Introduction
Accurately modeling gross primary productivity (GPP) is of great importance in diagnosing terrestrial carbon-climate feedbacks. Process-based terrestrial ecosystem models are often subject to substantial uncertainties, primarily attributed to inadequately calibrated parameters. Recent attention has identified carbonyl sulfide (COS) as a promising proxy of GPP, due to the close linkage between leaf exchange of COS and carbon dioxide (CO2) through their shared pathway of stomatal diffusion.
However, most of the current modeling approaches for COS and CO2 did not explicitly consider the vegetation structural impacts, i.e., the differences between the sunlit and shaded leaves in COS uptake. This study used ecosystem COS fluxes data from 7 sites to optimize GPP estimation across various ecosystems with the Boreal Ecosystem Productivity Simulator (BEPS), which was further developed for simulating the leaf COS uptake under its state-of-the-art ‘two-leaf’ framework.
Our results demonstrated the substantial improvement in GPP simulation across various ecosystems through the fusion of COS data into the ‘two-leaf’ model, with the ensemble mean of root mean square error (RMSE) for simulated GPP reduced by 18.99% to 66.64%. Notably, we also shed light on the remarkable identifiability of key parameters within the BEPS model, including the maximum carboxylation rate of Rubisco at 25°C (Vcmax25), minimum stomatal conductance (bH2O), and leaf nitrogen content (Nleaf), despite intricate interactions among COS-related parameters.
Furthermore, our global sensitivity analysis delineated both shared and disparate sensitivities of COS and GPP to model parameters and suggested the unique treatment of parameters for each site in COS and GPP modeling. In summary, our study deepened insights into the sensitivity, identifiability, and interactions of parameters related to COS, and showcased the efficacy of COS in reducing uncertainty in GPP simulations.
Importance of Terrestrial Ecosystem GPP Modeling
Over the past 5 decades, terrestrial ecosystems have been absorbing about 30% of anthropogenic carbon dioxide (CO2) emissions, playing a crucial role in mitigating climate change (Friedlingstein et al., 2022). Driven by the photosynthesis of terrestrial vegetation, gross primary productivity (GPP) is the largest terrestrial carbon flux and plays an important role in understanding terrestrial carbon-climate feedbacks (Luo, 2007; Wang et al., 2021).
However, as direct observations of GPP using atmospheric CO2 observations are confounded by respiration (Hilton et al., 2017), and the modeling of GPP is affected by a range of uncertainties such as poorly calibrated parameters (MacBean et al., 2022), the precise quantification of GPP in terrestrial ecosystems has been a major challenge (Canadell et al., 2000; Yuan et al., 2007).
COS as a Proxy for GPP Modeling
Carbonyl sulfide (COS) is the most abundant sulfur-containing trace gas in the atmosphere with a lifetime of about 2 years (Montzka et al., 2007; Karu et al., 2023). The tropospheric atmospheric mole fraction of COS is approximately 500 parts per trillion (ppt), exhibiting a typical seasonal amplitude of ~100-200 ppt (Montzka et al., 2007; Ma et al., 2021; Hu et al., 2021; Remaud et al., 2022, 2023; Ma et al., 2023).
In the past decade, COS has emerged as a promising tracer for terrestrial photosynthesis (Stimler et al., 2010; Asaf et al., 2013; Launois et al., 2015; Kooijmans et al., 2019) and stomatal conductance (Commane et al., 2015; Wehr et al., 2017; Sun et al., 2022) as the leaf exchange of COS and carbon dioxide (CO2) are tightly coupled through stomata (Sandoval-Soto et al., 2005; Seibt et al., 2010; Wohlfahrt et al., 2012; Whelan et al., 2018).
Unlike CO2, which is emitted back into the atmosphere via leaf respiration (Sun et al., 2022), COS is completely destroyed by a hydrolysis reaction catalyzed by carbonic anhydrase (Protoschill-Krebs et al., 1996) without back flux in leaves under normal conditions (Stimler et al., 2010). Consequently, the measurement of COS flux can provide a direct and independent way to estimate GPP (Sandoval-Soto et al., 2005; Abadie et al., 2023).
Modeling Approaches for COS and CO2
In most early studies, GPP was directly estimated by scaling measurement of plant COS uptake with the empirically derived leaf relative uptake (LRU) approach or its extensions that incorporate the effects of temperature, humidity, light, and CO2 concentration on stomatal conductance (Kohonen et al., 2022a; Sun et al., 2022; Abadie et al., 2023) because of the simplicity of this approach and the sufficiency of it in many cases (Sandoval-Soto et al., 2005; Whelan et al., 2018).
