See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/378743193

Reference soil groups map of Ethiopia based on legacy data and machine
learning-technique: EthioSoilGrids 1.0

Article in SOIL · March 2024

DOI: 10.5194/soil-10-189-2024

CITATIONS READS

3 448

35 authors, including:

Ashenafi Ali Teklu Erkossa

Addis Ababa University Deutsche Gesellschaft für Internationale Zusammenarbeit
15 PUBLICATIONS 130 CITATIONS 111 PUBLICATIONS 1,937 CITATIONS

SEE PROFILE SEE PROFILE

Kiflu Gudeta Wuletawu Abera

Independent Researcher Consultative Group on International Agricultural Research
3 PUBLICATIONS 10 CITATIONS 83 PUBLICATIONS 729 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Yihenew G.Selassie on 06 March 2024.

The user has requested enhancement of the downloaded file.

SOIL, 10, 189–209, 2024

SOIL
https://doi.org/10.5194/soil-10-189-2024
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Reference soil groups map of Ethiopia based

on legacy data and machine learning-technique:
EthioSoilGrids 1.0
Ashenafi Ali1,2,3,4 , Teklu Erkossa3 , Kiflu Gudeta2 , Wuletawu Abera4 , Ephrem Mesfin2 , Terefe Mekete2 ,
Mitiku Haile6 , Wondwosen Haile7 , Assefa Abegaz1 , Demeke Tafesse12 , Gebeyhu Belay7 ,
Mekonen Getahun8,9 , Sheleme Beyene10 , Mohamed Assen1 , Alemayehu Regassa11 , Yihenew G. Selassie9 ,
Solomon Tadesse12 , Dawit Abebe13 , Yitbarek Wolde13 , Nesru Hussien2 , Abebe Yirdaw2 , Addisu Mera2 ,
Tesema Admas2 , Feyera Wakoya2 , Awgachew Legesse2 , Nigat Tessema2,10 , Ayele Abebe14 ,
Simret Gebremariam2 , Yismaw Aregaw2 , Bizuayehu Abebaw2 , Damtew Bekele12 , Eylachew Zewdie7 ,
Steffen Schulz3 , Lulseged Tamene4 , and Eyasu Elias2,5
1 Department of Geography and Environmental Studies, Addis Ababa University (AAU), Addis Ababa, Ethiopia
2 Ministry of Agriculture (MoA), Addis Ababa, Ethiopia
3 Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ), Addis Ababa, Ethiopia
4 International Center for Tropical Agriculture (CIAT), Addis Ababa, Ethiopia
5 Center for Environmental Science, Addis Ababa University, Addis Ababa, Ethiopia
6 Land Resource Management and Environmental Protection, Mekelle University, Mekelle, Ethiopia
7 private consultant: Addis Ababa, Ethiopia
8 Amhara Design and Supervision Enterprise (ADSE), Bahir Dar, Ethiopia
9 Department of Natural Resources Management, BahirDar University (BDU), Bahir Dar, Ethiopia
10 School of Plant and Horticultural Science, Hawassa University (HU), Hawassa, Ethiopia
11 Department of Natural Resource Management, Jimma University (JU), Jimma, Ethiopia
12 Ethiopian Construction Design and Supervision Works Corporation (ECDSWCo), Addis Ababa, Ethiopia
13 Engineering Corporation of Oromia, Addis Ababa, Ethiopia
14 National Soil Testing Center, MoA, Addis Ababa, Ethiopia

Correspondence: Ashenafi Ali ([email protected], [email protected])

Received: 6 May 2022 – Discussion started: 23 May 2022

Revised: 21 November 2023 – Accepted: 9 January 2024 – Published: 5 March 2024

Abstract. Up-to-date digital soil resource information and its comprehensive understanding are crucial to sup-
porting crop production and sustainable agricultural development. Generating such information through conven-
tional approaches consumes time and resources, and is difficult for developing countries. In Ethiopia, the soil
resource map that was in use is qualitative, dated (since 1984), and small scaled (1 : 2 M), which limit its practical
applicability. Yet, a large legacy soil profile dataset accumulated over time and the emerging machine-learning
modeling approaches can help in generating a high-quality quantitative digital soil map that can provide better
soil information. Thus, a group of researchers formed a Coalition of the Willing for soil and agronomy data-
sharing and collated about 20 000 soil profile data and stored them in a central database. The data were cleaned
and harmonized using the latest soil profile data template and 14 681 profile data were prepared for modeling.
Random forest was used to develop a continuous quantitative digital map of 18 World Reference Base (WRB)
soil groups at 250 m resolution by integrating environmental covariates representing major soil-forming factors.
The map was validated by experts through a rigorous process involving senior soil specialists or pedologists
checking the map based on purposely selected district-level geographic windows across Ethiopia. The map is
expected to be of tremendous value for soil management and other land-based development planning, given its
improved spatial resolution and quantitative digital representation.

Published by Copernicus Publications on behalf of the European Geosciences Union.

190 A. Ali et al.: EthioSoilGrids 1.0

1 Introduction 2 Methods

Soils are important resources that support the development 2.1 The study area
and production of various economic, social, and ecosystem
services, and are useful in climate change mitigation and The study area covered the entire area of Ethiopia (1.14 ×
adaptation (Baveye et al., 2016). Data on the physical and 106 km2 ) located between 3 and 15° N, and between 33 and
chemical characteristics of soils and their spatial distribution 48° E (Fig. 1). The topography of the country is marked
are needed to define and plan their functions over time and by a large altitudinal variation, ranging from 126 m below
space, which are important steps toward sustainable use and sea level at Dalol in the northeast to 4620 m at Ras Dashen
management of soils (Elias, 2016; Hengl et al., 2017). Mountain in the northwest (Billi, 2015; Enyew and Steen-
In Ethiopia, soil surveys and mapping have been con- eveld, 2014). Ethiopia’s wide range of topography, climate,
ducted at various scales with varying scopes, approaches, parent material, and land use types created conditions for the
methodologies, qualities, and levels of detail (Abayneh, formation of different soil types (Abayneh, 2005; Berhanu
2001; Abayneh and Berhanu, 2007; Berhanu, 1994; Elias, and Ochtman, 1974; Donahue, 1972; Mesfin, 1998; Nyssen
2016; Zewdie, 2013). The most recent countrywide digital et al., 2019; Virgo and Munro, 1978; Zewdie, 2013, 1999).
soil mapping efforts focused primarily on soil characteristics More than 33 % of the country is covered by the central,
(Ali et al., 2020; Iticha and Chalsissa, 2019; Tamene et al., upper, and highland complex (Abegaz et al., 2022), which
2017), although soil class maps are equally important for al- embraces Africa’s most prominent mountain system (Hurni,
locating a particular soil unit for specific use (Leenaars et al., 1998).
2020a; Wadoux et al., 2020). Many attempts have been made The country’s complex topography strongly determines
to improve digital soil information systems (Hengl et al., both rainfall and temperature patterns, by modifying the in-
2021, 2017, 2015; Poggio et al., 2021). However, the initia- fluence of the large-scale ocean–land–atmosphere pattern,
tives were based on limited and unevenly distributed soil pro- thus creating diverse localized climates. Spatially, rainfall is
file data (e.g., 1.15 soil profiles per 1000 km2 for Ethiopia), characterized by a general decreasing trend in the direction
which restricts the accuracy and applicability of the products. from west to east, north, northeast, south and southeast. The
In Ethiopia, thousands of soil profile data have been col- lowlands in the southeast and northeast, covering approxi-
lected since the 1960s (Erkossa et al., 2022), but these data mately 55 % of the country’s land area, are characterized
were scattered across different institutions and individuals by arid and semi-arid climates. Annual rainfall ranges from
(Ali et al., 2020). Furthermore, countrywide quantitative and less than 300 mm in the southeastern and northwestern low-
gridded spatial soil-type information does not exist (Elias, lands to over 2000 mm in the southwestern highlands (south-
2016). The Ethiopian Soil Information System (EthioSIS) ern portion of the western highlands). The eastern lowlands
project attempted to develop a countrywide digital soil map get rain twice a year, in April–May and October–November,
focusing on topsoil characteristics, including plant nutri- with two dry periods in between. The total annual precipi-
ent content, but overlooked soil resource mapping (Ali et tation in this region varies from less than 500 to 1000 mm.
al., 2020; Elias, 2016), despite a strong need for a high- The driest of all regions is the Denakil Plain, which receives
resolution soil resource map (Mulualem et al., 2018). less than 500 mm of rain and sometimes none (Fazzini et
Ethiopia has an area of about 1.14 × 106 km2 consisting al., 2015). Temperatures are also greatly influenced by the
of varied environments, making its soils extremely heteroge- rapidly changing altitude, and the mean monthly values vary
neous. Capturing the heterogeneity using conventional soil from ∼ 35 °C in the northeast lowlands to less than 7.5 °C
survey and mapping approaches is an expensive and time- over the north and central highlands.
consuming endeavor (Hounkpatin et al., 2018). This can be The country is characterized by a wide variety of geo-
circumvented by using available legacy soil profile data accu- logical formations (Abyneh, 2005; Alemayehu et al., 2014;
mulated over decades and by tapping into the potential of ad- Elias, 2016; Zewdie, 2013). These include (i) recent and old
vanced analytical techniques to develop high-resolution digi- volcanic activities; (ii) the highlands consisting of igneous
tal soil maps (Hounkpatin et al., 2018; Kempen, 2012, 2009). rocks (mainly basalts); (iii) steep-sided valleys characterized
Therefore, the objectives of this study were to (i) develop a by strong colluvial and alluvial deposits; (iv) metamorphic
national legacy soil profile dataset that can be used as an in- rocks exposed by denudation process; and (v) various sedi-
put for various digital soil mapping exercises, and (ii) gener- mentary rocks such as limestone and sandstone in the rela-
ate an improved 250 m digital Reference Soil Groups (RSGs) tively lower areas.
map of Ethiopia. Diverse biophysical factors affecting the spatial distribu-
tion of vegetated land cover which in turn, both as single
and combined factors, result in diverse soil types and prop-
erties across Ethiopia’s landscapes (Hurni, 1998; Nyssen et
al., 2019; WLRC-AAU, 2018). The spatiotemporal vegeta-

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 191

Figure 1. Location map of Ethiopia (inset) and overview map of Esri World Topographic Map.

