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

Abstract

Soil property and class maps for the continent of Africa were so far only available at very generalised scales, with many countries not mapped at all. Thanks to an increasing quantity and availability of soil samples collected at field point locations by various government and/or NGO funded projects, it is now possible to produce detailed pan-African maps of soil nutrients, including micro-nutrients at fine spatial resolutions. In this paper we describe production of a 30 m resolution Soil Information System of the African continent using, to date, the most comprehensive compilation of soil samples ($$N \approx 150,000$$ N ≈ 150 , 000 ) and Earth Observation data. We produced predictions for soil pH, organic carbon (C) and total nitrogen (N), total carbon, effective Cation Exchange Capacity (eCEC), extractable—phosphorus (P), potassium (K), calcium (Ca), magnesium (Mg), sulfur (S), sodium (Na), iron (Fe), zinc (Zn)—silt, clay and sand, stone content, bulk density and depth to bedrock, at three depths (0, 20 and 50 cm) and using 2-scale 3D Ensemble Machine Learning framework implemented in the (Machine Learning in ) package. As covariate layers we used 250 m resolution (MODIS, PROBA-V and SM2RAIN products), and 30 m resolution (Sentinel-2, Landsat and DTM derivatives) images. Our fivefold spatial Cross-Validation results showed varying accuracy levels ranging from the best performing soil pH (CCC = 0.900) to more poorly predictable extractable phosphorus (CCC = 0.654) and sulphur (CCC = 0.708) and depth to bedrock. Sentinel-2 bands SWIR (B11, B12), NIR (B09, B8A), Landsat SWIR bands, and vertical depth derived from 30 m resolution DTM, were the overall most important 30 m resolution covariates. Climatic data images—SM2RAIN, bioclimatic variables and MODIS Land Surface Temperature—however, remained as the overall most important variables for predicting soil chemical variables at continental scale. This publicly available 30-m Soil Information System of Africa aims at supporting numerous applications, including soil and fertilizer policies and investments, agronomic advice to close yield gaps, environmental programs, or targeting of nutrition interventions.

Earth Observation Data-Driven Cropland Soil Monitoring: A Review

“…Nowadays, a series of conventional digital soil mapping approaches are used to produce coarse spatial resolution products and reflect the spatial variation of soil variables. For instance, SoilGrids 2.0 provides soil information (250 m) for the globe with quantified spatial uncertainty [71], while Hengl et al [72] recently provided African soil properties and nutrients mapped at 30 m resolution. Similarly, Fathololoumi et al [73] improved the digital soil maps in Iran by making use of multitemporal Landsat-8.…”

iSDAsoil: The first continent-scale soil property map at 30 m resolution provides a soil information revolution for Africa

“…The increase in map resolution has been driven by the incorporation of high-resolution satellite data from Sentinel 2 and Landsat Satellites, both of which provide images at 30 m or finer resolution. A number of different spectral wavelength bands from the Sentinel 2 satellite were important for predicting soil properties, including shortwave infrared (bands 9, 11, and 12; [ 4 ]). While these wavelengths are well known for monitoring of vegetation, their importance in predicting soil properties was previously less clear.…”

“…Many farms are smallholdings, and fields are smaller and more heterogenous when compared with farms in other continents such as North America. iSDAsoil is the world’s first continental-scale soil property map at 30 m resolution and provides soil information at a resolution that approaches the true variability on the ground [ 4 ]. For technical information on the workflow used, please visit our Technical Information page .…”

“…A selection of over 100,000 geolocated soil samples were used for model training, collected across more than 10 different datasets. An ensemble machine learning approach was used, alongside an extensive collection of more than 100 covariate layers [ 4 ] to predict soil properties for more than 24 billion pixels.…”

High Resolution Water Table Modeling of the Shallow Groundwater Using a Knowledge-Guided Gradient Boosting Decision Tree Model

“…Relationships between the target variable and the explanatory variables are established via the decision trees which, once trained, can be generalized to make predictions of the target variable at all grids. This framework has been successfully applied across the geosciences to model water chemistry indicators (Tesoriero et al, 2015;Erickson et al, 2021) soil properties (Møller et al, 2017;Hengl et al, 2021), subsurface redox conditions (Close et al, 2016;Koch et al, 2019a), water table depth (Bechtold et al, 2014;Koch et al, 2019b), and other variables. In such modeling frameworks, uncertainty can be quantified via quantile regression (López López et al, 2014;Tyralis et al, 2019a).…”