In contrast, the process-based model that mechanistically simulates COS plant uptake by incorporating stomatal transport processes has also been developed and widely evaluated (Maignan et al., 2021; Kooijmans et al., 2021). Concurrently, the significance of soil COS exchange has also been recognized, leading to the development of a suite of empirical or mechanistic COS soil exchange models (Kesselmeier et al., 1999; Berry et al., 2013; Launois et al., 2015; Sun et al., 2015; Ogée et al., 2016; Whelan et al., 2022).
The process-based COS plant uptake model and soil exchange models have been integrated into land surface models (LSMs) (Berry et al., 2013; Maignan et al., 2021; Kooijmans et al., 2021). Consequently, by constraining the model parameters of LSMs with COS through data assimilation, not only are the model variables like GPP expected to be improved, but our understanding of ecosystem processes is also expected to be enhanced.
Currently, several studies have been conducted to refine the model parameters of LSMs through assimilating the COS fluxes and thereby optimizing the modeling of water-carbon fluxes (Chen et al., 2023; Abadie et al., 2023; Zhu et al., 2023). Within a big-leaf framework, Abadie et al. (2023) demonstrated COS could provide mechanistic constraint on stomatal diffusion, and the joint assimilation of COS and GPP is able to improve the model performance of GPP and latent heat.
Limitations of Big-Leaf Models and the Need for Two-Leaf Modeling
Ecosystem carbon, water, and energy processes interact and are nonlinear, and the changes in one process could induce variations in the other processes. While COS assimilation has proven to be effective in constraining COS-related model parameters and optimizing GPP, there remains a gap in systematic understanding of the ability of COS to optimize model parameters from different processes.
For example, how effective is the assimilation of COS in reducing model prediction uncertainty in GPP and in the relevant ecosystem processes in different ecosystems? Due to the dissimilar illumination conditions, there is a significant variability of leaf photosynthesis between sunlit and shaded leaves (Chen et al., 1999; Pignon et al., 2017; Wang et al., 2018; Bao et al., 2022). It is now clearly recognized that big-leaf models are conceptually flawed and practically inaccurate, and sunlit-shaded leaf stratification is necessary to make accurate canopy-level photosynthesis estimation (Chen et al., 1999; Chen et al., 2012; Luo et al., 2018).
Consequently, in the process-based LSM that simulates COS plant uptake and photosynthesis in a coupled manner (Ball et al., 1987; Berry et al., 2013), the application of the two-leaf model shows promise for providing accurate simulation of plant COS uptake.
Objectives and Approach of This Study
In this context, we have further explored the capacity of COS to constrain the model parameters of an LSM and to optimize GPP within the two-leaf modeling framework. Our goal is to address the following questions:
- Which parameters is the COS simulation sensitive to, and what are the differences in parameter sensitivities between COS and GPP?
- How effective is COS assimilation in improving model prediction and reducing prediction uncertainty in GPP?
- Which process parameters can be well identified by the assimilation of COS?
- How do process parameters interact in COS modeling across diverse ecosystems?
To address these questions, we utilized ecosystem COS flux data to optimize GPP across various ecosystems based on the coupling of COS modeling with the two-leaf-based Boreal Ecosystem Productivity Simulator (BEPS).
Through Monte Carlo simulations, we conducted a global parameter sensitivity analysis to explore the sensitivity of COS and GPP simulations to model parameters related not only to photosynthesis but also to water and energy. The interaction and identifiability of these parameters were quantified using Monte Carlo-optimized parameter sets. Additionally, the effectiveness of COS in constraining model uncertainty in simulated COS and GPP is evaluated.
The Boreal Ecosystem Productivity Simulator (BEPS) Model
The BEPS model (Liu et al., 1997; Chen et al., 1999; Chen et al., 2012) used in this study is a process-based diagnostic model driven by remotely sensed vegetation parameters, including leaf area index (LAI), clumping index, and land cover type, as well as meteorological and soil data (Chen et al., 2019).
With the coupling among terrestrial carbon, water, and nitrogen cycles (He et al., 2021), it simulates photosynthesis, energy balance, and hydrological and soil biogeochemical processes at hourly time steps (Ju et al., 2006; Liu et al., 2015). For photosynthesis, it stratifies whole canopies into sunlit and shaded leaves and calculates GPP for each group of leaves by scaling Farquhar’s leaf biochemical model (Farquhar et al., 1980) up to canopy-level with a temporal and spatial scaling scheme (Chen et al., 1999).
In this study, the BEPS model stratifies the soil profile into five layers, and the model implicitly solves the soil water content values for these layers (Ju et al., 2010). Over the last few decades, the BEPS model has been continuously improved and has been used in a wide variety of terrestrial ecosystems (Schwalm et al., 2010; Liu et al., 2015).
