$

We divide the organic carbon content here by 10 to convert the organic carbon content from g/kg to % as required by the PTF. Note that one might want to use different PTFs for different soil layers. For mineral soils the bulk density of subsoil layers is often somewhat higher than for topsoil layers. For organic soils this typically is the inverse. For instance, Köchy, Hiederer, and Freibauer (2015) propose the following PTF for the subsoil (for layers with SOC > 3%):

$$$BLD = -0.32 * log({\rm ORC}[\%]) + 1.38 \tag{7.5}$$$

which gives slightly lower bulk density values. Another useful source for PTFs for organic soils is work by Hossain, Chen, and Zhang (2015). For illustrative purposes, we have here used only one PTF for all soil layers.

We can again fit mass-preserving splines and determine OCS for standard depth intervals by using the functions applied to profile 1. This ultimately gives the following estimates (Fig. 7.2):

• 0–30 cm: 24.8 kg / m-square

• 0–100 cm: 75.3 kg / m-square

• 0–200 cm: 114.5 kg / m-square (1145 tonnes / ha)

Note that only 3–4% of the total soil profiles in the world have organic carbon content above 8% (soils with ORC >12% are classified as organic soils or histosols in USDA and/or WRB classifications and are even less frequent), hence soil-depth functions of organic carbon content and derivation of OCS for organic soils are specific for patches of organic soils. On the other hand, organic soils carry much more total OCS. Precise processing and mapping of organic soils is, therefore, often crucial for accurate estimation of total OCS for large areas. Therefore, it is fairly important to use a good PTF to fill in missing values for BLD for organic soils. As a rule of thumb, organic soil will rarely exhibit a density greater than some minimum value e.g. 120 kg/m$$^3$$ because even though SOC content can be >50%, bulk density of such soil gets proportionally lower and bulk density is physically bound with how material is organized in soil (unless soils is artificially compacted). Also, using correct estimates for coarse fragments is important as otherwise, if CRF is ignored, total stock will be over-estimated (Poeplau, Vos, and Axel 2017).

A somewhat more straightforward way to estimate OCS for a list of soil profiles vs spline fitting is:

1. Fill in bulk densities using some PTF or global data,
2. Use information about the depth to bedrock to correct for shallow soils,
3. Use information on CRF to correct stocks for stony / skeletoidal component,
4. Aggregate non-standard horizon depth values using some simple rules (Fig. 7.3).

## 7.4 Estimation of Bulk Density using a globally-calibrated PTF

Where values for bulk density are missing, and no local PTF exists, WoSIS points (global compilation of soil profiles) can be used to fit a PTF that can fill-in gaps in bulk density measurements globally. A regression matrix extracted on 15th of May 2017 (and which contains harmonized values for BD, organic carbon content, pH, sand and clay content, depth of horizon and USDA soil type at some 20,000 soil profiles world-wide), can be fitted using a random forest model (see also Ramcharan et al. (2018b)):

dfs_tbl = readRDS("extdata/wosis_tbl.rds")
library(ranger)
fm.BLD = as.formula(
paste("BLD ~ ORCDRC + CLYPPT + SNDPPT + PHIHOX + DEPTH.f +",
paste(names(ind.tax), collapse="+")))
m.BLD_PTF <- ranger(fm.BLD, dfs_tbl, num.trees = 85, importance='impurity')
m.BLD_PTF
#> Ranger result
#>
#> Call:
#>  ranger(fm.BLD, dfs_tbl, num.trees = 85, importance = "impurity")
#>
#> Type:                             Regression
#> Number of trees:                  85
#> Sample size:                      98650
#> Number of independent variables:  70
#> Mtry:                             8
#> Target node size:                 5
#> Variable importance mode:         impurity
#> OOB prediction error (MSE):       32379
#> R squared (OOB):                  0.549

This shows somewhat lower accuracy i.e. an RMSE of ±180 kg/m$$^3$$ (R squared (OOB) = 0.54), but is still probably preferable to completely excluding all observations without bulk density from SOC assessment. A disadvantage of this model is that, in order to predict BD for new locations, we need to also have measurements of texture fractions, pH and organic carbon of course. For example, an Udalf with 1.1% organic carbon, 22% clay, pH of 6.5, sand content of 35% and at depth of 5 cm would result in an estimate for bulk density of:

ind.tax.new = ind.tax[which(ind.tax$TAXOUSDA84==1)[1],] predict(m.BLD_PTF, cbind(data.frame(ORCDRC=11, CLYPPT=22, PHIHOX=6.5, SNDPPT=35, DEPTH.f=5), ind.tax.new))$predictions
#> [1] 1526

Note also that the PTF above needs USDA suborder values per point location following the SoilGrids legend for USDA suborders, and formatted as in the ind.tax object. Unfortunately, the model above probably over-estimates bulk density for organic soils as these are usually under-represented i.e. often not available (consider for example a Saprist with 32% organic carbon):

ind.tax.new = ind.tax[which(ind.tax$TAXOUSDA13==1)[1],] predict(m.BLD_PTF, cbind(data.frame(ORCDRC=320, CLYPPT=8, PHIHOX=5.5, SNDPPT=45, DEPTH.f=10), ind.tax.new))$predictions
#> [1] 773

An alternative to estimating BLD is to just use ORC values, e.g. (see plot below):

m.BLD_ls = loess(BLD ~ ORCDRC, dfs_tbl, span=1/18)
predict(m.BLD_ls, data.frame(ORCDRC=320))
#>   1
#> 664

This gives about 30% lower value than the random forest-based PTF from above. Over-estimating BLD will always result in higher OCS, hence clearly accurate information on BLD can be crucial for any OCS monitoring project. This means that the PTF fitted using random forest above is likely over-estimating BLD values for organic soils, mainly because there are not enough training points in organic soils that have measurements data for all of ORC, BLD, soil pH and texture fractions (if ANY of the calibration measurements are missing, the whole horizons are taken out of calibration and hence different ranges of BLD could be completely misrepresented).

