Soil property maps of Africa at 250 m resolution

Share on:
Comparison of predicted soil organic carbon content (fine earth) for an area around the town of Arusha (Tanzania): SoilGrids1km (left) and AfSoilGrids250m (right)
End year
2015

Important notice:

To access the updated and globally integrated maps of soil properties and classes please refer to: www.soilgrids.org
 

Background

Over the period 2008–2014, the AfSIS project has compiled two soil profiles / samples datasets: the Africa Soil Profiles database [Leenaars, 2014] holding legacy soil profiles data and the Sentinel Sites database [Vagen et al, 2010] holding newly collected topsoil data, jointly consisting of ca. 28 thousand sampling locations (or over 85 thousand samples). Using these soil point observations & measurements and an extensive collection of global (SoilGrids1km) and continental (Africa) environmental covariates, ISRIC - World Soil Information, in collaboration with The Earth Institute, Columbia University, World Agroforestry Centre, Nairobi and the International Center for Tropical Agriculture (CIAT), has produced (February 2015) predictions of soil properties — organic carbon, pH, sand, silt and clay fractions, coarse fragments, bulk density, cation-exchange capacity, total nitrogen, exchangeable acidity, exchangeable bases (Ca, Mg, K, Na) and extractable aluminium — for the whole African continent at 250 m spatial resolution at either two or six standard soil depths. The predictions are obtained using an automated mapping framework (3D regression-kriging based on random forests). Compressed GeoTiffs of the soil property maps, together with all metadata are available for download from here. A web-mapping interface to the maps is available via: http://af.soilgrids.org/. Read more about how were these maps made

fig_africa_points_only.png?

Distribution of soil sample locations in Africa used to build spatial predictive models: (left) legacy soil profile observations (Africa Soil Profiles database, AfSP) showing ca. 18.5 thousand locations [Leenaars, 2014], and (right) AfSIS Sentinel Sites showing ca. 9.5 thousand locations, clustered at ca. 60 sentinel sites [Vagen et al, 2010].
 

Data download and metadata

For download of the data related to this project, please see the 'Related Links' section at the bottom of this page.

The AfSIS project has been funded by the Bill and Melinda Gates Foundation and the Alliance for a Green Revolution in Africa (AGRA). To learn more about the AfSIS project, visit the project website Africasoils.net and/or the AGRA website. Please cite as:

This data set is available under the Attribution-NonCommercial International CC BY-NC. This means that: you are free to share (copy and redistribute the material in any medium or format) and adapt (remix, transform, and build upon the material), as long as you give appropriate credit and provide a link to the SoilGrids.org homepage; you may not use these materials for commercial purposes.

Inputs: Africa Soil Profiles database (AfSIS), Africa Sentinel Sites soil database (AfSIS) Africa Soil covariates (250 m), SRTM DEM and derivatives (250 m), GlobLand30 land cover data (250 m), SoilGrids1km predictions.
Period (temporal coverage approximate): 1950–2012.
Spatial resolution (covariates): 250 m and 1 km.
Producers: ISRIC - World Soil Information, in collaboration with The Earth Institute (Columbia University), ICRAF - World Agroforestry Centre, Nairobi and the International Center for Tropical Agriculture (CIAT).
Data license (IP policy): Attribution-NonCommercial International CC BY-NC

 

Specifications AfSoilGrids250m

Output maps comply with the GlobalSoilMap.net/specifications (except for uncertainty quantification and spatial resolution, which has been set at 250 m). Each layer (one compressed GeoTiff) provides predictions of one of the soil properties at one of the standard depths. Maps of uncertainty (upper and lower limits of 90% prediction limits) are not provided. Only a summary accuracy measures (RMSE) is provided (see summary Table). Two examples illustrate the naming convention:

  • ORCDRC_T__M_sd1 - indicates predicted organic carbon content in permilles ("ORCDRC") for the first standard depth (0?5 cm) using a tiling system ("T");
  • EMGX_T__M_xd1 - indicates predicted exchangeable Mg in soil for the first non-standard depth (0?20 cm);

Each GeoTiff comes with a metadata XML file (following the Inspire metadata scheme), so that users are advised to obtain and use both the data and metadata files together:

The six standard depth intervals in the GlobalSoilMap.net/specifications are:

  • sd1: 0-5 cm
  • sd2: 5-15 cm
  • sd3: 15-30 cm
  • sd4: 30-60 cm
  • sd5: 60-100 cm
  • sd6: 100-200 cm

The two ‘non-standard’ depth intervals, as sampled at the AfSIS sentinel sites, are:

  • sx1: 0-20 cm
  • sx2: 20-50 cm

Typical GDALInfo output is:

  • rows        31505
  • columns     29501
  • bands       1
  • lower left origin.x        -3977625
  • lower left origin.y        -4321625
  • res.x       250
  • res.y       250
  • driver      GTiff
  • projection  +proj=laea +lat_0=5 +lon_0=20 +x_0=0 +y_0=0 +datum=WGS84 +units=m +no_defs
  • file        af_ORCDRC_T__M_sd1_250m.tif

 

Spatial reference AfSoilGrids250m

Because the interpolation is done in 3D, the grid cells are three dimensional blocks (or voxels) with different thicknesses for each standard depth (e.g. 250 m by 250 m by 5 cm for the first layer; 250 m by 250 m by 10 cm for the second). In all cases, the predictions refer to the centre of the voxel (e.g. "sd1" refers to a depth of 2.5 cm, while “sd2” refers to 10 cm depth).

