Sensitivity analysis to constrain soil microbial activities in a land model

Overview

This research applies variance-based sensitivity analysis and Gaussian process emulation to constrain soil microbial activity parameters in land surface models. Using five years of weekly field measurements—including soil organic carbon, nitrogen species, moisture, temperature, greenhouse gas fluxes (CO2, CH4, N2O), and NH3—from a corn field in Georgia under multiple agricultural practices (conventional, cover crop, intercrop) and three active ingredient application rates, the work aims to optimize microbial activity coefficients in the Community Land Model and Landscape DNDC model. The approach enables more accurate quantification of soil carbon sequestration potential and greenhouse gas emissions under varying management conditions.

Technical specifications

• Data collection: Weekly measurements of total organic carbon, nitrate, ammonium, soil moisture, soil temperature, CO2, CH4, N2O, and NH3 fluxes across multiple agricultural treatments over five growing seasons
• Modeling framework: Integration of carbon and nitrogen cycles in land surface models, building on prior work incorporating denitrifying bacteria growth rates for nitrous oxide flux quantification
• Sensitivity analysis: Variance-based global sensitivity analysis quantifying contributions of five model parameters and their interactions to surface and root-zone soil moisture uncertainty
• Emulation approach: Gaussian process emulation with Gaussian Emulation Machine for Sensitivity Analysis (GEM-SA), validated across parameter space (r > 0.6 for 10 of 12 sites)
• Key findings: Surface soil moisture uncertainty is largely attributed to shape parameter uncertainty; root-zone uncertainty is driven by porosity uncertainty
• Future analysis: Optimization of microbial activity coefficients within defined multi-dimensional uncertainty bounds to separate main effects from interaction effects among parameters

Technology readiness level

The research has demonstrated proof-of-concept through variance-based sensitivity analysis on soil moisture parameters, with the Gaussian process emulator successfully capturing land model behavior across diverse soil textures and seasons. The methodology has been validated at most sites (10 of 12) with strong correlation between emulator predictions and model outputs. Future work will extend this validated approach to optimize microbial activity coefficients for carbon sequestration potential, leveraging the established multi-year field dataset and refined emulation techniques.