Data assimilation methods for cost-effective soil organic carbon monitoring

Overview

Soil organic carbon (SOC) is a major carbon sink in agricultural soils, and carbon markets increasingly rely on SOC measurements to reward farmers for adopting soil-friendly practices. However, intensive soil sampling across crop production areas is expensive and represents a significant barrier to technology adoption and policy implementation.

This research addresses that barrier by applying advanced data assimilation (DA) methods to process-based SOC models. DA is a rigorous methodology that incorporates real SOC measurement data into model simulations, improving estimates and reducing forecast bias caused by model parameter uncertainties. The result is more accurate SOC estimation with significantly reduced sampling requirements, lowering monitoring costs and creating more favourable conditions for carbon market participation and public policy deployment.

Technical specifications

Core approach:

• State-of-the-art non-linear DA methods, including the Unscented Kalman Filter, Unscented Particle Filter, and Moving Horizon Estimation, will be used to incorporate SOC stocks data or chronosequences into two widely used SOC dynamics models: Century and RothC
• Models and DA filters will be implemented in Python for accessibility and reproducibility
• The accuracy of DA-enhanced estimates and forecasts will be benchmarked against raw measurements and against SOC model predictions without DA
• Data sources include EMBRAPA fields in Brazil, data provided by identified project partners, and published datasets

Key benefits:

• Significantly fewer soil samples needed for reliable SOC estimation, reducing monitoring costs
• Improved spatial and temporal SOC forecasts through model-data integration
• Rigorous quantification of forecast uncertainty for better-informed decision-making
• Flexible methodology applicable across diverse agricultural contexts and data availability scenarios

Technology readiness level

The research is at an early-to-mid stage of development. Existing literature applying DA to SOC estimation is limited; prior work using the Ensemble Kalman Filter with synthetic data demonstrated a 68% reduction in mean squared error compared to measurements alone. This project extends that foundation by comparing multiple DA filters and SOC models, and by validating against real-world field data from EMBRAPA and partner sources. The methodology is positioned for advancement toward practical deployment in carbon market monitoring frameworks and agricultural policy instruments.