Soil Bulk density (BLD) is an important soil property that is required to estimate stocks of nutrients especially soil organic carbon. Measurements of BLD are often not available and need to be estimated using some PTF or similar. Most PTF’s for BLD are based on correlating BLD with soil organic carbon, clay and sand content, pH, soil type and climate zone.

To fill-in missing values for BLD, a combination of two global Pedo-Transfer functions can be used for example: (1) PTF fitted using random forest model that locally predicts BLD as a function of organic carbon content, clay and sand content, pH and coarse fragments, and (2) a simpler model that predicts BLD based on only ORC. The average RMSE of these PTFs for BLD is about ±150 kg/m$$^3$$.

For mineral soils the relationship between soil organic carbon and soil depth follows a log-log relationship which can be also approximated with the following (global) model (R-square: 0.36; see Fig. 7.5):

$$$ORC (depth) = exp[ 4.1517 −0.60934 \cdot log(depth) ] \tag{7.6}$$$

This also illustrates that any organic carbon spatial prediction model can significantly profit from including depth in the statistical modelling.

In summary, PTFs can be used to efficiently estimate missing BLD values (BLD is usually highly correlated with organic carbon content and depth, texture fractions, soil classification and soil pH can also help improve accuracy of the PTFs). However, for organic soils there is, in general, less calibration data and therefore errors are potentially higher. Mistakes in estimating BLD can result in systematic and significant over/under-estimations of the actual carbon stock. On the other hand, removing all soil horizons from OCS assessment that lack BLD measurements leads also to reduced accuracy as fewer points are then available for training of the spatial prediction models. Especially for organic soils (>12% organic carbon), there is no easy solution for estimating missing values for BLD. consequently collecting additional (local) calibration points might become unavoidable. Lobsey and Viscarra Rossel (2016) recently proposed a method that combines gamma-ray attenuation and visible–near infrared (vis–NIR) spectroscopy to measure bulk density ex situ using samples that are freshly colllected under wet field conditions. Hopefully BLD measurements (or their complete absence) will become less of a problem in the future.

## 7.5 Generating maps of OCS

Most projects focused on monitoring OCS require an estimate of OCS to be provided for the entire area of interest, so that users can visually explore spatial patterns of OCS. In this tutorial we demonstrate how to generate maps of OCS using point samples and RS based covariates. The output of this process is usually a gridded map (SpatialPixelsDataFrame) covering the entire area of interest (plot, farm, administrative unit or similar). Once OCS is mapped, we can multiply OCS densities by the area of each pixel and summing up all numbers we can compute the total OCS in total tonnes using the formula above. Predicted OCS values can also be aggregated per land cover class or similar. If a series of OCS maps is produced for the same area of interest (time-series of OCS), these can then be used to derive OCS change through time per pixel.

In principle, there are three main approaches to estimating total OCS for an area of interest (Fig. 7.6):

• By directly predicting OCS, here called the the 2D approach to OCS mapping (this often requires vertical aggregation / modeling of soil variable depth curves as indicated above),

• By predicting ORC, BLD and CRF, and then deriving OCS per layer, here called the 3D approach to OCS mapping with ORC, BLD and CRF mapped separately,

• By deriving OCD (organic carbon density) and then directly predicting OCD and converting it to OCS, here called the 3D approach to OCS mapping via direct modeling of OCD,

Soil Organic Carbon stock can be mapped by using at least differing approaches: (1) the 2D approach where estimation of OCS is done at the site level, (2) the 3D approach where soil organic carbon content, bulk density and coarse fragments are mapped seperately, then used to derive OCS for standard depths at each grid cell, and (3) the 3D approach based on mapping Organic Carbon Density, then converting to stocks.

Although 2D prediction of OCS from point data seems to be more straightforward, many soil profiles contain measurements at non-standard depth intervals (varying support sizes also) and therefore 2D modeling of OCS can often be a cumbersome. In most situations where legacy soil profile data are used, 3D modeling of OCD is probably the most elegant solution to mapping OCS because:

• No vertical aggregation of values via spline fitting or similar is needed to standardize values per standard depths,

• No additional uncertainty is introduced (in the case of the 2D approach splines likely introduce some extra uncertainty in the model),

• Predictions of OCD/OCS can be generated for any depth interval using the same model (i.e. predictions are based on a single 3D model),

A disadvantage of doing 3D modeling of OCD is, however, that correlation with covariate layers could be less clear than if separate models are built for ORC, BLD and CRF. Because OCD is a composite variable, it can often be difficult to distinguish whether the values are lower or higher due to differences in ORC, BLD or CRF. We leave it to users to compare various approaches to OCS mapping and then select the method that achieves best accuracy and/or is most suitable for use for their applications.

