We’ve created predictions for a number of different chemical, physical and agronomic soil properties. Where we’ve predicted numerical values, we have also included estimates of uncertainty. For categorical data and some other third-party layers, no uncertainty measurements are available.

The below table refers to data available via the iSDAsoil API. For data available via bulk download, visit the iSDAsoil on AWS page.

A detailed description of our workflow is available on the technical information page.

A lot of the data that was the basis of iSDAsoil was created by grant-funded projects. We felt that the best thing to do was to make this data easily accessible and readily useable. Even as a profit-seeking company, we pursue a business model around advisory services built upon the foundations of publicly generated data. 

We are keen to create maps that are constantly improving. If you’re aware of some legacy soil data or have been involved in data collection, please let us know. We’ll be happy to integrate it and provide acknowledgment.

Contact us to share information about your dataset, and we’ll be in touch.

You can get in touch via email at our Contact us page

We have covered all areas suitable for agriculture across Africa. For the majority of properties, we have not made predictions for locations classed as deserts or water bodies. You can check the map to see where predictions were made for each layer.

We’ve created a partial implementation of the soil FCC framework, as developed by Sanchez et al. (2013), in order to assess soil quality in the tropics. The idea behind this layer is to inform users about possible constraints that may need to be addressed in order to make soil productive and fertile. As data was not available for all soil constraints, we’ve produced a subset of the total constraints, with minor modifications:

Shallow: Depth to bedrock less than 50cm
Gravel: Stone content more than 10%
Slope: slope angle more than 15°

High erosion risk: Textual discontinuity (any of the following):

• USDA Texture class of 0-20cm layer = Sand or Loamy Sand, and USDA Texture class of 20-50cm layer = Sandy clay, Silty Clay or Clay
• USDA Texture class of 0-20cm layer = Sand, Loamy sand, Sandy loam, Loam, Silty loam, Silt, or Sandy clay loam, and USDA Texture class of 20-50cm layer = Clay Loam, Silty clay loam, Sandy Clay, Silty Clay, or Clay

High erosion risk: Shallow depth (any of the following):

• USDA Texture class of 0-20cm layer = Clay Loam, Silty clay loam, Sandy Clay, Silty Clay or Clay, and depth to bedrock less than 50cm
• USDA Texture class of 0-20cm layer = Sand, Loamy sand, Sandy loam, Loam, Silty loam, Silt or Sandy clay loam, and depth to bedrock less than 50cm
• USDA Texture class of 0-20cm layer = Sand or Loamy Sand, and depth to bedrock depth less than 50cm

High erosion risk: Steep slope: slope more than 30°
Sulfidic: pH of 0-20cm or 20-50cm layer less than 3.5
Aluminium toxicity: Organic carbon less than 120g/kg in 0-20 or 20-50cm layer, and pH between 3.5 and 5.5 in 0-20cm or 20-50cm layer
Calcareous: pH more than 7.3 in 0-20 or 20-50cm layer
Low potassium reserves: extractable potassium less than 78ppm, and pH less than 7.5 in 0-20cm or 20-50cm layer
High leaching potential: Cation exchange capacity less than 4cmol/kg in 0-20cm or 20-50cm layer
No constraints: if none of the above are satisfied. Other constraints may be present that we have not mapped.

The majority of the soil property layers were created using machine learning and are therefore predictions. For each prediction we also include an estimate of the prediction error, or uncertainty – i.e. how certain the algorithms are that the predicted value is correct. We appreciate that higher resolution does not necessarily equate to higher accuracy, which is why we communicate the uncertainty. R-squared values for each model can be found here.

For users of the iSDAsoil API, uncertainty is returned at 3 different confidence intervals, representing 50%, 68% (1 standard deviation) and 90% quantile intervals. These confidence intervals are defined using an upper and lower boundary value, which captures the percentage chance that the predicted value falls between these two values. The closer the upper and lower boundary values to the predicted value, the more certain the prediction, with 50% representing the most stringent threshold, and 90% representing the least. 

When creating models for each soil property, we had minimal accuracy thresholds. Where the predictive models weren’t accurate enough to give useful data, the soil property was not included. 

The “Variability” displayed on the map refers to 1 standard deviation on either side of the predicted value, capturing 68% of the total variability of the prediction (see previous question for explanation of confidence intervals).

We have used harmonization methods based on the literature, as well as expert input from researchers at Rothamsted Research. Many of the data points were from projects that used similar analytical techniques and were therefore directly comparable. We have done as much as we can to reduce errors and to harmonise the data. 

For a more detailed description of the method used, see the technical information page. 

Yes, the API accepts lat/lon queries. You can also search on the iSDAsoil map by providing the lat,lon as comma-separated values.

Yes. We have uploaded all of the soil property layers are covariates used in the predictions to the AWS Registry of Open Data. For more information, visit https://www.isda-africa.com/isdasoil/isdasoil-on-aws/

Yes. You can use the data on the AWS Registry of Open Data to do this – visit the iSDAsoil on AWS page for more details.

All data are freely available under a Creative Commons Attribution 4.0 International License.

If you have any another questions, you can contact us via email.