You are on page 1of 79

helles blau (Balken)

Digital Soil Mapping


Strategies for Regional
Land Use Planning in the
North and South-West
Regions of Cameroon
Francis B. T. Silatsa and Dr Martin Yemefack

Technical Report
Authors: Francis B. T. Silatsa, Dr Martin Yemefack

Editorial Board: C. Wilczok, Dr D. Rückamp, Department B2 Groundwater and


Soil, Sub-Department B2.4 Soil as a Resource – Properties and
Dynamics

S. Thayne, Department B4 Geoscientific Information,


International Cooperation, Sub Department 4.1 International
Cooperation

Project: Project on Soil and Subsoil Resources of North and South-West


Regions, Cameroon (PRESS NO & SW)

BMZ N°: 2014.2472.0

BGR N°: 05-2388

Country of assignment: Cameroon

Date: September 2017

Funded by: Federal Ministry for Economic Cooperation and Development


(BMZ)

Publisher: Federal Institute for Geosciences and Natural Resources (BGR)


PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Table of Content

List of Figures ........................................................................................................................ iv

List of Tables .......................................................................................................................... v

Abbreviations ..................................................................................................... vi
Abstract ............................................................................................................... 1
1. INTRODUCTION ................................................................................................ 2
1.1 Rationale........................................................................................................................... 2

1.2 Objective ........................................................................................................................... 4

1.3 Importance of selected soil properties ............................................................................. 4

1.3.1 Particle size distribution ............................................................................................. 4


1.3.2 Soil pH (pH water) ..................................................................................................... 4
1.3.3 Soil organic carbon .................................................................................................... 5
2. STUDY AREA.................................................................................................... 6
2.1 North region of Cameroon ................................................................................................ 6

2.1.1 Climate ....................................................................................................................... 6


2.1.2 Geology ...................................................................................................................... 7
2.1.3 Hydrography .............................................................................................................. 7
2.1.4 Soils ........................................................................................................................... 8
2.1.5 Vegetation and land use ............................................................................................ 9
2.1.6 Relief ........................................................................................................................ 10
2.2 South-West region of Cameroon..................................................................................... 11

2.2.1 Climate ...................................................................................................................... 12


2.2.2 Hydrography ............................................................................................................. 13
2.2.3 Soils .......................................................................................................................... 14
2.2.4 Vegetation and Land use.......................................................................................... 14
2.2.5 Relief ......................................................................................................................... 15
3. METHODS ....................................................................................................... 17
3.1 Soil profile data ................................................................................................................ 17

3.1.1 Data harmonization and standardization.................................................................. 17


3.1.2 Quality assessment of legacy soil database ............................................................18
3.2 Environmental covariates ............................................................................................... 20
i
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

3.3 Spatial prediction of soil properties ................................................................................ 23

3.3.1 Digital soil mapping approach ................................................................................. 23


3.3.2 Soil properties splinning........................................................................................... 24
3.3.3 Random forest model .............................................................................................. 24
3.3.4 Selection of important covariates ............................................................................ 25
3.3.5 Model validation ....................................................................................................... 25
3.4 Improving the sampling network .................................................................................... 26

4. RESULTS AND DISCUSSION ............................................................................ 28


4.1 Spatial distribution of soil profiles in the studied regions ............................................... 28

4.2 Summary statistics of legacy soil database ................................................................... 28

4.3 Summary statistics of soil properties after splining........................................................ 29

4.4 Prediction performance ................................................................................................... 31

4.5 Variables of importance.................................................................................................. 35

4.6 Spatial prediction of soil properties ................................................................................ 36

4.6.1 Spatial prediction assessment in the North region.................................................. 36


4.6.2 Spatial prediction assessment in the South-West region ....................................... 38
4.7 Digital maps of soil properties ........................................................................................ 38

4.7.1 Soil organic carbon in the North and South-West regions ...................................... 38
4.7.2 Soil pH in the North and South-West regions ......................................................... 40
4.7.3 Clay content in the North and South-West regions ................................................. 42
4.7.4 Sand content in the North and South-West regions................................................ 44
4.7.5 Silt content in the North and South-West regions ................................................... 46
4.8 Sampling network design ............................................................................................... 48

5. CONCLUSION .................................................................................................. 51
Acknowledgements .............................................................................................................. 52

6. REFERENCES................................................................................................. 53
7. ANNEXES ....................................................................................................... 59
Annex A: prediction assessment of soil properties .............................................................. 59

Annex A1: Prediction assessment of soil properties in the North region ......................... 59
Annex A2: Prediction assessment of soil properties in the South-West region .............. 60
Annex B : Variation of soil properties with depth in the North region of Cameroon ............. 61

ii
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex B1: Variation of SOC with depth in the North region ............................................. 61
Annex B2: Variation of clay with depth in the North region ............................................... 61
Annex B3: Variation of sand with depth in the North region ............................................ 62
Annex B4: Variation of silt with depth in the North region ................................................ 62
Annex B5: Variation of pH water with depth in the North region ..................................... 63
Annex C : Variation of soil properties with depth in the South-West region of Cameroon . 64

Annex C1: Variation of SOC with depth in the South-West ............................................. 64


Annex C2: Variation of clay with depth in the South-West region ................................... 65
Annex C3: Variation of sand with depth in the South-West region .................................. 65
Annex C4: Variation of silt with depth in the South-West region ..................................... 66
Annex C5: Variation of soil pH with depth in the South-West region............................... 66
Annex D: R script (Example: Prediction of clay content) ..................................................... 67

iii
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

List of Figures

Figure 1: Administrative organization of the North Region....................................................... 6


Figure 2: Hydrographic network of the North Region with the main rivers .............................. 7
Figure 3: Majors soils types in the North region ....................................................................... 8
Figure 4: Land cover of the North region of Cameroon ............................................................ 9
Figure 5 : Relief of the North region of Cameroon ..................................................................11
Figure 6: Administrative organization of the South-West region, Cameroon......................... 12
Figure 7: Hydrographic Network of the South-West region of cameroon .............................. 13
Figure 8: Majors soils types in the South-West region ........................................................... 14
Figure 9: Land cover of the South-West region ...................................................................... 15
Figure 10 : Relief of the South-West region of Cameroon ..................................................... 16
Figure 11: Digital Soil Mapping steps for soil properties prediction .......................................23
Figure 12: Spatial distribution of soil profiles in the North and South-West regions ............. 28
Figure 13: Relative importance of covariates in the North region .......................................... 35
Figure 14: Relative importance of covariates in the South-West region ............................... 36
Figure 15: Distribution of SOC (g/kg) in North region of Cameroon (0 - 30 cm)................... 39
Figure 16: Distribution of SOC (g/kg) in South-west region of Cameroon (0 – 30 cm) ........ 40
Figure 17: Distribution of soil pH in the North region of Cameroon (0 - 30 cm)..................... 41
Figure 18: Distribution of soil pH in the South-West region of Cameroon (0 - 30 cm) ......... 42
Figure 19: Clay (%) distribution in the North region of Cameroon (0 - 30 cm) ..................... 43
Figure 20: Clay (%) distribution in the South-West region of Cameroon (0 - 30 cm) ........... 44
Figure 21: Sand (%) distribution in the Norh region of Cameroon (0 - 30 cm) ......................45
Figure 22: Sand (%) distribution in the South-West region of Cameroon (0 - 30 cm) .......... 46
Figure 23: Silt (%) distribution in the North region of Cameroon (0 - 30 cm) ......................... 47
Figure 24: Silt (%) distribution in the South-West region of Cameroon (0 - 30 cm) ............. 48
Figure 25: Sampling netwok in the North region ................................................................... 49
Figure 26: Sampling network in the South-West region ........................................................ 50

iv
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

List of Tables

Table 1: Differences between the common particle size classifications .................................19


Table 2: List of applied grigged environmental covariates ..................................................... 22
Table 3: Statistics of targeted soil properties in the North Region ......................................... 29
Table 4: Statistics of targeted soil properties in the South-West Region ............................... 29
Table 5: Summary statistics of soil properties after splinning in the North region ................. 29
Table 6: Summary statistics of soil properties after splinning in the South-West .................. 30
Table 7: Prediction performances of soil properties in the North region ................................ 32
Table 8: Prediction performances of soil properties in the South-West region ...................... 34
Table 9: Statistical feature values of predicted soil properties (30 cm) in the North .............. 37
Table 10: Statistical feature of predicted soil properties (30 cm) in the South-West ............. 38

v
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Abbreviations

AfSIS : Africa Soil Information Service


BGR : Federal Institute for Geosciences and Natural Resources
Camsodat : Cameroon Soil Database
CCC : Concordance Correlation Coefficient
DSM : Digital Soil Mapping
EVI : Enhanced Vegetation Index
GIS : Geographic Information System
GSP : Global Soil Partnership
IITA : International Institute of Tropical Agriculture
INC : National Institute of Cartography
INS : National Institute of Statistics
IRAD : Institute of Agricultural Research and Development
IRGM : Institute of Mining and Geological Research
ISO : International Organization for Standardization
LOCV : Leave One Out Cross Validation
MINADER : Ministry of Agriculture and Rural Development
MINEPAT : Ministry of Economy, Planning and Regional Development
MODIS : Moderate Resolution Imaging Spectroradiometer
MSSD : Mean Square Shortest Distance
NO : North Region, Cameroon
NIR : Near Infrared
ORSTOM : Office de la Recherche Scientifique et Technique Outre-Mer
OOB : Out of-Bag
pH : Potential of hydrogen
PRESS NO & : Project on Soil and Subsoil Resources of North and South-West
SW Regions, Cameroon
PNDP : Programme National de Développement Participatif
RFM : Random Forest Model
RMSE : Root Mean Square Error
SOC : Soil Organic Carbon
SOM : Soil Organic Matter
SW : South-West Region, Cameroon
UNSECO : United Nations Educational, Scientific and Cultural Organization
USDA : United State Department of Agriculture
GPS : Global Positioning System

vi
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Abstract

Good quality and spatially explicit soil data are required to support research and to inform
discussions and decisions on sustainable soil management to improve food security in
Cameroon. The goal of this study was to produce a robust quantitative framework, which is
updateable and spatially explicit, in order to generate and maintain functional soil properties
(pH, soil organic carbon (SOC), and Particle size distribution) information in the North and
South-West regions of Cameroon. We applied the random forest model coupled with
auxiliary gridded environmental variables to predict soil properties in various depths. Soil
data were obtained from the national database of soil profiles data (Camsodat 0.1).
Variables explaining the distribution of soil properties have been identified for each soil
property using a large compilation of raster images generated by the Africa Soil Information
Service (AfSIS) project and publicly available. The vertical distribution of each soil property
along the soil profiles was modelled with the mass preserving equal-area quadratic splines
at standard depths intervals (0 – 5, 5 – 15, 15 – 30, 30 – 60, 60 – 100, 100 – 200 cm) as
defined in GlobalSoilMap project. The model performance parameters showed promising
prediction results and a large influence of soil sampling depth, terrain attribute, climatic
parameters, and land cover in predicting soil properties was observed. The resulting
functional soil properties maps are available as gridded dataset at 250m spatial resolution.
These maps are useful input to many types of models, like simulation of crop yield potentials,
estimation of yield gaps, estimation of carbon stock, assessment of land suitability for
agricultural production etc. We then proposed a sampling network strategy, based on the
random spatial coverage of the regions studied, which will increase the quality of the
predictions obtained in this study.

Keywords: Digital soil mapping, soil properties, random forest, cross validation, Cameroon,
North region, South-West region, soil information, PRESS NO & SW project, GlobalSoilMap,
environmental covariates.

1
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

1. INTRODUCTION

1.1 Rationale

Numerous environmental and socio-economic models require soil parameters as inputs to


estimate and forecast changes in our future life conditions. However, the availability of soil
data is limited. Soil information remains either missing at the appropriate scale, or its
meaning is not well explained for reliable interpretation. Considering the amazing variability
and diversity of soils, knowledge about soil types and properties is vital to ensure sustainable
agricultural production and development planning. Policy makers and environmental
managers require accurate predictions of the variation of soil properties to assess the
suitability of the soil to perform its functions and the interaction between the soil and the
wider environment (Keesstra 2016). Such predictions allow government departments and
agencies to understand the current state of soils, how this is changing and the pressures
placed upon soil quality. Furthermore, accurate information on soil variation in space could
indicate preferentially suitable areas for certain land use types.

Solutions on how soils can be best managed at landscape level require accurate soil
information (e.g. soil maps) as basis for decision making. Easy to interpret and use soil
databases are mandatory to support decision making and modeling on the regional/national
scale (Keesstra 2016). However, the available databases often fail to provide the necessary
soil parameter for the users.

In order to fulfill the user requirements, a new generation of soil information has to be
initiated that makes use of the state-of-the-art data collection or compilation and spatial
prediction techniques. DSM provide continuous soil information instead of spatially discrete
point information. It is also cost and time effective and a promising technique in areas with
a huge lack of soil data availability and applicable to increasing computer performances. In
order to be in phase with current development trends and challenges, there is a consequent
need to be quick and concise in the production of relevant soil information. Digital soil
mapping is opportunely the right strategy.

Digital Soil Mapping (DSM) is defined as the creation and population of spatial soil
information systems by numerical models inferring the spatial and temporal variations of soil
types and soil properties from soil observation and knowledge and from related
environmental variables (Lagacherie 2008). Traditional soil survey and DSM do not differ
much basically in principle (Roecker et al. 2006). Both require a predefined model of soil
formation, data on soil properties and on other environmental variables that have significant
impact on soil formation and thus on the spatial distribution of the soil properties. Both
approaches need input data on soil and covariates characterizing the environment where

2
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

the soil formation takes place. The major difference is how the model derives the soil
information from the input data (Dobos et al. 2006).

Traditional soil survey models are based on empirical studies. They define qualitatively
correlation that formulates a mental model in the surveyor’s mind, used to understand and
characterize the soil resources (Dobos et al. 2006). This requires intensive field work.
Decisions are made mainly on the field, where all environmental covariates can be directly
observed and information on the soil can be deduced. The DSM approach is quite similar,
as it is based on hard soil data as well. Like in the traditional approach, profile information
is needed to train the models, and to understand the soil resources of the area.

The major differences, the strengths and also the limitations are coming from the way the
environmental covariates are represented in the procedure. Digital soil mapping requires
digital data sources (gridded raster layer) as input variables (covariates) for the quantitative
models. Then, DSM provides the means for supplying soil information in format and
resolution compatible with other fundamental data sets from remote sensing, terrain
analysis, and other systems for mapping, monitoring, and forecasting biophysical processes.
There is also growing agreement indicating that digital soil mapping techniques offer the
potential to greatly accelerate the rate of map production and update.

In accordance with the pressing necessity in Cameroon to have a national and regional
structure for geospatial data collection and sharing, given that it is the only way to produce
harmonized and shared data for sustainable development planning. The Project on Soil and
Subsoil Resources of North and South-West Regions (PRESS NO & SW) is a bilateral
technical project in the framework of German Cooperation between MINEPAT/DATZF and
BGR with participation of IRAD, IRGM, INS, INC and PNDP, financed by the German
Federal Ministry for Economic Cooperation and Development (BMZ). With the aim of
supporting land use planning in Cameroon with geospatial database contributions to
georessource aspects such as soils, water, geology etc. for the North and the South-West
regions. Thereby, the objective of PRESS NO & SW project is to provide harmonized,
reliable, and unbiased geospatial datasets in an exchangeable form between institutions
and put them on disposal to civil society as invaluable tools for coordinated development.