Modeling COS Fluxes in BEPS
The ecosystem COS flux includes both plant COS uptake, FCOS,plant, and soil COS flux exchange, FCOS,soil (Whelan et al., 2016). In this work, the canopy-level COS plant uptake, FCOS,plant (pmol/m2/s), was calculated by upscaling the resistance analog model of COS uptake (Berry et al., 2013) with the two-leaf upscaling scheme (Chen et al., 1999).
Considering the different responses of foliage to diffuse and direct solar radiation (Gu et al., 2002), FCOS,plant is calculated as:
FCOS,plant = FCOS,sunlit * LAIsunlit + FCOS,shaded * LAIshaded
where FCOS,sunlit and FCOS,shaded denote the leaf-level COS uptake rate (pmol/m2/s) for sunlit and shaded leaves. The leaf-level COS uptake rate, FCOS,leaf, is determined by the following formula (Berry et al., 2013):
FCOS,leaf = (COSa * gsw / (1.94 * gbw)) * (1 / (1 + (gsw / (1.56 * gCOS))))
where COSa represents the COS mole fraction in the bulk air. gsw and gbw are the stomatal conductance and leaf laminar boundary layer conductance to water vapor (H2O). The factors 1.94 and 1.56 account for the smaller diffusivity of COS with respect to H2O. gCOS indicates the apparent conductance for COS uptake from the intercellular airspaces, which combined the mesophyll conductance and the biochemical reaction rate of COS and carbonic anhydrase.
The soil COS fluxes are simulated by considering the abiotic and biotic components separately, as done by Whelan et al. (2016). We took the soil COS modeling scheme including the parameterizations from Whelan et al. (2016) and Whelan et al. (2022) in this study.
Study Sites and Data
The model was evaluated on seven sites distributed on the Eurasian and American continents in boreal, temperate, and subtropical regions based on field observations collected from several studies. Those sites were representative of different climate regions and land cover types (in the model represented by plant function types and soil textures, as depicted in Table 1).
Data used in this study include LAI, land cover type, meteorological and soil data, and CO2 and COS mole fraction data. The CO2 and COS mole fractions in the bulk air were regarded as spatially invariant over the globe but assumed to vary annually.
The ecosystem COS flux observations were utilized to perform optimization and to evaluate the optimization results. They were derived from existing studies with pre-processing with regard to the data quality check. To assess the model performance of GPP, the GPP observations were also collected from FLUXNET, AmeriFlux, and existing studies.
Parameter Optimization Approach
The Monte Carlo-based parameter optimization approach was employed here (Figure 1). The methodology calls for rejecting the concept of a unique global optimum parameter set within some particular model structure, instead recognizing the equifinality of parameter sets that exhibit similarly good performance in producing the observed data (Beven and Freer, 2001).
In a Monte Carlo simulation framework, a large number of random sets of parameters are derived across specified parameter ranges (Staudt et al., 2010) and employed to drive the model. Subsequently, model realizations are grouped into behavioral and non-behavioral model runs and associated parameter sets based on the values of the single or multiple performance measures and the predefined threshold value (Houska et al., 2014).
The former describes acceptable model realizations conditioned on the available observational data (Blasone et al., 2008; Beven and Binley, 2014). The latter describes parameter sets that produce behavior inconsistent with observed behavior. The deterministic model prediction is given by the ensemble mean of the 100 behavioral simulations.
Parameter Selection and Sensitivity Analysis
Based on current understanding of COS exchange (Wohlfahrt et al., 2012; Berry et al., 2013; Whelan et al., 2016; Whelan et al., 2018; Cho et al., 2023), photosynthesis (Ball et al., 1987; Raines, 2003; Blankenship, 2021), and related parameter sensitivity studies (Liu et al., 2011; Chen et al., 2012; Chen et al., 2023; Xing et al., 2023; Abadie et al., 2023; Zhu et al., 2023), 9 parameters were selected to be calibrated in this study (Table 2). These parameters are related to formulas describing four processes: (1) photosynthesis, (2) soil hydrology, (3) stomatal gas exchange, and (4) energy balance.
A density-based global sensitivity analysis approach (Plischke et al., 2013) was used to investigate the sensitivity of COS and GPP simulations to the selected model parameters via the Sensitivity Analysis Library (SALib) (Iwanaga et al., 2022).
Results and Discussion
Sensitivity of COS and GPP to Model Parameters
The sensitivity indexes of COS and GPP simulations with respect to the model parameters for the seven sites are illustrated in Figure 2. It can be seen that both COS and GPP simulations exhibit high sensitivity to leaf nitrogen content (Nleaf) and the maximum carboxylation rate of RuBisCO at 25°C (Vcmax25),