tion cover of the country has been characterized by a long soil and agronomy data access and sharing in the country
history of land use and land cover changes (WLRC-AAU, (Tamene et al., 2021).
2018). In terms of the type and spatial coverage of major land The CoW conducted a national soil and agronomy data
use and land cover classes, woody vegetation (forest, wood- ecosystem mapping which revealed that a plethora of legacy
land, and shrub and bush lands) covers about 57 % of the soil resource datasets exist across different institutions and
country in accordance with the national 2016 map (WLRC- individuals (Ali et al., 2020). The assessment also revealed
AAU, 2018). This is followed by cultivated land (20 %) and that a sizable proportion of the data holders were willing to
grasslands (12 %). Barren lands are estimated to cover about share the data in their custody, provided that some regula-
1/10 of the area of the country while other minor lands with tions were put in place to administer the data. The CoW de-
ecological significance (i.e., wetlands, water bodies, and sub- veloped and approved internal data-sharing guidelines (CoW,
afro-alpine and afro-alpine) cover about 1.2 % of the coun- 2020), and facilitated data collation campaigns, which in-
try’s land mass. volved both formal and informal approaches to data holders.
Through a data collation campaign, soil profile data col-
lected between the 1970s and 2021 were acquired from over
2.2 Legacy soil profile data collation and preparation 88 diverse sources (Ali et al., 2020; Tamene et al., 2021). Ini-
tially, 8000 profile data points were collated and subjected to
The soil profile data generated over decades through vari- improved modeling techniques to create a provisional WRB
ous soil survey missions were kept in a variety of formats reference soil group map of Ethiopia. This was presented to
with limited accessibility. There has been no institution with various partners and data-holding institutions to demonstrate
a mandate to coordinate the generation, collation, harmoniza- the power of data sharing. This created awareness and en-
tion, and sharing of soil profile data. This led to the formation abled us to mobilize and collate over 20 000 legacy soil pro-
of a group of individuals and institutions who were willing file data. These data were then added to the national data
to exchange soil and agronomy data. Established in 2018, repository.
the group known as the Coalition of the Willing (CoW) was
committed to addressing the challenges posed by the lack of

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

192 A. Ali et al.: EthioSoilGrids 1.0

The data had varying levels of completeness in terms of date environmental variables representing soil-forming fac-
soil field and environmental descriptions and laboratory anal- tors (climate, organisms, relief, parent material, and time)
ysis. These required a rigorous expert-based quality assess- were derived from diverse remote sensing products and the-
ment and standardization before being compiled into a har- matic maps (Hengl and MacMillan, 2019; McBratney et al.,
monized format. The expanded version of the Africa Soil 2003).
Profile (AfSP) database (Leenaars et al., 2014) template was Relief and topography-related covariates were derived
used for standardizing and harmonizing the data. Out of the from a 90 m Shuttle Radar Topography Mission (SRTM) dig-
collated soil profile data, 14 681 georeferenced data points ital elevation model (DEM) (Vågen, 2010). Climate-related
were extracted based on completeness and cleanness for the variables including long-term mean, minimum, maximum,
purposes of modeling. The cleaned soil profile data set con- and standard deviation temperature as well as precipita-
tained, at least, the reference soil group (RSG) nomenclature tion data for the period between 1983 and 2016 (Dinku
as outlined in the WRB legend. While the original soil pro- et al., 2014) were acquired from Enhancing National Cli-
file records were set in different coordinate systems, all were mate Services (ENACTS-NMA) initiatives with 4 km res-
projected into the adopted standard georeferencing system, olutions (Dinku et al., 2014). Moderate-resolution imag-
namely, WGS84, decimal degrees in the QGIS (3.20.2) en- ing spectroradiometer (MODIS) imagery raw bands and de-
vironment (QGIS Development Team, 2021). To verify their rived indices (Vågen, 2010) were downloaded from USGS
position, soil profile locations were plotted using a standard EarthExplorer (https://earthexplorer.usgs.gov/, last access:
WGS84 coordinate system to verify that points matched the 12 November 2021) to represent vegetation-related factors.
site description, geomorphological settings, and at the very National geological (Tefera et al., 1996) and land use and
least the source project boundary outline. land cover (WLRC-AAU, 2018) thematic maps of Ethiopia
The accuracy of the data depends on the quality and reli- were gathered to represent parent material and organisms, re-
ability of the survey data themselves, which in turn requires spectively.
expert knowledge and experience in soil description and clas- Downscaling (disaggregating) or upscaling (aggregating)
sification (Leenaars et al., 2020a). In this study, data clean- of rasters was also performed to match the target resolu-
ing, validation, reclassification, and verification were carried tion. A 250 m spatial resolution was chosen to accommodate
out by a team of prominent national pedologists and soil sur- both the spatial resolution of the major covariate inputs and
veyors, including those involved in the generation of some of make it applicable for large-scale analysis. All layers were
the soil profile data themselves (Fig. 2). masked for buildings and water bodies by the national bound-
In addition, the Ministry of Agriculture (MoA) soil sur- ary of Ethiopia and a stacked layer was created using the
vey and mapping experts and other volunteers validated the raster package (R Core Team, 2020) to extract covariate val-
legacy soil profile observations. This led to the reclassifica- ues at the locations of soil profiles. One-hot encoding using
tion of the soil types as deemed necessary. Such validation the dummyVars function available in Caret package (Kuhn,
and reclassification involved re-examining the geomorpho- 2008) was used to pre-process and convert categorical co-
logical setup of the soil profile locations using Google Earth variates into a binary vector. Each element of the binary
as well as reviewing the site and soil descriptions and the cor- vector represents the presence or absence of that category.
responding laboratory data, and reviewing the proposed soil One-hot encoding is beneficial because it enables machine-
type. The harmonized datasets in the database were used as learning algorithms to interpret categorical variables as nu-
input soil profile data for modeling and mapping IUSS WRB merical features. The covariate pre-processing, visual inspec-
reference soil groups. tion for inconsistencies, and resampling to a target grid of
250 m were conducted in QGIS [3.20.2] (QGIS Develop-
ment Team, 2021), SAGA GIS [7.8.2] (Conrad et al., 2015)
2.3 Preparation and selection of environmental and R [version 4.05] (R Core Team, 2020) software pack-
covariates ages. All input data were projected to a common Lambert
2.3.1 Covariate acquisition and preparation azimuthal equal-area projection with the latitude of origin
at 8.65 and center of meridian at 39.64, which is the cen-
In order to develop spatially continuous soil class and/or ter point for Ethiopia. This projection was selected since it is
type maps, data on environmental covariates that represent effective in minimizing area distortions over land. Each co-
directly or indirectly the soil-forming factors have to be inte- variate was adjusted to have an identical spatial resolution,
grated with soil profile data (Hengl and MacMillan, 2019). extent, and projection using two resampling methods. Con-
Environmental covariates are spatially explicit proxies of tinuous covariates were resampled using the bilinear spline
soil-forming factors based on the soil–environment relation- method, whereas categorical covariates were resampled us-
ship (McBratney et al., 2003; Shi et al., 2018). Acquisi- ing the nearest neighbor method.
tion and preparation of covariates represent a crucial step
in digital soil mapping using machine-learning algorithms
(McBratney et al., 2003). In this study, 68 potential candi-

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 193

Figure 2. Schematic presentation of data acquisition and workflow.

2.3.2 Covariate selection the RF model are its ability to handle numerical and categor-
ical data without any assumption of the probability distribu-
Selecting an optimal set of covariates to effectively repre- tion, and its robustness against nonlinearity and overfitting
sent the soil–environment relationship is a key step in digital (Breiman, 2001; Svetnik et al., 2003). While building the RF
soil mapping (DSM) since improper selection of covariates model, data were split into training (80 %) and testing (20 %)
will affect the quality of model outputs (Shi et al., 2018). In components using random sampling for training the model
this study, near-zero variance assessment was conducted us- and evaluating its performance, respectively (Kuhn, 2008).
ing the nearZeroVar function available in the R caret package Hyper-parameter optimization and repeated cross-validation
(Kuhn, 2008) to identify and remove environmental variables on the training dataset were performed for optimal model ap-
that have little or no variance. In addition, preliminary ran- plication using the ranger method of the Caret package. The
dom forest model training was performed to assess and iden- three tuning parameters for ranger method are mtry, splitrule,
tify covariates having high variable importance. After expert and .min.node.size. Generally this function is used to tune
judgment, a total of 27 environmental variables (24 contin- the parameters in modeling in an automated fashion, as this
uous and 3 categorical) were selected for modeling and pre- will automatically check all the possible tuning parameters
dicting RSGs. and return the optimized parameters on which the model
gives the best accuracy. Model tuning was performed with a
2.4 Modeling and mapping soil types or reference soil repeated 10-fold cross-validation procedure applying multi-
groups ple combinations of hyper-parameters for the ranger method.
This is a fast implementation of RF particularly suited for
2.4.1 Model tuning and quantitative evaluation
high-dimensional data (Wright and Ziegler, 2017). Then the
In digital soil mapping, machine-learning techniques have number of covariates used for the splits (mtry), splitting rules
been extensively used to determine the relationship between (splitrule), and minimum node size (min.node.size) were op-
soil types and environmental variables (McBratney et al., timized. The parameter ntree was adjusted to 1000 in the
2003). Many machine-learning models were developed in model, and mtry values (10, 15, 20), min.node.size values
the past decades for digital soil mapping to spatially pre- (5, 10, 15), and splitrule values (“variance”, “extratrees”, and
dict soil classes based on existing soil data and soil-forming “maxstat”) were fed for the optimization procedure. The ac-
environmental covariates (Heung et al., 2016). Random for- curacy of the testing dataset was related to the model per-
est (RF), a tree-based ensemble method, is one of the most formance for the new dataset, indicating the capacity of the
promising machine-learning techniques available for digital model to predict at the unsampled location. A confusion
soil mapping (Breiman, 2001; Heung et al., 2016). RF has matrix was also used to calculate a cross-tabulation of ob-
gained popularity due to its high overall accuracy and has served and predicted classes with associated statistics, i.e.,
been widely used in predictive soil mapping (Brungard et al., producer’s accuracy and user’s accuracy.
2015; Hengl et al., 2018). Examples of the main strengths of