Given that intensive field surveys campaigns covering the whole South-West and North
regions areas of Cameroon are very unlikely to happen in the near future and would be very
expensive as far as soil is concerned. Digital soil mapping technique is likely a viable tool
that can assist to develop the soil data on a more cost-effective basis. While the state-of-
the-art requirements on the quality assurance, accuracy assessment, GIS support, and
reported quantitative data development procedure are more easily fulfilled. These
advantages make DSM a possible crucial strategy for the project PRESS NO & SW to
provide continous soil information on a regional scale.
3
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

This report presents the practical framework for digital mapping of soil properties in the North
and South-West regions of Cameroon with the data of previous soil surveys conducted in
these regions. After a brief presentation of the regions studied, we will present the
methodological approach used, with particular emphasis on the soil database and the
method of prediction. Specific analyzes assessed the quality and accuracy of the prediction
of the soil properties at the depths considered. Then the spatial distribution of soil organic
carbon, soil pH, and soil particle size distribution are presented. These products can be used
in various regional planning activities.

1.2 Objective

The purpose of this study is to apply state-of-the-art Digital Soil Mapping (DSM) strategies
to generate spatial continuous soil maps of Cameroon’s North and South-West regions key
soil properties. Based on dicrete legacy soil data and gridded environmental covariates
(spatial resolution: 250m), the predicted soil properties (SOC, clay, sand, silt, and pH,) can
be e.g., used as input for sustainable regional land use planning.

1.3 Importance of selected soil properties

1.3.1 Particle size distribution

The particle size distribution (sand, silt and clay content) of the soil is one of its most
important characteristics. It strongly affects water and nutrient retention, infiltration,
drainage, aeration, SOC content, pH, and porosity and that affects many soil functions and
mechanical properties (Akpa et al. 2014). It is used in the diagnosis of some key epipedons
(Bockheim and Hartemink 2013), and also determines the suitability of the soil for a
particular use and management, waste disposal, and water management (Thompson et al.
2012).

1.3.2 Soil pH (pH water)

Soil pH measures the acidity and alkalinity of the soil, and provides information on growing
conditions for most agricultural plants (McCauley et al. 2017). All plants are affected by the
extremes of pH, but there is wide variation in their tolerance of acidity and alkalinity (Haling
et al. 2011). Some plants grow well over wide pH range, whilst others are very sensitive to
small variations in acidity or alkalinity (Munns 1986). Microbial activity in the soil is also
affected by soil pH (Rousk et al. 2009). Where the extremities of acidity or alkalinity occur,
various species of earthworms and nitrifying bacteria disappear. Soil pH can alter the
chemical nature of molecules and substances in the soil; hence it affects the availability of
nutrients and how the nutrients react with each other (Clemens et al. 1990). The changes in

4
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

the availability of nutrients cause the majority of effects on plant growth attributed to acid
soils. Consequently, knowledge of the soil pH is vital for proper soil management.

1.3.3 Soil organic carbon

Soil organic carbon (SOC) is the main constituent of soil organic matter (SOM), formed by
the biological, chemical and physical decay of organic materials that enter the soil system
from sources above ground (e.g. leaf fall, crop residues, animal wastes and remains) or
below ground (e.g. roots and soil biota). The elemental composition of SOM varies, with
values up to 50 per cent carbon in undecayed wood. Soil carbon plays a vital role in
regulating climate, water supplies and biodiversity, and therefore providing the ecosystem
services that are essential to human well being (Hombegowda et al. 2016; Mulder et al.
2016; McBratney et al. 2014). The distribution of SOC reflects climate distribution, with
greater accumulations of carbon in more humid areas. Temperature also plays a secondary
role in global SOC distribution. This is illustrated by the occurrence of deep peat deposits in
both tropical and polar humid areas. Within climatic zones the amount of SOC is determined
by soil moisture, which in turn is influenced by relief, soil texture and clay type. The carbon
content of soils under different land cover types varies substantially.

5
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2. STUDY AREA

2.1 North region of Cameroon

The North region extends from latitudes 7°00’N to 10°20’N and longitudes 12°10’E to
15°40’E, and cover and area of approximately 66 090 km². Neighbouring territories include
the Far North Region to the north, the Adamaoua Region to the south, Nigeria to the west,
Chad to the east, and Central African Republic to the southeast (Figure 1).

Figure 1: Administrative organization of the North Region

2.1.1 Climate

The North region is in the Tropical Sudanian climatic zone with medium pluviometry, high
evaporation during the dry season, and high average temperature (Humbel 1968). There
are two distinct seasons, with the dry season from mid-October to April and a wet season
from May to October with maximum rain in August. Inter annual variations of rainfall are
large from one place to another, with values between 537 and 1340 mm/yr at Garoua, and
1037 and 1873 mm/yr at Poli (Martin 1962). The average annual temperature is 28 °C, with
monthly mean maxima observed from March to May and the minimum in December and
January.

6
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2.1.2 Geology

The geology of the North region has been among others described in many studies (Cratchle
1984; Kamguia 2005; Martin 1962), and is mainly composed of:

 Old and new alluvial deposits of the Benue and Faro valleys
 Sandy sedimentary formations of the upper and middle cretaceous, which extend on
both side of the Benue, as well as in the valleys of Mayo Rey and Vina
 Lower Cretaceous clay sedimentary formations
 Precambrian metamorphic formations of the old African basement, including in
particular the schists and micachists of the Poli series and the ectinites and
migmatists of the basic complex
 Plutonic formations, including granites of different ages and composition
 Volcanic formations of various ages (trachyte, basalt, andesite, and rhyolite) exist in
the region, but are of little pedological importance. they are found in the form of rocky
peaks

2.1.3 Hydrography

The hydrographic network is linked to the morphology of the region, and the two main rivers
are Benue and Vina (Figure 2).

Figure 2 : Hydrographic network of the North Region with the main rivers (OpenStreetMap
2018)
7
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Vina, is one of the main branches of the river Logone, originates in Adamaoua and flows
eastwards to the border of Chad. All the rest of the region is drained by the Benue and its
tributaries (Martin 1962). Benue river begins in the cliff that borders the Adamaoua plateau,
then flows on the ancient base until Boukma, where the sedimentary deposits begin (Humbel
1965). Its main tributaries are: The Mayo Kebbi formed by the Mayo Louti and Mayo Kebbi;
The Faro and its tributary the Mayo Deo, which comes from the Adamaoua. The largest
open-water stretch in the region is the Lagdo Dam built between August 1977 and July 1982
for electricity supply (Ngatcha et al. 2001).

2.1.4 Soils

In the North region of Cameroon, the most occuring soil groups are presented in the figure
3. They include Lixisols (22%), Luvisols (18%), Nitisol (16%), Plinthosols (12 %), Vertisols
(11%) (Jones et al. 2013).

Figure 3: Majors soils types in the North region (adapted from Jones et al. 2013)

According to Yerima and Van Ranst (2005), Luvisols exist around Garoua (West, East and
North of Garoua), Vertisols occur in the North of Garoua through Maroua, and Regosols
occur in the South of Garoua. Most of these soils have good agricultural potentials (Luvisols,
Nitisols and vertisols with some specific physical properties management constraints). In
general, Lixisols and Plinthosols are soils with poor natural fertility (WRB 2014).

8
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2.1.5 Vegetation and land use

As reported by Martin (1962), climatic variations explain better the main variations of the
vegetation of the North region. The Benue valley and Mayo Godi form a boundary between
the typically Sudanian woodland and wooded savannahs in which Sahelian elements
already appear. According to the sentinel land cover, the main land cover units in the north
region are tree cover area, grassland/savannah and cropland (Figure 4).

Figure 4 : Land cover of the North region of Cameroon (Adapted from ESA 2016)

There are also many protected areas in the North Region, which contribute to the economic
development of the region (Ndamé 2003). They were created between 1932 and 1980, and
consist of three national parks (Benoue 180 000 ha, Faro 330 000 ha, Bouba Ndjidda 220
000 ha) and 27 hunting areas (ZIC) or game reserves, 23 of which are leased to
predominantly expatriate professional hunting guides. All this vast network of protected
areas represents nearly three million hectares, that is 44% of the total territory of the region
(Ndamè 2007).

9
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

In the North region of Cameroon, many ethnic groups farm on small plots for subsistence.
Sorghum, millet (both fast- and slow-growing), and maize are the staple crop throughout
most of the region, and rice is especially popular in cities. Other crops include yams along
the Lagdo reservoir and at Tcholliré and groundnutsin the Mayo-Rey division. Farmers often
create their fields by burning an area of its vegetation during the dry season. Only fruit trees
or trees useful for animal fodder or firewood are kept, such as baobab, faidherbia, and karita.

2.1.6 Relief

The Benue depression constitutes the North Region's primary land feature. This basin runs
along the Mayo Kébi and Benue River and has an elevation of between zero and 200 metres
(Figure 5). The valleys surrounding the various rivers that feed the Kébi and later Benue
reach elevations only slightly higher than this, averaging 200–500 metres in the north and
500–1000 metres in the south. Garoua lie at about 235 metres. The third significant land
feature is the Mandara Mountains and their southern extension. These chains form most of
the western edge of the region, with peaks as high as 1000 metres. The mountains continue
north into the Far North region and Nigeria, though their elevations gradually drop to as low
as 500 metres. The surrounding terrain is hilly.

10
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 5 : Relief of the North region of Cameroon (STRM 30m)

2.2 South-West region of Cameroon

The South-West region extends from latitudes 3°50’N to 6°30’N and longitudes 8°30’E to
10°00’E (Figure 6), covering an area of about 25 410 km². The human capital and richness
of the earth, coupled with abundant land resources, makes this region attractive to both
large-scale and small-scale intensive agricultural businesses (Business in Cameroon 2013).
The main economics activities of the South-West region include agriculture, tourism, trade,
fishing and hunting (Molongue 2016). It is one of the two anglophone (English-speaking)
regions of Cameroon.

11
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 6: Administrative organization of the South-West region, Cameroon (MINADER)

2.2.1 Climate

The area has an equatorial and sub-equatorial climate characterized by heavy rainfall
(average of more than 2000 mm/yr), a long rainy season (at least three months), high relative
humidity (generally about 85%) and high temperatures (above 22°C on average) (Molua
2006). The climate is attractive to diverse agricultural commodities and the near temperate
climate at the foot of Mount Cameroon and the tropical climate along its coastline make the
region suitable for lucrative business endeavours (Business in Cameroon 2013)

12
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2.2.2 Hydrography

The South-West Region possesses a high density of hydrographical network (Figure 7)


characterised by two principal basins: the Manyu and the Moungo basins. The Manyu basin,
drains in the Mamfe watershed, River Manyu and its tributaries, with the River Munaya being
the most important.

Figure 7: Hydrographic Network of the South-West region of Cameroon (Data from


MINFOF)

Other small and average flowing water courses are found around forest massif. The water
acquifers are usually not far from the surface. Lake Ejagham is also another water body that
provides along side the other rivers, proteinous food for the local people and serves as a
source of potable water. Its 250 km long coastline and dense river network also account for
this region substantial fish and seafood potential.

13
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2.2.3 Soils

In the South-West region of Cameroon, major’s soil groups are Nitisols (56%), Ferralsols
(23%), Gleysols (7%), Umbrisols (6%), Leptosols (4%) and Andosols (3%) (Jones et al.
2013). Andosols, from volcanic material occur around Mount Cameroon stretching through
the Bakossi area (Yerima and Van Ranst 2005). Nitisols exist around the mountain chain,
stretching from Mount Cameroon to the Bambouto area. Ferralsols exist along the border
with the North-West region.

Figure 8: Majors soils types in the South-West region (Jones et al. 2013)

2.2.4 Vegetation and Land use

The South-West region is gifted with high yields of both cash crops and food crops. Over
38% of the total surface area in the South-West is under cultivation (MINADER 2013).
Perennial crops include cacao, palms, bananas, tea, coffee, citrus and rubber. Commonly
grown food and vegetable crops include cassava, maize, yams, cocoyams, groundnuts,
pepper, plantains etc. Export crops such as oil palm, and rubber are grown industrially in
plantations by companies such as Cameroon Development Corporation (CDC). However,
many cash crops such as cacao and oil palm are grown and sold by individuals along with
several food crops. Food crops are used for personal consumption as well as for sale in
local, national and international markets including Nigeria, Equatorial Guinea and Europe
14
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

As reported by Yerima and Van Ranst (2005), this region belongs to the dense equatorial
forest, which can be divided into three types:

- The mangrove forest from Bimbia through Tiko;


- The rain forest, variously known as tropical rain forest lies inland adjacent to the
mangrove forest. The forest is characterized by a multilayer structure composed of
several tree layers and a continuous ground herb understory;
- The mountain forest is found at high altitudes on the slopes of mountainous massifs
where island of dense forest can be found. This is typical of the Cameroon mountain
and the Kupe Manenguba mountains.

Many protected areas exist in the south west region of Cameroon. Four national parks
(Mount Cameroon, Bakossi, Korup, Takamanda and Ndongere), and other protected
reserve (Rumpi hill, Banyang Mbo, Ejagham)

The main land cover units as classified by Sentinel Land cover are presented in figure 9.
The two main land units are tree cover and cropland in the South-West region.

Figure 9: Land cover of the South-West region (Adapted from ESA 2016)

2.2.5 Relief

In a global sense, the relief of the South-West region shows three distinct aspects:

15
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

 The littoral plain (Tiko basin in Rio del Rey) that is interrupted by Mount Cameroon;
 The Ndian basin (low altitude region that is in contact with the sea);
 The Mamfe caldera (depression zone, which is more or less encircled by the
western highlands to the east, the Akwaya plateau to the north and the Rumpi
mountains to the south).

It is also in this region that the mountain chain of Cameroon begins. This chain includes
Mount Cameroon, Mount Manengouba, Mount Koupe, and the Bambouto Mountains.

Figure 10 : Relief of the South-West region of Cameroon (STRM 30m)

16
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

3. METHODS

3.1 Soil profile data

Legacy soil profile data were obtained from the Cameroon soil profile database (Camsodat
0.1) that we collated from reports of many decades of soil surveys and research conducted
in Cameroon (Silatsa et al. 2017). Camsodat 0.1 is the result of the joint efforts of the local
initiative (Cameroon), and the International Soil Reference Information Center (ISRIC). The
database is compiled from reports of soil studies carried out in Cameroon since 1950 under
multiple bilateral cooperation with Cameroon and other countries or institutions. Soil profiles
data were listed from these reports and only those that could be located were recorded, in
order to establish and model the relationship between soil data and auxiliary information.
The data were then harmonized and standardized following the GlobalSoilMap
specifications when feasible (see below, section 3.1.1). In Cameroon, Camsodat 0.1 is
currently the only harmonized soil profile database to which a number of essential soil
attributes (SOC, pHwater, pHKCl, Exchangeable bases (Ca2+, Mg2+, K+, Na+), Cation exchange
capacity (CEC), Base saturation, total nitrogen, and exchangeable phosphorus) are
attached.

