Stacking

Combines multiple ML models (stacking/averaging) to reduce errors and improve stability of soil maps and uncertainty.

This project improves digital soil mapping by combining multiple models instead of relying on a single algorithm. We train diverse base learners and fuse them using model averaging or Super Learner (stacking) to reduce variance and stabilize predictions across regions, properties, and sampling designs. Evaluation uses spatially informed validation to ensure realistic performance and to avoid optimistic bias. Outputs include improved soil maps and, where feasible, uncertainty indicators based on ensemble spread.

Main focus

  • Reduce single-model risk by combining complementary learners.
  • Improve robustness and stability of soil predictions across settings.
  • Provide reliability signals using ensemble-based uncertainty summaries.

Objectives

  1. Train a diverse set of base models for key soil properties.
  2. Combine models via stacking/averaging under spatially robust validation.
  3. Deliver improved prediction maps plus uncertainty from ensemble behavior.

Graphical abstract

:::: {.row .justify-content-sm-center} ::: {.col-sm-10 .mt-3 .mt-md-0}

::: ::::

::: caption Graphical abstract summarizing the workflow and key results. :::

Papers (selected)

Paper 1 — Local Ensemble Modeling of Cropland SOC

DOI




Paper 2 — Stacked ML Improves SOC Mapping

DOI




Paper 3 — Minimum-Variance Model Averaging for Digital Soil Mapping

DOI



References