Phenology Constrains the Predictability of Soil Respiration in Cropping Systems Revealed Through Interpretable Machine Learning

ORCID

Czarnecki: https://orcid.org/0000-0001-6810-8227

MSU Affiliation

College of Arts and Sciences; Department of Geosciences; Geosystems Research Institute

Creation Date

2026-10-02

Abstract

Soil respiration (Rs) links canopy carbon assimilation with belowground carbon cycling; however, its drivers and predictability vary substantially across crop development. Most traditional, season-aggregated approaches implicitly treat the growing season as homogeneous, thereby obscuring phenology-dependent controls on soil CO2 efflux. To address this limitation, we applied stage-explicit modeling to examine how the predictability and dominant drivers of Rs shift across the life cycles of cotton (Gossypium hirsutum L.) and corn (Zea mays L.) in a no-till dryland system. Using this framework, we found that both Rs predictability and dominant controls were strongly stage-dependent. In particular, predictive skill was confined to specific phenological stages (R2 up to 0.52) and consistently collapsed during reproductive transitions, indicating reduced coupling between Rs and the measured surface predictors during these developmental periods. Within these stage-dependent regimes, vegetation indices (VIs) provided informative signals for Rs prediction during select phenological periods, whereas their predictive utility diminished during transitional phases. By contrast, cover crop (CC) residual effects primarily functioned as a magnitude filter, modulating the baseline intensity of Rs without fundamentally altering its sensitivity to environmental drivers (∆R2 ≤ 0.04). To further interpret these patterns, interpretable analyses (SHAP/PDP/ICE) revealed crop-specific nonlinearities in predictor-response relationships. Most notably, non-monotonic VI-Rs trajectories emerged during late-season stages, indicating that increases in canopy greenness or cover did not consistently correspond to higher soil respiration later in crop development. Taken together, these results demonstrate that the governing logic of Rs changes systematically with phenology. Consequently, stage-explicit modeling provides a framework for interpreting phenology-dependent shifts in soil respiration predictability and identifying developmental windows of enhanced or reduced flux detectability, informing phenology-aware data collection and model evaluation.

Keywords

soil respiration, carbon flux, phenology, non-stationarity, cover crops, interpretable machine learning

Publication Date

8-30-2026

Publication Title

Agricultural and Forest Meteorology

Publisher

Elsevier

Rights

© 2026 The Author(s)

Link to Full Text

Share

COinS
 

Digital Object Identifier (DOI)

https://doi.org/10.1016/j.agrformet.2026.111443