3.1.1 Data harmonization and standardization

The cross-border description of soil condition status and change requires the availability of
harmonized and comparable soil data relating to the site and general area of the soil profile.
Harmonization is the conversion of a value observed or measured by a recorded non-
standard method, to a target value as if observed or measured by a specific standard method
(Leenaars et al. 2014).

Data standardization however, is the critical process of bringing data into a common format
that allows for collaborative research and large-scale analytics (Bader et al. 1999). Despite
the growing use of standardized terminologies in soil science, the same concept may be
represented in a variety of ways from one setting to the next.

However, there is generally no universal equation for converting property values from one
method to another in all situations (GlobalSoilMap 2015). The Global Soil Partnership (GSP)
proposed the establishment of geographical sensitive systems that will allow applying these
specific conversion functions. This implies that each region or continent will need to develop
and apply specific conversion functions (Batjes et al. 2017). But in the meantime, some
functions have been recommended as standard reference methods and are currently widely
used for this purpose (Baritz et al. 2014). For this study, harmonization and standardization
followed the main steps defined by Nelson et al. (2017).

17
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

1rst step: We observed and clean the data. This was done during the data entry phase.
The various reports were first scrutinized for the quality of the data they contained.
Understanding the data was the hardest and time-consuming part, because we
needed sometimes to understand the reason why some data were so different from
others, and in other cases we spend a lot of time to discover that it was not possible
to correctly spatially localize some soil profiles.
2nd Step: We have established the traceability of the reference data to always keep
track of the data entry point. At this level, we created a reference bibliography
containing all the documents that have been used for the construction of our database
using the unique ID for each document. Some of these documents can be consulted
online.
3rd step: We applied the suitable and comparable standard to present the data in a
common and unique format: Global soil map standard have been adopted
(GlobalSoilMap 2015).

3.1.2 Quality assessment of legacy soil database

3.1.2.2 Precision of geolocation

Several of collected legacy soil profiles were sampled and described before the era of Global
Positionning Systems (GPS). In most cases, the spatial location of the soil profiles was
estimated from the descriptions of the physical environment encountered in these reports.
The GoogleEarth1 software was used to visually assess the physical environment described
in these reports and to facilitate decision-making on positioning these profiles. The error
associated with the spatial location of the soil profiles is thus very scattered and depends on
the precision with which the physical environment of each soil profile has been described.
We estimated an overall error of about fifty meters for cases where the description of the
physical environment is well documented, and an overall error of about twenty kilometers in
cases where the description of the physical environment is brief.

This positioning error will surely propagate during the process of overlaying the soil attribute
data with the gridded environmental covariates. Because in some cases, the corresponding
soil attribute data will not overlay with the pixel values of the covariates at its real position in
the field. This will have the effect of reducing the goodness of fit of the model and in the
most extreme situation, this could even have influence on the overall quality of the prediction.
However, investigations are still necessary to understand how in these conditions, the error
on the spatial location of the soil profiles will propagate and hamper the entire process. The
spatial stochastic simulation (e.g. Monte Carlo) is a good strategy for such investigation.

1
https://www.google.co.uk/earth/download/ge/index.html
18
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

3.1.2.3 Laboratory methods

As the data comes from several sources or reports, it was not surprising that the analytical
methods or procedures differ from one report to another. The soil properties that we have
mapped were subjected to the following processes

 Soil pH: pH data values were standardized (GlobalSoilMap 2015) into reference
values suggested for reporting pH for the GlobalSoilMap project (ISO 10390 2005).
The standard specifies an instrumental method for the routine determination of pH
using a glass electrode in a 1:5 (volume fraction) suspension of soil in water (pH in
H2O). We used predefined equations defined by Aitken and Moody (1991) for the
conversions of the values to fit the standarts (e.g.: y = 1.28 (x) – 0.613, with x = source
method value of pHwater (1:5 water) and y = pH (soil solution)).

 Soil Organic Carbon: It has not been possible to find an appropriate method for the
harmonization of soil organic carbon data into standard reference method (dry
combustion to at least 900°C; ISO 10694). Soil organic carbon data were mostly
analysed with the classic Walkley and Black method. The units were standardized in
g/kg.

 Particle size classes: The standard reference method adopted by the


GlobalSoilMap project for reporting particle size classes of sand, silt and clay, is the
USDA Soil Survey Laboratory Methods. Where necessary, the data of the silt class
were harmonized from other particles size classification (Table 1) by the equation
developed by Minasny and McBratney (2001).

Table 1: Differences between the common particle size classifications


Size fraction USDA FAO
Clay < 2 µm < 2 µm
Silt 2 - 50 µm 2 - 63 µm
Sand 50-2000 µm 63-2000 µm
USDA = United State Department of Agriculture; FAO = Food and Agriculture Organisation

3.1.2.4 Soil classification systems

The main soil classification systems that have received considerable international
recognition and have been widely used in Cameroon and most of the tropical regions
according to Yerima and Van Ranst (2005) are;

 The French soil classification system: The French CPCS (Commission de Pédologie
et de Classification des Sols) system (CPCS 1967), was developed based on
morphological soil characteristics, from 1963 to 1967. Under the auspice of INRA, a
new French soil classification system was developed and called “Referentiel
19
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Pedologique” (RP). The first version was published in 1987 (AFES and INRA 1987).
It is the result of a long and faithful evolution to the same morpho-genetic conceptions
of soils, with new ideas considered as well as the experience acquired since 1967 by
mapping soil in France (AFES 2008). The latest version was published in 2008 (AFES
2008).

 The U.S soil classification system (Soil Taxonomy): It was developed from 1951
onwards by the soil conservation service of the U.S. Department of Agriculture
(USDA) (Soil survey staff 1975). It is based on morphological characteristics and
related soil properties that can be objectively observed or measured (Yerima and Van
Ranst 2005). The system went throught a series of approximations and after
substancial revisions, was published in 1975. Since then, it has been subject to
regular revisions, which are published as the well-known “Key to soil taxonomy”,
currently at its twelfth edition (Soil Survey Staff 2014)

 The World Reference Base (WRB) system: The WRB system has been developed to
serve as an international legend, which aimed to be common denominator of existing
national schemes while adequately accommodating the major’s soil paterns of the
global soil cover to ensure its geographic relevance (Yerima and Van Ranst 2005).
The system was envisaged to be use as a basis to revise the legend of the soil map
of the world (FAO-UNESCO 1974). The project was initiated in 1982 and the revised
legend of the FAO/UNESCO Soil Map of the World (FAO 1988) was used as a basis
for the development of the WRB, in order to take advantage of the international soil
correlation that had already been conducted through this project and elsewhere
(WRB 2015). The first edition of the WRB was published in 1998, the second in 2006
and the third and most recent edition in 2014 (WRB 2015).

3.2 Environmental covariates

In addition to the soil point data, a large collection of raster images prepared and available
at ISRIC was integrated as environmental covariate layers to fit predictive models (total of
27 data sets). These raster images come from many sources as described in Hengl et al.
(2017) and were selected to represent the major soil forming factors and surface
characteristics unerlaying scorpan model. Furthermore, Hengl et al. (2015) indicated a
promising potential for (SoilGrids2) directly as a covariate in regional-scale mapping. For
processing the covariates, a combination of Open Source GIS software, primarily SAGA GIS
(Conrad et al. 2015), R packages raster (Hijmans et al. 2012), sp (Pebesma and Bivand
2005), GSIF and GDAL (Mitchell and Developers 2014) for reprojecting, mosaicking and

2 https://soilgrids.org/
20
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

merging tiles. These covariates were resampled at the same target resolution (250 m) prior
to model fitting.

21
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Table 2: List of applied grigged environmental covariates


N° Variables Codes Data source Original Target
scale scale
1 Average soil and sedimentary ASSDAC3 Average soil and sedimentary-deposit thickness in meters 1 km 250 m
2 Mean diurnal range [°C] at 1 km B02CHE3 Mean diurnal range [°C] at 1 km 1 km 250 m
3 Mean monthly cloud cover Feb C02MCF5 Long-term averaged monthly cloud cover Feb 1 km 250 m
4 Land surface elevation DEMENV5 DEM based on 100 m resolution from EarthEnv-DEM90 100 m 250 m
5 Entropy MODIS ENTENV3 Entropy Disorderliness of EVI 1 km 250 m
6 SD monthly MODIS EVI JanFeb ES1MOD5 Long-term s.d. of the monthly MODIS Enhanced Vegetation Index (EVI). 250 m 250 m
7 Evenness of MODIS EVI EVEENV3 Evenness of MODIS EVI 1 km 250 m
8 Mean monthly MODIS EVI EX1MOD5 Long-term averaged mean monthly Enhanced Vegetation Index (EVI) 250 m 250 m
9 Global Water Table Depth GTDHYS3 Global Water Table Depth in meters based on Fan and Miguez-Macho (2015) 1 km 250 m
10 Mean monthly MODIS NIR band 4 I01MOD4 Long-term averaged mean monthly surface reflectance (NIR) band 4 MODIS 500 m 250 m
11 Mean monthly MODIS MIR band 7 M08MOD4 Long-term averaged mean monthly surface reflectance (MIR) MODIS. 500 m 250 m
12 Mean annual cloud cover MANMCF5 Long-term averaged mean cloud cover 1 km 250 m
13 Maximum MODIS EVI MAXENV3 Maximum Dominance of EVI combinations between adjacent pixels 1 km 250 m
14 SD monthly MODIS LST N03MSD3 Long-term s.d. of the monthly surface temperature (nighttime) MODIS 1 km 250 m
15 Negative Topographic Openness NEGMRG5 Negative Topographic Openness based on DEMMRG5 250 m 250 m
16 Landsat Band 4 NIRL00 Landsat Band 4 (NIR) for year 2000 30 m 250 m
17 Mean monthly precipitation at 1 km P01CHE3 Mean monthly precipitation at 1 km (based on CHELSA Climate) for January 1 km 250 m
18 pH SoilGrids PHIHOX SoilGrids predicted soil pH 250 m 250 m
19 Range MODIS EVI RANENV3 Range of EVI 1 km 250 m
20 Sand content SoilGrids SNDPPT SoilGrids predicted sand content 250 m 250 m
21 Silt content SoilGrids SLTPPT SoilGrids predicted silt content 250 m 250 m
22 SD monthly MODIS LST T01MSD3 Long-term s.d. of the monthly surface temperature (daytime) MODIS 1 km 250 m
23 Mean monthly MODIS LST T07MOD3 Long-term averaged mean monthly surface temperature (daytime) MODIS. 1 km 250 m
24 Multiresolution Index of Valley VBFMRG5 Multiresolution Index of Valley Bottom Flatness (MRVBF) 250 m 250 m
25 Valley depth VDPMRG5 Valley depth based on DEMMRG5 i.e. vertical distance to a channel network 250 m 250 m
26 Monthly Precipitable Water Vapor VW1MOD1 Long-term averaged mean monthly MODIS Precipitable Water Vapor in cm. 10 km 250 m
27 Clay content SoilGrids CLYPPT SoilGrids predicted clay content 250 m 250 m

22
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

3.3 Spatial prediction of soil properties

3.3.1 Digital soil mapping approach

Digital Soil Mapping (DSM) can be defined as “the creation and population of spatial soil
information systems by numerical models inferring the spatial and temporal variations
of soil types and soil properties from soil observation and knowledge and from related
environmental variables” (Lagacherie and McBratney 2007).

Figure 11: Digital Soil Mapping steps for soil properties prediction (Hengl et al. 2017)

DSM thus requires digital data sources as input variables for the quantitative models.
Jenny’s (1941) well known equation identified 5 major factors in the soil formation,
namely the climate, organism, relief, parent material and time (Cl, o, r, p, t). This
approach was followed, refine and extended by McBratney et al. (2003), who identified
7 factors for soil spatial prediction and formulated the so-called SCORPAN equation:

23
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

S = f (S, C, O, R, P, A, N)

With S = Soil properties at the same location is a function of C = Climate, O = Organism,


R = Relief, P = Parent material, A = Age, time and N = Geographic position.

SCORPAN is a conceptual model of soil spatial inference (McBratney et al. 2003). In


practice, the factors of the models are obtained as images or maps that come from
different sources, different companies or technologies. A common spatial prediction
technique that can be used to apply SCORPAN model is the machine learning algorithm:
Random Forest (Hengl et al. 2015). DSM techniques can be used throughout the
framework drawn in Figure 11.

3.3.2 Soil properties splinning

In conventional soil survey, the soil profile is divided into horizons. The number of
horizons and the position of each horizon are generally based on attributes easily
observed in the field, such as morphological soil properties (Bishop et al. 1999). A bulk
sample is usually collected from these horizons and it is assumed to represent the
average value for a soil attribute over the depth interval from which it is sampled (Malone
et al. 2017). In order to apply the DSM on soil variables using legacy soil data,
standardization at a specified depth is essential.

The vertical distribution of each soil property (Sand, Silt, Clay, pH and SOC) along the
soil profiles was modelled with the mass preserving equal-area quadratic splines to
generate continuous soil data (Malone et al. 2009). From which, a weighted-average
value of these properties was derived at standard depth intervals (0–5, 5 – 15, 15 – 30,
30 – 60, 60 – 100, 100 – 200cm). A useful feature of the spline function is that it is mass
preserving, or in other words the original data is preserved and can be retrieved again
via integration of the continuous spline (Malone et al. 2009). Compared to exponential
decay functions where the goal is in defining the actual parameters of the decay function,
the spline parameters are the values of the soil attribute at the standard depths that are
specified by the user. This is a useful feature, because one can harmonize a whole
collection of data. After using the spline function to determine values at standard
intervals, it is still possible to have the value of a soil property at cumulative depth
intervals (e.g. 0 – 30 cm) using the weighted averages of resulting splined values.

3.3.3 Random forest model

An increasingly popular machine learning algorithm in DSM and soil sciences in general
is the Random Forests Model (RFM). It is a non-parametric multivariate technique,
developed as an extension of regression tree and boosted regression tree model to
respectively improve their prediction accuracy and to reduce model over fitting (Breiman
24
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

2001; Liaw and Wiener 2002). It is an assemblage of a number of classification or


regression trees using two levels of randomization for every tree in the forest (Breiman
2001), and consists of a combination of tree predictors, each grown on a bootstrapped
sub sample of the training data. Each tree is trained using a bootstrap sample of the
training data, and at each node the best split is selected from among a random subset
of the predictor variables. The process ensures that each tree utilizes the training data
and predictor variables in a different way, reducing its statistical dependence on the
other trees. The data excluded from the construction of the model are called out of bag
(OOB) and are used to evaluate the performance of that tree, and provides a way of
quantifying the “importance” of each predictor variable. These importance assessments
are quite useful in selecting predictors from a large set of candidates. The RFM has
several advantages over other prediction models. The most common are the
insensitivity to noise or weak prediction variables and to missing values or outliers (Craig
and Huettmann 2008), as it selects the most important variable at each node split (Okun
and Priisalu 2007), and the reasonable predictive performance with noisy predictive
variables (Diaz-Uriarte and de Andres 2006). Random forest is becoming more and
more applied in soil science, with demonstrated strong performance (Ließ et al. 2012;
Sreenivas et al. 2014).

