Summary
This paper presents a multivariate random forest approach to digital soil mapping that simultaneously models multiple soil properties whilst preserving their interdependence relationships. By comparing MRF with independent univariate random forest models applied to SOC and TN mapping, the authors demonstrate that the multivariate approach produces more realistic C:N ratios without sacrificing prediction accuracy. The work addresses a notable gap in machine learning applications to soil mapping, showing that accounting for covariance between soil properties improves both consistency and plausibility of multi-property soil maps.
UK applicability
The methodology could be applicable to UK soil survey and monitoring programmes, particularly for characterising multiple soil properties in spatially variable landscapes. However, applicability would depend on availability of relevant environmental covariates and calibration data specific to UK soil types and conditions.
Key measures
Root mean square error (RMSE) for SOC, TN, and C:N ratios; conditional distributions estimated via quantile regression forest; dependence structure preservation between soil properties
Outcomes reported
The study compared multivariate random forest (MRF) models with univariate random forest models for mapping soil organic carbon (SOC) and total nitrogen (TN) simultaneously. The MRF approach better preserved the dependence structure between SOC and TN, resulting in more realistic carbon-to-nitrogen (C:N) ratios whilst maintaining comparable prediction accuracy.
Topic tags
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.