Simulating canopy carbonyl sulfide uptake of two forest stands through an improved ecosystem model and parameter optimization using an ensemble Kalman filter

The uptake of carbonyl sulfide (COS) by plant canopy closely correlated with photosynthetic CO2 uptake as they share the same diffusion pathway. Process-based ecosystem models have been used to simulate the CO2, H2O and COS fluxes between terrestrial ecosystems and atmosphere. Some model parameters often vary seasonally and interannually. In this study, the hourly Boreal Ecosystem Productivity Simulator (BEPS) was first improved to simulate the canopy COS uptake rate. Then, an ensemble Kalman filter was designed to optimize the key pa-rameters of the hourly BEPS model, considering the errors in the input, parameters and measurements. The optimized parameters include leaf maximum carboxylation rate at 25 degrees C at top of canopy (V0cmax25) and a cali-bration term (alpha) used to scale the apparent conductance for COS uptake from the intercellular airspace (gcos) to Vcmax. The fluxes of gross primary productivity (GPP), latent heat (LE) and COS uptake by canopy (Fcos_veg) measured using eddy covariance method at a temperate deciduous forest during 2012-2013 and a boreal conifer forest during 2013-2017 were used for data assimilation. Parameters were optimized daily and presented as 10 -day averages. It was found that the parameters showed considerable seasonally and interannually variations. With parameter optimization using EnKF, the improved hourly BEPS model could explain 86%, 68% and 55% of variations in measured daily GPP, LE and Fcos_veg at US-Ha1 site during 2012 and 2013 and captured 90%, 74% and 69% of variations in the measured GPP, LE and Fcos_veg at FI-Hyy site during 2013-2017, respectively. The simulated daily Fcos_veg with the parameters optimized using EnKF could capture the day-to-day variations of Fcos_veg at the two studied sites and performed better than the simulations with calibrated parameter values. Further efforts were required to investigate the impacts of model structure and different parameter optimization methods on the optimized parameter values.