3.3.4 Selection of important covariates

The relative importance of the predictor variables in modelling soil properties was
assessed using the “importance” function in the “randomForest” R package (Breiman
2001). In fact, another important metric that the Random Forest algorithm provides is
the variable importance measure, which indicates the relative importance of input
variables used to build the model. The key advantage of using RFM in this study is
because random forest “spreads” the importance of predictors in the model across all
the predictor variables (Cutler et al. 2007). RFM estimates the relative importance of the
predictor variables, based on how bad the prediction would be if the data for a particular
variable were permuted randomly (Prasad et al. 2006). This approach guards against
the elimination of good predictors variables which may be pedologically important,
although are highly correlated with each other.

3.3.5 Model validation

The regional models were calibrated with data for each region. To evaluate the
prediction performance using the models, we used the leave-one-out-cross-validation
(LOOCV) approach. From n initial number of the sample, we subseted n-1 of these data,
and fit a model. Using this model, we make a prediction for the single data that was left
out of the model (and saved the residual). This is repeated such that each observation
in the sample is used once as the validation data. LOOVC provides an effective mean

25
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

of model selection (Cawley and Talbot 2004), but involves a little more computation
(Brendan et al. 2017). The following parameters were computed for model accuracy
evaluation:

 Percentage of variance explained (R2): is a measure of the percentage of


variation explained, with Pi =predicted values and Oi = observed values.

̅ 𝒊)
∑𝒏𝒊=𝟏(𝑷𝒊 − 𝑶
𝟐
𝑹 = 𝒏
̅ 𝒊)
∑𝒊=𝟏( 𝑶𝒊 − 𝑶

 Root-mean-squared-error (RMSE): is frequently used as measure of the


differences between predicted and observed values.

∑𝒏𝒊=𝟏(𝑶𝒊 − 𝑷𝒊 )𝟐
𝟐
𝑹𝑴𝑺𝑬 = √
𝒏

Where 𝑂𝑖 is the observed soil property, 𝑃𝑖 is the predicted soil property from a
given model, and n is the number of observations i.

 The Lin’s concordance correlation coefficient (CCC), a measure of the strength


of the agreement between the observed and predicted soil properties values. It
is an index of how well a new measurement reproduces a standard measurement,
and quantifies the agreement between these two measures of the same variable
(Huiman 2012). Like a correlation, ρc ranges from -1 to 1, with perfect agreement
at 1. In terms of predicting accuracy, our results showed a good agreement
between the predicted values and the measured values of soil properties

𝟐𝝆𝝈𝒑 𝝈𝒐
𝝆𝒄 =
𝝈𝟐𝒑 + 𝝈𝟐𝒐 + (𝝁𝒑 − 𝝁𝒐 )𝟐

Where 𝝁𝒑 and 𝝁𝑶 are the means of the predicted and observed values respectively.𝝈𝟐𝑷
and 𝝈𝟐𝑶 are the corresponding variances. 𝝆 is the correlation coefficient between the
predictions and observations.

3.4 Improving the sampling network

Sampling concerns selection of a subset of individuals from within a population, to


estimate characteristics of the whole population (Wang et al. 2012). The aims of spatial
sampling methods are to get results of a higher quality at lower cost. Commonly applied
constraints are the cost constraint that do not exceed a given budget and the quality
26
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

constraint that the result meets a given minimum requirement (Cochran 1977). For
mapping or estimating spatial distribution of a soil property, the accuracy of the result
will usually be increased by dispersing the sample locations so that they cover the study
area as uniformly as possible (Walvoort et al. 2010).

In DSM, the input data are usually legacy soil data which measurements may have left
large spaces unsampled, and we would like to fill in because there, the greatest gain in
accuracy can be achieved. Several methods for optimization of the pattern of sample
locations have been described in the literature. The methods differ with respect to the
objective function, and in the way the method searches for the optimal pattern
(optimization algorithm). In model-based sampling, the objective function explicitly
defined in terms of the prediction error variance is minimized (van Groenigen et al.
1999), whereas in spatial coverage sampling, an objective function is defined in terms
of the distance between the sample locations and the nodes of a fine interpolation grid
(Royle and Nychka 1998).

In a design based sampling strategy, spreading of the sample locations can be achieved
by sampling on a randomly placed regular grid. An alternative is stratified random
sampling, using geographically compact subareas as strata. By using these compact
sub-areas as strata, spatial clustering of the sample locations can be avoided, which
usually increases the accuracy of the estimated spatial mean. In this study, we used the
stratified random sampling approach strategy to propose the spatial localization of the
new potential soil profiles to be collected in order to spatially cover the study regions
and thus increase the precision of the predictions. We performed the analysis in the
“spcosa” R package (Walvoort et al. 2010).

27
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

4. RESULTS AND DISCUSSION

4.1 Spatial distribution of soil profiles in the studied regions

Figure 12 shows the spatial distribution of legacy soil profiles in the Camsodat v01 soil
database (300 in the North and 130 in the South-West region) used in this study. The
spatial distribution of data in the North region covers a relatively large part of the region
in comparison with data from the South-West region, which mainly covers the south of
the region. Therefore, there is a problem related to the representativeness of the data
for the South-West region. For the actual validation of this work, it will be necessary to
acquire other data covering more representatively the region of the South-West in order
to have more reliable results.

In general, there are also spots in the southern part of the North region, with a low spatial
coverage of the data. Areas with little soil data density or no soil data represent where
studies were conducted at a very large scale, or areas that have not yet been explored
by soil surveys.

Figure 12: Spatial distribution of soil profiles in the North and South-West regions

4.2 Summary statistics of legacy soil database

Before carrying out modeling analyses, we presented here the extent of variation of the
soil properties throughout the database in the North region (Table 3) and South-West
regions (Table 4).

28
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Table 3: Statistics of targeted soil properties in the North Region


Min Max Mean Median SD N
SOC (g/kg) 0.0 119.2 8.2 5.2 10.66 940
Clay (%) 1.0 97.0 26.9 23.0 18.19 1107
Sand (%) 1.0 98.0 50.5 55.0 24.90 1009
Silt (%) 1.0 88.0 22.0 19.0 13.07 1009
pH 4.5 10.2 6.9 6.6 1.01 1115
Min = Minimum, Max = Maximum, SOC = Soil organic carbon, SD = Standard deviation, N = Sample size

These tables contain also the calculated standard deviations of soil properties, which
reflect the general trend of variation. Exploratory analyses of soil properties as
calculated from the database showed that SOC ranged from 0.0 to 119.2 g/kg, clay
content ranged from 01% to 97%, silt content varied from 01% to 88%, sand contents
oscillated between 01% and 98% and soil pH varies from 4.5 to 10.2 (Table 3) in the
north region.

Table 4: Statistics of targeted soil properties in the South-West Region


Min Max Mean Median SD N
SOC (g/kg) 1.0 101.0 14.2 9.3 11.7 627
Clay (%) 1.0 89.0 42.6 41.0 23.3 579
Sand (%) 0.0 98.0 39.3 35.0 27.4 564
Silt (%) 0.0 71.0 18.1 14.0 15.0 564
pH 2.6 6.8 5.1 5.0 0.6 600
Min = Minimum, Max = Maximum, SOC = Soil organic carbon, SD = Standard deviation, N = Sample size

These results illustrate a high variability of soil properties in the North region. In the
South-West region, these were from 1.0 to 101 g/kg for SOC, 1.0% to 89% for clay, from
0.0 to 71% and 98% respectively for silt and sand content, and between 2.6 and 6.8 for
pH (Table 4).

4.3 Summary statistics of soil properties after splining

In order to make prediction of soil properties at specified depth, we applied the equal
area spline function to each soil property. Then, we extracted the values of soil
properties at standard depth defined in the GlobalSoilMap project.

Table 5: Summary statistics of soil properties after splinning in the North region
Properties Depth Min Max Mean Median SD Kurt Skew N
0–5 1.78 124.27 14.39 10.50 13.35 1.39 0.34 297
5 – 15 1.50 96.46 11.22 8.28 9.51 1.41 0.38 296
SOC 15 – 30 0.09 44.83 6.38 5.23 5.04 1.17 0.23 296
(g/kg) 30 – 60 0.01 19.32 3.75 3.29 2.67 1.35 0.13 280
60 – 100 0.16 10.95 2.75 2.20 2.08 1.22 0.26 201
100 – 200 0.19 20.60 2.14 1.27 2.63 1.65 0.40 96

0–5 1.51 97.35 24.53 16.40 20.61 1.20 0.43 298


5 – 15 1.99 95.43 25.42 18.15 19.77 1.22 0.38 298
29
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Clay 15 – 30 1.75 94.53 27.68 22.72 18.95 1.00 0.28 298


(%) 30 – 60 0.95 86.66 30.07 27.29 17.93 1.33 0.09 292
60 – 100 1.58 86.83 30.22 25.78 18.46 1.32 0.24 245
100 – 200 1.33 80.74 28.03 26.72 17.67 1.26 -0.08 158

0–5 1.37 92.98 51.16 60.01 27.94 0.86 -0.42 275


5 – 15 1.48 91.19 50.81 58.95 26.58 0.97 -0.42 275
Sand 15 – 30 2.15 87.47 49.69 55.44 25.35 0.98 -0.27 275
(%) 30 – 60 1.94 97.77 48.68 52.03 24.02 1.12 -0.18 270
60 – 100 1.27 98.86 49.98 52.67 25.07 1.26 -0.15 228
100 – 200 2.06 98.36 51.11 53.66 24.98 1.28 -0.19 144

0–5 0.88 85.68 23.00 19.57 14.23 1.08 0.25 275


5 – 15 1.56 86.30 22.59 19.83 13.45 1.03 0.20 275
Silt 15 – 30 3.10 87.86 21.88 19.283 13.12 1.06 0.19 275
(%) 30 – 60 0.95 88.06 21.03 18.55 12.86 1.01 0.19 270
60 – 100 1.40 62.02 20.63 18.29 12.10 1.11 0.19 228
100 – 200 0.71 61.96 21.28 18.49 12.33 1.20 0.24 144

0–5 4.54 8.96 6.58 6.54 0.82 1.55 0.02 300


5 – 15 4.84 9.54 6.61 6.52 0.77 1.51 0.01 300
pH 15 – 30 4.61 9.74 6.69 6.54 0.84 1.35 0.13 300
30 – 60 4.42 9.52 6.90 6.71 0.99 1.45 0.12 299
60 – 100 4.57 10.17 7.22 6.99 1.23 1.09 0.17 251
100 – 200 5.21 10.15 7.36 7.28 1.09 0.93 0.13 161
SD = Standard deviation, Kurt = Kurtosis, Skew = Skewness, N = samples size

Table 5 and table 6 summarizes the statistics of each soil properties at specified soil
depth respectively in the North and the South-West regions.

Table 6: Summary statistics of soil properties after splinning in the South-West


Properties Depth Min Max Mean Median SD Kurt Skew N
0–5 2.84 104.41 27.80 23.94 17.69 1.24 0.17 137
5 – 15 3.09 90.16 21.59 17.53 14.31 1.35 0.18 137
SOC 15 – 30 3.36 70.97 13.80 11.26 10.77 1.45 0.08 137
(g/kg) 30 – 60 0.48 48.75 9.29 7.28 7.47 1.78 0.10 125
60 – 100 0.63 31.20 7.44 5.44 5.79 1.67 0.39 121
100 – 200 0.00 30.93 6.47 4.87 5.11 2.25 0.40 98

0–5 3.19 81.02 35.66 29.89 20.16 1.03 0.38 127


5 – 15 3.00 82.00 38.48 34.95 21.08 1.04 0.16 127
Clay 15 – 30 2.3 87.32 41.64 38.27 22.20 1.06 0.18 127
(%) 30 – 60 2.27 89.01 44.81 43.53 23.28 1.08 0.12 115
60 – 100 0.75 89.19 45.18 48.05 23.36 1.09 -0.21 111
100 – 200 1.22 88.65 43.45 45.17 24.23 1.16 -0.16 92

0–5 1.89 92.26 45.16 41.08 27.56 0.80 0.19 127


5 – 15 2.64 93.36 43.13 40.89 27.01 0.81 0.09 127
Sand 15 – 30 0.00 97.24 41.46 40.23 26.94 0.93 0.06 127
(%) 30 – 60 0.83 94.51 38.62 35.24 26.66 1.09 0.13 114
30
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

60 – 100 0.77 95.96 37.93 34.41 27.12 0.86 0.07 106


100 – 200 1.39 96.93 37.77 35.71 27.59 0.78 -0.03 87

0–5 0.99 59.61 19.19 14.22 14.34 1.33 0.43 127


5 – 15 0.14 60.80 18.42 15.81 13.68 1.24 0.12 127
Silt 15 – 30 0.02 63.08 17.24 4.92 13.33 1.64 -0.02 127
(%) 30 – 60 0.94 64.53 16.47 13.18 13.10 1.17 0.25 115
60 – 100 0.12 70.82 16.51 13.18 14.27 1.24 0.16 111
100 – 200 0.00 73.18 17.70 14.08 15.29 1.29 0.19 92

0–5 3.70 7.10 4.98 4.87 0.78 1.09 0.12 132


5 – 15 3.70 6.59 4.97 4.92 0.68 1.16 -0.03 132
pH 15 – 30 3.72 6.72 4.01 4.97 0.66 1.10 -0.05 132
30 – 60 4.07 6.58 5.09 5.06 0.60 1.10 0.01 119
60 – 100 3.86 6.89 5.08 5.05 0.61 1.21 -0.01 110
100 – 200 3.34 6.56 5.06 5.03 0.58 0.99 0.12 88
SD = Standard deviation, Kurt = Kurtosis, Skew = Skewness, N = samples size, Depth in cm

In the North region, the frequency distributions of the particle size fraction data are
typical, given that clay and silt are positively skewed (Table 5), whereas sand is skewed
slightly negative as reported elsewhere (Adhikari et al. 2013; Akpa et al. 2014). In the
South-West region, we did not have the same trend and the distribution of particles size
fraction becomes negatively skewed with soil depth (Table 6). The clay content
increases from the top 30 cm depth with a peak at the 60-100 cm, likely caused by clay
illuviation as reported in other studies (Ayuba et al. 2007; Sharu et al. 2013).

Soil pH showed very different patterns of variation when moving from the South-West
region to the North region. Indeed, the soils are mostly acidic in our two regions, and
this acidity is more accentuated in the region of South-West, with pH varying from 3.3
to 7.1 (Table 6). In fact, soil pH decreases over time in a process called soil acidification,
due to leaching from rainfall (Slessarev et al. 2016). Sandy soils commonly have low
organic matter content, resulting in a low buffering capacity, high rates of water
percolation and infiltration making them more vulnerable to acidification. Irrespective of
the region, there is a tendency for the pH to increase with the depth of the soil. The
standard deviation of soil pH gradually decreases with soil depth in the North region,
varying from 0.8 in the topsoil (0 – 5 cm) to 0.6 in the subsoil (200 cm), while an increase
in standard deviation of soil pH was observed with the soil depth in the north region, with
variation from 0.8 in the topsoil (0 – 5 cm) to 1.2 in the subsoil (200 cm). Such variation
of soil pH is probably due to the distribution of rainfall within these two regions as
reported by Slessarev et al. (2016).

4.4 Prediction performance

The model performance parameters (RMSE, R2, and ρc) were used to assess the quality
of prediction of soil properties in the North region (Table 7) and the South-West region
31
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

(Table 8). Results showed that the combination of the various covariates was able to
explain substantial percentatge of variance explained by the prediction models of soil
properties in these two regions.

