Pulse Brain · Growing Health Evidence Index
Tier 3 — Observational / field trialPeer-reviewed

African soil properties and nutrients mapped at 30 m spatial resolution using two-scale ensemble machine learning

Tomislav Hengl, Matt Miller, Josip Križan, Keith Shepherd, Andrew Sila, Milan Kilibarda, Ognjen Antonijević, Luka Glušica, Achim Dobermann, Stephan M. Haefele, S. P. McGrath, Gifty Acquah, Jamie Collinson, Leandro Parente, Mohammadreza Sheykhmousa, Kazuki Saito, Jean‐Martial Johnson, Jordan Chamberlin, Francis B.T. Silatsa, Martin Yemefack, John Wendt, R.A. MacMillan, Ichsani Wheeler, Jonathan H. Crouch

Scientific Reports · 2021

Read source ↗ All evidence

Summary

This study presents a continental-scale soil information system for Africa at 30 m spatial resolution, synthesising approximately 150,000 field soil samples with satellite Earth observation data and machine learning to predict 21 soil properties and classes at three depths. Using a two-scale 3D ensemble machine learning framework, the authors achieved varying accuracy levels across properties, with soil pH showing excellent predictability (CCC = 0.900) whilst extractable phosphorus and sulphur proved more challenging (CCC = 0.654–0.708). The resulting publicly available maps represent a substantial advance in soil data availability for the continent and identify climatic variables and Sentinel-2/Landsat spectral bands as the most important predictive covariates.

UK applicability

This study is primarily applicable to African soil mapping and agricultural planning. The methodological approach and machine learning framework could potentially be adapted for UK soil mapping, though the covariate importance (particularly climate) and sample density differ markedly from UK conditions, and UK soils are already characterised by existing national databases such as the National Soil Inventory.

Key measures

Soil pH, organic carbon, total nitrogen, total carbon, effective cation exchange capacity, extractable phosphorus, potassium, calcium, magnesium, sulphur, sodium, iron, zinc, silt, clay, sand, stone content, bulk density, and depth to bedrock at 0, 20, and 50 cm depths; model accuracy expressed as concordance correlation coefficient (CCC); spatial cross-validation performance

Outcomes reported

The study produced 30 m resolution maps of 21 soil properties and classes across Africa, including pH, organic carbon, nitrogen, micronutrients (zinc, iron, etc.), texture, and bulk density at three depths. Predictions were generated using ~150,000 soil samples and Earth observation data, with accuracy varying from soil pH (CCC = 0.900) to extractable phosphorus (CCC = 0.654).

Theme
Measurement & metrics
Subject
Soil health assessment & monitoring
Study type
Research
Study design
Field trial / Observational mapping study
Source type
Peer-reviewed study
Status
Published
Geography
Africa
System type
Other
DOI
10.1038/s41598-021-85639-y
Catalogue ID
MGmos8byf9-p227of

Topic tags

Pulse AI · ask about this record

Dig deeper with Pulse AI.

Pulse AI has read the whole catalogue. Ask about this record, its theme, or how the findings apply to UK farming and policy — every answer cites the underlying studies.