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

194 A. Ali et al.: EthioSoilGrids 1.0

2.4.2 Software and computational framework improvements were made, e.g., in the areas where Vertisols,
Fluvisols, and Leptosols were overestimated. Further, under-
In this study, various open-source software packages that estimated RSGs (Alisols, Solonetz, Planosols, Acrisols, Lix-
provide a comprehensive set of tools and diverse capabilities isols, Phaeozems, and Gleysols) showed a slight increase in
were used for data preparation, analysis, and visualization. area coverage and pattern improvements. However, the total
Data pre-processing and preparation were performed using area of Leptosols and Cambisols increased from the first run
QGIS (QGIS Development Team, 2021) and SAGA GIS due to the partial exclusion of the mask layer used in the first
(Conrad et al., 2015). For statistical analysis and machine- round of modeling. The mask layer used in the first run was
learning modeling, R (R Core Team, 2020) and relevant li- criticized for quality issues as it excluded significant soil ar-
braries were installed on a Windows server, 2016 standard eas and due to its weakness in capturing non-soil areas such
with 250 GB of working memory, to handle the challenges as rock outcrops, salt flats, swamps, and sand dunes. Never-
associated with large-scale data processing and analysis. theless, the spatial patterns of these soils occurring across
previously considered “non-soil areas” were examined by
2.4.3 Expert evaluation of spatial patterns of the the panel of experts. In parallel, geospatial and soil experts
beta-version soil map checked the raster map of the RSGs in the GIS environ-
ment to ensure areas with “no concern” before re-running
Visual inspection of the DSM output over the terrain was the model are kept the same or changes are accepted by the
used to identify abnormalities and assess how effectively it panel of experts. The map from the second run is presented
depicts landscape components (Rossiter et al., 2022). For in this paper as EthioSoilGrids version 1.0 product.
this, we employed an expert-based qualitative assessment of
the model output. This technique was used to complement
3 Results and discussion
model-based accuracy assessment and confirm agreement
soil specialists or pedologists checking the map based on 3.1 Soil profile datasets
purposely selected district-level geographic windows across
Ethiopia, representing different agro-ecological zones known Using the IUSS WRB (2015), the preliminary identified
to have diverse soil occurrences, and that were familiar to 14 742 georeferenced legacy soil profiles were classified
the panel of experts. Accordingly, an expert validation work- and/or reclassified into 23 RSGs. Nearly 90 % of the soil pro-
shop was conducted using the first version of the reference file points represented Vertisols, followed by Luvisols, Cam-
soil groups (RSGs) map. About 45 multi-disciplinary scien- bisols, Leptosols, Fluvisols, and Nitisols, which were found
tists including soil surveyors, pedologists, geologists, and ge- to be the dominant soil types in Ethiopia (Fig. 3). The re-
omorphologists were drawn from national and international maining 10 % represented the Regosols, Alisols, Andosols,
research, development, and higher-learning institutions to re- Arenosols, Calcisols, Solonetz, Lixisols, Phaeozems, Solon-
view the draft RSG map in plenary discussions. This was chaks, Acrisols, Planosols, Gleysols, Umbrisols, Ferralsols,
followed by breakout sessions where groups of experts eval- Gypsisols, Plinthosols, and Stagnosols.
uated the map based on their experience and knowledge of According to this study, about 72 % of the IUSS
soil–landscape relations of the country and examined geo- WRB (2015) RSGs were confirmed to occur in Ethiopia. This
graphic windows. reconfirms the characterization of Ethiopia as a land of soil
Most importantly, disagreements regarding RSG occur- diversity being endowed with a diverse range of soil types
rence and patterns of the modeling outputs across topo- (Elias, 2016; Mishra et al., 2004). One of the limitations
sequences and contrasting soil-forming factor sequences with legacy soil data in categorical mapping is the imbal-
were identified and discussed. Further, inferences on parts anced soil samples, in that all classes are not equally repre-
of the DSM framework that require improvement were rec- sented (Wadoux et al., 2020). For this study, soil profiles with
ommended. After finalizing the evaluation at the group-level fewer than 30 observations were objectively excluded from
assessment, each group presented the results in the plenary the model after examining the accuracy and spatial distribu-
followed by a discussion to get feedback from other par- tion of each RSG. Five RSGs (Umbrisols, Ferralsols, Gyp-
ticipants. Following the plenary discussions, the participants sisols, Plinthosols, and Stagnosols) were excluded from the
created a group of six senior pedologists to work on the rec- model and the EthioSoilGrids version 1.0 map.
ommendations including changing the quality mask layer, After excluding the built-up and water surface areas, the
validating the additional data obtained during the event, and average soil profile density was 13.1 per 1000 km2 (Fig. 4),
assessing the re-modeling outputs. but the actual density varied across the different parts of
After the second model was re-run, the group of senior the country. The variation tends to follow river basins, sub-
pedologists together with geospatial experts re-evaluated the basins, and agricultural land-use type-based studies from
output using the selected districts based on the feedback which most of the legacy data were pulled. For instance, in 30
from the first review, which was mainly on areas where there intervention districts of the Capacity Building for Scaling up
were “minor” and “major” concerns. Consequently, some of Evidence-Based Best Practices in Agricultural Production

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 195

Figure 3. Number of soil profile points per WRB reference soil groups.

The distribution of the soil profiles across the 32 agro-

ecological zones (AEZ) of Ethiopia revealed that all, except
two – tepid per-humid mid-highland (0.13 % landmass) and
very cold sub-humid sub-afro-alpine to afro-alpine (0.03 %
landmass) – were represented by soil profile observations.
Furthermore, about 95 % of the profile observations repre-
sented 91 % of the AEZ aerial coverage (Appendix A). The
distribution of legacy soil profiles varied across AEZs. In
general, the top-ranked lowland AEZs with roughly 56 %
area coverage were represented by 23 % of the total profile
observations, whereas top-ranked highland AEZs with 20 %
area coverage received 47 % of profile observations. For in-
stance, warm desert, warm moist, hot arid, and warm sub-
moist lowlands with area coverage of around 20 %, 15 %,
11 %, and 10 %, were represented roughly by 3 %, 11 %,
2 %, and 7 % of the total profiles, respectively. Tepid moist
Figure 4. Spatial distribution of collated legacy soil profile data. mid-highlands (8 % area coverage), tepid sub-humid mid-
highlands (7 % area coverage), and tepid sub-moist mid-
highlands (5 % area coverage) each were represented by
20 %, 15 %, and 12 % of the profiles, respectively.
in Ethiopia (CASCAPE) project, the average profile density
was about 87 profiles per 1000 km2 for a total area of about
26 830 km2 (Leenaars et al., 2020a). Similarly, semi-detailed 3.2 Modeling and mapping
soil mapping missions in 15 districts conducted through the 3.2.1 Variable importance
Bilateral Ethiopia–Netherlands Effort for Food, Income and
Trade (BENEFIT)-REALISE project generated about 217 The RSG spatial pattern is primarily influenced by long-
observations per 1000 km2 (Leenaars et al., 2020b). term average surface reflectance, flow-based DEM indices,
A soil type and depth map compilation and updating mis- and precipitation. Figure 5 shows variables of importance
sion at a 1 : 250 000 scale by the Water Land Resource Cen- for determining RSG spatial prediction. The top-ranked vari-
ter (WLRC) of Addis Ababa University collated and used ables were (i) long-term MODIS near-infrared (NIR) re-
about 3949 legacy soil profiles for the entire country (Ali et flectance, (ii) multiresolution index of valley bottom flatness,
al., 2020), which is approximately 3.5 profiles per 1000 km2 . (iii) long-term mean day–land surface temperature, (iv) long-
Although the distribution is not even and the eastern low- term mean soil moisture, (v) standard deviation of long-term
lands are sparsely represented, the number of data used in precipitation, (vi) long-term mean precipitation, and (vii) to-
this study is 8.5 times higher than the 1712 legacy soil pro- pographic wetness index.
files data currently existing in the Africa soil profile database MODIS long-term mean spectral signatures showed high
(Batjes et al., 2020; Leenaars et al., 2014). relative importance. According to Hengl et al. (2017), ac-

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

196 A. Ali et al.: EthioSoilGrids 1.0

counting for seasonal vegetation fluctuation and inter-annual fusion matrix at validation/testing points, i.e., 20 % of the ob-
variations in surface reflectance, long-term temporal signa- servation. Further, the matrix indicates the producer’s accu-
tures of the soil surface, derived as monthly averages from racy (class representation of observed versus predicted) and
long-term MODIS imagery, were more effective. Further- user’s accuracy were not similar for all RSGs. The map pu-
more, Hengl and MacMillan (2019) explained that long-term rity is in the order of Lixisols, Calcisols, Alisols, Phaeozems,
average seasonal signatures of surface reflectance provide a Vertisols, Andosols, Solonchaks, Fluvisols, Arenosols, Lep-
better indication of soil characteristics compared with only a tosols, Luvisols, Nitisols, and Cambisols. However, Verti-
single snapshot of surface reflectance. sols, Calcisols, and Andosols are the observed classes that
The multi-resolution valley bottom flatness index, a DEM- are best represented by the map followed by Fluvisols, Al-
derived topography index, is the second top-ranked covariate isols, Nitisols, Leptosols, Luvisols, and Cambisols.
driving soil variability across Ethiopia. This hydrological/- Global soil grids at 250 m resolution used machine-
soil removal and accumulation or deposition index is used learning algorithms to map the global WRB RSGs with map
to distinguish valley floor and ridgetop landscape positions purity and weighted kappa of 28 % and 42 %, respectively
(Soil Science Division Staff, 2017) greatly responsible for (Hengl et al., 2017). The SoilGrids 250 m WRB soil groups/-
multiple soil-forming processes to operate over a particu- classes prediction output–spatial soil patterns were not eval-
lar landscape, resulting in a wide range of soil development. uated based on expert knowledge while in this study we did
The influence of topography on spatial soil variation is man- an extensive back-and-forth qualitative assessment by a panel
ifested in every landscape of Ethiopia (Belay, 1997; Mesfin, of pedologists. The quantitative accuracy in the present study
1998; Nyssen et al., 2019; Zewdie, 2013). (about 56 %) coupled with an expert-based qualitative eval-
Long-term daily mean land surface temperature, mean soil uation of the predicted maps indicated the development and
moisture, rainfall standard deviation, and mean annual rain- achievement of a substantially enhanced national product for
fall were among the top-ranked covariates for predicting the users of spatial soil resource information. This finding is a
spatial variation of RSGs across the country. In Ethiopia, dif- step forward and acceptable considering that SoilGrids maps
ferent soil genesis studies revealed that climate has a signif- are not expected to be as accurate as locally produced maps
icant influence on soil development and properties and is, and models that use many more local-point data and finer lo-
therefore, responsible for the existence of widely varying cal variables (Mulder et al., 2016). Further, the data and find-
soils in the country (Abayneh, 2005; Abayneh et al., 2006; ings in this study can help improve the soil maps of Africa
Fikru, 1988, 1980; Zewdie, 2013). as they partially address the concern by Hengl et al. (2017),
Among the most important covariates for predicting RSGs who recognized that WRB RSGs modeling in the global Soil-
in the Ethiopian highlands are monthly average soil mois- Grids 250 m is critically uncertain for parts of Africa. This is
ture for January (ranked third), long-term average soil mois- mainly attributed to limited access to more local point data by
ture (ranked fourth), and monthly average soil moisture for regional and global modeling initiatives, unlike the present
August (ranked fifth) (Leenaars et al., 2020a). In the current study which accessed a large number of legacy soil profile
study, soil moisture was among the 10 top-ranked covariates datasets.
in modeling and explaining long-distance soil type variabil-
ity across the country. 3.2.3 Modeling and mapping: EthioSoilGrids version 1.0
In this study, lithology showed a relatively low influence
on soil variability that may be due to the use of a coarse-scale The study identified 18 RSGs in Ethiopia, mapped at 250 m
and less detailed lithology map, which may not sufficiently resolution (Fig. 6). The model prediction showed that seven
capture the spatial variability of the parent materials. soil reference groups including Cambisols, Leptosols, Ver-
tisols, Fluvisols, Nitisols, Luvisols, and Calcisols covered
3.2.2 Model performance
nearly 98 % of the total land area of the country (Fig. 7). Five
soil reference groups (Solonchaks, Arenosols, Regosols, An-
The parameter optimization process resulted in mtry = 20, dosols, and Alisols) were estimated to cover about 2 % of
split rule = extra trees and minimum node size = 5. The over- the land area, while trace coverages of Solonetz, Planosols,
all accuracy of the model was 56.24 % which ranged between Acrisols, Lixisols, Phaeozems, and Gleysols were also found
54.43 % and 58.1 % with a 95 % confidence interval. The in some pocket areas.
kappa values based on the internal cross-validation and test- In terms of spatial distribution, Nitisols and Luvisols dom-
ing dataset showed that the overall model performance pro- inated the northwestern and southwestern highlands while
duced using 10-fold cross-validation with the repeated fitting the southeastern lowlands were dominantly covered by Cam-
was 48 %. Considering similar area-based digital soil class bisols, Calcisols, and Fluvisols with some Solonchaks. The
mapping efforts, the overall accuracy was in line with the Vertisols extensively cover the north and southwestern low-
accuracies that were typically reported for soil class maps lands along with the Ethiopia–Sudan border areas and cen-
developed with RF models (Leenaars et al., 2020a) and sta- tral highland plateaus. The probability of occurrence of each
tistical methods (Heung et al., 2016). Table 1 shows the con- RSG was mapped (Appendix C) in each modeling spatial