The respectives models performed significantly better at the top 30 cm (0-5, 5-15 and
15-30 cm) compared to the lower layers (30-60, 60-100 and 10-200 cm). This could be
attributed to the nature of the environmental variables used and the effect of lower data
density with depth. Most environmental covariates used in this study are based on land
surface characteristics and are likely to have stronger relationship with topsoil than
subsoil properties. Similar results have been reported by several others (Minasny et al.
2006; Malone et al. 2009; Kempen et al. 2011). As reported by Akpa et al. (2014), the
inclusion of soil depth as a predictor variable significantly increases the accuracy output
of the prediction model. The integration of corresponding SoilGrids soil properties
(PHIHOX, SNDPPT, SLTPPT, CLYPPT) for each soil parameter positively enhanced
the prediction of soil properties in the South West region, as suggested by Hengl et al.
(2016), excluding that of SOC.

Table 7: Prediction performances of soil properties in the North region


Properties Depth RMSE R2 ρc
0–5 8.87 0.73 0.64
5 – 15 5.19 0.73 0.82
SOC 15 – 30 4.04 0.37 0.48
30 – 60 3.01 0.25 0.48
60 – 100 1.88 0.35 0.54
100 – 200 2.32 0.34 0.49

0–5 9.37 0.80 0.87


5 – 15 7.15 0.88 0.92
Clay 15 – 30 7.97 0.83 0.90
30 – 60 9.61 0.71 0.83
60 – 100 11.64 0.61 0.77
100 – 200 13.99 0.41 0.63

0–5 12.22 0.81 0.82


5 – 15 8.99 0.89 0.93
Sand 15 – 30 9.37 0.86 0.92
30 – 60 10.03 0.82 0.90
60 – 100 14.47 0.67 0.80
100 – 200 18.09 0.49 0.68

0–5 6.52 0.79 0.87


5 – 15 3.69 0.93 0.95
Silt 15 – 30 4.27 0.89 0.93
30 – 60 5.33 0.82 0.90

32
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

60 – 100 6.86 0.69 0.81


100 – 200 8.06 0.59 0.74

0–5 0.53 0.59 0.75


5 – 15 0.28 0.87 0.92
pH 15 – 30 0.46 0.72 0.82
30 – 60 0.62 0.63 0.72
60 – 100 0.73 0.62 0.68
100 – 200 0.74 0.59 0.67
RMSE = Root Mean Square Error; pc = Linc’s concordance coefficient

However, it is important to mention the low agreement of the SOC values in the North
region (Table 7), which is certainly due to the low ability of the auxiliary variables to
capturing and modeling its distribution and its dynamic. To increase the quality of the
result in this case, we can consider the inclusion of covariates such as Gamma-
radiometric or electromagnetic induction (Rawlins et al. 2009; Priori et al. 2014) as
suggested by Akpa et al. (2014), which are unfortunately more cost effective.

33
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Table 8: Prediction performances of soil properties in the South-West region


Properties Depth RMSE R2 ρc
0–5 10.10 0.79 0.75
5 – 15 6.12 0.85 0.88
SOC 15 – 30 5.76 0.75 0.80
30 – 60 3.65 0.81 0.89
60 – 100 3.92 0.73 0.81
100 – 200 4.27 0.81 0.77

0–5 9.78 0.81 0.88


5 – 15 6.48 0.91 0.94
Clay 15 – 30 6.31 0.92 0.95
30 – 60 6.97 0.92 0.94
60 – 100 9.07 0.85 0.91
100 – 200 12.07 0.75 0.85

0–5 12.10 0.81 0.83


5 – 15 6.80 0.94 0.96
Sand 15 – 30 7.04 0.93 0.95
30 – 60 8.31 0.90 0.94
60 – 100 10.94 0.83 0.90
100 – 200 13.31 0.77 0.86

0–5 6.57 0.77 0.85


5 – 15 4.08 0.90 0.94
Silt 15 – 30 4.01 0.89 0.93
30 – 60 4.91 0.84 0.90
60 – 100 6.72 0.73 0.84
100 – 200 6.88 0.74 0.84

0–5 0.47 0.63 0.76


5 – 15 0.24 0.88 0.92
pH 15 – 30 0.24 0.86 0.92
30 – 60 0.25 0.82 0.90
60 – 100 0.27 0.79 0.88
100 – 200 0.30 0.74 0.85
RMSE = Root Mean Square Error; pc = Linc’s concordance coefficient

The sand content had the highest RMSE values across all depths and regions, whereas
the lowest RMSE was associated with the prediction of soil pH at all depth intervals.
This trend corroborates the reports of Niang et al. (2013) on soil texture, using similar
modelling approach.

34
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

4.5 Variables of importance

Out of all the covariates considered in this study (27 in total for each soil parameter), the
10 most important environmental variables in modelling each soil properties are
appended as Figure 13 for the North region and Figure 14 for the South-West. In the
North region, the environmental covariates showed a varying level of importance in the
corresponding models (Figure 13). There was a large contribution of soil sampling
depth, terrain attribute (DEMENV5), climatic element (Long term average monthly cloud
cover), and land cover (Mean monthly MODIS Enhanced vegetation index) in predicting
soil properties in the North region. In the South-West region, among the main predictors
(Figure 14), we have the SoilGrids estimates (Sand, Silt, Clay and pH), the soil depth,
climatic element (Mean monthly precipitation), the land cover (MODIS Enhanced
vegetation index, Surface reflectance and MODIS NIR band 4), the terrain attribute
(Valley depth, land surface elevation). However, the relative importance of these
variables varies with depth and from one soil property to another.

Sand OrgC
OrgC
SOC OrgC
pHOr
Sand
Sand Silt OrgC
SOCSOC OrgC
pHpH
SiltSilt

%% increase
increase MSE
inin MSE
Figure 13: Relative importance of covariates in the North region (Depth = Soil depth;
DEMENV5 = Digital elevation Model; CxxMCF5 = Average Monthly cloud cover; T04MSD3
= Mean Monthly surface temperature; ESxMOD5 = Enhanced vegetation Index (EVI);
VDPMRG5 = Valley depth; ENTENV3 = entropy disorder line of EVI; ASSDAC3 = Soil
sedimentary deposits; PxxCHE3 = Monthly precipitations)

Other studies have also reported the relationship between terrain attributes and soil
properties (Moore et al. 1993; Thompson et al. 2006; Ließ et al. 2012; Akpa et al. 2014),
especially with terrain attributes explaining between 20% and 88% of the variation in soil

35
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

properties (Thompson et al. 2006). This could be attributed to their impact on vertical
and lateral movement of soil particles through erosion and disposition. Akpa et al. (2014)
showed that the inclusion of soil sampling depth significantly improves the performance
of random forest model by 67-100% in predicting soil properties. Other studies also
reported the usefulness of land cover and climatic element in predicting soil properties
(Sreenivas et al. 2016; Adhikari et al. 2014; Ließ et al. 2016). Soil parameters as
predictors in predicting soil properties have been proven to be valuable (Law-Ogbomo
and Nwachokor 2010). However, this study shows that including SoilGrids as covariates
only shows high importance for some soil properties in the the South-West region and
not for the North region.

Silt OrgC pH
SOC

% Increase in MSE % Increase in MSE


Figure 14: Relative importance of covariates in the South-West region (Depth = Soil depth;
SLTPPT = SoilGrid silt content; NEGMRG5 = Topographic openness; BxxCHE3 = Mean
diurnal range; VWXMOD1 = Mean monthly precipitable water vapor; IxxMOD4 = Mean
monthly surface reflectance; EVEENV3 = Evenness of MODIS; PHIHOX = SoilGrid pH;
DEMENV5 = Digital elevation Model; CxxMCF5 = Average Monthly cloud cover; ESxMOD5
= Enhanced vegetation Index (EVI); VDPMRG5 = Valley depth; ENTENV3 = entropy disorder
line of EVI; PxxCHE3 = Monthly precipitations)

4.6 Spatial prediction of soil properties

4.6.1 Spatial prediction assessment in the North region

In this study, about 300 soil point observations from the North region were collected and
used with environmental covariates for spatial prediction. Analysis showed that the

36
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

predicted values of soil properties had wide ranges, which is an indication of high
diversity in landscape and soil types in the considered areas.

Since soil data generally used for decision making, especially agronomically, are
sampled within the top 0 - 30 cm of soil, we have determined the characteristics of the
different soil properties studied in this respect. At 30 cm depth, these were 2.4 – 61.1
g/kg for SOC, 5.2 – 9.2 for soil pH, 3.7 – 81.8 % for clay content, 4.5 – 87.4 % for sand
content and 4.0 – 84.4 % for silt content (Table 9). The minimum standard deviation
(SD) was 0.3 for pH in the top 30 cm of soil, and the maximum SD was 9.4% for sand
content which was far more than other elements. The SD of soil properties were in order
of Sand>Clay>Silt> SOC> soil pH.

Table 9: Statistical feature values of predicted soil properties (30 cm) in the North
Properties Min Max Mean Median SD Kurt Skew
SOC (g/kg) 2,35 61,08 10,80 9,71 4,01 1,17 0,25
pH 5,15 9,16 6,60 6,55 0,28 1,37 0,12
Clay (%) 3,66 81,79 20,64 19,07 6,20 1,44 0,20
Sand (%) 4,45 87,42 56,33 58,71 9,37 1,52 -0,23
Silt (%) 3,98 84,44 22,51 21,41 5,58 1,16 0,28
Min = Minimum; Max = Maximum; SD = Standard deviation; Kurt = Kurtosis; Skew = Skewness; SOC = Soil organic Carbon.

The mean predicted value of SOC concentration at 30 cm depth was 10. 1 g/kg, with a
standard deviation of 4.0 g/kg and the median value close to 10 g/kg (Table 9). There
was a gradual decrease of SOC with soil depth, varying between 14.6 to 4.0 g/kg from
the first layer (0 – 5 cm) up to 200 cm depth. The spatial distribution of SOC revealed a
positive kurtosis and skewness on all layers, with values close to those of a normal
distribution (Annex A1).

At 30 cm depth, the mean predicted value of soil pH was 6.6, with a standard deviation
of 0.3 and the median value of 6.6 (Table 9). There is a gradual increase of soil pH with
soil depth with mean value varying between 6.6 – 7.1. The distribution of soil pH
revealed a positive kurtosis and skewness in all layers, with values close to those of a
normal distribution (Annex A1). The values of the mean and median also indicated a
normal distribution for predicted soil pH values.

Still at 30 cm soil depth, the mean predicted value of soil clay, sand and silt content were
respectively 19.1, 58.7, and 21.4 % (Table 9). This result implies dominance of the sand
fraction in the landscape of the North region of Cameroon, and the tendency toward the
accumulation of clay with soil depth (between 60 and 100 cm). The general trend of
sand and silt shows a gradual decline with depth, while that of clay illustrates a horizon
of clay accumulation (Annex A1).

37
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

4.6.2 Spatial prediction assessment in the South-West region

In the South-West region, 130 soil point observations were used with environmental
covariates for spatial prediction of soil properties. Analysis showed that the predicted
values of soil properties had wide ranges. At 30 cm depth, these were 5.2 – 70.1 g/kg
for SOC, 3.9 – 6.6 for soil pH, 3.2 – 81.1 % for clay content, 2.7 – 94.2 % for sand
content and 1.0 – 64.1 % for silt content (Table 10). The minimum standard deviation
(SD) was 0.3 for pH in surface soil (30 cm depth), the maximum SD was 8.9 % for SOC
which was together with sand SD (8.2) far more than other elements. The SD of soil
properties were in order of SOC > Sand > Clay > Silt > soil pH.

Table 10: Statistical feature of predicted soil properties (30 cm) in the South-West
Properties Min Max Mean Median SD Kurt Skew
SOC (g/kg) 5,19 70,11 23,66 21,75 8,91 1,30 0,10
Clay (%) 3,24 81,07 34,83 34,51 6,22 1,28 -0,03
Sand (%) 2,74 94,19 50,03 50,64 8,26 1,20 -0,05
Silt (%) 0,96 64,13 16,27 14,18 5,81 1,28 0,43
pH 3,89 6,64 5,15 5,17 0,25 1,19 -0,10
Min = Minimum; Max = Maximum; SD = Standard deviation; Kurt = Kurtosis; Skew = Skewness; SOC = Soil organic Carbon.

The SOC and silt were positively skewed, whereas percentage of sand and clay content,
together with soil pH had a negative skewness (Table 10). The mean predicted value of
SOC concentration at 30 cm depth was 23.7 g/kg, with standard deviation of 8.9 g/kg
and the median value close to 21.8 g/kg (Table 10). The trend of gradual decrease of
SOC with soil depth was observed, with mean value varying between 30.8 and 12.6 g/kg
from the first layer (0 – 5 cm) up to 200 cm depth (Annex A2).

At 30 cm depth, the mean predicted value of soil pH was identical to median (5.2), with
a standard deviation of 0.3. The mean predicted value of soil clay, sand and silt content
were respectively 34.8, 50.0, and 16.3 % (Table 10). Statistical feature values of the
variation of predicted soil properties with depth are summarized in Annex A2.

4.7 Digital maps of soil properties

4.7.1 Soil organic carbon in the North and South-West regions

The maps of soil organic carbon (SOC) distribution in the top 30 cm is given in Figure
15 for the North region and Figure 16 for the South-West region. Irrespective of the
region, the spatial pattern of SOC depends only on few environmental covariates, the
most important being the soil depth and the land surface elevation (see section 4.5). In
the North region, the SOC concentration is low as reported by Combeau (1955). There
is a south to north gradient decrease of SOC (Figure 15), probably because the volcanic
chain of Cameroon ends south of the North region.
38
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 15: Distribution of SOC (g/kg) in North region of Cameroon (0 - 30 cm)

The main features of this distribution of SOC have been reported in other studies. Curtis
and Martin (1957) reported low SOC content (~ 10 g/kg) in the area around Lam, (north-
east of Guider), with a spatial variation strongly influenced by the type of vegetation and
anthropogenic activities. The values reported by Humbel (1965), near Touboro (East of
the region) are also close to the values predicted in this study. The general spatial
predicted trend of SOC is also consistent with that described by Martin (1962), with
samples covering the whole region.

There is a gradual decrease in SOC with soil depth (Kips et al. 1986; Humbel 1965; FAO
1977) as shown in Annex B1 for the North region and Annex C1 for the South-West
region, also as reported in several others studies (Mulder et al. 2016; Adhikari et al.
2014). The spatial distribution of SOC is also strongly influenced by the changes in
altitudes at the soil surfaces in the studied regions, which decreases from high to low
altitude. The influence of altitude is as expected, amplified in the South-West region,
where high altitudes belong to the volcanic mountain chain of Cameroon. Our model
predicted high concentration of SOC in the coastal area south of the region (around
Mount Cameroon) as repoted in FAO (1977) near Debundscha, and Bakingili.

39
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 16: Distribution of SOC (g/kg) in South-west region of Cameroon (0 – 30 cm)

Also, along the ridge bordering the north-eastern of the region (Mount Manengoumba
and Bamboutos Mountains) as observed in Tombel (FAO 1977). However, Kips et al.
(1986) reported low organic carbon in Tiko plain area, which is corroborated in this study.