## 7.6 Predicting OCS from point data (the 2D approach)

The geospt package contains 125 samples of OCS from Colombia already at standard depth intervals, hence this data set is suitable and ready for 2D mapping of OCS. The data set consists of tabular values for points and a raster map containing the borders of the study area:

load("extdata/COSha10.rda")
str(COSha30)
#> 'data.frame':    118 obs. of  10 variables:
#>  $ID : Factor w/ 118 levels "S1","S10","S100",..: 1 44 61 89 100 110 2 9 15 21 ... #>$ x         : int  669030 669330 670292 669709 671321 670881 670548 671340 671082 670862 ...
#>  $y : int 448722 448734 448697 448952 448700 448699 448700 448969 448966 448968 ... #>$ DA30      : num  1.65 1.6 1.5 1.32 1.41 1.39 1.51 1.39 1.55 1.63 ...
#>  $CO30 : num 0.99 1.33 1.33 1.09 1.04 1.19 1.21 1.36 1.09 1.19 ... #>$ COB1r     : Factor w/ 6 levels "Az","Ci","Cpf",..: 5 5 2 5 2 5 2 2 2 5 ...
#>  $S_UDS : Factor w/ 19 levels "BJa1","BQa1",..: 12 5 12 5 11 12 12 12 12 12 ... #>$ COSha30   : num  49.2 64 59.8 43.1 44.2 ...
#>  $Cor4DAidep: num 43.3 56.3 54 37.9 39.9 ... #>$ CorT      : num  1.37 1.39 1.38 1.36 1.36 ...

where COSha10 = 0–10 cm, COSha30 = 0–30 cm in tons / ha are values for OCS aggregated to standard soil depth intervals, so there is no need to do any spline fitting and/or vertical aggregation. We can also load the raster map for the area by using (Fig. 7.7):

load("extdata/COSha30map.rda")
proj4string(COSha30map) = "+proj=utm +zone=18 +ellps=WGS84 +datum=WGS84 +units=m +no_defs"
str(COSha30map@data)
#> 'data.frame':    10000 obs. of  2 variables:
#>  $var1.pred: num 39.9 39.8 39.9 40.3 40.7 ... #>$ var1.var : num  1.91e-05 6.39e-05 1.05e-04 1.39e-04 1.66e-04 ...

which shows predictions and kriging variances for COSha30.

We can import a number of RS-based covariates into R by (these were derived from the global 30 m layers listed previously):

covs30m = readRDS("extdata/covs30m.rds")
proj4string(covs30m) = proj4string(COSha30map)
names(covs30m)
#>  [1] "SRTMGL1_SRTMGL1.2_cprof"
#>  [2] "SRTMGL1_SRTMGL1.2_devmean"
#>  [3] "SRTMGL1_SRTMGL1.2_openn"
#>  [4] "SRTMGL1_SRTMGL1.2_openp"
#>  [5] "SRTMGL1_SRTMGL1.2_slope"
#>  [6] "SRTMGL1_SRTMGL1.2_twi"
#>  [7] "SRTMGL1_SRTMGL1.2_vbf"
#>  [8] "SRTMGL1_SRTMGL1.2_vdepth"
#>  [9] "SRTMGL1_SRTMGL1.2"
#> [10] "COSha30map_var1pred_"
#> [11] "GlobalForestChange2000.2014_first_NIRL00"
#> [12] "GlobalForestChange2000.2014_first_REDL00"
#> [13] "GlobalForestChange2000.2014_first_SW1L00"
#> [14] "GlobalForestChange2000.2014_first_SW2L00"
#> [15] "GlobalForestChange2000.2014_treecover2000"
#> [16] "GlobalSurfaceWater_extent"
#> [17] "GlobalSurfaceWater_occurrence"
#> [18] "Landsat_bare2010"

This contains a number of covariates from SRTM DEM derivatives, to Global Surface Water occurrence values and similar (see section 4.1.1 for more details). All these could potentially prove useful for mapping OCS. We can also derive buffer distances from observations points and use these (as measures of spatial context) to improve predictions (T. Hengl, Nussbaum, et al. 2018):

proj4string(COSha30map) = "+proj=utm +zone=18 +ellps=WGS84 +datum=WGS84 +units=m +no_defs"
coordinates(COSha30) = ~ x+y
proj4string(COSha30) = proj4string(COSha30map)
covs30mdist = GSIF::buffer.dist(COSha30["COSha30"], covs30m[1], as.factor(1:nrow(COSha30)))

We can convert the original covariates to Principal Components, also to fill in all missing pixels:

covs30m@data = cbind(covs30m@data, covs30mdist@data)
sel.rm = c("GlobalSurfaceWater_occurrence", "GlobalSurfaceWater_extent", "Landsat_bare2010", "COSha30map_var1pred_")
rr = which(names(covs30m@data) %in% sel.rm)
fm.spc = as.formula(paste(" ~ ", paste(names(covs30m)[-rr], collapse = "+")))
proj4string(covs30m) = proj4string(COSha30)
covs30m.spc = GSIF::spc(covs30m, fm.spc)
#> Converting covariates to principal components...
ov.COSha30 = cbind(as.data.frame(COSha30), over(COSha30, covs30m.spc@predicted))

