Distinct diurnal drivers of estuarine pCO₂ revealed by interpretable machine learning

Pronounced diurnal differences in the partial pressure of CO₂ (pCO₂) in surface estuary waters are observed, and resolving these disparities is a prerequisite for improving regional carbon budget estimates. However, underlying drivers of pCO₂ at the sub-daily scale remain poorly understood mainly due to nocturnal data gaps, which inherently restrict our predictive capacity. This study leverages automated high-frequency observations and constructs an XGBoost-SHAP framework based on Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) to decouple environmental drivers of pCO₂ during the daytime and nighttime in the Jiulong River Estuary, China. The developed diel-specific models transparently disentangled the diurnal distinction of environmental drivers (coefficient of determination R²: 0.676 for the daytime model and 0.653 for the nighttime model). Results reveal that while daytime dynamics are predominantly governed by thermodynamic forcing, nocturnal dynamics revert to a baseline hydrology-dominated structure where salinity-tracked water mass transport becomes the primary determinant. Furthermore, rather than representing internal ecosystem memory, historical states serve as highly effective predictors by physically capturing the time-lags of advective transport. These findings confirm that carbon regulation drivers undergo a structural transformation characterized by asymmetric diurnal disparities. Therefore, accurate estuarine carbon modeling requires diel-partitioned frameworks that integrate time-lagged hydrodynamic processes, a step that is essential to enhancing the predictive capacity for coastal carbon flux estimates.