Modeling and sampling optimization of soil microbial biomass for soil health assessment in agricultural systems

Abstract

Soil health assessment increasingly emphasizes biological indicators, yet their integration into routine monitoring remains limited by high analytical costs, spatial variability, and interpretive challenges. Soil microbial biomass, a key indicator of soil biological functioning, plays a central role in organic matter dynamics and nutrient cycling but is typically measured using resource-intensive methods such as phospholipid fatty acid (PLFA) analysis. This thesis addresses these limitations by integrating predictive modeling and sampling optimization to develop scalable approaches for biological soil health assessment. Chapter 1 synthesizes current knowledge on soil health frameworks, emphasizing the importance of biological indicators, the influence of spatial variability, and the limitations of existing sampling and modeling approaches. It identifies a critical gap in linking biological indicators with efficient sampling strategies and predictive modeling. Chapter 2 evaluates the extent to which soil microbial biomass can be predicted from routinely measured soil physicochemical properties using machine learning. Results show that soil properties, particularly organic matter, carbon, nitrogen, and texture, explain a substantial proportion of variability in microbial biomass, with XGBoost achieving strong predictive performance (R² ≈ 0.80). The inclusion of enzyme activity provided only modest improvements, indicating that microbial biomass can be effectively estimated using commonly measured soil variables. Chapter 3 investigates whether reduced-intensity sampling strategies can reliably capture field-scale variability in soil health indicators. Results demonstrate that reducing sampling density from one sample per acre to one per five acres preserved field-scale means and spatial structure for most indicators. Additionally, a six-point management zone-based sampling design estimated microbial biomass with less than 2% bias, demonstrating that targeted sampling can substantially reduce sampling effort while maintaining accuracy. Together, these findings establish a tiered framework for soil health assessment that integrates dense physicochemical sampling, predictive modeling, and targeted biological validation. This approach enables the incorporation of biological indicators into scalable soil monitoring systems, supporting precision agriculture and sustainable land management in spatially heterogeneous agricultural landscapes.