By using the above listed covariates, we can fit a spatial prediction 2D model using an available model, such as ranger (Wright and Ziegler 2017), xgboost and/or gamboost:

library(caret)
library(ranger)
fm.COSha30 = as.formula(paste("COSha30 ~ ", paste(names(covs30m.spc@predicted), collapse = "+")))
fm.COSha30
#> COSha30 ~ PC1 + PC2 + PC3 + PC4 + PC5 + PC6 + PC7 + PC8 + PC9 +
#>     PC10 + PC11 + PC12 + PC13 + PC14 + PC15 + PC16 + PC17 + PC18 +
#>     PC19 + PC20 + PC21 + PC22 + PC23 + PC24 + PC25 + PC26 + PC27 +
#>     PC28 + PC29 + PC30 + PC31 + PC32 + PC33 + PC34 + PC35 + PC36 +
#>     PC37 + PC38 + PC39 + PC40 + PC41 + PC42 + PC43 + PC44 + PC45 +
#>     PC46 + PC47 + PC48 + PC49 + PC50 + PC51 + PC52 + PC53 + PC54 +
#>     PC55 + PC56 + PC57 + PC58 + PC59 + PC60 + PC61 + PC62 + PC63 +
#>     PC64 + PC65 + PC66 + PC67 + PC68 + PC69 + PC70 + PC71 + PC72 +
#>     PC73 + PC74 + PC75 + PC76 + PC77 + PC78 + PC79 + PC80 + PC81 +
#>     PC82 + PC83 + PC84 + PC85 + PC86 + PC87 + PC88 + PC89 + PC90 +
#>     PC91 + PC92 + PC93 + PC94 + PC95 + PC96 + PC97 + PC98 + PC99 +
#>     PC100 + PC101 + PC102 + PC103 + PC104 + PC105 + PC106 + PC107 +
#>     PC108 + PC109 + PC110 + PC111 + PC112 + PC113 + PC114 + PC115 +
#>     PC116 + PC117 + PC118 + PC119 + PC120 + PC121 + PC122 + PC123 +
#>     PC124 + PC125 + PC126 + PC127 + PC128 + PC129 + PC130 + PC131 +
#>     PC132
rf.tuneGrid <- expand.grid(.mtry = seq(2, 60, by=5),
.splitrule = "maxstat",
.min.node.size = c(10, 20))
gb.tuneGrid <- expand.grid(eta = c(0.3,0.4),
nrounds = c(50,100),
max_depth = 2:3, gamma = 0,
colsample_bytree = 0.8,
min_child_weight = 1, subsample=1)
fitControl <- trainControl(method="repeatedcv", number=4, repeats=1)
mFit1 <- train(fm.COSha30, data=ov.COSha30, method="ranger",
trControl=fitControl, importance='impurity',
tuneGrid=rf.tuneGrid)
mFit1
#> Random Forest
#>
#> 118 samples
#> 132 predictors
#>
#> No pre-processing
#> Resampling: Cross-Validated (4 fold, repeated 1 times)
#> Summary of sample sizes: 88, 90, 89, 87
#> Resampling results across tuning parameters:
#>
#>   mtry  min.node.size  RMSE  Rsquared  MAE
#>    2    10             11.2  0.0492    8.68
#>    2    20             11.1  0.1059    8.60
#>    7    10             11.1  0.0640    8.54
#>    7    20             11.1  0.0794    8.53
#>   12    10             11.1  0.0905    8.47
#>   12    20             11.1  0.1044    8.49
#>   17    10             11.1  0.0977    8.52
#>   17    20             11.1  0.0890    8.50
#>   22    10             11.0  0.1066    8.49
#>   22    20             11.0  0.1184    8.41
#>   27    10             11.0  0.0653    8.45
#>   27    20             11.0  0.0849    8.42
#>   32    10             11.1  0.0605    8.44
#>   32    20             11.0  0.1119    8.37
#>   37    10             11.0  0.0865    8.46
#>   37    20             11.0  0.1104    8.41
#>   42    10             10.9  0.1165    8.37
#>   42    20             11.0  0.0930    8.47
#>   47    10             11.0  0.1065    8.41
#>   47    20             10.9  0.1204    8.34
#>   52    10             11.0  0.0834    8.44
#>   52    20             11.0  0.1159    8.39
#>   57    10             10.9  0.1303    8.35
#>   57    20             11.0  0.1192    8.39
#>
#> Tuning parameter 'splitrule' was held constant at a value of maxstat
#> RMSE was used to select the optimal model using the smallest value.
#> The final values used for the model were mtry = 47, splitrule =
#>  maxstat and min.node.size = 20.
mFit2 <- train(fm.COSha30, data=ov.COSha30, method="xgbTree",
trControl=fitControl, tuneGrid=gb.tuneGrid)
mFit2
#> eXtreme Gradient Boosting
#>
#> 118 samples
#> 132 predictors
#>
#> No pre-processing
#> Resampling: Cross-Validated (4 fold, repeated 1 times)
#> Summary of sample sizes: 88, 88, 89, 89
#> Resampling results across tuning parameters:
#>
#>   eta  max_depth  nrounds  RMSE  Rsquared  MAE
#>   0.3  2           50      12.3  0.0460     9.77
#>   0.3  2          100      12.2  0.0459     9.75
#>   0.3  3           50      12.5  0.0570     9.79
#>   0.3  3          100      12.5  0.0571     9.79
#>   0.4  2           50      13.1  0.0210    10.28
#>   0.4  2          100      13.1  0.0216    10.28
#>   0.4  3           50      12.2  0.0232     9.72
#>   0.4  3          100      12.2  0.0232     9.72
#>
#> Tuning parameter 'gamma' was held constant at a value of 0
#>  0.8
#> Tuning parameter 'min_child_weight' was held constant at a value of
#>  1
#> Tuning parameter 'subsample' was held constant at a value of 1
#> RMSE was used to select the optimal model using the smallest value.
#> The final values used for the model were nrounds = 50, max_depth = 3,
#>  eta = 0.4, gamma = 0, colsample_bytree = 0.8, min_child_weight = 1
#>  and subsample = 1.

