Integrating vegetation constraints to reduce parameter uncertainty in watershed modeling of in-stream organic carbon

A newly developed Soil and Water Assessment Tool-Carbon (SWAT-C) model has been widely adopted for watershed-level carbon cycling, but its parameter uncertainty has been rarely explored. This study quantified how additional vegetation constraints (leaf area index [LAI] and crop yield) reduce parameter uncertainty in SWAT-C simulations of in-stream organic carbon within an agricultural watershed. Using a sequential constraint approach, the SWAT-C model was calibrated under various configurations of constraints: standalone streamflow (PAR-F); streamflow and POC/DOC (PAR-FP/ PAR-DP); streamflow with POC/DOC and RS-LAI (PAR-FPL/ PAR-FDL); streamflow with POC/DOC, RS-LAI, and crop yield (PAR-FPLC/ PAR-FDLC). The model performance was assessed using Nash–Sutcliffe efficiency (NSE) and percent bias (P-bias). This study counted the number of behavioral parameter sets for individual constraint configurations to quantify the parameter uncertainty. The results showed that introducing these constraints notably decreased the number of behavioral parameter sets from 17 (PAR-FP) to 2 (PAR-FPLC) and from 63 (PAR-FD) to 4 (PAR-FDLC), reducing parameter uncertainty. The incorporation of vegetation constraints improved the model’s performance in simulating in-stream organic carbon. The NSE of the PAR-FPLC was 0.02 – 0.03 greater than PAR-FP and PAR-FPL, indicating an enhancement in the accuracy of POC predictions. While the validation performance of PAR-FDLC was lower than PAR-FD and PAR-FDL, additionally constraining the model with vegetation dynamics helped identify reliable globally optimal parameters. Overall, our findings suggested that incorporating additional vegetation constraints enhanced the predictive capability of the SWAT-C model, introducing a more reliable way to simulate in-stream organic carbon within an agricultural landscape.