4.7.2 Soil pH in the North and South-West regions

Based on the classification of Soil Survey Division Staff (1993), the pH distribution in the
North region varies from strongly acid to moderately alkaline (Figure 17) when moving
northward, with mean values varying from 5.2 to 9.2. Annex B5 shows a slight decrease
of soil pH in the first layers, followed by an increase in subsoil as reported by Curtis and
Martin (1956). This observation confirms the soil pH dependence at the sampling depth
as illustrated by the variables of importance (see section 4.5). As described by Humbel
(1968), the soil pH is slightly acid in the area around Poli, and usually oscillates between
6 and 7. There is however an overestimated soil pH in the area around Sanguere, south

40
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

of Garoua (Brabant and Fardin 1979). The figure in Annex B5 shows the vertical
distribution of soil pH all over the North region.

Figure 17: Distribution of soil pH in the North region of Cameroon (0 - 30 cm)

In the South-West region, soil pH varies from very strongly acid to slightly alkaline
(Figure 18). These low pH values follow the high rainfall pattern in this region. Annex C5
shows a slight increase of soil pH with soil depth in the region. Tiko plain area is strongly
acid with pH between 4.5 and 5.0 as shown by Kips et al. (1986), and also the area
around Ekondo Titi, Mundemba and Kumba. Predicted values in this study are close to
the values measured by Awah (1985) around Mbonge Area.

41
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 18: Distribution of soil pH in the South-West region of Cameroon (0 - 30 cm)

4.7.3 Clay content in the North and South-West regions

Figure 19 shows the maps of predicted soil clay in the North region. There are patches
of high to medium clay content around lakes and rivers due to soil erosion in upper
catchment and then, followed by deposit of fine materials during flooding (Amusan et al.
2005), as also described by Olowolafe (2002) from two separate catchments in Nigeria.
There is an increase in the clay content with depth, and the magnitude of this vertical
increase in clay content differs from one location to another (Annex B2). At some
locations this is steady and gradual while it is abrupt in others, giving rise to a bulge of
clay with depth. The gradual increase of clay content with depth has also been reported
42
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

in some area in the Northern Cameroon (Curtis and Martin 1957; Humbel 1968). This
pattern is the result of vertical clay movement (eluviation/illuviation), faunal perturbation
and movement of clay particles (Yerima and Van Ranst 2005; Sharu et al. 2013).

Figure 19: Clay (%) distribution in the North region of Cameroon (0 - 30 cm)

The spatial distribution of clay in the South-West region of Cameroon is described in the
Figure 20. As reported by FAO (1977), we predicted in this study a high content of clay
in the soils around Kumba. The model also showed high quantity of clay in the east side
of the Mount Cameroon (Buea). There is a slight non-significant variation of clay with
depth in the South-West region. The clay content remains steady in the north of the
region after Kumba. The values predicted here are close to those found by Awah (1985)
in the area of Mbonge.

43
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 20: Clay (%) distribution in the South-West region of Cameroon (0 - 30 cm)

4.7.4 Sand content in the North and South-West regions

The sand content at 30 cm depth (Figure 21) of soils in the North region is relatively high
compared to clay and silt across the entire region. This can be attributed to aeolian
deposition (Yerima and Ranst 2005). There is evidence of soils with high to medium
sand content (Figure 21) which is probably caused by deposition of sand from the
Sahara Desert as reported in Nigeria (McTainsh 1984). The sand content decreases
with soil depth in the north region (Annex B3). The soils are moderately sandy in the
south west part of the region.

44
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 21: Sand (%) distribution in the Norh region of Cameroon (0 - 30 cm)

In the South-West region, the sand content (Figure 22) varies gently from the West to
the East of the region, with high values all along the western border with Nigeria. As
also reported by Akpa et al. (2014). The sand content decreases from the surface soil
to the deep soil layers, as shown in Annex C3. Kips et al. (1985) reported medium sand
content in the area around Tiko as predicted here. The sand content is also very low in
the area around Kumba.

45
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 22: Sand (%) distribution in the South-West region of Cameroon (0 - 30 cm)

4.7.5 Silt content in the North and South-West regions

The silt content of the soils in the North region is relatively low (Figure 23) as has been
reported previously (Martin 1962; Laplante 1961). However, soils with medium silt
content occur from Risso to Touboro axis and areas between Garoua and Figuil, as
reported in previous studies (Brabant 1970; Humbel 1965). Annex B4 shows how silt
content decreases with soil depth in the north region. There are some spots of high silt
content in the area north-west of Guider, probably because of the effect of valley depth
in the prediction of silt content as reported by Ogban and Babalola (2003).

46
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 23: Silt (%) distribution in the North region of Cameroon (0 - 30 cm)

In the South-West region, the silt content decreases from the west to the east of the
region (Figure 24), high propancy of high silt content around the hilly areas (Mount
Cameroon, Mount Manengouba, Mount Bamboutos, and Rumpi hills). This is surely
captured by the negative topographic openness identifed as important covariate (see
section 4.5) in predicting silt content in this region. With soil depth, the silt distribution
follows the same pattern, with a slight decrease of values in the deepest horizons (Annex
C4). As reported by Kips et al. (1986), this study predicted low silt content in the Tiko
plain area. The silt content is also low in the area around Kumba, as reported in FAO
(1977).

47
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Figure 24: Silt (%) distribution in the South-West region of Cameroon (0 - 30 cm)

4.8 Sampling network design

The key elements to consider when developing a sampling strategy are budget and level
of accuracy. Although these two parameters are distinct, one often very significantly
influences the other. Without having a priori information on these two parameters, we
chosen to increase one hundred profiles to those existing in each region to improve the
quality of our prediction. This number has been chosen for the needs of the exercise
and can be adjusted according to the budgetary constraints.

48
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

In the North region, we calibrated the spatial coverage model to contain a maximum of
400 soil profiles, spatially covering the entire surface of the region. With the 300 existing
soil profiles, the model proposes 100 new profiles locations to be collected. The process
consists in dividing randomly the entire surface of the region into 400 uneven surface
geographical strata, a priori containing the soil profiles already collected (Walvoort et al.
2010). Figure 25 illustrates the spatial location of the new profiles to be collected and
those existing already.

• New points
∆ Prior points

Figure 25: Sampling netwok in the North region

In the South-West region (Figure 26), the model was calibrated to contain a maximum
of 250 soil profiles, with 120 new point locations to visit.

49
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

• New points
∆ Prior points

Figure 26: Sampling network in the South-West region

50
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

5. CONCLUSION

This work resulted in a quantitative and spatially explicit framework for assessing soil
properties distribution in the North and South-West regions of Cameroon at a spatial
resolution of 250 m. This can be relatively easily updated as more and higher quality
geo-referenced soil data become available. This functional soil information can be used
as input for land-use planning in these regions. Input soil data (soil data base), soil
property maps and derived functional soil information, as well as the parameterisation
of the rules and thresholds to model soil properties distribution, can easily be updated.
The framework allows the processing of soil data of fragmented and heterogeneous
nature, compiled from various sources and from various areas, into complete and
consistent soil information (digital soil maps) which is applicable throughout studied
regions in a coherent manner.

Soil properties are mapped with an accuracy assessed from cross-validation (LOOCV)
and results appear promising. Based on the accuracy assessment, it is concluded that
the accuracy of some soil properties needs to be further improved. More input data, and
a better quality of their distribution and spatial location, better quality and identification
of covariates (resolution and information content) and better prediction modelling
techniques are needed.

The smoothening effect of machine learning (Random Forest) approach is an issue


deserving attention for coming updates of these maps of soil properties. We have to
handle the overestimated low values and underestimated high values of soil properties
as reported in this study. More attention must be paid to search and find covariates that
are likely relevant for predicting certain soil properties, which can also explain the
variation in soil properties with soil depth, based on soil scientific knowledge.

We estimate the spatial distribution of SOC, soil pH, and particle size distribution (sand,
silt and clay content) in the North and South-West regions of Cameroon, with varying
levels of prediction accuracy and varying importance of covariates for each soil property.
Although predictions were generally acceptable, some soil properties revealed more
accurate models than others. While the representativeness of the data is acceptable for
the North region, the same is not true for the South-West region, where the data are
mainly concentrated in the south of the region. This can hinder the robustness of the
model in this region when the whole region is considered. Additional sampling would be
necessary for the validation of these results over the entire region.

Efforts to collect and compile additional soil profile data, either from existing data
sources or new in the field, in support to updating the current estimate of these soil
properties is essential for improving the quality and usability of these products. It would

51
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

also be important to consider a similar study for mapping these soil properties to 100 m
resolution. The present resolution of 250 m does not permit precise use of the maps for
local application because these maps produce predictions of soil properties over an
average area of 6.3 ha. However, this study is a proof of concept that digital soil mapping
is an interesting tool that can enable the efficient and rapid production of maps for
different soil properties. It would therefore be more interesting to make predictions on
the average acreage of one ha. Within PRESS NO&SW project, old geological maps
(1:500 00) and soil maps (1:200 000) are being digitized. These data can be integrated
to improve the prediction at the same time produce even more reliable and accurate
estimates.

Concerning soil input data, an accurate evidence-based final product at high resolution
is the most cost efficiently and rapidly produced on the basis of using a combination of
legacy soil data and new soil data. Where the legacy soil data prove cost-efficient input
for accurate mapping at specially reduced resolution, the accurately georeferenced,
analyzed by modern laboratory methods and clustered new soil data are expensive but
necessary as additional input to achieve an accurate result at high resolution.

Acknowledgements

We wish to thank the organisations and individuals who contributed to the inventory and
collection of soil data sources in any format. These primary soil datasets are key to
producing the soil maps underpinning this study.

Particular thanks go to the BGR (Bundesanstalt für Geowissenschaften und Rohstoffe)


through the project PRESS NO & SW, which agreed on the importance of exploring
digital soil mapping strategies for mapping soil properties and provided targeted means
for training at ISRIC Wageningen. We appreciate the whole team project.

Furthermore, we express our gratitute to the ISRIC team (Johan Leenaars, Gerard
Heuvelink, Tom Hengl, Maria Ruiperez Gonzalez and Marocs Angelini) who contributed
and assisted in several major issues. For example, the training required for the analysis
of the data and also to the availability of the covariates ready for use at 250 m resolution
for these analyzes. Finally, we want to appreciate Miss Vera Sham for her time spent in
editing this document.

52
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

6. REFERENCES
Adhikari K., Hartemink A. E., Minasny B., Kheir R. B., Greve M. B., Greve M. H. 2014. Digital
Mapping of Soil Organic Carbon Contents and Stocks in Denmark. PLOS ONE, Volume 9
Adhikari K., Kheir R. B., Greve M. B., Bøcher P. K., Malone B. P., Minasny B.,
McBratney A.B. & Greve M.H. 2013. High-Resolution 3-D mapping of soil texture
in Denmark. Soil Science Society of America Journal, 77(3), 860-876.
AFES. 2008. Referentiel pedologique. Edition QUAE
AFES, INRA. 1987. Referentiel Pedologique Francais. 1ère Proposition.
Aitken R. L., Moody P. W. 1991. Interrelations between soil pH measurements in various
electrolytes and soil solution pH in acidic soils. Australian Journal of Soil Research 29(4).
Akpa S. I. C., Odeh I. O. A., Bishop T. F. A., Hartemink A. E. 2014. Digital Mapping of Soil
Particle Size Fractions for Nigeria. Soil Science Society of America Journal
Amusan A. A., Olayinka A., Oyedele D. J. 2005. Genesis, classification, andmanagement
requirements of soils formed in windblown material in the GuineaSavanna area of Nigeria.
Communications in soil science and plant analysis, 36(15 -16), 2015-2031.
Awah E. T. 1985. Semi – detailed soil survey of Mbonge ubber Estate of the Cameroon
development corporation. FAO Soil Ressouces Project Technical report. Cameroon.
Ayuba S. A., Akamigbo F. O. R. Itsegha, S. A. 2007. Properties of soils in River KatsinaAla
catchments area, Benue State, Nigeria. Nigerian Journal of Soil Science
17(1):24-29
Bader J., Hayward C., Razo J., Madnick S., Siegel M. 1999. An Analysis of Data Standardization
across a Capital Markets/Financial Services Firm. Massachusetts Institute of Technology
Sloan School of Management Cambridge. MA 02139. Working paper.
Batjes N. H., Ribeiro E., Oostrum A. V., Leenaars J., Hengl T., Mendes de JesusJ. 2017. WoSIS:
providing standardised soil profile data for the world. Earth Syst. Sci. Data, 9, 1–14, 2017.
Baritz R., Erdogan H., Fujii K., Takata Y., Nocita M., Bussian B., Batjes N. H., Hempel J., Wilson
P., Vargas R. 2014. Harmonization of methods, measurements and indicators for the
sustainable management and protection of soil resources (Providing mechanisms for the
collation, analysis and exchange of con-sistent and comparable global soil data and
information), Global Soil Partnership, FAO. 44 pp.
Bishop T. F. A., McBratney A. B., Laslett G. M. 1999. Modelling soil attribute depth functions
with equal-area quadratic smoothing splines. Geoderma 91:27–45
Bockheim J. G., Hartemink A. E. 2013. Distribution and classification of soils with taxonomically
defined clay-enriched horizons in the USA: A review. Geoderma 209–210 :153–160.
Brabant P. 1970. Reconnaissance pédologique du bassin versant du Risso à Ndok (Nord
Cameroun). ORSTOM. Centre ORSTOM de yaoundé. P. 179.
Breiman L. 2001. Random forests. Machine Learning, 45(1), 5–32
Business in Cameroon. 2013. South west region of Cameroon. N°2 March 2013
Cawley G. C., Talbot N. L. C. 2004. Fast leave-one-out cross-validation of sparse least-squares
support vector machines. Neural Networks, 17(10):1467–1475.
Clemens D. F., Whitehurst B. M., Whitehurst G. B. 1990. Chelates in agriculture. Fertilizer
Research. 25:127-131.
Conrad O., Bechtel B., Bock M., Dietrich H., Fischer E., Gerlitz L., et al. 2015. System for
Automated Geoscientific Analyses (SAGA) v. 2.1.4. Geoscientific Model Development.
8(7):1991–2007.
CPCS. 1967. Classification des sols. Ecole Nationale Superieure agrionomique, Grignon, 87P

53
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Cratchley C. R., Louis P., Ajakaiye D. E. 1984. Geophysical and geological evidence for the
Benue-Chad Basin Cretaceous rift valley system and its tectonic implications. Journal of
African Earth Sciences. Vol. 2, No.2. 141-150.
Craig E., Huettmann F. 2009. Using “blackbox” algorithms such as TreeNET and Random
Forests for data-mining and for finding meaningful patterns, relationships and outliers in
complex ecological data: An overview and example using G. In: Wang H (ed.) Intelligent
data analysis: developing new methodologies through pattern discovery and recovery.
Information Science Reference, Hershey, pp 65–84.
Cochran W. G. 1977. Sampling Techniques, 3rd ed. Wiley, New York
Combeau A. 1955. Les sols du reboisement de Garoua. ORSTOM/IRCAM
Cutler D. R., Edwards Jr., Beard T. C., Cutler K.H., Hess A., Gibson K.T., LawlerJ. J. 2007.
Random forests for classification in ecology. Ecology, 88(11), 2783-2792.
Curtis M., Martin D. 1957. Carte pédologique du canton de Lam, (subdivision de guider).
ORSTOM/IRCAM.
Diaz-Uriarte, R. and de Andres, S.A. 2006. Gene selection and classification of microarray data
using random forest. BMC Bioinformatics 7, 3.
Dobos E., Carré F., Hengl T., Reuter H. I., Tóth G. 2006. Digital Soil Mapping as a support to
production of functional maps. EUR 22123 EN. European Comission Joint Research
Centre. Luxemburg.
ESA. 2016. CCI LAND COVER - S2 prototype Land Cover 20m map of Africa 2016. ESA Climate
Change Initiative - Land Cover project 2017
FAO. 1977a. Soil survey and Land avaluations for the second developement programme of the
Cameroon development corporation (CDC). Summary report, Annexes. Technical report
N°7 FAO/UNDP/ Soil rsoource project.
FAO. 1977b. Soil survey and Land avaluations for the MINAGRI Coffee developement project
Ikiliwindi – South West province. Technical report N°8 FAO/UNDP/ Soil ressource project.
FAO-UNESCO. 1974. Soil map of the world 1: 5 000 000. Vol.1. Legend. UNESCO. Paris,
France, 59p.
FAO-Unesco-ISRIC. 1990. Soil map of the world. Revised legend. Soil bulletin 60. FAO, Rome,
119p.
GlobalSoilMap. 2015. Specifications Tiered 1GlobalSoilMap products. Release 2.4 Science
Committee.
Haling R. E., Simpson R. J., Culvenor R. A., Lambers H., Richardson A. E. 2011. Effect of soil
acidity, soil strength and macropores on root growth and morphology of perennial grass
species differing in acid-soil resistance. Plant, Cell and Environment (2011) 34, 444–456
Hengl T., Mendes de Jesus J., Heuvelink G. B. M., Ruiperez Gonzalez M., Kilibarda M., Blagotić
A., Shangguan W., Wright M. N., Geng X., Bauer-Marschallinger B., Guevara M. A.,
Vargas R., MacMillan R. A., Batjes N. H., Leenaars J. G. B., Ribeiro E., Wheeler I., Mantel
S., Kempen B. (2017). SoilGrids 250m: Global gridded soil information based on machine
learning. PLoSONE 12(2): e0169748.
Hijmans R. J., van Etten J. 2012. raster Geographic data analysis and modeling. Available from:
http:// CRAN.R-project.org/package=raster.

54
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Hombegowda H. C., van Straaten O., Köhler M., Hölscher D. 2016. On the rebound: soil organic
carbon stocks can bounce back to near forest levels when agroforests replace agriculture
in southern India. SOIL, 2, 13–23.
Huiman X. B., Haber M., Song J. 2002. Overall Concordance Correlation Coefficient for
Evaluating Agreement Among Multiple Observers. Biometrics 58, 1020-1027.
Humbel F. X. 1965. Etude pedologique du bassin versant du bome pres de touboro (benoue).
ORSTOM
Humbel F. X. 1968. Contribution à l'étude des sols à horizon caillouteux du Nord-Cameroun.
ORSTOM.
Jenny H. 1941. Factors of soil formation - a system of quantitative pedology. McGraw-Hill, New
York, 281 pp.
Jones A., Breuning-Madsen H., Brossard M., Dampha A., Deckers J., Dewitte O., Gallali T.,
Hallett S., Jones R., Kilasara M., Le Roux P., Micheli E., Montanarella L., Spaargaren O.,
Thiombiano L., Van Ranst E., Yemefack M., Zougmoré R., (eds.), 2013, Soil Atlas ofAfrica.
European Commission, Publications Office of the European Union, Luxembourg. 176 pp.
Kamguia J., Manguelle-Dicoum E., Tabod C. T., Tadjou J. M. 2005. Geological models
deduced from gravity data in the Garoua basin, Cameroon. J. Geophys. Eng. 2 147–152.
Keesstra S. D., Bouma J., Wallinga J., Tittonell P., Smith P., Cerdà A., Montanarella L., Quinton
J. N, Pachepsky Y., van der Putten W. H., Bardgett R. D., Moolenaar S., Mol G., Jansen
B., Fresco L. 2016. The significance of soils and soil science towards realization of the
United Nations Sustainable Development Goals. Soil, 2, 111–128.
Kempen B., Brus D., Stoorvogel J.J. 2011. Three-dimensional mapping of soil organicmatter
content using soil type–specific depth functions. Geoderma, 162, 107–123.
Kips A. P., Moukam A., Van Ranst E. 1986. Exchange characteristics, clay – silt minaralogy and
classification of some yellowish sedimentary soils in the Tiko plain area, South-West
Cameroon. Regional seminar on laterites, Douala Cameroon, 21-27 January 1986.
Klarer A. J. 2014. The Evolution and Expansion of Cacao Farming in South West Cameroon
and its Effects on Local Livelihoods. Master thesis. SupAgro IRC, Montpellier,
Copenhagen University, Copenhagen.
Lagacherie P., 2008. Digital Soil Mapping: A State of the Art, in: Hartemink, A.E., McBratney,
A.B., Mendonça-Santos, M.L. (Eds.), Digital Soil Mapping with Limited Data. Springer, pp.
3–14
Lagacherie P., McBratney A. B. 2007. Chapter 1. Spatial soil information system and spatial soil
inference systems: Perspectives for Digital Soil Mapping In: P. Lagacherie, A. B.
McBratney and M. Voltz (Eds.), Digital soil mapping, an introductory perspective.
Developments in Soil science, Vol. 31. Elsevier, Amsterdam, pp. 3 – 24.
Laplante A. 1961. Prospection pedologiue dans la region de Garoua. ORSTOM/IRCAM.
Law-Ogbomo K. E., Nwachokor M. A. 2010. Variability in selected soil physic-chemical
properties of five soils formed on different parent materials in southeastern Nigeria.
Research Journal of Agriculture and Biological Sciences, 6(1), 14-19
Leenaars J. G. B., A. J. M. Van Ostrum and M. Ruiperez Gonzalez, 2014. Africa Soil Profile
Database, version 1.2. A compilation of georeferenced and standardized legacy soil
profiles data for Sub Saharan Africa (Witn dataset). ISRIC Report 2014/01. Africa Soi

55
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Information Service (AfSIS) Project and ISRIC – World Soil Information, Wageningen, the
Netherlands. 162 pp.
Liaw A., Wiener M. 2002. Classification and regression by randomForest. R news volume 2(3):
18-21
Ließ M., Glaser B., Huwe B. 2012. Uncertainty in the spatial prediction of soil texture:
Comparison of regression tree and Random Forest models. Geoderma, 170, 70-79.
Ließ M., Schmidt J., Glaser B. 2016. Improving the Spatial Prediction of Soil Organic Carbon
Stocks in a Complex Tropical Mountain Landscape by Methodological
Specifications in Machine Learning Approaches. PLOS ONE.
Malone B. P., McBratney A. B., Minasny B., Laslett G. M. 2009. Mapping continuous depth
functions of soil carbon storage and available water capacity. Geoderma 154:138–152.
Malone B. P., Minasny B., McBratney A. B. 2017. Using R for Digital Soil Mapping. Springer
International Publishing Switzerland 2017.
Martin D. 1962. Reconnaissance pédologique dans le département de la Bénoué.
ORSTOM/IRCAM.
McBratney A., Field D. J., Koch A. 2014. The dimensions of soil security. Geoderma 213,
203–213.
McBratne A. B., Mendoça Santos M. L., Minasny B. 2003. On digital soil mapping. Geoderma,
117(1-2): 3-52.
McCauley A., Jones C., Olson-Rutz K. 2017. Soil pH and organic matter. Nutrient Management.
Module N°8. March 2017 4449-8.
McTainsh G. H. 1984. The nature and origin of the Aeolian mantles of central northern
Nigeria. Geoderma, 33, 13-37.
Minasny B., McBratney A. B., Mendonça-Santos M. L., Odeh I. O. A., Guyon B. 2006. Prediction
and digital mapping of soil carbon storage in the Lower NamoiValley. Soil Research, 44(3),
233-244.
MINADER 2013. Annual Report for the Ministry of Agriculture and Rural Development, South
West Region, Meme Divisional Delegate.
Mitchell T., Developers G. 2014. Geospatial Power Tools: GDAL Raster & Vector Commands.
Locate Press
Molongue K. T. 2016. Growth potentials and constrains of micro, small and medium sized
enterprises in the southwest region of Cameroon. Master Thesis. Pan African Institute for
developement – West Africa. Buea Cameroun.
Molua E. L. 2006. Climatic trends in Cameroon : implications for agricultural management. Clim.
Res. Vol. 30 : 255–262.
Moore I. D., Gessler P. E., Nielsen G. A. Peterson G. A. 1993. Soil attributes prediction
using terrain analysis. Soil Science Society of America Journal, 57, 443–452.
Mulder V. L., Lacoste M., Richer-de-Forges A. C., Martin M. P., Arrouay D. 2016. National versus
global modelling the 3D distribution of soil organic carbon in mainland France. Geoderma
263 (2016) 16–34.
Munns D. N. 1986. Acid soil tolerance in legumes and rhizobia. Advances in Plant
Nutrition. 2:63-91.
Ndamè J. P. 2003. Natural linking areas and local communities in the north province of
Cameroon, in T. Bassett et M.C. Cormier-Salem (dir.), Nature as Local Heritage in Africa
56
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

: New Approaches to Biodiversity Conservation, Territory, and Identity, 29th Annual Spring
Symposium of Centre for African studies /IRD, April 5-10 2003, University of Illinois at
Urbana Champaign, 7 p.
Ndamè J. P. 2007. L'aménagement difficile des zones protégées au Nord Cameroun, Autrepart
2007/2 (n° 42), p. 145-161.
Nelson A., Fusaro J., and Johnson A. 2017. CRM REHAB: How to standardize your data.
RingLead. DATAhttps://www.ringlead.com/4-steps-data-standardization.
Ngatcha N. B., Njitchoua R., Naah, E. 2001. Le barrage de Lagdo (Nord-Cameroun) Impact sur
les plaines d'inondation de la Bénoué. Gestion intégrée des zones inondables tropicales.
455-474.
Niang M. A., Nolin M. C., Jégo G., Perron I. 2014. Digital Mapping of Soil Texture
Using RADARSAT-2 Polarimetric Synthetic Aperture Radar Data. Soil Science
Society of America Journal, 78(2), 673-684.
Ogban P. I., Babalola O. 2003. Soil characteristics and contsraints to crop production ininland
valley bottoms in southwestern Nigeria. Agricultural Water Management, 61, 13-28.
Okun O., Priisalu H. 2007. Random forest for gene expression-based cancer classification:
overlooked issues. In: Martı, J., J.M. Benedı, A.M. Mendonc and J. Serrat, (Eds.). Pattern
Recognition and Image Analysis: Third Iberian Conference, Lecture Notes in Computer
Science, Springer Berlin Heidelberg, p 483-490.
Olowolafe E. O. 2002. Soil parent materials and soil properties in two separate catchmentareas
on the Jos Plateau, Nigeria. GeoJournal, 56, 2001-212.
OpenStreetMap. 2018. l'encyclopédie libre. 10 sept. 2018, 17:27 UTC. 10 sept. 2018
Pebesma E. J., Bivand R. S. 2005. Classes and methods for spatial data in R. R news. 5(2):9–
13.
Prasad A. M., Iverson L. R., Liaw A. 2006. Newer classification and regression tree
techniques: bagging and random forests for ecological prediction. Ecosystems, 9,
181-199.
Priori S., Bianconi N. Costantini E. A. 2014. Can γ-radiometrics predict soil textural data
and stoniness in different parent materials? A comparison of two machine-learning
methods. Geoderma, 226, 354-364.
Rawlins B. G., Marchant B. P., Smyth D., Scheib C., Lark R. M., Jordan, C. 2009.
Airborne radiometric survey data and a DTM as covariates for regional scale
mapping of soil organic carbon across Northern Ireland. European Journal of Soil
Science, 60(1), 44-54.
Roecker S. M., Howell D. W., Haydu-Houdeshell C. A., Blinn C. 2006. A quantitative comparison
of conventional soil suvey and digital soil mapping approaches. In: Digital soil mapping
Bringing Research, Environmental Application, and Operation. ISBN 978-90-481-8862-8.
Rousk J., Brookes P. C., Bååth E. 2009. Contrasting soil pH effects on fungal and bacterial
growth suggest functional redundancy in carbon mineralization. Applied & Environmental
Microbiology. 75:1589- 1596.
Royle J. A., Nychka D. 1998. An algorithm for the construction of spatial coverage design with
implementation in SPLUS. Computers & Geosciences 24, 479–488.
Silatsa T. F. B., Yemefack M., Tabi F. O. 2017. Digital soil mapping of soil propertiesCameroon.
Cameroon soil database (CAMSODAT.01).
57
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Sharu M. B., Yakubu M., Noma, S. S. Tsafe A. I. 2013. Land Evaluation of an Agricultural
Landscape in Dingyadi District, Sokoto State, Nigeria. Nigerian Journal of Basic and
Applied Sciences, 21(2), 148-156.
Slessarev E. W., Lin Y., Bingham N. L., Johnson J. E., Dai Y., Schimel J. P., Chadwick O. A.
2016. Water balance creates a threshold in soil pH at the global scale. Nature 540, 567–
569.
Soil Survey Staff. 2014. Keys to soil taxonomy, 12th edition. USDA Natural Resources
Conservation Service.
Sreenivas K., Dadhwal V.K., Kumar S., Sri Harsha G., Mitran T., Sujatha G., Suresh G. J. R.,
Fyzee M. A., Ravisankar T. 2016. Digital mapping of soil organic and inorganic carbon
status in India. Geoderma 269 (2016) 160–173.
Thompson J. A., Pena-Yewtukhiw E. M., Grove J. H. 2006. "Soil–landscape modeling across a
physiographic region: Topographic patterns and model transportability." Geoderma,
133(1), 57-70.
Thompson J. A., Roecker S., Grunwald S., Owens P. R. 2012. Digital soil mapping: Interactions
with and applications for hydropedology. In: H. Lin, editor, Hydopedology. 1sted. Academic
Press, Amsterdam. p. 665–709.
van Groenigen J. W., Siderius W., Stein A., 1999. Constrained optimisation of soil sampling for
minimisation of the kriging variance. Geoderma 87, 239–259.
Walvoort D. J. J., Brus D. J., Gruijter J. J. 2010. An R package for spatial coverage sampling
and random sampling from compact geographical strata by k-means. Computers &
Geosciences 36 (2010) 1261–1267.
Wang J. F., Stein A., Gao B., Ge Y. 2012. A review of spatial sampling. Spatial Statistics.
WRB. 2015. World Reference Base for soil Ressources 2014.International soil classification
system for naming soils and creating legends for soil maps. Update 2015. FAO Rome,
Italy.
Yerima B. P. K., Van Ranst E. 2005. Majors soil classification systems used in the tropics: Soil
of Cameroon. Trafford Publishing, Canada.

