Physics-Informed and Explainable AI for Soil Health Modelling

Artificial intelligence (AI) and machine learning (ML) are revolutionizing the way we understand and predict soil processes. Yet, while data-driven models excel at pattern recognition, they often act as “black boxes,” offering little transparency into how predictions are made and sometimes ignoring the physical principles that govern soil systems. To address these challenges, this Research Topic focuses on the integration of physics-informed and explainable AI (XAI) approaches to advance soil health modelling—linking data science with process understanding to build reliable and interpretable models for sustainable soil and crop management.

Physics-informed AI combines the flexibility of machine learning with established knowledge of soil physics, chemistry, and biology. By embedding mechanistic laws—such as mass balance, energy conservation, and nutrient cycling—into data-driven architectures, these models maintain consistency with real-world processes while improving prediction accuracy and generalization. Meanwhile, explainable AI enhances transparency and trust by revealing the underlying drivers of model outcomes, enabling users to interpret how soil properties, environmental variables, and management practices influence soil health indicators.

This Research Topic invites contributions that explore, develop, or apply hybrid, physics-informed, or explainable AI frameworks for modelling soil health, nutrient dynamics, and ecosystem functions across multiple scales. We especially encourage studies that integrate remote sensing, proximal sensing, laboratory measurements, and digital soil maps to improve predictions of soil organic carbon, nutrient availability, and biological indicators. Research linking soil modelling to agronomic decision support, fertiliser efficiency, or climate-smart agriculture is highly encouraged, particularly those demonstrating practical or industry applications.

Potential contributions may include:
• Physics-informed neural networks and hybrid models for soil carbon and nutrient cycling
• Explainable machine learning techniques for soil property prediction and soil health index development
• Multi-source data fusion and digital twin approaches for soil–plant–atmosphere modelling
• Model transferability and uncertainty quantification across environmental gradients and management systems
• Case studies integrating AI-based soil models into digital farming platforms or agronomic advisory tools

By bringing together soil scientists, agronomists, AI researchers, and industry experts, this Research Topic seeks to accelerate the development of transparent and physically consistent AI systems for soil health assessment and management. These models will not only enhance scientific understanding of soil processes but also provide interpretable and actionable insights for practitioners—supporting sustainable land management, improved nutrient use efficiency, and global soil health resilience.