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 197

Figure 5. Random forest covariate relative importance for modeling RSGs. Note: prep = precipitation; prep_sd = standard devia-
tion of precipitation; tmax = maximum temperature; tmin = minimum temperature; trange = temperature range; tav_sd = standard devi-
ation of average temperature; pet = potential evapotranspiration; lstd = land surface temperature–day; lstn = land surface temperature–
night; soil_moist = soil moisture; soil_temp = soil temperature; DEM = digital elevation model (elevation); twi = topographic wet-
ness index; aspect = topographic aspect; curv = topographic curvature; conv = topographic convergence index; ls = slope length and
steepness factor (ls_factor); morph = terrain morphometry; mrvbf = multiresolution index of valley bottom flatness; slope = slope class
(%); ndvi = normalized difference vegetation index (NDVI); evi = enhanced vegetation index (EVI); lulc = land use/land cover; lithol-
ogy = geology; ref1 = red band; ref2 = near-infrared; ref7 = mid-infrared.

window (i.e., the cell size of 250 m × 250 m). The dominant servations, the modeling approach, and most importantly
RSGs were aggregated based on the most probable RSGs in the level of detail and scale/resolution of the environmen-
each spatial modeling window. There was high correspon- tal variables used in this study. We used the currently avail-
dence between the seven top-ranked prediction probabilities able coarse-resolution national geological map and hence
and observed soil types as confirmed visually by overlaying soil parent material might be inadequately represented in
observed classes and prediction probabilities. the model, which probably resulted in irregular RSG se-
The overall occurrence and the relative position of each quences. For instance, the main driving factors to establish
of the RSGs along the topo-sequence and its association and explain the soil-landscape variability in the May-Leiba
with other RSGs agree with previous works (Abayneh et catchment of northern Ethiopia were geology (soil parent
al., 2006; Ali et al., 2010; Abdenna et al., 2018; Asma- material) and different mass movements (Van de Wauw et
maw and Mohammed, 2012; Belay, 2000, 1998, 1997, 1996; al., 2008). These factors led to Cambisols–Vertisols cate-
Driessen et al., 2001; Elias, 2016; FAO, 1984a; Fikre, 2003; nas on basalt and Regosols–Cambisols–Vertisols catenas on
Mitiku, 1987; Mohammed and Belay, 2008; Mohammed and limestone formations. Similar studies identified parent mate-
Solomon, 2012; Mulugeta et al., 2021; Nyssen et al., 2019; rial strongly determines the soil type (e.g., Vertisol, Luvisol,
Sheleme, 2017; Shimeles et al., 2007; Tolossa, 2015; Zewdie, Cambisol) (Nyssen et al., 2019). In general, in areas where
2013). However, in some cases, the position of the RSGs there is complex soil diversity and distribution of soils, one
along the topo-sequence and the association with other RSGs of the most important parameters is to identify parent ma-
require further investigation. The disparities observed might terial including effective techniques to capture and delineate
be attributed to the positional accuracy of legacy point ob- mass movement bodies, and human-induced soil erosion and

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

198 A. Ali et al.: EthioSoilGrids 1.0

Table 1. Confusion matrix of random forest RSG prediction (at validation/testing observations).

Prediction Reference

User accuracy
Solonchaks
Phaeozems
Cambisols
Arenosols

Leptosols

Solonetzs
Planosols
Andosols

Calcisols

Fluvisols

Regosols

Vertisols
Gleysols

Luvisols
Acrisols

Lixisols

Nitisols
Alisols

Total
Acrisols 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0.33 3
Alisols 0 40 0 0 0 0 1 1 0 0 9 4 0 0 2 0 0 2 0.68 59
Andosols 0 0 28 1 1 3 5 0 2 0 2 0 0 0 0 0 1 1 0.64 44
Arenosols 0 0 0 11 0 2 1 0 0 0 5 0 0 0 0 0 0 1 0.55 20
Calcisols 0 0 0 0 21 0 1 0 0 0 2 0 0 0 0 0 0 5 0.72 29
Cambisols 2 3 6 9 1 197 28 2 35 2 47 16 5 1 16 3 3 28 0.49 404
Fluvisols 1 0 3 5 1 34 144 0 9 0 15 7 0 0 1 5 5 17 0.58 247
Gleysols 0 0 0 0 0 0 1 2 0 0 1 0 0 1 0 0 0 0 0.40 5
Leptosols 0 1 4 3 3 47 11 0 176 0 27 7 1 0 32 0 0 24 0.52 336
Lixisols 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1.00 1
Luvisols 2 16 3 8 0 34 13 2 33 3 216 30 3 0 25 1 0 41 0.50 430
Nitisols 6 8 0 0 1 23 8 3 18 8 29 132 0 1 8 0 1 21 0.49 267
Phaeozems 0 0 0 0 0 0 0 0 0 0 1 0 2 0 0 0 0 0 0.67 3
Planosols 0 0 0 0 0 0 0 0 0 0 1 1 0 5 1 0 0 1 0.55 9
Regosols 0 0 0 0 0 7 1 0 7 1 8 1 0 0 22 0 0 5 0.42 52
Solonchaks 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 3 1 0 0.60 5
Solonetzs 0 0 0 0 1 4 1 0 0 0 0 0 0 0 0 1 6 0 0.46 13
Vertisols 3 1 3 5 5 92 32 2 61 3 81 31 5 5 25 2 6 641 0.64 1003
Producer 0.07 0.58 0.60 0.26 0.62 0.44 0.58 0.17 0.51 0.06 0.49 0.58 0.13 0.38 0.17 0.20 0.25 0.81 0.56 –
accuracy
Total 15 69 47 42 34 443 247 12 342 18 445 229 16 13 132 15 24 787 – 2930

deposition areas (Leenaars et al., 2020a; Nyssen et al., 2019; 3.3 Expert validation of the soil map
Van de Wauw et al., 2008).
Expert knowledge of soil–landscape relations and soil dis-
Considering the third position of Cambisols in the or-
tribution remains important for evaluating the predictive soil
der of frequency of occurrence of RSGs per point observa-
mapping results and assessing whether the predicted spatial
tions (following Vertisols and Luvisols), these soils seem to
patterns make sense from a pedological viewpoint (Hengl
be over-represented on the map (ranked first) apparently at
et al., 2017; Poggio et al., 2021; Rossiter et al., 2022). An
the expense of Vertisols and Luvisols, and to some extent
important step in qualitative model evaluation is, therefore,
in places of Leptosols and other RSGs. This might be at-
expert assessment, whereby professionals with broad expe-
tributed to the fact that Cambisols create a geographical con-
rience in soil survey and mapping can evaluate and improve
tinuation with Vertisols and/or Luvisols at the lower slopes
the quality of the soil resource map. This can highlight ar-
and Leptosols/Regosols at the higher slopes, suggesting the
eas of agreement or concern across the landscape (Rossiter
presence of some bordering soil qualities in respective tran-
et al., 2022). The expert validation workshop provided use-
sitional zones (Ali et al., 2010; Asmamaw and Mohammed,
ful insights and tangible improvements to the development of
2012; Sheleme, 2017; Zewdie, 2013).
the map. While the plenary discussion provided an overview
The proportion of area mapped as Cambisols (34 %) re-
of the approaches followed in developing the map, the group
vealed new insights compared with the information from
discussions helped to have an in-depth review of the selected
the most cited spatial soil maps: Cambisols ranked second
polygons of the map assigned to them. Participants were split
(21 %), second (16 %), fourth (9 %), and fourth (8 %) as re-
into five groups (with 8–10 members each) and chose up to
ported by Berhanu (1980), FAO (1984b, 1998), and Soil-
60 polygons representing areas with which at least one of
Grids – Hengl et al. (2017), respectively. This might be
the group members has sufficient information, including data
due to (i) the number and distribution of profile observa-
sources. Overall, the groups checked a total of 126 polygons
tions, which is more extensive than the previous ones; (ii) the
(Fig. 8), which were fairly evenly distributed across the coun-
type and level of details of covariates considered; (iii) vari-
try.
ations and rearrangements in the keys for classification of
The group members displayed the polygons one by one
the RSGs among soil classification versions used in previous
in a GIS environment and discussed the predicted dominant
studies and misclassification/confusion of Vertisols with Ver-
and associated soil RSGs and labeled them in one of three
tic Cambisols, as legacy soil profile data come from diverse
confirmation categories: (1) confirmed with “no concern”,
sources.
(2) confirmed with “minor concern”, and (3) confirmed with
“major concern”. Confirmation with “no concern” was made

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 199

Figure 6. Major reference soil groups of Ethiopia (EthioSoilGrid V1.0).

presence of another soil type, other than the predicted types,

was noted.
All three groups rated the accuracy of the map at 60+ %;
of the 126 polygons, they expressed no concern for 63 %,
minor concern for 23 %, and major concern for 14 % of
the polygons. Furthermore, differences in the prevalence of
RSGs and patterns of the modeling outputs across different
soil-forming factor sequences, as well as inferences about
which areas of the DSM framework still need work, were
identified and elaborated on by the expert input and are pre-
sented in the subsequent sections.
Figure 7. The area coverage (in %) for the major WRB RSGs. Note:
the remaining 10 RSGs-Arenosols (0.44 %), Regosols (0.35 %), An- 3.4 Evaluation of results, limitations, and future direction
dosols (0.31 %), Alisols (0.16 %), Solonetzs (0.04 %), Planosols
(0.04 %), Acrisols (0.02 %), Lixisols (0.02 %), Phaeozems (0.02 %), Up-to-date soil resource spatial information is critically
and Gleysols (0.01 %) were not plotted because of their relatively missing at the required scale and extent in Ethiopia. As a re-
small area coverage. sult, resource management strategies miss their targets. Fur-
thermore, the absence of such data at a required resolution
and extent forced developers of decision support tools to pick
and use the data they can access and afford. As a result,
when all members of a group agreed on the types, the rela- model outputs appear more site-specific or representation be-
tive coverage, and the patterns of the predicted soils within comes homogeneous over the very heterogeneous landscapes
the polygon. Confirmation with “minor concern” was made that exist in reality. On the other hand, in large areas and com-
when all or some of the team members agreed on the pre- plex landscapes such as Ethiopia, it is very difficult to address
dicted soil types within the polygons but did not agree on the the demand for reasonably accurate and detailed soil-type
order of abundance or the probability occurrence of one or maps using a conventional approach due to the costs involved
two soils including observed spatial patterns. Confirmation and to the resources and time this requires. For instance,
with “major concern” was made when all members of the given the vastness of the country and the heterogeneous land-
team did not agree on the predicted soil type, or when the scapes, a new conventional soil survey mission requires at

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

200 A. Ali et al.: EthioSoilGrids 1.0

Figure 8. The spatial distribution of districts validated by stakeholders and feedback categories according to the level of concerns raised.

least 170 000 profile point observations to map the entire ter- prominent contrasting landscapes of Ethiopia. This qualita-
restrial land mass of Ethiopia at a scale of 1 : 250 000 with tive assessment indicated areas of concern in terms of how
at least one observation per square centimeter. Moreover, well EthioSoilGrids version 1.0 represents soil geography
the soil profile data requirement definitely could have been across a mosaic of the country’s landscapes. For instance, in
much higher as we increase the scale of mapping and den- the northeastern lowlands of Ethiopia, mainly along the “De-
sity of observations. In the present study, machine-learning nakil” depression, Fluvisols, Cambisols, and Vertisols were
techniques combined with expert input were implemented to found on the map in areas where normally other soil types
produce a countrywide soil resource map of Ethiopia at rea- were expected to occur. In this area, the expected prediction
sonably higher accuracy and with less time and cost com- and area coverage of Leptosols has probably been overshad-
pared with conventional methods. In addition, rescue, com- owed by Fluvisols and Cambisols. Similarly, in some parts of
pilations, and standardization of about 14 681 geo-referenced western Ethiopia landscapes, the prediction of Vertisols over-
legacy soil profiles that can be included in the National Soil shadows other RSGs, which resulted in an underestimation
Information System (NSIS) of Ethiopia and the World Soil of the area coverage of Fluvisols (along the “Akobo”, “Gilo”,
Information Center will support future national, regional, and and “Baro” rivers and their tributaries) and Alisols. Likewise,
global DSM efforts. The approach used here demonstrates in the central parts of northwestern Ethiopia, the prediction of
the power of data and analytics to map the soil resources of Nitisols was overshadowed by Vertisols and Luvisols, result-
Ethiopia, and the output is an exemplary use case for similar ing in a likely underestimation of the Nitisols area coverage.
digital content development efforts in Ethiopia and beyond. The relatively low model performance and some classifi-
Moreover, in this study the quality-monitoring processes cation errors in some of the examined geographic windows
and methods were followed to filter dubious soil profiles (e.g., the Denakil depression, along Akobo, Baro, and Gilo
as well as soil classification and harmonization protocols. rivers and the Somali region) are probably due to the paucity
Thereafter, the study followed a robust modeling framework of samples from those areas (Fig. 4), the inadequacy of the
and generated new insights into the relative area coverage of dataset by RSGs, and over-representation of the dataset by
WRB RSGs of Ethiopia. In addition, the study provided co- some RSGs, such as Vertisols, Luvisols, and Cambisols. Bal-
herent and up-to-date digital quantitative gridded spatial soil anced datasets are ideal to allow decision tree algorithms
resource information to support the successful implementa- to produce better classification but for datasets with uneven
tion of various digital agricultural solutions and decision sup- class size, the generated classification model might be biased
port tools (DSTs). toward the majority class (Hounkpatin et al., 2018; Wadoux
The spatially explicit limitation of the present study is re- et al., 2020). In addition, uncertainty around the quality of
vealed by expert-based qualitative evaluation of spatial pat- the covariates included, not the covariates considered in the
terns across objectively selected geographic windows and modeling process including management, use of validation

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 201

methods that do not sufficiently control the effect of clustered resented with the lowest number (63) of profiles. The mod-
samples, and small sample size for some RSGs could have eling revealed that the most important covariates for predict-
possibly biased the modeling results in some geographic ar- ing RSGs in Ethiopia are MODIS long-term reflectance, mul-
eas. tiresolution index of valley bottom flatness, land surface tem-
To improve the modeling performance, future studies perature, soil moisture, long-term mean annual rainfall, and
could explore (i) adding data for under-represented geo- wetness index of the landscape.
graphic areas, land uses, and covariate spaces; (ii) opportu- Our 10-fold spatial cross-validation result showed an over-
nities to include other covariates (parent material and man- all accuracy of about 56 % with varying accuracy levels
agement) that could capture the variability of the country’s among RSGs. The modeling result revealed that seven major
heterogeneous landscapes; (iii) dimension reduction of co- soil reference groups including Cambisols (34 %), Leptosols
variates; (iv) use of remedial measures for imbalances in (20 %), Vertisols (18 %), Fluvisols (10 %) Nitisols (7 %), Lu-
sample sizes; (v) comparing different cross-validation meth- visols (6 %), and Calcisols (3 %) covered nearly 98 % of
ods; (vi) use of an ensemble modeling approach and/or the total land area of the country, while minor coverage of
robust modeling technique that accommodates neighbor- other RSGs (Solonchaks, Arenosols, Regosols, Andosols,
hood size and connectivity analyses; (vii) use of a better- Alisols, Solonetzs, Planosols, Acrisols, Lixisols, Phaeozems,
resolution/quality mask layer to segregate non-soil areas and Gleysols) was also detected in some areas. Compared
(rock outcrops, salt flats, sand dunes, and water bodies) with the existing soil resource map, the coverage of the first
from mapping areas; and (viii) implementation of quantita- three major soil groups has substantially increased, which is
tive and qualitative comparisons of national, regional, and related to the increased availability of soil profile data cover-
global legacy soil maps/soil grids with new DSM products in ing larger areas of the country, implying that these soils were
terms of how well DSM products represent soil geography. previously underestimated. Cambisols and Vertisols which
In addition, future digital soil mapping strategies in Ethiopia together represent nearly half of the total land area are rel-
may require consideration of new soil sampling missions atively young with inherent fertility, suggesting a high agri-
in under-represented areas; adoption of standard soil sam- cultural potential for the country. However, given their limi-
pling, description guidelines, and soil classification systems tations, these and the other soil types require the implementa-
including soil physicochemical and mineralogical analysis; tion of suitable land, water, and crop management techniques
and a combination of local soil nomenclature/classification to sustainably exploit their potential.
systems with RSGs and development of a map of RSGs The EthioSoilGrids version 1.0 product from this first
with qualifiers. At the moment the under-sampled and under- countrywide RSGs modeling effort requires complementary
represented areas are the Somali region, the Denakil, and the activities. These include modeling and mapping that should
western and northwestern border areas of Ethiopia (Fig. 4). go beyond RSGs and need to include second-level classifica-
Despite these limitations, and to the best of our knowledge, tions including principal and supplementary qualifiers. Fur-
the EthioSoilGrids v1.0 product provides the most complete thermore, a soil atlas of Ethiopia with details of the soil
soil information available for Ethiopia. physicochemical properties needs to be prepared together
with the map, which the authors and/or others responsible
need to prioritize in their future research endeavors.
4 Conclusions

Coherent and up-to-date countrywide digital soil information

is essential to support digital agricultural transformation ef-
forts. This study involved collation, cleaning, harmonization,
and validation of the legacy soil profile datasets, involving
soil scientists with different backgrounds individually and in
groups. To develop the 250 m digital soil resource map, a
machine-learning modeling approach and expert validation
were applied to the harmonized soil database and environ-
mental covariates affecting soil-forming processes. Accord-
ingly, about 20 000 soil profile data were collated, out of
which about 14 681 were used for the modeling and map-
ping of 18 RSGs out of the 23 RSGs identified. Although
unevenly distributed, the legacy soil profile data used in the
modeling covered most of the agro-ecologies of the country.
Among the 18 RSGs mapped, the highest number of ob-
served (3935) profiles represent Vertisols, followed by Lu-
visols, Cambisols, and Leptosols, while Gleysols were rep-

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

202 A. Ali et al.: EthioSoilGrids 1.0

Appendix A: Legacy soil profile data distribution

Table A1. Distribution of legacy soil profile data by agroecology zones.

Major agroecological zones AEZ area Profiles

coverage (%)a observation (%)b
Warm arid lowland plains 19.76 3.40
Warm moist lowlands 15.12 10.74
Hot arid lowland plains 10.79 2.44
Warm sub-moist lowlands 9.63 6.94
Tepid moist mid highlands 8.05 20.21
Warm sub-humid lowlands 7.11 5.69
Tepid sub-humid mid highlands 6.63 15.26
Tepid sub-moist mid highlands 5.17 12.39
Warm semi-arid lowlands 2.75 3.23
Tepid humid mid highlands 2.65 2.48
Warm humid lowlands 2.29 0.45
Cool moist mid highlands 1.74 4.15
Hot sub-humid lowlands 1.67 0.07
Cool sub-moist mid highlands 1.16 3.00
Cool humid mid highlands 0.82 1.01
Warm per-humid lowlands 0.68 0.01
Hot moist lowlands 0.59 3.56
Hot sub-moist lowlands 0.56 0.03
Cool sub-humid mid highlands 0.52 1.38
Tepid arid mid highlands 0.43 0.39
Hot semi-arid lowlands 0.40 2.05
Tepid semi-arid mid highlands 0.19 0.67
Cold moist sub-afro-alpine to afro-alpine 0.07 0.16
Cold sub-moist mid highlands 0.07 0.04
Cold sub-humid sub-afro-alpine to afro-alpine 0.06 0.03
Cold humid sub-afro-alpine to afro-alpine 0.06 0.01
Very cold humid sub-afro-alpine 0.04 0.02
Very cold sub-moist mid highlands 0.02 0.02
Very cold moist sub-afro-alpine to afro-alpine 0.01 0.03
Hot per-humid lowlands 0.01 0.15
Tepid perhumid mid highland 0.13 0
Very cold sub-humid sub-afro-alpine to afro-alpine 0.03 0

Note: a total area of Ethiopia 1.14 × 106 km2 ; b total number of profiles 14 681.

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 203

Appendix B: Environmental covariates

Table B1. List, description, spatial and temporal extent, and source of covariates used in modeling the reference soil groups.