58
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

7. ANNEXES

Annex A: prediction assessment of soil properties


Annex A1: Prediction assessment of soil properties in the North region

Properties Depth Min Max Mean Median SD Kurt Skew


0–5 2.70 77.21 14.60 13.15 5.56 1.12 0.24
5 - 15 2.51 76.61 13.93 12.44 5.41 1.11 0.28
SOC 15 - 30 2.12 45.34 7.44 6.75 2.56 1.23 0.24
30 - 60 1.11 42.74 4.99 4.17 2.33 1.28 0.39
60 - 100 0.79 42.07 4.19 3.35 2.35 1.34 0.37
100 - 200 0.77 42.01 4.01 3.18 2.36 1.34 0.35

0–5 3.56 81.54 19.34 17.61 6.51 1.53 0.20


5 - 15 3.74 81.70 19.65 17.97 6.45 1.49 0.20
Clay 15 - 30 3.64 81.93 21.73 20.28 5.93 1.38 0.19
30 - 60 3.87 81.87 25.07 24.12 4.69 1.36 0.13
60 - 100 4.24 81.31 25.37 24.58 4.44 1.35 0.10
100 - 200 4.33 78.43 24.93 24.24 4.34 1.35 0.10

0–5 4.48 86.92 57.80 60.59 9.72 1.54 -0.26


5 - 15 4.47 87.31 56.04 58.34 9.62 1.54 -0.25
Sand 15 - 30 4.43 87.65 56.04 58.32 9.09 1.49 -0.21
30 - 60 4.42 89.15 53.54 55.24 8.14 1.36 -0.15
60 - 100 5.04 90.38 53.77 55.17 7.67 1.29 -0.12
100 - 200 5.54 90.53 54.48 55.87 7.46 1.27 -0.14

0–5 3.92 84.43 22.86 21.73 5.64 1.16 0.29


5 - 15 3.89 84.53 22.73 21.59 5.64 1.17 0.29
Silt 15 - 30 4.05 84.38 22.24 21.18 5.51 1.16 0.27
30 - 60 3.87 83.98 21.39 20.38 5.22 1.78 0.27
60 - 100 3.23 83.07 20.86 19.96 4.96 1.15 0.25
100 - 200 3.31 81.91 20.59 19.69 4.86 1.15 0.27

0–5 5.07 9.06 6.61 6.57 0.27 1.10 0.15


5 - 15 5.10 9.13 6.61 6.56 0.28 1.19 0.13
pH 15 - 30 5.20 9.22 6.59 6.54 0.29 1.58 0.10
30 - 60 5.10 9.27 6.69 6.65 0.36 1.45 0.10
60 - 100 5.41 9.45 6.92 6.89 0.46 1.19 0.09
100 - 200 5.47 9.50 7.03 7.04 0.47 1.09 0.01

59
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex A2: Prediction assessment of soil properties in the South-West


region
Properties Depth Min Max Mean Median SD Kurt Skew
0–5 5.70 77.18 30.84 29.24 9.78 1.17 0.09
5 - 15 5.20 76.55 28.53 26.53 9.56 1.25 0.13
SOC 15 - 30 5.01 63.45 18.01 16.06 8.19 1.37 0.08
30 - 60 2.74 53.23 14.16 12.79 7.51 1.18 0.02
60 - 100 2.04 51.59 13.087 11.61 7.32 1.19 0.05
100 - 200 1.98 51.13 12.75 11.31 7.15 1.22 0.04

0–5 3.29 78.71 33.71 33.39 6.02 1.26 -0.01


5 - 15 3.24 79.92 34.39 34.08 6.18 1.28 -0.01
Clay 15 - 30 3.23 82.62 35.50 35.17 6.32 1.29 -0.04
30 - 60 3.01 84.55 36.61 36.27 6.41 1.33 -0.04
60 - 100 3.27 85.15 37.11 36.76 6.43 1.32 -0.05
100 - 200 3.21 85.09 36.85 36.44 6.52 1.33 -0.04

0–5 3.81 93.70 49.51 49.69 8.08 1.20 -0.04


5 - 15 2.66 93.89 50.46 51.08 8.26 1.19 -0.04
Sand 15 - 30 2.44 94.56 49.91 50.66 8.31 1.21 -0.06
30 - 60 2.34 94.46 49.20 50.01 8.19 1.21 -0.08
60 - 100 2.40 94.33 48.80 49.67 8.17 1.23 -0.07
100 - 200 2.59 94.37 48.88 49.90 8.11 1.22 -0.10

0–5 1.39 63.67 16.78 14.71 5.89 1.25 0.42


5 - 15 0.93 63.90 16.50 14.39 5.85 1.27 0.44
Silt 15 - 30 0.83 64.43 15.95 13.86 5.75 1.30 0.43
30 - 60 0.81 64.89 15.47 13.43 5.59 1.13 0.43
60 - 100 0.76 66.63 15.45 13.44 5.55 1.32 0.43
100 - 200 0.82 66.80 15.63 13.59 5.62 1.31 0.43

0–5 3.88 6.63 5.15 5.17 0.23 1.15 -0.09


5 - 15 3.87 6.63 5.14 5.16 0.23 1.15 -0.07
pH 15 - 30 3.91 6.64 5.16 5.18 0.26 1.22 -0.12
30 - 60 3.99 6.64 5.17 5.19 0.22 1.23 -0.13
60 - 100 4.01 6.66 5.17 5.19 0.22 1.23 -0.13
100 - 200 4.01 6.64 5.17 5.18 0.21 1.21 -0.12

60
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex B : Variation of soil properties with depth in the North region


of Cameroon
Annex B1: Variation of SOC with depth in the North region

Annex B2: Variation of clay with depth in the North region

61
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex B3: Variation of sand with depth in the North region

Annex B4: Variation of silt with depth in the North region

62
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex B5: Variation of pH water with depth in the North region

63
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex C : Variation of soil properties with depth in the South-West


region of Cameroon
Annex C1: Variation of SOC with depth in the South-West

64
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex C2: Variation of clay with depth in the South-West region

Annex C3: Variation of sand with depth in the South-West region

65
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex C4: Variation of silt with depth in the South-West region

Annex C5: Variation of soil pH with depth in the South-West region

66
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

Annex D: R script (Example: Prediction of clay content)


1. Splinning soil data

## Load the packages


library(devtools)
library(ithir)
library(aqp)

## import the data files


profile <- read.csv("profiles.csv")
layer <- read.csv("Clay.csv")
summary(layer)

## Merge the data


data_clay<- merge(layer, profile, by.x = "ProfileID", by.y = "ProfileID", all.x = TRUE)
write.csv(data_clay, "data_clay.csv")
data <- read.csv("dat.csv")

## Check if the data contains NA values


which(is.na(data))

## Transform the data from dataframe to soilprofilecollecetion class


depths(data) <- ProfileID ~ UpDpthSample + LowDpthSample
## Apply the splinning to the pHwater data
data_spline <- ea_spline(data, var.name = "Clay",
d = t(c(0,5,15,30,60,100,200)),lam = 0.1, vlow = 0,
show.progress = TRUE)
names(data_spline)

## remove the data frame with satndard depht soil profile from the list created during splining
std_Clay<- data_spline[[1]]
names(std_Clay)

## Joint the profiles coordinates to the splinned data


s_data <- merge(std_Clay, profile, by.x = "id", by.y = "ProfileID", all.x = TRUE)

## Change the name of the columns


names(s_data)
colnames(s_data) <- c("profileID","L1", "L2", "L3", "L4", "L5", "L6", "soil depth","PrObj", "X", "Y",
"XYAccur", "ObsDpth")
names(std_Clay)
summary(s_data)
hist(log(s_data$L1), col ="green")

## Save the data as csv File


write.csv(s_data, "spline_Clay_data.csv")

2. Check and select the important covariates for the model from the whole set of covariates

## Load the packages


library(plotKML)
library(sp)
library(raster)
library(randomForest)
67
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

### Import the splined Clay data


spline_Clay <- read.csv("spline_Clay_data .csv")
names(spline_Clay)

### Change the data into spaial point data


pnts <- spline_Clay[, c("L1", "X", "Y")]
names(pnts)
coordinates(pnts) <- ~X + Y
proj4string(pnts) <- CRS("+proj=longlat +datum=WGS84")
plotKML(pnts)

### Check for coordinates duplicates and removes if any


coordinates(spline_Clay) <-~ X+Y
print(zerodist(spline_Clay))
spline_Clay <- remove.duplicates(spline_Clay, remove.second=T)
zerodist(spline_Clay)

# LIdentify the path where we can access the covariates and asign names to the path
files <- list.files(path = "/home/silatsaf/Documents/Data_analysis/Data_for_DSM/Covar_OrgC",pattern =
"\\.tif$", full.names = TRUE)

# Now, we are going to stack these raster files


beginCluster(4)
stack_Clay<- raster(files[1])
for (i in 2:length(files)) {
stack_clay <- stack(stack_Clay, files[i])
}
endCluster()

# Now, we are going to make the soil point data intersection (4 minutes)
beginCluster(4)
my_data <- as.data.frame(extract(stack_Clay, spline_Clay, sp = 1, method = "simple"))
endCluster()
names(my_data)

#### Save the file as csv and remove the NA values in the covariates
write.csv(my_data, "data_covar.csv")

#### Import the data rearanged in libre office without NA in the covariates as my_data1
my_data1 <- read.csv("Clay_covar.csv")
names(my_data1)

3. Prepare the prediction grid

### Load the packages


library(rgdal)
library(GSIF)
library(aqp)
library(sp)
library(maptools)
library(gstat)
library(rgeos)
library(raster)
library(mvtnorm)
library(spatstat)
68
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

library(randomForest)
library(plotKML)
library(psych)
m <-
readGDAL(paste0("/home/silatsaf/Documents/Data_analysis/Data_for_DSM/Prediction_grid_NorthR/so
uth_west.tif"), silent=TRUE)
fullgrid(m) <- FALSE
gridded(m) <- FALSE

# adjust projection
t.proj <- "+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0"
proj4string(m) <- CRS(as.character(NA)) # removes projection
proj4string(m) <- CRS(t.proj)

# convert to data.frame
beginCluster(4)
m <-as.data.frame(m)
m$id <- c(1:nrow(m))
coordinates(m) <- ~x+y
proj4string(m) <- (CRS(t.proj))
endCluster()
cov.lst <- list.files(path=(paste("./sw_covar/", sep=",")), pattern=glob2rx("*.tif$"), full.names=TRUE) ##
glob2rx("*.tif$"))
beginCluster(4)
coords <- coordinates(m)
grid <- extract(m, cov.lst,
path="./", ID = "id",
method = "simple")
names <- strsplit(names(grid), paste0(".tif")[[1]][1])
names(grid)<- names
grid <- grid[, 2:ncol(grid)]
grid <- cbind(coords,grid)
grid <- na.omit(grid)
endCluster()
write.csv(grid, "grid.csv")

4. Spatial prediction of soil properties

## Load the packages


library(aqp)
library(sp)
library(rgdal)
library(maptools)
library(gstat)
library(rgeos)
library(raster)
library(spatstat)
library(randomForest)
library(psych)
library(ithir)
library(fBasics)
library(nortest)

## Import the soil data layers with overlaid with covariates and the prediction grid
Clay_model <- read.csv("Clay_model.csv")

69
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

## Define the random forest model for this soil property


Clay_RF <- randomForest(Clay~ Depth + ASSDAC3 + B02CHE3+ C02MCF5 + DEMENV5 + ENTENV3
+ ES1MOD5 + EVEENV3 + EX1MOD5 + GTDHYS3 + I01MOD4 + M08MOD4 + MANMCF5 + MAXENV3
+ N03MSD3 + NEGMRG5 + NIRL00 + P01CHE3 + RANENV3 + T01MSD3 + T07MOD3 + VBFMRG5 +
VDPMRG5 + VW1MOD1 + CLYPPT, data = clay_dataL1, importance = TRUE, ntree = 500, na.action =
NULL)

varImpPlot(Clay_RF, n.var = 10) # show only the 10 most important variables


print(Clay_RF)
Clay_RF$predicted
clay_dataL1$predicted <- Clay_RF$predicted
write.csv(clay_dataL1, "Clay_data_validation.csv")
grid <- read.csv("grid.csv")

###Analysis on the first layer


## Add the first depth values to the grid
grid$Depth <- 2.5
head(grid)
tail(grid)

### Predicting and summary statistics on the first layer


clay_L1 <- predict(Clay_RF, newdata = grid)
grid$clayL1 <- clay_L1
summary(grid$clayL1)
sd(grid$clayL1)
sampleSKEW(grid$clayL1)
sampleKURT(grid$clayL1)

5. Rasterization and maps of soil properties after prediction

## import the raster of the study area


m1 <-
readGDAL(paste0("/home/silatsaf/Documents/Data_analysis/Data_for_DSM/Prediction_grid_NorthR/so
uth_west.tif"), silent=TRUE)
mx <- raster(m1)
coordinates(grid) <- ~x + y
names(grid)
which(is.na(grid$VW4MOD))
clayL1_raster <- rasterize(x = grid, y = mx, field = "ClayFL1", background = NA)
plot(clayL1_raster)
writeRaster(clayL1_raster, filename="ClayL1.tif", format="GTiff", overwrite=TRUE)
my.colors = colorRampPalette(c("#f2f0f7", "#cbc9e2", "#9e9ac8", "#756bb1", "#54278f"))
plot(clayL30_raster, frame.plot=T, axes = T, box = T, legend.width = 1, legend.shrink = 1, col =
my.colors(255))

6. Validation processes

library(ithir)
library(MASS)
library(ggplot2)

### Clay data


##Import the clay validation data
Clay_validation <- read.csv("Clay_data_validation.csv")

70
PRESS NO & SW
Projet Ressources du Sol et du Sous-sol des Régions du Nord et du Sud-Ouest
Project on Soil and Subsoil Resources of North and South-West Regions

names(Clay_validation)
Clay_validationL1 <- subset(Clay_validation, Layer == "L1")
names(Clay_validationL1)[4] <- paste("Observed")
names(Clay_validationL1)[184] <- paste("Predicted")
goof(observed = Clay_validationL1$Observed, predicted = Clay_validationL1$Predicted)
ggplot(Clay_validationL1, aes(Observed, Predicted) ) + geom_point() + geom_smooth(method=lm)

7. Script for optimization of sampling design

### Load the packages


library(rgdal)
library(sp)
library(maptools)
library(spcosa)

### Import the shapefile of the region


shp <-readOGR(".","Southwest_cam", verbose = TRUE, p4s = NULL)

#### Import the csv file containing the coordinate of exixting soil profiles in the region
point <- read.csv("spatial_point.csv")

#### Convert the point into spatial pointsclass data


point <- SpatialPoints(point, proj4string=CRS(as.character(NA)))

### Match the projection of the shape file and the profiles coordinates
t.proj <- "+proj=longlat +datum=WGS84 +no_defs +ellps=WGS84 +towgs84=0,0,0"
proj4string(point) <- CRS(as.character(NA)) # removes projection
proj4string(point) <- CRS(t.proj)

### Stratify the area of the regionn and define the sampling pattern
new_stratification <- stratify(shp, nStrata = 250, priorPoints = point, nTry = 10, equalArea = F, verbose
= T)
mySamplingPattern <- spsample(new_stratification)

## Plot the strata containing the points


plot(new_stratification, mySamplingPattern)

71

You might also like