Introducing the Russell Review ‘Artificial Intelligence in Soil Science’ by Alexandre M.J.-C. Wadoux

Abstract

Scientific progress is often driven by technological advancements, and soil science is no exception. Over recent decades, innovations in laboratory techniques—such as DNA sequencing, stable isotope analysis and spectroscopy—have greatly advanced soil chemistry and soil biology. Similarly, remote sensing, GIS and geocomputation have revolutionized soil mapping and enabled novel spatial analyses in soil geography more broadly. Sensor technologies, including electromagnetic induction and ground-penetrating radar, have significantly contributed to soil physics and soil hydrology and, when combined with robotics, have facilitated precision agriculture.

Computers, perhaps the most prominent technological advancement of all, have long since become indispensable in soil science, with many researchers spending more time in front of screens than in the field or laboratory—much to the regret of many. Until recently, computers primarily served as tools for storing data and maps, processing and analysing measurements, running mathematical and statistical models, and supporting the production of graphical and written reports. However, with the rise of Artificial Intelligence (AI), this is set to change dramatically. Computers are becoming ‘intelligent’, taking on more roles in soil science—whether we welcome it or not.

This Russell Review by Alexandre Wadoux examines the role of AI in soil science over past decades and explores how AI will transform soil science in the future. Through a systematic literature review, Alexandre identifies the soil science domains where AI has made the most significant contributions. The review begins by defining AI and categorizing its applications in soil science. While AI is used in soil science in diverse ways, its most notable sub-field is machine learning, which involves the development and application of complex, non-linear statistical regression and classification models. These models are trained on paired observations of the dependent and independent variables and subsequently used to predict outcomes for new, unseen cases. Today, machine learning is the dominant technique in digital soil mapping and is increasingly used in pedometrics for other purposes, such as developing pedotransfer functions and soil spectroscopy models. There is also the recent rise of physics-informed machine learning, which embeds physical laws and constraints into machine learning models and holds much promise for process-based, mechanistic modelling in soil science.

The senior editorial team was delighted when Alexandre accepted our invitation to publish this Russell Review as part of the EJSS 75th anniversary celebrations. A talented young pedometrician, he has already made a substantial impact on the field. He earned his doctorate from Wageningen University in 2019, focusing on spatial sampling design optimization, and then spent 4 years as a postdoctoral researcher at the University of Sydney. He is now based at INRAE in Montpellier, France. Alexandre received the 2021 Margaret Oliver Award, which recognizes emerging talents in pedometrics, currently chairs the Pedometrics Commission of the International Union of Soil Sciences, and has been an Associate Editor of EJSS since 2021. His contributions to explainable machine learning in digital soil mapping and other areas mark him as a leading figure in pedometrics.

Beyond highlighting AI's contributions, Alexandre's Russell Review also addresses its risks and limitations, including data quality concerns and ethical issues related to data sharing and harvesting. Toward the end of the review, Alexandre explores potential AI-driven advancements that could reshape soil science. He envisions AI playing a far greater role in scientific discovery, writing: ‘In the future, it is likely that soil science will be driven by AI-augmented researchers, where AI assists and accelerates nearly all the tasks involved in the scientific process’. However, he also reassures human researchers, stating: ‘Ultimately, the vision remains with the human, while the execution increasingly falls to AI’.

I sincerely hope Alexandre is right. When EJSS celebrates its 100th anniversary in 25 years, I hope human soil scientists remain at the forefront, guiding and advancing soil science research. Time will tell.

Gerard B. M. Heuvelink: conceptualization, writing – original draft, writing – review and editing.