Categories Covariates Descriptions Spatial Temporal Source

resolution resolution
Climate prep Precipitation 4 km 1981–2016 ENACTS (Dinku et al., 2014)
prep_sd Standard deviation of 4 km 1981–2016 Derived from ENACTS (Dinku et
precipitation al., 2014)
tmax Maximum temperature 4 km 1983–2016 ENACTS (Dinku et al., 2014)
tmin Minimum temperature 4 km 1983–2016 ENACTS (Dinku et al., 2014)
trange Temperature range 4 km 1983–2016 ENACTS (Dinku et al., 2014)
tav_sd Standard deviation of average 4 km 1983–2016 Derived from ENACTS (Dinku et
temperature al., 2014)
pet Potential evapotranspiration 4 km 1981–2016 Derived from ENACTS (Dinku et
al., 2014) using modified Penman
method
lstd Land surface temperature–day 1000 m 2002–2018 AfSISa
(Aqua MODIS-MYD11A2, time
series monthly average)
lstn Land surface temperature–night 1000 m 2002–2018 AfSIS
(Aqua MODIS-MYD11A2, time
series monthly average)
soil_moist Soil moisture (derived from one- 4 km 1981–2016 Ethiopian Digital AgroClimate
dimensional soil water balance) Advisory Platform (EDACaP)
soil_temp Soil temperature 30 km 1979–2019 ERA 5-Reanalysis ECMWF
datab
Topography DEM Digital elevation model 90 m – SRTM-DEM (Vågen, 2010)
(Elevation)
twi Topographic wetness index 90 m – SAGA GIS-based
SRTM-DEM derivative
aspect Topographic aspect 90 m – SAGA GIS-based
SRTM-DEM derivative
curv Topographic curvature 90 m – SAGA GIS-based
SRTM-DEM derivative
conv Topographic convergence index 90 m – SAGA GIS-based
SRTM-DEM derivative
ls Slope length and steepness 90 m – SAGA GIS-based
factor (ls_factor) SRTM-DEM derivative
morph Terrain morphometry 90 m – SAGA GIS-based
SRTM-DEM derivative
mrvbf Multiresolution index of valley 90 m – SAGA GIS-based
bottom flatness SRTM-DEM derivative
slope Slope class (%) 90 m – SAGA GIS-based
SRTM-DEM derivative

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

204 A. Ali et al.: EthioSoilGrids 1.0

Table B1. Continued.

Categories Covariates Descriptions Spatial Temporal Source

resolution resolution
Vegetation ndvi Normalized difference 250 m 2000–2021 AfSISa
vegetation index (NDVI)
(MODIS-MODIS MOD13Q1,
time series monthly average)
evi Enhanced vegetation index 250 m 2000–2021 AfSIS
(EVI) (MODIS-MODIS
MOD13Q1, time series
monthly average)
lulc Land use/landcover 30 m 2010 Water and land resource
Center–Addis Ababa University
(WLRC-AAU, 2018)
parent lithology Geology/parent material 1 : 2 000 000 1996 The Ethiopian Geological Survey
material (Tefera et al., 1996)
MODIS ref1 Red band (MODIS-MODIS 250 m 2000–2018 AfSISa
spectral MOD13Q1, time series monthly
reflectance average)
ref2 Near-infrared (MODIS-MODIS 250 m 2000–2018 AfSIS
MOD13Q1, time series monthly
average)
ref7 Mid-infrared (MODIS-MODIS 250 m 2000–2018 AfSIS
MOD13Q1, time series monthly
average)
a Africa Soil Information Service (AfSIS). b Fifth generation European Centre for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis of the global climate.

Appendix C: Probability of occurrence of reference

soil groups

Figure C1. Occurrence probability maps of Cambisols, Leptosols, Vertisols, and Fluvisols.

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 205

Figure C2. Occurrence probability maps of Nitisols, Luvisols, and Calcisols.

Data availability. Full data will be available upon request based Acknowledgements. We sincerely appreciate the Coalition of
on the CoW guideline (CoW, 2020; https://ethioagridata.com/, last the Willing (CoW) members who are instrumental in providing,
access: 7 November 2023) and the MoA “Soil and Agronomy Data collating, cleaning, standardizing, and harmonizing the legacy soil
Management, Use and Sharing” directive No. 974/2023 Ethiopia profile data used in generating the soil resource map of Ethiopia
(https://nsis.moa.gov.et/, last access: 7 November 2023). at 250 m resolution. The CoW team also deserves credit for inspir-
ing many to share data and develop an integrated national database
related to agronomy and soil profile data. The leadership of the
Author contributions. AshA, TE, KG, WA, and LT conceived Natural Resource Development Sector and Soil Resource Informa-
and designed the study, performed the analysis and wrote the first tion and Mapping Directorate of the Ministry of Agriculture (MoA)
draft with substantial input and feedback from all authors. EM, TM, played a crucial role. This includes assigning experts from the min-
NH, AY, AM, TA, FW, AL, NT, AyeA, SG, YA, and BA contributed istry and other organizations who worked on collating, encoding,
to input data preparation, data encoding, and harmonization. Legacy harmonizing, and processing the soil survey legacy data; and mod-
data validation and review of subsequent versions of the paper were eling and prediction of reference soil groups using robust machine-
performed by MH, WH, AssA, DT, GB, MG, SB, MA, AR, YGS, learning algorithms and high-performance computing servers are
ST, DA, YW, DB, EZ, SS, and EE. the foundation for the soil resource map. Various institutions, as
well as the late and current soil surveyors and pedologists, deserve
special recognition for their contributions to the generation and
sharing of soil profile data. We owe a debt of gratitude to ISRIC and
Competing interests. The contact author has declared that none
the bilateral Ethiopia–Netherlands projects (cascape and BENEFIT-
of the authors has any competing interests.
REALISE) funded by the Directorate-General for International Co-
operation (DGIS) of the Netherlands Ministry of Foreign Affairs
through the Netherlands Embassy in Ethiopia, which have been
Disclaimer. Publisher’s note: Copernicus Publications remains crucial in providing capacity building to the MoA, and national
neutral with regard to jurisdictional claims made in the text, pub- soil and geospatial experts. Many thanks are due to Eyasu Elias,
lished maps, institutional affiliations, or any other geographical rep- Arie van Kekem, Tewodros Tefera, Mulugeta Diro, Johan Leenaars,
resentation in this paper. While Copernicus Publications makes ev- Bas Kempen, Stephan Mantel, and Maria Ruiperez Gonzalez who
ery effort to include appropriate place names, the final responsibility have been organizing and providing training on soil classifica-
lies with the authors. tions and digital soil mapping to the MoA, as well as national
soil and geospatial experts, during the Ethiopia–Netherlands bilat-
eral projects period. The senior pedologists and soil surveyors who
provided invaluable support to check and harmonize thousands of
soil profiles and laboratory results are sincerely appreciated. They
worked very hard with positive energy, for which we are very grate-
ful. In addition, the same group of experts and additional experts
who supported the validation of the preliminary soil resource map

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

206 A. Ali et al.: EthioSoilGrids 1.0

deserve credit for their commitment to contributing their expertise. Abegaz, A., Ashenafi, A., Tamene, L., Abera, W., and Smith, J. U.:
We thank Degefe Tebebe, Sileshi Gudeta, and Neil Munro for sup- Modeling long-term attainable soil organic carbon sequestration
port in the extraction of climate covariates as well as for providing across the highlands of Ethiopia, Environ. Dev. Sustain., 24, 131–
critical technical support and comments that helped improve the pa- 5162, https://doi.org/10.1007/s10668-021-01653-0, 2022.
per. Our sincere appreciation also goes to the continued and persis- Alemayehu, R., Van Daele, K., De Paepe, P., Dumon, M., Deckers,
tent support of GIZ-Ethiopia mainly through the project Support- J., Asfawossen, A., and Van Ranst, E.: Characterizing weathering
ing Soil Health Interventions in Ethiopia (SSHI), which supported intensity and trends of geological materials in the Gilgel Gibe
and facilitated the activities of the CoW. The Alliance of Biover- catchment, southwestern Ethiopia, J. Afr. Earth Sci., 99, 568–
sity and CIAT is greatly acknowledged for coordinating CoW and 580, https://doi.org/10.1016/j.jafrearsci.2014.05.012, 2014.
its efforts and for supporting the implementation of activities that Ali, A., Abayneh, E., and Sheleme, B.: Characterizing soils of
are of high national importance. We would also like to sincerely Delbo Wegene watershed, J. Soil Sci. Environ. Manage., 1, 184–
thank the Excellence in Agronomy (EiA) CGIAR Initiative, which 199, 2010.
has made huge contributions to this project in terms of funding and Ali, A., Tamene, L., and Erkossa, T.: Identifying, Cataloguing, and
building skills of the various teams. The Water, Land and Ecosys- Mapping Soil and Agronomic Data in Ethiopia, CIAT Publi-
tems (WLE) and Climate Change, Agriculture and Food Security cation No. 506, International Center for Tropical Agriculture
(CCAFS) programs of the CGIAR also provided support in vari- (CIAT), Addis Ababa, Ethiopia, https://hdl.handle.net/10568/
ous forms. Recently, our work has benefited from the Accelerating 110868 (last access: 21 November 2021), 2020.
Impacts of CGIAR Climate Research in Africa (AICCRA) project Asmamaw, L. and Mohammed, A.: Characteristics and classifica-
supported by the World Bank in terms of data, analytics, and re- tion of the soils of Gerado catchment, Northeastern Ethiopia,
sources to support data linkage and integration. Ethiopian Journal of Natural Resources, 12, 1–22, 2012.
Batjes, N. H., Ribeiro, E., and van Oostrum, A.: Standardised
soil profile data to support global mapping and modelling
Financial support. This work is financially supported by the (WoSIS snapshot 2019), Earth Syst. Sci. Data, 12, 299–320,
Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) https://doi.org/10.5194/essd-12-299-2020, 2020.
through the project “Supporting Soil Health Interventions in Baveye, P. C., Jacques, B., and John, G.: Soil “Ecosys-
Ethiopia,” funded by the Bill & Melinda Gates Foundation. This tem” Services and Natural Capital: Critical Appraisal of Re-
work was supported, in whole or in part, by the Bill & Melinda search on Uncertain Ground, Front. Environ. Sci., 4, 41,
Gates Foundation (INV-005460). Under the grant conditions of the https://doi.org/10.3389/fenvs.2016.00041, 2016.
Foundation, a Creative Commons Attribution 4.0 Generic License Belay, T.: Characteristics and Landscape relationships of Vertisols
has already been assigned to the Author Accepted Manuscript ver- and Vertic Luvisols of Melbe, Tigray, Ethiopia, SINET, 19, 93–
sion that might arise from this submission. 115, 1996.
Belay, T.: Variabilities of Soil Catena on Degraded Hill Slopes of
Wtiya Catchment, Wello, Ethiopia, SINET, 20, 151–175, 1997.
Review statement. This paper was edited by Kristof Van Oost Belay, T.: Pedogenesis and soil-geomorphic relationships on the
and reviewed by Skye Wills and one anonymous referee. Piedmont slopes of Wurgo Valley, South Welo, Ethiopia, SINET,
21, 91–111, 1998.
Belay, T.: Characteristics and classification of soils of Gora Daget
forest, South welo highlands, Ethiopia, SINET, 23, 35–51, 2000.
Berhanu, D.: A survey of studies conducted about soil resources ap-
References praisal and evaluation for rural development in Ethiopia, Institute
of Agricultural Research, Addis Ababa, Ethiopia, 1980.
Abayneh, E.: Application of Geographic Information System (GIS) Berhanu, D.: The soils of Ethiopia: Annotated bibliography, Re-
for soil resource study in Ethiopia, in: Proceedings of the Na- gional Soil Conservation Unit (RSCU), Swedish International
tional Sensitization Workshop on Agro metrology and GIS, 17– Development Authority (SIDA), Tech. handbook no. 9, 1994.
18 December 2001, Addis Ababa, Ethiopia, 162–169, 2001. Berhanu, D. and Ochtman, L.: Soil resource appraisal and evalu-
Abayneh, E.: Characteristics, Genesis and Classification of Reddish ation studies for rural development in Ethiopia, meeting of the
Soils from Sidamo Region of Ethiopia, PhD thesis, Universiti east African sub-committee for soil correlation and land evalua-
Putra Malaysia, 2005. tion, Nairobi, Kenya, FAO World Soil Resources Rep. 46, 63–70,
Abayneh, E. and Berhanu, D.: Soil Survey in Ethiopia: Past, Present 1974.
and the Future, in: Proceedings of the 8th Conference of the Billi, P.: Geomorphological landscapes of Ethiopia, in: Landscapes
Ethiopian Society of Soil Science, Soils for sustainable devel- and Landforms of Ethiopia, World Geomorphological Land-
opment, 27–28 April, 2006, Addis Ababa, Ethiopia, 2007. scapes, Springer, Dordrecht, 3–32, https://doi.org/10.1007/978-
Abayneh, E., Zauyah, S., Hanafi, M. M., and Rosenani, A. B.: Gene- 94-017-8026-1_1, 2015.
sis and classification of sesquioxidic soils from volcanic rocks in Breiman, L.: RandomForests, Mach. Learn., 45, 5–32,
sub-humid tropical highlands of Ethiopia, Geoderma, 136, 682– https://doi.org/10.1023/A:1010933404324, 2001.
695, https://doi.org/10.1016/j.geoderma.2006.05.006, 2006. Brungard, C. W., Boettinger, J. J., Duniway, M. C., Wills, S. A.,
Abdenna, D., Yli-Halla, M., Mohamed, M., and Wogi, L.: Soil clas- and Edwards, W. T. C.: Machine learning for predicting soil
sification of humid Western Ethiopia: A transect study along classes in three semi-arid landscapes, Geoderma, 239–240, 68–
a toposequence in Didessa watershed, Catena, 163, 184–195, 83, https://doi.org/10.1016/j.geoderma.2014.09.019, 2015.
https://doi.org/10.1016/j.catena.2017.12.020, 2018.