This example illustrates that no significant spatial prediction models (with an R-square exceeding 10%) can be fitted using these data. It is very common in soil mapping projects for models to explain only low amounts of the total variation, resulting in large average errors of prediction and/or wide prediction intervals. This can occur because of high measurement errors, and/or because there are missing covariates, but it coud also happen because the natural complexity of soils in the area is simply high.

Note that the absolute values of our predictions of OCS are somewhat different than those produced by the geospt package authors, although the main patterns are comparable.

We can compare the difference between mean predicted OCS and measured OCS:

mean(COSha30.pr$COSha30map_RF, na.rm=TRUE); mean(COSha30$COSha30, na.rm=TRUE)
#> [1] 48.6
#> [1] 50.6
## 48 tonnes/ha vs 51 tonnes / ha

and derive the total SOC in tonnes:

sum(COSha30.pr$COSha30map_RF*30^2/1e4, na.rm=TRUE) #> [1] 102230 ## 7.7 Deriving OCS from soil profile data (the 3D approach) In the following example, we will demonstrate, using a known data set, (Edgeroi, from Australia) which has been well documented in the literature (Malone et al. 2009), how to derive OCS in t/ha using soil profile data and a 3D approach to spatial prediction based on mapping the Organic Carbon Density (OCD) in kg/m-cubic. The Edgeroi data set is a typical example of a soil profile data set that is relatively comprehensive, but still missing BLD measurements. The Edgeroi data set can be loaded from the GSIF package: library(GSIF) data(edgeroi) edgeroi.sp = edgeroi$sites
coordinates(edgeroi.sp) <- ~ LONGDA94 + LATGDA94
proj4string(edgeroi.sp) <- CRS("+proj=longlat +ellps=GRS80 +towgs84=0,0,0,0,0,0,0 +no_defs")
edgeroi.sp <- spTransform(edgeroi.sp, CRS("+init=epsg:28355"))

This data set comes with a list of covariate layers which can be used to model the distribution of soil organic carbon:

load("extdata/edgeroi.grids.rda")
gridded(edgeroi.grids) <- ~x+y
proj4string(edgeroi.grids) <- CRS("+init=epsg:28355")
names(edgeroi.grids)
#> [1] "DEMSRT5" "TWISRT5" "PMTGEO5" "EV1MOD5" "EV2MOD5" "EV3MOD5"

Because some of the covariate layers are factors e.g. PMTGEO5 (parent material map) and because random forest requires numeric covariates, we can convert factors to numeric PCs by using:

edgeroi.spc = spc(edgeroi.grids, ~DEMSRT5+TWISRT5+PMTGEO5+EV1MOD5+EV2MOD5+EV3MOD5)
#> Converting PMTGEO5 to indicators...
#> Converting covariates to principal components...

Note that Edgeroi completely lacks any BLD values, therefore before we can compute OCD values, we need to estimate BLD values for each corresponding horizon. Here the easiest option is probably to use BLD values sourced from SoilGrids250m predictions (and which you can dowload from the SoilGrids FTP). Matching between the irregularly distributed soil horizons and SoilGrids BLD at standard depths can be implemented in three steps. First, we overlay the points and SoilGrids250m GeoTIFFs to get the BLD values at standard depths:

sg <- paste0("/mnt/nas/ftp.soilgrids.org/data/recent/BLDFIE_M_sl",1:7,"_250m_ll.tif")
ov.edgeroi.BLD <- raster::extract(stack(sg),
spTransform(edgeroi.sp, CRS("+proj=longlat +datum=WGS84")))
str(ov.edgeroi.BLD)
#>  num [1:359, 1:7] 1249 1091 1255 1262 1241 ...
#>  - attr(*, "dimnames")=List of 2
#>   ..$: NULL #> ..$ : chr [1:7] "BLDFIE_M_sl1_250m_ll" "BLDFIE_M_sl2_250m_ll" "BLDFIE_M_sl3_250m_ll" "BLDFIE_M_sl4_250m_ll" ...

Second, we derive averaged estimates of BLD for standard depth intervals:

ov.edgeroi.BLDm = data.frame(BLD.f=as.vector(sapply(2:ncol(ov.edgeroi.BLD),                                                function(i){rowMeans(ov.edgeroi.BLD[,c(i-1,i)])})),
DEPTH.c=as.vector(sapply(1:6, function(i){rep(paste0("sd",i),
nrow(edgeroi$sites))})), SOURCEID=rep(edgeroi$sites$SOURCEID, 6)) str(ov.edgeroi.BLDm) #> 'data.frame': 2154 obs. of 3 variables: #>$ BLD.f   : num  1270 1175 1296 1286 1278 ...
#>  $DEPTH.c : Factor w/ 6 levels "sd1","sd2","sd3",..: 1 1 1 1 1 1 1 1 1 1 ... #>$ SOURCEID: Factor w/ 359 levels "199_CAN_CP111_1",..: 1 2 3 4 5 6 7 8 9 10 ...

