Soil Mapping
Maps soil properties from samples using covariates and spatial validation, producing high-resolution predictions with confidence.
This project converts point-based soil observations into continuous, high-resolution maps that capture soil variability more realistically than traditional polygon mapping. It integrates soil measurements (e.g., SOC, texture, pH, salinity-related indicators, CaCO₃, CEC, and hydraulic properties) with environmental covariates (terrain, climate, land cover, geology, and remote sensing) to train machine-learning models that learn nonlinear soil–landscape relationships. Performance is evaluated using spatially informed validation to ensure realistic generalization, and outputs can include uncertainty/confidence layers to highlight where predictions are reliable and where additional sampling is most valuable.
Main focus
- Robust and scalable spatial prediction of multiple soil properties.
- High-resolution soil maps for management and environmental applications.
- Reliability-aware outputs (confidence/uncertainty) for trustworthy decisions.
Objectives
- Integrate multi-source covariates with soil observations to build prediction-ready datasets.
- Train and evaluate ML models using spatially structured validation to avoid optimistic bias.
- Produce prediction maps (and optional confidence layers) that support decision-making and sampling design.
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 — Imbalanced Soil Texture Class Prediction
Paper 2 — UMAP-Enhanced Machine Learning for Soil Class Delineation
Paper 3 — Data Mining Classifiers for USDA Soil Family Mapping