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

A. Ali et al.: EthioSoilGrids 1.0 207

CoW (Coalition of the Willing): Coalition of the Willing for soil Fikru, A.: Soil resources of Ethiopia, in: Natural Resources Degra-
and agronomy data access, management and sharing, Data Shar- dation a Challenge to Ethiopia, First Natural Resources Conser-
ing Guidelines, Ethiopian Institute of Agricultural Research vation Conference, Institute of Agricultural Research (IAR), 7–8
(EIAR), Addis Ababa, Ethiopia, 28 pp., https://hdl.handle.net/ February 1980, Addis Ababa, Ethiopia, 1980.
10568/107988 (last access: 5 December 2021), 2020. Fikru, A.: Need for Soil Survey Studies, in: Proceedings of the first
Conrad, O., Bechtel, B., Bock, M., Dietrich, H., Fischer, E., Gerlitz, soils science research review workshop, 11–14 February 1987,
L., Wehberg, J., Wichmann, V., and Böhner, J.: System for Auto- Addis Ababa, Ethiopia, 1988.
mated Geoscientific Analyses (SAGA) v. 2.1.4, Geosci. Model Hengl, T. and MacMillan, R. A.: Predictive Soil Mapping with R,
Dev., 8, 1991–2007, https://doi.org/10.5194/gmd-8-1991-2015, OpenGeoHub foundation, Wageningen, the Netherlands, https:
2015. //soilmapper.org/ (last access: 14 September 2021), ISBN 978-0-
Dinku, T., Block, P., Sharoff, J., Hailemariam, K., Osgood, D., del 359-30635-0, 2019.
Corral, J., Rémi Cousin, R., and Thomson, M. C.: Bridging crit- Hengl, T., Heuvelink, G. B. M., Kempen, B., Leenaars, J. G. B.,
ical gaps in climate services and applications in Africa, Earth Walsh, M. G., Shepherd, K. D., Sila, A., MacMillan, R. A.,
Perspectives, 1, 1–13, https://doi.org/10.1186/2194-6434-1-15, Mendes de Jesus, J., Tamene, L., and Tondoh, J. E.: Mapping soil
2014. properties of Africa at 250 m resolution: random forest signifi-
Donahue, R. L.: Ethiopia: Taxonomy, cartography and ecology cantly improve current predictions, PLoS ONE, 10, e0125814,
of soils, Michigan State Univ., African Stud. Center and Inst. https://doi.org/10.1371/journal.pone.0125814, 2015.
Int. Agric., Comm., Ethiopian Stud., Occasional Papers Series, Hengl, T., Mendes de Jesus, J., Heuvelink, G. B., Ruiperez
Monograph 1, 1972. Gonzalez, M., Kilibarda, M., Blagotić, A., Shangguan, W.,
Driessen, P. M., Deckers, J., Spaargaren, O., and Nachtergaele, F.: Wright, M. N., Geng, X., Bauer-Marschallinger, B., Gue-
Lecture notes on the major soils of the world, world soil re- vara, M. A., Vargas, R., MacMillan, R. A., Batjes, N. H.,
sources reports No. 94, FAO, Rome, https://edepot.wur.nl/82729 Leenaars, J. G., Ribeiro, E., Wheeler, I., Mantel, S., and
(last access: 12 December 2021), 2001. Kempen, B.: SoilGrids250m: Global gridded soil informa-
Elias, E.: Soils of the Ethiopian Highlands: Geomorphology and tion based on machine learning, PloS one, 12, e0169748,
Properties, CASCAPE Project, ALTERRA, Wageningen UR, the https://doi.org/10.1371/journal.pone.0169748, 2017.
Netherlands, https://library.wur.nl/WebQuery/isric/2259099 (last Hengl, T., Nussbaum, M., Wright, M. N., Heuvelink, G. B. M., and
access: 11 November 2021), 2016. Gräler, B.: Random forest as a generic framework for predic-
Enyew, B. D. and Steeneveld, G. J.: Analysing the impact of tive modeling of spatial and spatio-temporal variables, PeerJ, 6,
topography on precipitation and flooding on the Ethiopian e5518, https://doi.org/10.7717/peerj.5518, 2018.
highlands, J. Geol. Geosci., 3, https://gert-jan.steeneveld.wur.nl/ Hengl, T., Miller, M., Križan, J., Shepherd, K. D., Sila, A.,
enyewsteeneveld2014.pdf (last access: 13 August 2021), 2014. Kilibarda, M., Antonijevi´c, O., Glušica, L., Dobermann, A.,
Erkossa, T., Laekemariam, F., Abera, W., and Tamene, L.: Evolu- Haefele, S. M., McGrath, S. P., Acquah, G. E., Collinson,
tion of soil fertility research and development in Ethiopia: From J., Parente, L., Sheykhmousa, M., Saito, K., Johnson, J. M.,
reconnaissance to data-mining approaches, Exp. Agr., 58, E4, Chamberlin, J., Silatsa, F., Yemefack, M., Wendt, J., MacMil-
https://doi.org/10.1017/S0014479721000235, 2022. lan, R. A., Wheeler, I., and Crouch, J.: African soil prop-
FAO: Assistance to Land Use-Planning, Ethiopia: Provisional Soil erties and nutrients mapped at 30 m spatial resolution using
Association Map of Ethiopia, Field document No. 6, The United two-scale ensemble machine learning, Sci. Rep., 11, 6130,
Nations Development Programme and Food and Agriculture Or- https://doi.org/10.1038/s41598-021-85639-y, 2021.
ganization, FAO, Rome, https://www.fao.org/3/ar767e/ar767e. Heung, B., Hung, C. H., Zhang, J., Knudby, A., Bulmer, C. E.,
pdf (last access: 5 July 2021), 1984a. and Schmidt, M. G.: An overview and comparison of machine-
FAO: Assistance to Land Use-Planning, Ethiopia: Geomorphology learning techniques for classification purposes in digital soil
and soils, Field Document AG DP/ETH/78/003, The United Na- mapping, Geoderma, 265, 62–77, 2016.
tions Development Programme and FAO, FAO, Rome, 1984b. Hounkpatin, K. O. L., Schmidt, K., Stumpf, F., Forkuor, G.,
FAO: The Soil and Terrain Database for north-eastern Africa, Crop Behrens, T., Scholten, T., Amelung, W., and Welp, G.: Pre-
production systems zones of the GAD sub region, Land and wa- dicting reference soil groups using legacy data: A data
ter digital media series no. 2, FAO, Rome, Italy, 1998. pruning and Random Forest approach for tropical environ-
FAO: Guideline for Soil Description, 4th Edn., FAO, ment (Dano catchment, Burkina Faso), Sci. Rep., 8, 9959,
Rome, Italy, https://www.fao.org/publications/card/en/c/ https://doi.org/10.1038/s41598-018-28244-w, 2018.
903943c7-f56a-521a-8d32-459e7e0cdae9/ (last access: 23 Hurni, H.: Agro-ecological Belts of Ethiopia: Explanatory Notes on
February 2021), 2006. three maps at a scale of 1 : 1,000,000, Soil Cons. Res. Pro., Uni-
Fazzini, M., Bisci, C., and Billi, P.: The Climate of Ethiopia, in: versity of Bern, (Switzerland) in Association with the Ministry
Landscapes and Landforms of Ethiopia, World Geomorpholog- of Agriculture, Addis Ababa, https://edepot.wur.nl/484855 (last
ical Landscapes, edited by: Billi, P., Springer, Dordrecht, the access: 6 June 2021), 1998.
Netherlands, 65–87, https://doi.org/10.1007/978-94-017-8026- Iticha, B. and Chalsissa, T.: Digital soil mapping for site-
1_3, 2015. specific management of soils, Geoderma, 351, 85–91,
Fikre, M.: Pedogenesis of major volcanic soils of the southern cen- https://doi.org/10.1016/j.geoderma.2019.05.026, 2019.
tral Rift Valley region, Ethiopia, MSc Thesis, 270 pp., University IUSS WRB (IUSS Working Group): World Reference Base for
of Saskatchewan, Saskatoon, Canada, 2003. Soil Resources 2014, update 2015 International soil classifi-
cation system for naming soils and creating legends for soil

