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

Summary

This study presents a continental-scale Soil Information System for Africa at 30 m spatial resolution, integrating approximately 150,000 field soil samples with Earth Observation data through a two-scale 3D ensemble machine learning framework. The resulting maps provide predictions for 18 soil properties at three depths, with accuracy varying by property—soil pH predictions were most reliable (CCC = 0.900), whilst extractable phosphorus and sulphur were less predictable. Climate-derived covariates proved most important for predicting soil chemical variables, whilst spectral and topographic data from Sentinel-2, Landsat and digital terrain models dominated as the most important 30 m resolution variables.

UK applicability

This African soil mapping framework and methodology may inform development of more detailed soil property mapping systems for the UK, though direct application of the trained models would be limited to regions with similar soil-climate-geology characteristics. The approach demonstrates best practice for continental-scale soil information systems and could support efforts to enhance UK soil monitoring infrastructure, particularly given recent policy emphasis on soil health and natural capital assessment.

Key measures

Soil pH, organic carbon, total nitrogen, total carbon, effective Cation Exchange Capacity (eCEC), extractable phosphorus, potassium, calcium, magnesium, sulphur, sodium, iron, zinc, silt, clay, sand, stone content, bulk density, depth to bedrock; predictions at 0, 20 and 50 cm depths; concordance correlation coefficient (CCC) for cross-validation accuracy

Outcomes reported

The study produced 30 m spatial resolution maps of 18 soil properties and characteristics across the African continent, including pH, organic carbon, total nitrogen, extractable nutrients (P, K, Ca, Mg, S, Na, Fe, Zn), texture, bulk density and depth to bedrock at three soil depths. Predictive accuracy varied by property, with soil pH achieving the highest accuracy (CCC = 0.900) and extractable phosphorus and sulphur performing more poorly (CCC = 0.654 and 0.708 respectively).

Topic tags