Third, we match BLD values by matching horizon depths (center of horizon) with the standard depth intervals sd1 to sd6:

edgeroi$horizons$DEPTH = edgeroi$horizons$UHDICM + (edgeroi$horizons$LHDICM - edgeroi$horizons$UHDICM)/2
edgeroi$horizons$DEPTH.c = cut(edgeroi$horizons$DEPTH, include.lowest=TRUE, breaks=c(0,5,15,30,60,100,1000), labels=paste0("sd",1:6))
summary(edgeroi$horizons$DEPTH.c)
#> sd1 sd2 sd3 sd4 sd5 sd6
#> 100 314 356 408 391 769
edgeroi$horizons$BLD.f = plyr::join(edgeroi$horizons[,c("SOURCEID","DEPTH.c")], ov.edgeroi.BLDm)$BLD.f
#> Joining by: SOURCEID, DEPTH.c

Now that we have a rough estimate of the bulk density for all horizons, we can derive OCD in kg/m-cubic by using:

edgeroi$horizons$OCD = edgeroi$horizons$ORCDRC/1000 * edgeroi$horizons$BLD.f
summary(edgeroi$horizons$OCD)
#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.    NA's
#>     0.1     2.5     7.3     9.5    13.2   110.0     297

This shows that OCD values range from 0–110 kg/m-cubic, with an average of 9.5 kg/m-cubic (this corresponds to an average organic carbon content of about 0.8%).

For further 3D spatial prediction of OCD we use the ranger package, which fits a random forest model to this 3D data. We start by overlaying points and rasters so that we can create a regression matrix:

ov2 <- over(edgeroi.sp, edgeroi.spc@predicted)
ov2$SOURCEID = edgeroi.sp$SOURCEID
h2 = hor2xyd(edgeroi$horizons) m2 <- plyr::join_all(dfs = list(edgeroi$sites, h2, ov2))
#> Joining by: SOURCEID
#> Joining by: SOURCEID

The spatial prediction model can be fitted using:

fm.OCD = as.formula(paste0("OCD ~ DEPTH + ", paste(names(edgeroi.spc@predicted), collapse = "+")))
fm.OCD
#> OCD ~ DEPTH + PC1 + PC2 + PC3 + PC4 + PC5 + PC6 + PC7 + PC8 +
#>     PC9 + PC10 + PC11 + PC12
m.OCD <- ranger(fm.OCD, m2[complete.cases(m2[,all.vars(fm.OCD)]),],
quantreg = TRUE, importance = "impurity")
m.OCD
#> Ranger result
#>
#> Call:
#>  ranger(fm.OCD, m2[complete.cases(m2[, all.vars(fm.OCD)]), ],      quantreg = TRUE, importance = "impurity")
#>
#> Type:                             Regression
#> Number of trees:                  500
#> Sample size:                      4858
#> Number of independent variables:  13
#> Mtry:                             3
#> Target node size:                 5
#> Variable importance mode:         impurity
#> OOB prediction error (MSE):       17.3
#> R squared (OOB):                  0.703

Which shows that the average error with Out-of-bag training points is ±4.2 kg/m-cubic. Note that setting quantreg = TRUE allows us to derive also a map of the prediction errors (Fig. 7.10), following the method of Meinshausen (2006).

To derive OCS in tons/ha we can compute OCD at two depths (0 and 30 cm) and then take the mean value to produce a more representative value:

for(i in c(0,30)){
edgeroi.spc@predicted$DEPTH = i OCD.rf <- predict(m.OCD, edgeroi.spc@predicted@data) nm1 = paste0("OCD.", i, "cm") edgeroi.grids@data[,nm1] = OCD.rf$predictions
OCD.qrf <- predict(m.OCD, edgeroi.spc@predicted@data,
type="quantiles", quantiles=c((1-.682)/2, 1-(1-.682)/2))
nm2 = paste0("OCD.", i, "cm_se")
edgeroi.grids@data[,nm2] = (OCD.qrf$predictions[,2] - OCD.qrf$predictions[,1])/2
}

so that the final Organic carbon stocks in t/ha is (see Eq. ??):

#>    Min. 1st Qu.  Median    Mean 3rd Qu.    Max.
#>    20.7    38.1    46.4    47.3    55.1   101.6

Note that deriving the error map in the ranger package can be computationally intensive, especially if the number of covariates is high. It is therefore not yet recommended for large rasters.