All maps are projected in the pan-African equal area coordinate system with the following proj4 string:

"+proj=laea +lat_0=5 +lon_0=20 +x_0=0 +y_0=0 +units=m +ellps=WGS84 +datum=WGS84"

Predictions have been made only for areas with vegetation cover. i.e. all deserts and shifting sand areas were excluded using a soil mask. We derived the soil mask for Africa by using the GlobeLand30 data set, by removing all pixels that have >10% of bare land (class 90 in GlobeLand30) and/or >30% of water cover (class 60 in GlobeLand30). The non-vegetated land areas were masked out because soil prediction models could not reliably be obtained from the very few soil point observations non-vegetated land. Extrapolation of models calibrated on vegetated land would lead to unrealistic and erroneous predictions.

comparisonpredictedsoilco2.png

Comparison of predicted soil organic carbon content (fine earth) for an area around the town of Arusha (Tanzania): SoilGrids1km (left) and AfSoilGrids250m (right). Vector lines data source: OpenStreetMap. See also a visual comparison in this video.

 

Mapping accuracy AfSoilGrids250m

The accuracy of the maps (random forests part of the model) was assessed using 5-fold cross-validation, with model re-fitting, and reported using RMSE and % variance explained. The latter is defined as 1-MSE/sigma-square, where MSE is the mean square error at cross-validation points and sigma-square is the variance of the target variable. Some soil properties were log-transformed prior to prediction using regression kriging. In such cases the maps are the back-transformed regression kriging maps, but the % variance explained was derived in the transformed (trans-Gaussian) space which differ from the % variance explained in the original space. The figure and table below shows scatter plots and summary statistics of cross-validation results.

mappingaccuracy.png

Example of a mapping accuracy assessment plot for soil organic carbon, pH and bulk density (obtained with 5-fold cross-validation).

Table: Mapping accuracy for AfSoilGrids250m maps assessed using 5-fold cross-validation (using only random forests models).

Code

Soil property name

Units

Sample size

Range
(99% prob.)

RMSE

% variance
explained

ORCDRC

Organic carbon

g kg-1

64,010

0.9–42‰

10.6

61.3

PHIHOX

Soil pH in H2O

-

68,458

4.4–8.7

0.67

66.9

CRFVOL

Coarse fragments volumetric

m3 100 m-3

39,206

7–90%

18.4

20.3

SNDPPT

Soil texture fraction sand

g 100 g-1

55,578

7–94%

15.9

61.1

SLTPPT

Soil texture fraction silt

g 100 g-1

54,164

1–47%

8.3

56.1

CLYPPT

Soil texture fraction clay

g 100 g-1

54,167

3–73%

13.7

52.4

BLD

Bulk density (fine earth)

kg dm-3

8732

0.9–1.9

0.14

70.4

CEC

Cation Exchange Capacity

cmol+ kg-1

47,875

1.2–57

7.9

66.3

NTO

Total nitrogen

g kg-1

50,997

0.1–3.1‰

0.69

61.0

ALUM3S

Extractable Aluminium

mg kg-1

18,055

150–1800

160

86.3

EACKCL

Exchangeable acidity

cmol+ kg-1

24,242

0–6.4

1.30

77.3

ECAX

Exchangeable Calcium

cmol+ kg-1

65,158

0.1–46

12.7

67.2

EXKX

Exchangeable Potassium

cmol+ kg-1

64,518

0.01–2.4

0.60

58.6

EMGX

Exchangeable Magnesium

cmol+ kg-1

63,261

0.04–15

2.5

60.0

ENAX

Exchangeable Sodium

cmol+ kg-1

42,986

0–8.3

3.6

46.7

EXBX

Sum of exchangeable bases

cmol+ kg-1

64,270

0.31–66

11.0

68.8

 

Updates and improvements

The Africa soil property maps at 250 m resolution were produced using an automated spatial prediction framework, which has been fully documented and allows for reproducible research. The maps will be regularly updated and improved using an extended list of covariates and improved spatial prediction methods. It is therefore important to refer to the date of access when using these maps for modelling purposes, as there will be differences between different versions of the maps. Before using these maps for decision making purposes, please read the general disclaimer. If you discover any inconsistencies, gross errors or artifacts please report these to the SoilGrids production team.

We are aware of multiple artifacts and inconsistencies in the version of AfSoilGrids250m predictions released on February 10th 2015 and are working on updating maps in periods of less than 3 months. A detailed quality control report can be downloaded from here.

fig_AfSoilGrids250m_scheme.png?

Spatial prediction scheme used to produce AfSoilGrids250m data. Spatial predictions in the case of an automated soil mapping system can be continuously updated by adding new soil field observations and new covariates..

Possible uses of these maps include, but are not limited to:

  • Soil-environmental modelling, land degradation studies, soil-landscape planning, biodiversity assessment (continental scale or country scale models);
  • General assessment of soil characteristics (baselines) for the African continent (e.g. total carbon stock) and further planning of soil surveys / new soil sampling campaigns;
  • Downscaling and/or merging of coarse resolution maps with finer resolution maps (100 m, 30 m).
Scale