Biplots for understanding machine learning predictions in digital soil mapping

Aug 2, 2026

Upload by Aurelia

Digital soil mapping increasingly relies on machine learning for flexible, high-accuracy predictions, yet many models remain black-box and offer limited interpretability. Explainable machine learning (XML) visual methods such as partial dependence, independent conditional expectation, and Shapley-based plots can help, but they assume uncorrelated covariates, provide limited views of covariates, and lack readily available goodness-of-fit metrics. This study proposes a principal component analysis biplot as a model-agnostic approach that supports interpretation even with correlated covariates and provides an analytically derived goodness-of-fit metric, illustrated with random forest soil organic carbon mapping in South Africa.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Soil maps and digital soil mapping (DSM)

Machine learning models and interpretability challenges

Explainable machine learning (XML)

Model-agnostic vs model-specific XML methods

Ecological Informatics 84 (2024) 102892 | Biplots for understanding machine learning predictions in digital soil mapping
Stephan van der Westhuizena,b,e,∗, Gerard B.M. Heuvelinkb,c, Sugnet Gardner-Lubbe a,e, Catherine E. Clarke d
a Department of Statistics and Actuarial Science, Stellenbosch University, Stellenbosch, South Africa b Soil Geography and Landscape Group, Wageningen University, Wageningen, The Netherlands c ISRIC-World Soil Information, Wageningen, The Netherlands
d Department of Soil Science, Stellenbosch University, Stellenbosch, South Africa
e Centre for Multi-dimensional Data Visualisation (MuViSU), Stellenbosch University, Stellenbosch, South Africa | | | | | --- | --- | --- | --- | | A R T I C L E I N F O | | A B S T R A C T | | | Keywords:
Accumulated local effect Partial dependence
Shapley
Interpretable machine learning Principal component analysis XAI | | In digital soil mapping, machine learning is gradually replacing traditional statistical models because of their greater flexibility and better prediction performance. However, unlike traditional models, a notable drawback of machine learning models is that they are ‘‘black-box’’ in nature due to their limited ability to provide comprehensive interpretations for their predictions. Explainable machine learning (XML) methods provide visualisations that can be used to aid in understanding predictions made by machine learning models. Popular model-agnostic visualisation methods include partial dependence plots, independent conditional expectation curves, and partial dependence plots produced with Shapley values. These methods require that covariates are uncorrelated which could be restrictive. For cases where covariates are correlated, an alternative approach is the Accumulated Local Effect plot, which however is limited to depicting one or two covariates at a time. Another disadvantage of the above mentioned methods is that no readily available goodness-of-fit metric is available. In this paper we propose the use of a principal component analysis biplot as a model-agnostic method to gain insight into machine learning predictions in digital soil mapping. A biplot is a powerful visualisation tool that is used to seek patterns in multivariate data. A biplot does not require covariates included in the visualisation to be uncorrelated, and furthermore, an analytically derived goodness-of-fit metric is provided which allows the user to evaluate the accuracy of the approximation. We present examples from a case study in South Africa in which soil organic carbon is mapped with a random forest model. Our findings show that biplots can provide meaningful interpretations for predictions, making it a worthy addition to the XML toolkit. | | 1. Introduction Soil maps play a crucial role in various fields by providing valuable information about the spatial distribution of soil properties. A widely used tool for generating these maps is digital soil mapping (DSM) (McBratney et al., 2003), which often makes use of machine learning models like the random forest (RF) model (Minasny and McBratney, 2016). The reason for this wide use is that machine learning models can effectively capture complex nonlinear relationships between soil properties and environmental covariates, leading to more accurate soil maps compared to traditional statistical models such as multiple linear regression and geostatistical models (Wadoux et al., 2020a). However, unlike traditional models, a notable drawback of machine learning models is that they are often ‘‘black-box’’ in nature due to their limited ability to provide comprehensive interpretations for their predictions. In this paper, we adopted the definition provided by Belle and Papantonis (2021) for ‘‘black-box’’ machine learning models, i.e., models that are not simulateable by a human, lack decomposability, and are algorithmically nontransparent. This definition includes models like RF, support vector machin

FAQs

Why do machine learning models pose interpretability challenges in digital soil mapping?

Machine learning models such as random forests are often black-box, meaning they do not readily provide decomposable, human-simulateable explanations for how predictions are produced.

What limitations affect common model-agnostic XML visualization methods?

Many popular methods require covariates to be uncorrelated, may be limited in depicting only one or two covariates at a time, and do not offer an immediately available goodness-of-fit metric.

How does the proposed principal component analysis biplot improve understanding and evaluation of predictions?

The biplot is model-agnostic, does not require covariates to be uncorrelated, and includes an analytically derived goodness-of-fit metric to assess approximation accuracy.

Aug 2, 2026

Upload by Aurelia