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

208 A. Ali et al.: EthioSoilGrids 1.0

maps, World Soil Resources Reports No. 106, FAO, Rome, https: Mulugeta, T., Seid, A., Kefyialew, T., Mulugeta, F., and Tadla, G.:
//www.fao.org/3/i3794en/I3794en.pdf (last access: 11 February Characterization and Classification of Soils of Askate Subwater-
2019), 2015. shed, Northeastern Ethiopia, Agri. For. Fisheries, 10, 112–122,
Kempen, B., Brus, D. J., Heuvelink, G. B. M., and Stoorvogel, J. J.: https://doi.org/10.11648/j.aff.20211003.13, 2021.
Updating the 1 : 50,000 Dutch soil map using legacy soil data: A Nyssen, J., Tielens, S., Tesfamichael, G., Tigist, A., Kassa,
multinomial logistic regression approach, Geoderma, 151, 311– T., Wauw, J., Degeyndt, K., Descheemaeker, K., Kassa,
326, https://doi.org/10.1016/j.geoderma.2009.04.023, 2009. A., Mitiku, H., and Amanuel, Z.: Understanding spatial
Kempen, B., Brus, D. J., Stoorvogel, J. J., Heuvelink, G. B. M., and patterns of soils for sustainable agriculture in northern
de Vries, F.: Efficiency comparison of conventional and digital Ethiopia’s tropical mountains, PLoS ONE, 14, e0224041,
soil mapping for updating soil maps, Soil Sci. Soc. Am. J., 76, https://doi.org/10.1371/journal.pone.0224041, 2019.
2097–2115, https://doi.org/10.2136/sssaj2011.0424, 2012. Poggio, L., de Sousa, L. M., Batjes, N. H., Heuvelink, G. B. M.,
Kuhn, M.: Building predictive Models in R using the caret package, Kempen, B., Ribeiro, E., and Rossiter, D.: SoilGrids 2.0: pro-
J. Stat. Softw., 28, 1–26, https://doi.org/10.18637/jss.v028.i05, ducing soil information for the globe with quantified spatial un-
2008. certainty, SOIL, 7, 217–240, https://doi.org/10.5194/soil-7-217-
Leenaars, J. G. B., van Oostrum, A. J. M., and Ruiperez, G. M.: 2021, 2021.
Africa Soil Profiles Database, Version 1.2. A compilation of QGIS Development Team: QGIS Geographic Information System,
georeferenced and standardised legacy soil profile data for Sub- Open Source Geospatial Foundation Project, https://qgis.org/en/
Saharan Africa (with dataset), ISRIC Report 2014/01, Africa Soil site/ (last access: 17 August 2021), 2021.
Information Service (AfSIS) project and ISRIC – World Soil R Core Team: R: A Language and Environment for Statistical Com-
Information, Wageningen, https://library.wur.nl/WebQuery/isric/ puting, R Foundation for Statistical Computing, Vienna, http:
2259472 (last access: 7 August 2023), 2014. //www.R-project.org/ (last access: 14 September 2021), 2020.
Leenaars, J. G. B., Elias, E., Wösten, J. H. M., Ruiperez- Rossiter, D. G., Poggio, L., Beaudette, D., and Libohova, Z.:
González, M., and Kempen, B.: Mapping the ma- How well does digital soil mapping represent soil geogra-
jor soil-landscape resources of the Ethiopian High- phy? An investigation from the USA, SOIL, 8, 559–586,
lands using random forest, Geoderma, 361, 114067, https://doi.org/10.5194/soil-8-559-2022, 2022.
https://doi.org/10.1016/j.geoderma.2019.114067, 2020a. Sheleme, B.: Topographic positions and land use impacted soil
Leenaars, J. G. B., Ruiperez, M., González, M., Kempen, B., and properties along Humbo Larena-Ofa Sere toposequence, South-
Mantel, S.: Semi-detailed soil resource survey and mapping of ern Ethiopia, Journal of Soil Science and Environmental Man-
REALISE woredas in Ethiopia, Project report to the BENEFIT- agement, 8, 135–147, https://doi.org/10.5897/JSSEM2017.0643,
REALISE programme, December 2020, ISRIC – World Soil In- 2017.
formation, Wageningen, The Netherlands, https://www.isric.org/ Shi, J., Yang, L., Zhu, A.-X., Qin, C., Liang, P., Zeng,
projects/realise-survey-and-mapping-soil-resources (last access: C., and Pei, T.: Machine-Learning Variables at Different
18 October 2021), 2020b. Scales vs. Knowledge-based Variables for Mapping Multi-
McBratney, A. B., Santos, M. M., and Minasny, B.: On digital soil ple Soil Properties, Soil Sci. Soc. Am. J., 82, 645–656,
mapping, Geoderma, 117, 3–52, 2003. https://doi.org/10.2136/sssaj2017.11.0392, 2018.
Mesfin, A.: Nature and Management of Ethiopian Soils, 272 pp., Shimeles, D., Mohamed, A., and Abayneh, E.: Characteristics and
Alamaya University of Agriculture, Alamaya, Ethiopia, 1998. classification of the soils of Tenocha Wenchacher Micro catch-
Mishra, B. B., Gebrekidan, H., and Kibret, K.: Soils of Ethiopia: ment, South west Shewa, Ethiopia, Ethiopian Journal of Natural
Perception, appraisal and constraints in relation to food secu- Resources, 9, 37–62, 2007.
rity, International journal of food, agriculture and environment, Soil Science Division Staff: Soil survey manual, edited by:
2, 269–279, 2004. Ditzler, C., Scheffe, K., and Monger, H. C., USDA Hand-
Mitiku, H.: Genesis, characteristic and classification of the Central book 18, Government Printing Office, Washington, D.C.,
Highland soils of Ethiopia, PhD Thesis, 399 pp., State University USA, https://www.nrcs.usda.gov/sites/default/files/2022-09/
of Ghent, Belgium, 1987. The-Soil-Survey-Manual.pdf (last access: 6 October 2020),
Mohammed, A. and Belay, T.: Characteristics and classification 2017.
of the soils of the Plateau of Simen Mountains National Park Svetnik, V., Liaw, A., Tong, C., Culberson, J. C., Sheridan, R. P., and
(SMNP), Ethiopia, SINET, 31, 89–102, 2008. Feuston, B. P.: Random forest: a classification and regression tool
Mohammed, A. and Solomon, T.: Characteristics and fertility qual- for compound classification and QSAR modeling, J. Chem. Inf.
ity of the irrigated soils of Sheneka, Ethiopia, Ethiopian Journal Comp. Sci., 43, 1947–1958, https://doi.org/10.1021/ci034160g,
of Natural Resources, 12, 1–22, 2012. 2003.
Mulder, V. L., Lacoste, M., Richer de Forges, A. C., and Arrouays, Tamene, L., Erkossa, T., Tafesse, T., Abera, W., and Schultz, S.:
D.: GlobalSoilMap France: high resolution spatial modelling the A coalition of the willing powering data-driven solutions for
soils of France up to two meter depth, Sci. Total Environ., 573, Ethiopian agriculture, CIAT Publication No. 518, CIAT, Addis
1352–1369, 2016. Ababa, Ethiopia, 2021.
Mulualem, A., Gobezie, T. B., Kasahun, B., and Demese, M.: Tamene, L. D., Amede, T., Kihara, J., Tibebe, D., and Schulz, S.: A
Recent Developments in Soil Fertility Mapping and Fertilizer review of soil fertility management and crop response to fertil-
Advisory Services in Ethiopia, A Position Paper, https://www. izer application in Ethiopia: towards development of site- and
researchgate.net/publication/327764748/ (last access: 7 October context-specific fertilizer recommendation, CIAT Publication
2021), 2018. No. 443, International Center for Tropical Agriculture (CIAT),

SOIL, 10, 189–209, 2024 https://doi.org/10.5194/soil-10-189-2024

 A. Ali et al.: EthioSoilGrids 1.0 209

Addis Ababa, Ethiopia, https://hdl.handle.net/10568/82996 (last Wadoux, A. M. J. C., Minasny, B., and McBratney, A. B.: Ma-
access: 17 July 2021), 2017. chine learning for digital soil mapping: Applications, chal-
Tefera, M., Chernet, T., and Workineh, H.: Geological Map of lenges and suggested solutions, Earth Sci. Rev., 210, 103359,
Ethiopia, Addis Ababa, Ethiopia: Federal Democratic Republic https://doi.org/10.31223/osf.io/8eq6s, 2020.
of Ethiopia, Ministry of Mines and Energy, Ethiopian Institute of Westphal, E.: Agricultural Systems in Ethiopia, Agricultural Re-
Geological Surveys, Addis Ababa, Ethiopia, 1996. search Report 826, https://edepot.wur.nl/361350 (last access: 19
Tolossa, A. R.: Vertic Planosols in the Highlands of South-Western March 2021), 1975.
Ethiopia: Genesis, Characteristics and Use, Ghent University, WLRC-AAU (Water and Land Resource Centre-Addis Ababa Uni-
Faculty of Sciences, Ghent, Belgium, http://hdl.handle.net/1854/ versity): Land use/land cover mapping, change detection and
LU-5991501 (last access: 23 June 2021), 2015. characterization of Ethiopia, Water Land Resource Centre, Ad-
Vågen, T. G.: Africa Soil Information Service: Hydrologically Cor- dis Ababa University, Addis Ababa, Ethiopia, 2018.
rected/Adjusted SRTM DEM (AfrHySRTM), International Cen- Wright, M. N. and Ziegler, A.: Ranger: A fast implementation of
ter for Tropical Agriculture – Tropical Soil Biology and Fertil- random forests for high dimensional data in C++ and R, J. Stat.
ity Institute (CIAT-TSBF), World Agroforestry Centre (ICRAF), Softw., 77, 1–17, https://doi.org/10.18637/jss.v077.i01, 2017.
Center for International Earth Science Information Network Zwedie, E.: Selected physical, chemical, and mineralogical charac-
(CIESIN), Columbia University, https://cmr.earthdata.nasa.gov/ teristics of major soils occurring in Chercher highlands, Eastern
search/concepts/C1214155420-SCIOPS (last access: 18 Febru- Ethiopia, Ethiopian Journal of Natural Resources, 1, 173–185,
ary 2021), 2010. 1999.
Van de Wauw, J., Baert, G., Moeyersons, J., Nyssen, J., De Geyndt, Zewdie, E.: Properties of major Agricultural Soils of Ethiopia, Lam-
K., Nurhussen, T., Amanauel, A., Poesen, J., and Deckers, J.: bert Academic Publishing, Germany, 2013.
Soil-landscape relationships in the basalt-dominated highlands
of Tigay, Ethiopia, Catena, 75, 117–127, 2008.
Virgo, K. J. and Munro, R. N.: Soil and erosion features of the Cen-
tral Plateau region of Tigrai, Ethiopia, Geoderma, 20, 131–157,
1978.

https://doi.org/10.5194/soil-10-189-2024 SOIL, 10, 189–209, 2024

View publication stats