Next, we can derive the total soil organic carbon stock per land use class (2007). For this we can use the aggregation function from the plyr package:

library(rgdal)
edgeroi.grids$LandUse = readGDAL("extdata/edgeroi_LandUse.sdat")$band1
#> extdata/edgeroi_LandUse.sdat has GDAL driver SAGA
#> and has 128 rows and 190 columns
edgeroi.grids$LandUseClass = paste(join(data.frame(LandUse=edgeroi.grids$LandUse), lu.leg, match="first")$LU_NSWDeta) #> Joining by: LandUse OCS_agg.lu <- plyr::ddply(edgeroi.grids@data, .(LandUseClass), summarize, Total_OCS_kt=round(sum(OCS.30cm*250^2/1e4, na.rm=TRUE)/1e3), Area_km2=round(sum(!is.na(OCS.30cm))*250^2/1e6)) OCS_agg.lu$LandUseClass.f = strtrim(OCS_agg.lu$LandUseClass, 34) OCS_agg.lu$OCH_t_ha_M = round(OCS_agg.lu$Total_OCS_kt*1000/(OCS_agg.lu$Area_km2*100))
OCS_agg.lu[OCS_agg.lu$Area_km2>5,c("LandUseClass.f","Total_OCS_kt","Area_km2","OCH_t_ha_M")] #> LandUseClass.f Total_OCS_kt Area_km2 OCH_t_ha_M #> 2 Constructed grass waterway for wat 54 11 49 #> 3 Cotton 41 8 51 #> 4 Cotton - irrigated 795 203 39 #> 5 Cropping - continuous or rotation 1737 402 43 #> 6 Cropping - continuous or rotation 227 59 38 #> 10 Farm dam 52 10 52 #> 11 Farm Infrastructure - house, machi 87 18 48 #> 12 Grazing - Residual strips (block o 47 10 47 #> 13 Grazing of native vegetation. Graz 662 129 51 #> 14 Grazing of native vegetation. Graz 66 13 51 #> 16 Irrigation dam 63 16 39 #> 21 Native forest 222 37 60 #> 26 Research facility 38 9 42 #> 27 River, creek or other incised drai 67 11 61 #> 28 Road or road reserve 114 23 50 #> 29 State forest 418 83 50 #> 32 Volunteer, naturalised, native or 1359 238 57 #> 33 Volunteer, naturalised, native or 62 16 39 #> 34 Volunteer, naturalised, native or 74 14 53 #> 35 Volunteer, naturalised, native or 460 99 46 #> 37 Wide road reserve or TSR, with som 453 90 50 Which shows that, for the Cropping - continuous or rotation, which is the dominant land use class in the area, the average OCS is 43 tons/ha for the 0–30 cm depth. In this case, the total soil organic carbon stock for the whole area (for all land use classes) is ca 7154 thousand tons of C. There do not appear to be large differences in OCS between the natural vegetation and croplands. ## 7.8 Deriving OCS using spatiotemporal models Assuming that measurements of ORC have also been referenced temporally (at least to the year of sampling), point data can be used to build spatiotemporal models of soil organic carbon. Consider, for example, the soil profile data available for the conterminous USA: OCD_stN <- readRDS("extdata/usa48.OCD_spacetime_matrix.rds") dim(OCD_stN) #> [1] 250428 134 This data shows that there are actually sufficient observations spread through time (last 60+ years) to fit a spatiotemporal model: hist(OCD_stN$YEAR, xlab="Year", main="", col="darkgrey")

In fact, because the data set above represents values of OCD at variable depths, we can use this data to fit a full 3D+T spatiotemporal model in the form:

$$${\rm OCD}(xydt) = d + X_1 (xy) + \ldots + X_k (xy) + \ldots + X_p (xyt) \tag{7.7}$$$

where $$d$$ is the depth, $$X_k (xy)$$ are static covariates i.e. the covariates that do not change in time, and $$X_p (xyt)$$ are spatiotemporal covariates i.e. covariates that change with time. Here we can assume that the static covariates are mainly landform and lithology: these have probably not changed significantly in the last 100 years. Land cover, land use and climate, on the other hand, have probably changed drastically in the last 100 years and have to be represented with a time-series of images. There are, indeed, several time-series data sets now available that can be used to represent land cover dynamics:

• HYDE 3.2 Historic land use data set (Klein Goldewijk et al. 2011): portrays the distribution of major agricultural systems from 10,000 BC (pre-historic no landuse condition) to the present time. 10 categories of land use have been represented: total cropping, total grazing, pasture (improved grazingland), rangeland (unimproved grazingland), total rainfed cropping, total irrigated cropping, with further subdivisions for rice and non-rice cropping systems for both rainfed and irrigated cropping.

• CRU TS2.1 climatic surfaces for period 1960–1990 (Harris et al. 2014).

• UNEP-WCMC Generalised Original and Current Forest cover map showing global dynamics of forest cover.

All these layers are available only at a relatively coarse resolution of 10 km, but then cover longer time spans. Note also that, since these are time-series images, spatiotemporal overlay can take time as spatial overlay must be repeated for each time period. The spatiotemporal matrix file already contains the results of overlay, so that we can focus directly on building spatiotemporal models of OCD e.g.:

pr.lst <- names(OCD_stN)[-which(names(OCD_stN) %in% c("SOURCEID", "DEPTH.f", "OCDENS", "YEAR", "YEAR_c", "LONWGS84", "LATWGS84"))]
fm0.st <- as.formula(paste('OCDENS ~ DEPTH.f + ', paste(pr.lst, collapse="+")))
sel0.m = complete.cases(OCD_stN[,all.vars(fm0.st)])
## takes >2 mins
rf0.OCD_st <- ranger(fm0.st, data=OCD_stN[sel0.m,all.vars(fm0.st)],
importance="impurity", write.forest=TRUE, num.trees=120)

the most important covariates being:

xl <- as.list(ranger::importance(rf0.OCD_st))
print(t(data.frame(xl[order(unlist(xl), decreasing=TRUE)[1:10]])))

which shows that the most important soil covariate by far is soil depth, followed by elevation, grazing, MODIS cloud fraction images, cropland and similar. For a full description of codes please refer to Sanderman, Hengl, and Fiske (2018) (also available in this table).

Finally, based on this model, we can generate predictions for 3–4 specific time periods and for some arbitrary depth e.g. 10 cm. The maps below clearly show that ca 8% of soil organic carbon has been lost in the last 90 years, most likely due to increases in grazing and croplands. The maps also show, however, that some areas in the northern latitudes are experiencing an increase in SOC, possibly due to higher rainfall i.e. based on the CRU data set.

This demonstrates that, as long as there is enough training data spread through time, and as long as covariates are available for the corresponding time ranges, machine learning can also be used to fit full 3D+T spatiotemporal prediction models (Gasch et al. 2015). Once we produce a time-series of images for some target soil variable of interest, the next step would be to implement time-series analysis methods to e.g. detect temporal trends and areas of highest apparent soil degradation. An R package that is fairly useful for such analysis is the greenbrown package, primarily used to map and quantify degradation of land cover (Forkel et al. 2015).

We can focus on the time-series of predicted organic carbon density for USA48:

library(greenbrown)
library(raster)
setwd()
tif.lst <- list.files("extdata/USA48", pattern="_10km.tif", full.names = TRUE)
g10km <- as(readGDAL(tif.lst[1]), "SpatialPixelsDataFrame")
for(i in 2:length(tif.lst)){ g10km@data[,i] = readGDAL(tif.lst[i], silent=TRUE)$band1[g10km@grid.index] } names(g10km) = basename(tif.lst) g10km = as.data.frame(g10km) gridded(g10km) = ~x+y proj4string(g10km) = "+proj=longlat +datum=WGS84" to speed up processing we can subset grids and focus on the State of Texas: library(maps) library(maptools) states <- map('state', plot=FALSE, fill=TRUE) states = SpatialPolygonsDataFrame(map2SpatialPolygons(states, IDs=1:length(states$names)), data.frame(names=states$names)) proj4string(states) = "+proj=longlat +datum=WGS84" ov.g10km = over(y=states, x=g10km) txg10km = g10km[which(ov.g10km$names=="texas"),]
txg10km = as.data.frame(txg10km)
gridded(txg10km) = ~x+y
proj4string(txg10km) = "+proj=longlat +datum=WGS84"
spplot(log1p(stack(txg10km)), col.regions=SAGA_pal[[1]])
g10km.b = raster::brick(txg10km)

We can analyze this time-series data to see where the decrease in organic carbon is most significant, for example the slope of the change:

trendmap <- TrendRaster(g10km.b, start=c(1935, 1), freq=1, breaks=1)
## can be computationally intensive
plot(trendmap[["SlopeSEG1"]],
col=rev(SAGA_pal[["SG_COLORS_GREEN_GREY_RED"]]),
zlim=c(-1.5,1.5), main="Slope SEG1")

which shows that loss of soil organic carbon is especially distinct in the southern part of Texas. The slope coefficient map is, on average, negative, which indicates that most of the state has lost organic carbon for the period of interest. Note that running such time-series analysis is not trivial as a sufficient number of observations in time (if possible: repetitions) is needed to be able to extract significant patterns. Also TrendRaster function can be quite computationally intensive, hence some careful planning of the processing steps / processing infrastructure is usually recommended.

## 7.9 Summary points

Based on all the examples and discussion above, the following key points can be emphasized:

1. OCS for an area of interest can be derived either using 2D or 3D approachs. 3D approaches typically include modeling ORC, BLD and CRF separately (and then deriving OCS per pixel), or modeling OCD for standard depths and then converting to OCS.
2. Publicly available RS-based covariates (SRTM / ALOS DEM, Landsat, Sentinel satellites) are available for improving the mapping accuracy of OCS. Improving the accuracy of OCS maps is becomming less expensive, given the increasing availability of RS data.
3. PT (Pedo-Transfer) rules can be used to fill in (estimate) missing BLD values and to estimate ORC for deeper soil depths. Also global maps with predictions of BLD and CRF can be used to supply missing values, if there are no other alternatives.
4. Machine learning techniques such as Random Forest, neural nets, gradient boosting and similar, can be used to predict soil organic carbon in 2D, 3D and in spatiotemporal modeling frameworks. The accuracy of these predictions is improved relative to linear statistical models, especially where the relationship between soil organic carbon distribution and climatic, land cover, hydrological, relief and similar covariates is complex (i.e. non-linear).
5. Global estimates of ORC, BLD and CRF can be used as covariates so that consistent predictions can be produced (as explained in Ramcharan et al. (2018a)).
6. By producing spatial predictions of OCS for specific time periods, one can derive estimates of OCS change (loss or gain).
7. Most of the statistical / analytical tools required for running spatial analysis, time series analysis, export and visualization of soil carbon data are available in R, especially thanks to the contributed packages: aqp, caret, ranger, xgboost, GSIF, greenbrown and similar.