Developing regional soil micronutrient management strategies through ensemble learning based digital soil mapping

Summary

This study developed a hybrid ensemble machine learning approach to produce fine-resolution digital soil maps of four micronutrients across the Indo-Gangetic Plain using 1778 soil samples and 52 environmental covariates. The ensemble model outperformed all 14 individual machine learning algorithms tested, with particularly low prediction uncertainty for zinc and iron. The work demonstrates potential to identify regional hotspots for targeted biofortification interventions and to support nutrient-based policy decisions in resource-constrained settings.

UK applicability

The ensemble machine learning methodology for digital soil micronutrient mapping could be adapted to UK soil conditions and agro-climatic regions; however, the direct applicability of findings is limited as UK soils, crop varieties, and micronutrient deficiency patterns differ substantially from the Indo-Gangetic Plain. The policy framework advocating nutrient-based subsidies and micronutrient recommendations may inform UK soil health and food security discussions.

Key measures

Spatial prediction of soil available Zn, Cu, Fe, and Mn at 150 m resolution; machine learning model accuracy comparison (14 base learners and ensemble); R² values for soil-grain micronutrient relationships (0.52–0.63 for Zn and Fe); prediction uncertainty maps

Outcomes reported

The study produced high-resolution digital soil maps of available zinc, copper, iron, and manganese across four agro-climatic regions of the Indo-Gangetic Plain, and explored relationships between soil micronutrient concentrations and rice grain biofortification potential.

Topic tags