Domain Adaptation for Sustainable Soil Management using Causal and Contrastive Constraint Minimization

TL;DR

This work tackles the challenge of mapping soil organic matter (OM) across regions when sensor data are costly by combining remote sensing with sparse soil observations. The authors introduce a domain adaptation framework based on causal constraint minimization (CACM) and contrastive learning, using sensor data as an auxiliary training signal to preserve causal relations and capture location-specific structure for improved out-of-distribution generalization. Key contributions include extending CACM to continuous attributes, integrating contrastive embeddings to exploit spatial heterogeneity, and performing interpretability analyses to identify soil attributes driving OM changes. The approach enables scalable OM estimation across data-poor regions and informs targeted data collection and soil-management decisions, with practical implications for climate resilience and sustainable land stewardship.

Abstract

Monitoring organic matter is pivotal for maintaining soil health and can help inform sustainable soil management practices. While sensor-based soil information offers higher-fidelity and reliable insights into organic matter changes, sampling and measuring sensor data is cost-prohibitive. We propose a multi-modal, scalable framework that can estimate organic matter from remote sensing data, a more readily available data source while leveraging sparse soil information for improving generalization. Using the sensor data, we preserve underlying causal relations among sensor attributes and organic matter. Simultaneously we leverage inherent structure in the data and train the model to discriminate among domains using contrastive learning. This causal and contrastive constraint minimization ensures improved generalization and adaptation to other domains. We also shed light on the interpretability of the framework by identifying attributes that are important for improving generalization. Identifying these key soil attributes that affect organic matter will aid in efforts to standardize data collection efforts.

Figures (9)

• Figure 1: Domain Adaptation framework for Organic Matter Modeling using Causal and Contrastive Constraint Minimization. Conditioning the satellite image embedding using soil attribute embeddings provides additional context or guidance on how the underlying soil properties impact OM.
• Figure 2: Domain Adaptation framework for Organic Matter Modeling using Causal and Contrastive Constraint Minimization. The bi-level optimization scheme first enforces causal independence constraints and then modifies the embeddings via contrastive learning.
• Figure 3: Standardized MSE Gain on Variable Removal
• Figure 4: Sites with Soil Attribute Information
• Figure 5: Pairplot showing relationships between different soil variables.