SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty

← Back to papers

Physical Sciences › Earth and Planetary Sciences › Earth-Surface Processes

★

☆

SHARE

Overview

Paper Summary

Conflicts of Interest

Identified Weaknesses

Rating Explanation

Good to know

Topic Hierarchy

File Information

Paper Summary

Paperzilla title

SoilGrids 2.0: A Better Global Soil Map (But Still Needs More Data!)

SoilGrids 2.0 uses machine learning and a large global dataset of soil observations with environmental covariates to produce updated maps of soil properties like pH, organic carbon content, and texture. The maps include estimates of uncertainty and show improved performance compared to previous versions, but highlight the need for more soil data, particularly at high latitudes.

Possible Conflicts of Interest

The authors acknowledge core funding from the Netherlands Government and support from the European Union's EU H2020 Research and Innovation Programme for a related project. However, these are general funding sources and there is no indication of specific influence on the research presented.

Identified Weaknesses

Potential bias in cross-validation results

The evaluation metrics derived from cross-validation may not be unbiased estimators of the true numeric evaluation metrics due to non-uniform distribution of soil observations across the globe.

Simplistic approach to vertical soil variability

The vertical dimension of soil variability is addressed using depth of observation as a covariate, which may be too simple and lead to inconsistency over the predicted depth sequence, especially for datasets where short-range spatial variability is similar to vertical variability.

Use of legacy soil data

The study uses legacy soil data, which may not reflect current soil conditions due to changes in land use or management practices, particularly for dynamic soil properties like pH and soil organic matter content.

Lack of consideration for age of observations

The study doesn't address the age of observations, which could impact the accuracy for dynamic soil properties significantly affected by changes in land use or management.

Uncertainty quantification limitations

The uncertainty assessment uses quantile random forests, which may overestimate uncertainty for coarse fragments and underestimate it for sand. The uncertainty for silt and clay might be overestimated. The transformation method used to derive prediction intervals for texture components could contribute to these inaccuracies.

Limited use at local scale

The study acknowledges that SoilGrids250m is not intended for detailed, subnational, or local-scale use as more comprehensive local datasets are usually available for those applications.

Uncertainty visualization limitations

The visualization of uncertainty as maps using the interquartile range over the median ratio might not be fully accurate, particularly for properties with values near zero, like coarse fragments.

Limited scope of soil properties

The study only models some primary soil properties, and more work is needed to map other crucial information such as soil thickness, hydrologic soil properties, and complex properties like carbon stocks. These properties are important for modeling soil functions and Earth system modeling.

Rating Explanation

This paper presents a significant improvement in global digital soil mapping using updated methodology and data. Despite limitations related to cross-validation, uncertainty estimations, and reliance on legacy data, the study provides valuable global soil property maps at a medium resolution with quantified uncertainty, valuable for large-scale applications. The transparent acknowledgment of limitations and proposed future directions further strengthen the research.

Good to know

This is our free standard analysis. Paperzilla Pro fact-checks every citation, researches author backgrounds and funding sources, and uses advanced AI reasoning for more thorough insights.

Explore Pro →

Topic Hierarchy

Domain:

Physical Sciences

Field:

Earth and Planetary Sciences

Subfield:

Earth-Surface Processes

File Information

Original Title:

SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty

File Name: