Wetland ecosystems are an important component in global carbon (C) cycles and may exert a large influence on global climate change. Predictions of C dynamics require us to consider interactions among many critical factors of soil, hydrology, and vegetation. However, few such integrated C models exist for wetland ecosystems. In this paper, we report a simulation model, Wetland-DNDC, for C dynamics and methane (CH4) emissions in wetland ecosystems. The general structure of Wetland-DNDC was adopted from PnET-N-DNDC, a process-oriented biogeochemical model that simulates C and N dynamics in upland forest ecosystems. Several new functions and algorithms were developed for Wetland-DNDC to capture the unique features of wetland ecosystems, such as water table dynamics, growth of mosses and herbaceous plants, and soil biogeochemical processes under anaerobic conditions. The model has been validated against various observations from three wetland sites in Northern America. The validation results are in agreement with the measurements of water table dynamics, soil temperature, CH4 fluxes, net ecosystem productivity (NEP), and annual C budgets. Sensitivity analysis indicates that the most critical input factors for C dynamics in the wetland ecosystems are air temperature, water outflow parameters, initial soil C content, and plant photosynthesis capacity. NEP and CH4 emissions are sensitive to many of the tested input variables. By integrating the primary drivers of climate, hydrology, soil and vegetation, the Wetland-DNDC model is capable of predicting C biogeochemical cycles in wetland ecosystems.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Wetlands are an important component of the terrestrial landscapes that exert a great influence over global carbon (C) cycle and climate change. Wetlands contain 15–22% of the global terrestrial carbon [Eswaran et al., 1995; Gorham, 1991] and contribute 15–20% of the global methane (CH4) emissions to the atmosphere [Aselmann and Crutzen, 1989; Matthews and Fung, 1987]. Wetland ecosystems have unique characteristics affecting C dynamics. For example, the high water table and its fluctuation are the primary factors driving soil organic carbon (SOC) decomposition [Moore and Dalva, 1997; DeBusk and Reddy, 1988], plant C fixation [Bubier, 1995], CH4 production and consumption [Moore and Roulet, 1993; Bellisario et al., 1999], and other biogeochemical processes in the wetlands. Small changes in water table or temperature can perturb the C balance in the peatlands due to alterations in soil organic matter decomposition and/or plant production [Bubier et al., 1998; Silvola et al., 1996; Shurpali et al., 1995]. Accordingly, it is essential to quantify the processes and controls on wetland C dynamics, including CH4 emission, for assessing the impacts of climate change or management alternatives on wetland ecosystems.
 Carbon cycles and CH4 emission in wetland ecosystems are regulated by a series of interacting processes between soil, hydrology, and vegetation. For example, hydrological processes have a great impact on soil thermal dynamics; soil thermal and hydrological conditions influence both plant growth and soil C dynamics (e.g., decomposition, CH4 production, and oxidation); plant growth affects hydrological processes through evapotranspiration and intercepting precipitation; and soil, water, and plants work in conjunction to affect CH4 production and transport. For handling these complex interactions, process-oriented models should be the most productive approach to synthesize our knowledge. There are few existing wetland models which are comprehensive enough to integrate most of the important processes for wetland ecosystems [Mitsch et al., 1988] although many C models have been developed for upland ecosystems and used for global-scale C fluxes without explicitly identifying wetland ecosystems [Heimann et al., 1998]. Trettin et al.  evaluated 12 popular soil C models and found that they did not adequately account for anoxia, alternating hydroperiods, complex interactions of soil chemistry and abiotic factors, and CH4 processes that are important to wetland C cycling.
 Existing wetland-related models generally fall into three categories: long-term peat accumulation models, empirical CH4 emission models, and process-based CH4 emission models. The peat accumulation models developed by Clymo , Frolking et al. , and others have been reviewed by Yu et al. . Frolking et al.  developed a peat decomposition model (PDM) to calculate long-term peat accumulation based on vegetation conditions (NPP and rooting) and decomposition dynamics. This type of model focuses on long-term (several centuries to several millennia) peatland development and peat accumulation, and the effects of water table, vegetation, and climate are parameterized based on their average conditions, usually for specific sites. Empirical CH4 emission models have been developed by directly correlating the observed CH4 fluxes to controlling factors, such as water table, soil temperature, plant primary productivity, or ecosystem productivity [e.g., Bellisario et al., 1999; Frolking and Crill, 1994; Moore and Roulet, 1993; Whiting and Chanton, 1993; Crill et al., 1988]. These empirical relationships cannot be extrapolated to other sites where conditions are different from the experimental sites. Process-based models simulate CH4 emissions with different degrees of complexity and integration with other processes [Cao et al., 1996; Christensen et al., 1996; Potter, 1997; Arah and Stephen, 1998; Grant, 1998; Walter and Heimann, 2000; Li, 2000]. Christensen et al.  estimated CH4 emissions as a constant ratio of soil heterotrophic respiration under steady state conditions. Potter  simulated CH4 production (CH4/CO2 ratio) and CH4 oxidation (fraction of CH4) as functions of water table depth. Walter and Heimann  and Arah and Stephen  predicted CH4 production and oxidation based on Michaelis-Menten kinetics and simulated CH4 transportation in soil and via plants. Their maximum CH4 production and oxidation rates were parameterized without integration with vegetation and soil decomposition processes, and the models need water table as input. Cao et al.  considered the effects of environmental factors and substrates and the integrated processes of water table level, soil, and vegetation C dynamics in their simulation of CH4 production and oxidation, although they used the Terrestrial Ecosystem Model (TEM) [Raich et al., 1991] developed for upland ecosystems to handle the vegetation and soil C dynamics in their wetland studies. Grant  simulated CH4 emissions based on stoichiometries and energetics of the transformations mediated by each microbial community. His model may be the most complex CH4 emission model, but was developed mainly for agricultural ecosystems. Li  modified the DNDC model with detailed algorithms for simulating soil redox potential, substrate concentrations and CH4 production, consumption and transport but only for rice paddies. We have adopted several of the above-listed approaches in the development of Wetland-DNDC, a more comprehensive wetland model.
 The purpose of developing Wetland-DNDC is to predict both CO2 and CH4 emissions driven by hydrology, soil biogeochemistry, and vegetation processes in wetland ecosystems. The model can run from a year to several decades with a primary time step of 1 day. This temporal scale allows us to directly use field observations to validate the model, and to answer questions about climatic change and management practices. The general structure of the model was adopted from PnET-N-DNDC, a process-oriented model simulating C and N dynamic and trace gas emissions in upland forest ecosystems [Li et al., 2000]. PnET-N-DNDC was developed based on a basic biogeochemical concept, biogeochemical fields, which integrates the ecological drivers, environmental factors and geochemical and biochemical reactions into a dynamic system [Li et al., 2000]. In comparison with the other 11 published biogeochemical models, PnET-N-DNDC provides a better framework for our development of a wetland C model [Trettin et al., 2001]. For example, the ecological level and degree of complexity simulated by PnET-N-DNDC is a good match to the kinetic approaches adopted by Walter and Heimann  and Cao et al. [1995, 1996] for modeling CH4 fluxes from wetlands. In this paper, our discussion focuses on the new features developed in Wetland-DNDC, although Wetland-DNDC has inherited many existing functions from the DNDC model family (e.g., Crop-DNDC and PnET-N-DNDC). The distinguishing features of Wetland-DNDC include simulations of water table dynamics, effects of soil properties and hydrologic conditions on soil temperature, C fixation by mosses and herbaceous plants, and effects of anaerobic conditions on decomposition, CH4 production and consumption, and other biogeochemical processes. Details of the model are described in section 2, followed by validation and sensitivity analysis in section 3.
2. Model Description
 Wetland-DNDC consists of four components: hydrological conditions, soil temperature, plant growth, and soil C dynamics (Figure 1). These four components and their processes interact closely with each other. For example, soil thermal and hydrological conditions influence plant growth and soil C dynamics (e.g., decomposition, CH4 production, and oxidation). Plant growth influences evapotranspiration and precipitation interception and, thus, hydrological processes. Plant growth also affects CH4 emissions by providing C substrates for CH4 production and by providing conduits for CH4 transport. Wetland-DNDC explicitly considers all these processes and their interactions (Figure 1). The state variables are expressed as mass per unit area or relative content, representing a spatially homogeneous area or site as defined by the input data. The model input includes initial conditions (e.g., plant biomass, soil porosity, soil C content, water table position), model parameters (e.g., lateral inflow/outflow parameters, maximum photosynthesis rate, respiration rate), and climate drivers (e.g., daily maximum and minimum temperature, precipitation, solar radiation). The model output includes C pools and fluxes (e.g., C in plants and soil, photosynthesis, plant respiration, soil decomposition, CH4 emissions, and net ecosystem productivity), and thermal/hydrological conditions (e.g., soil moisture, water table position, water fluxes, soil temperature profile). Soil C pools and their decomposition processes are described in detail in DNDC [Li et al., 1992], and the dynamics of woody stratum are described in PnET [Aber et al., 1996; Aber and Federer, 1992]. Below we describe the major improvements and the new developments of Wetland-DNDC in (1) hydrology, (2) soil thermal dynamics, (3) growth of mosses and herbaceous plants, and (4) anaerobic processes (decomposition, CH4 emissions).
 The hydrological submodel was developed to simulate water table dynamics explicitly. The soil profile is divided into layers of different characteristics (e.g., organic soils and mineral soils). The soil layers are then grouped into two zones: the unsaturated zone above the water table and the saturated zone below it. The hydrological submodel considers water table dynamics, aboveground water input (e.g., precipitation, surface inflow, snow/ice melt) and output (evaporation, transpiration), and water movement in the unsaturated zone.
2.1.1. Soil Moisture and Water Table Dynamics
 The soil moisture content is determined for the unsaturated and the saturated zones separately. In the unsaturated zone, the soil moisture is determined by
where ΔSWl is the change of soil moisture (cm3·cm−3) in layer l in the unsaturated zone, Hl is the thickness (cm) of the layer, Fl is net water input to the layer (cm water) through infiltration, gravity drainage, and matric redistribution, and ESl and TPl are water uptake (cm) from this layer through evaporation and transpiration, respectively. In the top saturated layer where water table resides (i.e., layer l0), the soil moisture is estimated by
where is the soil moisture (cm3·cm−3) of layer l0, is the porosity (cm3·cm−3), is the field capacity (cm3·cm−3), WT′ is the water table position in layer l0 (cm above the bottom of layer l0), and is the thickness (cm) of the layer.
 Water table dynamics are determined directly by the water budget of the saturated zone, which includes water input from the unsaturated zone through infiltration and gravity drainage, capillary uptake through matric redistribution, evaporation and transpiration uptake from this zone, and outflow. The water budget is given by
where ΔWT is the change of water table position (cm), is net water input (cm water) to the saturated zone from the above layers, and n is the total number of soil layers in the saturated zone. ESl and TPl are the same as those in equation (1). Yield (cm3·cm−3) and outflow (cm water) are defined as
where and are the same as those in equation (2), and WT is the water table position (cm) in reference to the soil surface. WT is positive when the water table is above the soil surface, and negative otherwise. ΔWT > 0 means that WT becomes higher. a1, a2, D1, and D2 are calibrated parameters for outflows. D1 and D2 represent two critical levels (cm) of WT. Outflow increases linearly when WT is higher than these levels. a1 and a2 are the rates of increase. Because D1 is usually close to the surface and D2 is deeper along the soil profile, a1(WT-D1) and a2(WT-D2) are regarded as surface outflow and ground outflow, respectively.
2.1.2. Aboveground Water Input and Output
 Aboveground water input includes precipitation, snow/ice melt, and surface inflow. We followed the work of Running and Coughlan  to determine plant interception of precipitation and snowmelt
where Pint is the daily plant interception (cm), P is the daily precipitation (cm), and LAI is the leaf area index. Wmelt is the amount of snow or ice melted in 1 day (cm in water), SNOW is the snowpack accumulated above the surface (cm in water), and Tm is the daily mean air temperature (°C). For depressional wetlands and most peatlands, surface lateral inflow usually comes from surface runoff of the watershed [Mitsch and Gosselink, 1993], and can be expressed as
where Sin is the daily surface inflow (cm) expressed as water depth in the wetland, r is the ratio of the area of the watershed to the area of the wetland, and Rp is a hydrologic response coefficient. Rp represents the fraction of precipitation in the watershed that contributes directly to surface runoff. In the model, we combined (r − 1)Rp as one parameter (α0).
 Both the potential evapotranspiration rate and water availability are considered in simulating evaporation and transpiration. We used the Priestley-Taylor approach [Priestley and Taylor, 1972; Ritchie et al., 1988] to estimate potential evapotranspiration because it requires only daily solar radiation and temperature as input climate data. Potential evapotranspiration (ETp) is separated into potential soil evaporation (ESp) and potential plant transpiration (TPp) [Ritchie et al., 1988]
where ESp and TPp are in units of cm·day−1. Evapotranspiration will first consume water intercepted by plants or water on the soil surface, and then water from the soil profile. The actual water uptake from a soil layer for transpiration is determined by the potential demand and the root uptake rate
where TPl is the water uptake rate from layer l (cm water·day−1), Rmax is the maximum water uptake rate of a unit root length density from a cubic centimeter soil (cm water·day−1·cm−1 root·cm−3 soil), RLDl is the root length density in layer l (cm root·cm−3 soil), Hl is the thickness (cm) of layer l, and fET,l is a scalar. RLDl is converted from root biomass using a specific root length of 2.1 cm·mg−1 [Eissenstat and Rees, 1994]. fET,l ranges from 0 to 1, representing the effects of soil moisture on evaporation and transpiration
where WPl is the soil moisture (cm3·cm−3) of layer l at the wilting point, FCl is the field capacity (cm3·cm−3), and Tl is the soil temperature of layer l.
2.1.3. Water Movement in the Unsaturated Zone
 The hydrological submodel considers three types of water movement in the unsaturated zone: infiltration, gravity drainage, and matric redistribution. Daily infiltration is a function of the infiltration capacity and the amount of water available on the soil surface. If water available for infiltration is more than the infiltration capacity, the excessive water will stay on the surface as ponds. The infiltration capacity depends further on water table position, saturated conductivity, and the frozen layer depth. If no frozen layer exists and the water table is low, the infiltration capacity is estimated as the amount of water infiltrated in a period of 24 h. Otherwise, the infiltration capacity is the amount of water saturating all the layers above the frozen layer or above the water table level. Gravity drainage refers to the downward movement of water when soil moisture is higher than field capacity. In our model, we assume that a fraction of water above field capacity will move to the next layer each day. Matric redistribution refers to the water movement driven by the gradient of matric potential between layers. The movement can be upward or downward. Matric redistribution is estimated based on the soil moisture difference of the two adjacent layers [Ritchie et al., 1988]. This procedure can include the capillary uptake of water from water table.
2.2. Soil Thermal Dynamics
 The soil temperature submodel estimates the daily average temperature of each soil layer by numerically solving the one-dimensional (vertical) heat conduction equation
where C is the heat capacity (J·cm−3·°C−1), T is the soil temperature (°C), t is time (s), λ is the thermal conductivity (W·cm−1·°C−1), and Z is the soil depth (cm). The effects of soil water conditions and organic matter content on temperature can be expressed as their effects on C and λ
where fi is the fraction (volumetric ratio) of a given soil component i (i.e., minerals, water, ice, organic matter, and air), and Ci and λi are the heat capacity and the thermal conductivity of component i, respectively.
 The temperature of the top and the bottom soil layers defines the boundary conditions needed to solve equation (13). The top layer can be snowpack (when a snowpack exists), or water (when there is no snowpack and the water table is above the surface), or soil (when a snowpack does not exist and the water table is below the surface). The temperature of the top layer is estimated based on the daily air temperature [Zheng et al., 1993]
where T0 and T′0 are the temperatures of the top layer on the current day and the previous day, respectively, Tm is the air temperature on the current day, LAI is the leaf area index, and K is the light extinction coefficient. When a snowpack exists, mosses and herbaceous plants beneath the snowpack can effectively insulate heat conduction. Such insulation effects are considered by assuming a 5 cm moss/herbaceous layer with bulk density of 33.3 (kg m−3) and water content of 0.4 (g water·g−1 biomass) [Frolking et al., 1996]. The depth of the snowpack is estimated based on snow accumulation (snowfall and snowmelt) and snow density. We assume that precipitation will be in the form of snow when daily air temperature (Tm) is below 0°C. Snowpack is considered as one layer. Snow density increases each day by 0.001Tm when Tm is higher than 0°C, based on an approximation of a detailed hourly snow model [Kongoli and Bland, 2000]. Snow density is set to 0.1 g·cm−3 for fresh snow, and limited to 0.3 g·cm−3 as the upper value [Verseghy, 1991]. Snowmelt is a function of temperature [Running and Coughlan, 1988] (equation (7)). Thermal conductivity and heat capacity of snow are estimated based on snow density [Mellor, 1977].
 The bottom boundary temperature is estimated by
where is the soil temperature of the bottom layer with a depth of Z0 (cm) on day JD, Taa and Tam are the annual average and amplitude of air temperature, respectively, JD is the Julian date, and JD0 is the Julian day when solar altitude is the highest (i.e., 200th and 20th for the Northern and the Southern hemispheres, respectively). D is the damping depth (cm), given by
where Cm is the average heat capacity of the soil profile (J·cm−3·°C−1), and λm is the average thermal conductivity (W·cm−1·°C−1) of the soil profile.
2.3. Growth of Mosses and Herbaceous Plants
 Mosses and herbaceous plants (hereafter, ground vegetation) are much more important for C fixation in wetlands compared to upland forest ecosystems. Therefore, we added algorithms for C fixation by ground vegetation to the vegetation submodel. Photosynthesis is estimated similarly to the moss simulation model of SPAM [Frolking et al., 1996]
where GPPg is the daily gross photosynthesis of ground vegetation (kg C·ha−1·day−1), Amax,g is the maximum photosynthesis rate per unit of effective photosynthetic biomass per hour (kg C·kg−1 C·h−1), Bg is the effective photosynthetic biomass of ground vegetation (kg C·ha−1), DL is the day-length (hr·day−1), and fg,L, fg,T, and fg,W are scalars that quantify the effects of light, temperature, and soil moisture, respectively. Bg is a function of the maximum aboveground biomass and the growing degree-days of the site; Bg represents changes in the amount of photosynthetically active tissues and the photosynthesis capacity of ground vegetation with time or phenology [Skre and Oechel, 1981; Williams and Flanagan, 1998]. We assume that the total daily respiration of plants is proportional to the daily GPPg. Annual litter fall of ground vegetation is estimated as its annual net primary productivity (NPP) [Frolking et al., 1996]. The growth of woody strata is simulated based on PnET [Aber et al., 1996; Aber and Federer, 1992].
2.4. Anaerobic Processes
 In wetland ecosystems, soil C pools and fluxes and decomposition processes are strongly controlled by anaerobic conditions and, thus, hydrology. Anaerobic processes, such as CH4 production and oxidation, are unique features of wet soils and are critical to our understanding and prediction of C dynamics in wetlands. To deal with soil anaerobic conditions, existing models use water table either to define the boundary between the anoxic and the oxic zones [Walter and Heiman, 2000], or to modify CH4 production and oxidation rates [Cao et al., 1996; Potter, 1997]. However, after the soil has been inundated, soil anaerobic status changes with time, and CH4 production has a time delay of about 1–2 weeks [Patrick and Reddy, 1977]. In contrast, redox potential is a direct indicator of soil anaerobic status, and is closely related to the soil biochemical reactions [Mitsch and Gosselink, 1993; Fiedler and Sommer, 2000]. In Wetland-DNDC, we used the redox potential of the soil layers in the saturated zone to simulate the anaerobic effects on decomposition and CH4 production and oxidation.
2.4.1. Redox Potential and SOC Decomposition
 Redox potential (Eh) is used to quantify the relative degree of anaerobic status for the soil layers near and below water table. Eh is estimated based on its general variation patterns in soils with a fluctuating water table [Sigren et al., 1997] and in soils under continuously submerged conditions [Patrick and Reddy, 1977; Fiedler and Sommer, 2000]
where ΔEhl is the daily variation of redox potential of layer l (mv·day−1), CR is the rate of change (i.e., 100 mv·day−1), Al is the aerenchyma factor of the layer, and wfpsl is the fraction of water filled pore space when the water table is below the top of this layer. Al defines the plant-mediated gas transport (i.e., equivalent to the fraction of pore space for gas diffusion).
where FRD is the area of the cross section of a typical fine root (cm2), PA is a scalar for the degree of gas diffusion from root to the atmosphere, and RLDl is the root length density (cm root·cm−3 soil) in layer l. FRD is assumed to be a constant of 0.0013 (cm2) [Barber and Silberbush, 1984]. PA ranges from 0 (plants without aerenchyma) to 1 (plants with well developed aerenchyma). Grasses and sedges are good gas transporters (PA = 1), whereas trees are poor ones (PA = 0.5). Mosses are not considered for this effect because they are not vascular plants (PA = 0).
where fdec is a scalar for the anaerobic effects on decomposition. For the soil layers above water table, soil moisture is used to estimate fdec [Li et al., 1992]. Figure 2 shows the change of fdec with changes of soil moisture and redox potential.
2.4.2. Methane Production, Oxidation, and Emissions
 To simulate CH4 production, oxidation, and transport in Wetland-DNDC, we linked a process-based CH4 emission model [Walter and Heiman, 2000] directly to soil thermal and hydrological conditions, soil redox potential, decomposition, and vegetation dynamics. Following the study by Walter and Heiman , the change of CH4 content in each layer (ΔM) is given by
where MPRD and MOXD are the CH4 production rate and the oxidation rate, respectively, MDFS is the diffusion between layers or to the atmosphere, and MEBL and MPLT are the CH4 emissions through ebullition and plant-mediated transport from the layer, respectively. All these terms are in unit of kg C·ha−1·day−1.
 Methane production occurs in all soil layers if there are enough substrates and if environmental conditions are favorable [Fieldler and Sommer, 2000]. We simulated CH4 production from each layer, using an approach similar to the ones used by Cao et al. [1995, 1996] and Walter and Heiman , but with explicit consideration of the effects of redox potential
where CM is the amount of simple C substrates (from soil decomposition and root systems) available for CH4 production, and fpH, fT,MP, fEh,MP are scalars for the effects of temperature, pH, and redox potential on CH4 production, respectively. The C substrate from roots is estimated as 45% of the C transferred to roots from photosynthesis [Cao et al., 1995]. We calculated the pH effect based on the study of Cao et al. , but used a minimum pH of 4.0 (instead of 5.5) because CH4 emissions have been observed when pH is below 5.5 in wetlands [Crill et al., 1988; Valentine et al., 1994]. Walter and Heimann  use a Q10 value of 6 to represent the effects of temperature on CH4 production. We used a Q10 value of 3 because the temperature effects on CM for methanogenesis have already been included in the calculation for soil organic carbon decomposition [Li et al., 1992]. Methanogenesis requires a very low redox potential. Based on measurements and the literature review by Fiedler and Sommer , we used −200 and −100 mv as the two critical Eh values that define the effects of redox potential on CH4 production (Figure 3).
 Methane oxidation is primarily controlled by the CH4 concentration, redox potential, and temperature [Segers, 1998]. Methane oxidized in a soil layer is estimated by
where M is the amount of CH4 in a soil layer (kg C·ha−1), and fM, fT,MO, and fEh,MO are scalars, representing the effects of CH4 concentration, temperature, and redox potential, respectively. Based on the work of Walter and Heiman , the effects of CH4 concentration are expressed as
 The CH4 diffusion process is estimated with empirical relationships. In a daily time step, the CH4 concentration gradient between two adjacent layers in the saturated zone decreases by about 70%, and is fully mixed in air filled space. Actual diffusion rates between layers and from the top layer to the atmosphere were estimated based on soil water content.
 Methane in each layer can be directly emitted to the atmosphere through ebullition and plant−mediated emission [Walter and Heiman, 2000]. Ebullition emission is considered when the soil CH4 concentration in a layer exceeded a threshold concentration of 750 μmol·L−1 [Walter and Heiman, 2000]. The plant-mediated emission is estimated based on the plant aerenchyma factor (i.e., Al defined in equation (21))
where MPLT is plant-mediated emission from a soil layer (l), and Pox is the fraction of CH4 oxidized during the plant-mediated transport (i.e., 0.5) [Walter and Heimann, 2000].
3. Model Testing
 We tested Wetland-DNDC against field observations, including water table dynamics, soil temperature, CH4 emissions, NEP, and annual C budget. A sensitivity analysis was also performed to determine critical parameters.
3.1. Sites and Data
 We selected three northern wetland sites where extensive measurements are available: one in Saskatchewan, Canada, and two in Minnesota, USA. The first wetland site is a minerotrophic fen located about 115 km northeast of Alberta, Saskatchewan, Canada (53°57′N, 105°57′W). This site (referred hereafter as SSA-Fen) is in the southern study area of the Boreal Ecosystem-Atmosphere Study (BOREAS) [Sellers et al., 1997]. Suyker et al. [1996, 1997] give detailed descriptions of the site. Fluxes of CO2 and CH4 were measured in the growing seasons (from mid May to early October) of 1994 and 1995 by the eddy covariance technique, together with measurements of water table positions, plant species composition, and LAI [Suyker et al., 1996, 1997]. All the data were obtained from the BOREAS CD-ROM [Newcomer et al., 1999]. The second wetland site is a forested bog located in the Marcell Experimental Forest in Itasca County, Minnesota, USA (47°32′N, 93°28′W). This site (referred hereafter as MEF-Bog) is managed by the USDA Forest Service North Central Research Station. Dise  and Verry and Urban  give detailed descriptions of the site. Dise  measured CH4 emissions at MEF-Bog from 1989 to 1990 using open-bottom chambers, together with soil temperature and water table positions. In addition, the daily water table dynamics have been monitored since 1961 [Verry and Urban, 1991]. The third wetland site is a bog lake peatland located about 2 km from the MEF-Bog site. This site (referred hereafter as BLP-Fen) is characterized as a poor fen with carpet-forming mosses (Sphagnum papillosum) dominating the vegetation. Detailed descriptions of the site can be found in the study by Shurpali et al. [1993, 1995] and Kim and Verma . Both CO2 and CH4 fluxes were measured in 1991 and 1992 by the eddy covariance technique [Shurpali et al., 1993, 1995].
3.2. Initial Conditions and Model Parameters
 The initial conditions and parameter values used in the model testing are given in Table 1 (SSA-Fen), Table 2 (MEF-Bog), and Table 3 (BLP-Fen). First, we identified the difference in vegetation covers at the three sites. For SSA-Fen, the ground vegetation alone was used to determine the overall plant C fixation (Table 1) because of low tree density [Suyker et al., 1997]. For MEF-Bog, both woody plants and ground vegetation were simulated explicitly (Table 2). For BLP-Fen, the same ground vegetation as for MEF-Bog was used, but woody plants were absent at this site (Table 3). Second, the active soil profile at the three sites was assumed to be 2 m in-depth and composed of peat. The distributions of bulk density, field capacity, and wilting point are given in Table 4, on the basis of the study by Zoltai et al.  and Paavilainen and Paivanen . Third, the parameters of lateral water flow were determined by comparing the simulated and the measured water table (i.e., using the first 3 years' measurements at MEF-Bog and the 2 years' measurements at the other sites). Fourth, the micro-topographic effects (i.e., hollows and hummocks) on the overall C fluxes measured by the eddy covariance technique can be assessed based on the relative height of the peat surface [Clement et al., 1995]. At MEF-Bog, the data for the microsites of hollows and hummocks were reported separately; thus, we ran the model for each micro site directly using the measured water table as input. At SSA-Fen and BLP-Fen, however, the micro-topographic effects were not reported. To better reflect these effects on C flux predictions, we obtained the average water table for the average hollow site, ran the model with this average water table and with four different surface heights above the hollow surface (i.e., 0, 10, 20, and 30 cm), calculated the average C fluxes from these four runs, and used the average C fluxes to compare with the tower measured C fluxes.
Table 1. The Initial Conditions, Model Parameters, and Data Sources at SSA-Fen
Lateral water flow parameters were calibrated by comparing the simulated and measured water table Plant respiration parameter was determined by comparing the simulated and the measured annual NPP. Minimum, optimum and maximum growing degree days (GDD) were determined based on the phenology of the plant growth (beginning, maximum, and senescence) and climate data.
 Sensitivity analysis was conducted with the data from BLP-Fen in 1992 (Tables 3 and 4) as the baseline conditions. The model parameters, initial conditions, and climate drivers were changed one at a time to determine their effects on model predictions. The response variables used in the sensitivity analysis are water table, NPP, soil microbial respiration, CH4 emissions, and NEP.
3.3. Results and Discussions
 The long-term measurements of water table dynamics from 1961 to 1999 at MEF-Bog provide an excellent data set for model validation. The model predictions correspond well to the trends and the temporal variations of the water table measurements (Figure 4; R2 = 0.54, N = 14244); most of the unexplained variance may be a result of mismatches in the exact timing and exact values of water table change. The model also captured the variation patterns of surface outflow (Figure 5; R2 = 0.52, N = 468), but the simulated surface outflow was about 46% higher than the measured stream outflow, perhaps due to water loss from places other than the measuring station at the stream outlet. The model estimates a ground outflow of 2–2.7 cm·month−1 (aside from a low value of 1.3 cm·month−1 in 1977). Although there was no direct measurement of ground outflow, a slow, steady seepage must have existed because the water table was perched several meters above the regional water table [Verry and Urban, 1992]. The model also works well in predicting evapotranspiration (R2 = 0.62, N = 263. Years 1994 and 1996) and snowpack (R2 = 0.83, N = 1021. Years 1994–1996) at SSA-Fen where measurements were available. These results indicate that the hydrologic model can quantify water table dynamics and water fluxes of wetlands at the watershed scale.
3.3.2. Soil Temperature
Figure 6 shows comparisons between simulated and measured soil temperature at different depths of the hollow site at MEF-Bog. The model accurately predicts the trajectories of the measured soil temperature along the soil profile, with R2 values ranging from 0.88 to 0.91 (with sample sizes of 45–59). Similar results are also obtained for the hummock site at MEF-Bog and for SSA-Fen (results not shown here). The model captures the effects of snow and soil organic matter on soil temperature. For example, due to the insulating effects of the moss layer and snowpack, soil temperature stays near or above 0°C in winter and spring although air temperature could be as low as −30°C during this period. Soil temperature in deeper layers (e.g., 40 cm) may be 4°C lower than in the top layer in summer, but about 1°C higher than in the top layers in winter.
3.3.3. Methane Emissions
 The CH4 emission model was tested at all three sites. The model captures general patterns of CH4 emissions, including the annual total, inter-annual differences, and the effects of water table positions (Figure 7). Values of R2 ranged from 0.37 to 0.76 with sample sizes of 47–214, except at MEF-Bog hummock, where R2 = 0.03 (N = 42), because CH4 emission is very low. At SSA-Fen, CH4 emissions are largely correlated with plant growth and soil decomposition because the water table is almost always above the surface. The simulated annual CH4 emissions (kg C·ha−1·yr−1) are 122.1 for 1994 and 121.0 for 1995 (Figure 7A). The estimated CH4 emissions from the tower flux measurements were 163 kg C·ha−1 for the period from May 17 of 1994 to October 7 of 1994 [Suyker et al., 1996]. The difference between simulated and observed CH4 emissions in 1994 may be caused by an underestimation of peak CH4 emissions and late growing season emissions. Although the water table was above the hollow surface, it fluctuated in a wider range in 1994 than in 1995. In 1994, the water table increased from 5 cm in late May to 26 cm in late July, then declined to 3 cm in late September. In 1995, the water table declined from 25 cm in late May to 14 cm in late September. In general, the model overestimates CH4 emissions when water table increases, but underestimates CH4 emissions when water table decreases. We suspect that these simulation errors may be caused by: (1) combination effects of micro-topography and water table fluctuation, (2) trapping and releasing of CH4 during water table fluctuations, and (3) effects of water layer thickness since the model assumes that the water table will have the same effects on anaerobic conditions and CH4 transport when water table is above the surface, regardless of depth.
 At MEF-Bog, CH4 emissions from the hummock site are much smaller than those from the hollow site (Figure 7B, C) because the average water table for the hummocks is 38 cm below the surface compared to 7 cm below the surface for the hollows. The simulated annual CH4 emissions (kg C·ha−1·yr−1) for the hollow site are 157.1 for 1989 and 127.0 for 1990, while they are 8.6 for 1989 and 14.0 for 1990 at the hummock site. The estimated CH4 emissions (kg C·ha−1) from observations from April 1 of 1989 to March 31 of 1990 were 103.5 for the hollows and 26.3 for the hummocks [Dise, 1991].
 At BLP-Fen, the model predictions display good agreement with the observations (Figure 7D). Note that in Figure 7D, the CH4 measurements of 1991 are the daytime totals reported by Shurpali et al. , whereas the measurements for 1992 are daily totals reported by Clement et al. . The simulated annual CH4 emissions are 129.2 kg C·ha−1·yr−1 for 1991, which is within the range of the observed values of 120.0–146.3 kg C·ha−1·yr−1 reported for 1991 [Shurpali et al., 1993]. The simulated annual CH4 emissions for 1992 are 87.8 kg C·ha−1·yr−1. The lower CH4 value in 1992 may be due to the relatively stable and high water table that limits soil decomposition and, therefore, reduces C substrates for CH4 production.
3.3.4. CO2 Flux and C Budget
 Wetland-DNDC also predicts other C fluxes and the annual C budget in addition to CH4 emissions. Figure 8 shows the variation patterns of NEP, NPP, and soil decomposition (i.e., soil microbial CO2 emissions) at SSA-Fen and BLP-Fen. There were no observed daily fluxes of NPP and soil decomposition at these two sites. However, we present the simulated NPP and soil decomposition to aid in understanding the variation patterns of NEP and the annual C budget.
 At SSA-Fen, predicted NEP was compared to the measured daily NEP during the period of 1994–1995. The model captures the general pattern of NEP (Figure 8A) (R2 = 0.49, N = 266). The simulated total NEP from mid–May to early October in 1994 is 958.4 kg C·ha−1, which agrees closely with the measured NEP of 880 kg C·ha−1 for the same period [Suyker et al., 1997]. NEP is a function of the combined effects of plant C fixation and soil C decomposition (Figure 8B). The simulated annual NPP (kg C·ha−1·yr−1) is 2589.8 for 1994 and 2878.1 for 1995, while the field measured aboveground NPP in 1994 was 1950 kg C·ha−1·yr−1 [Suyker et al., 1996]. The simulated soil decomposition (kg C·ha−1·yr−1) is 2128.3 for 1994 and 2023.9 for 1995. The annual C sequestration (kg C·ha−1·yr−1), which included CO2 exchange and CH4 emissions, is calculated as 339.4 for 1994 and 733.2 for 1995. The average annual C sequestration for these 2 years is 536.3 kg C·ha−1·yr−1. This value is higher than the average long-term wetland C accumulation rate (210 kg C·ha−1·yr−1) estimated by Clymo et al. , perhaps because this site is more productive than moss-dominated wetlands and because the high water table during these 2 years may have reduced soil decomposition.
 At BLP-Fen, the model also captures the general pattern of NEP, displaying good agreement between the predicted and the measured daily NEP during the period of 1991–1992 (Figure 8C) (R2 = 0.59, N = 40). The simulated NEP (kg C·ha−1) for the period from mid May to mid October is −910.2 for 1991 and 931.0 for 1992, while the measured NEP for the same period was −710 for 1991 and 320 for 1992 [Shurpali et al., 1995]. The simulated NEP for early 1992 is much higher than the measurements largely because the model fails to capture the sizable C release during the pre-leaf period when C trapped in soils from decomposition in late fall and winter was released as soil thawed [Lafleur et al., 1997]. The simulated NPP of the ground vegetation shows a rapid decline and then an early termination in the fall of 1991 (Figure 8D) because of drought effects. During the period of May–October, total precipitation in 1991 was about 30% less than that in 1992, and the mean temperature in 1991 was 1.5°C higher than that in 1992. The simulated annual NPP (kg C·ha−1·yr−1) is 1754.2 for 1991 and 2436.9 for 1992. The simulated soil decomposition rate also reflects drought effects (i.e., high temperature, low precipitation, low water table) in 1991; the peak soil decomposition rate in 1991 was twice that in 1992 (Figure 8D). The simulated soil decomposition for the period from May to October of 1991 is 2939.3 kg C·ha−1, which is comparable to the measured value of 3654.5 kg C·ha−1 (including root respiration) [Kim and Verma, 1992]. The simulated annual NEP (kg C·ha−1·yr−1) is −1581.8 for 1991 (source) and 470.7 for 1992 (sink). The annual C sequestration (kg C·ha−1·yr−1), including CO2 exchange and CH4 emissions, is −1711.0 for 1991 and 382.9 for 1992.
 There is no woody stratum at SSA-Fen and BLP-Fen. The annual change in soil C therefore is the same as NEP because we assume that the annual NPP of ground vegetation should be the same as its litter production. Table 5 shows the simulated annual C budget at MEF-Bog. Both the hollow and hummock sites are sinks of atmospheric CO2 with the C accumulation occurring primarily in the woody strata. Although the mortality of the trees may greatly reduce this sequestration, as measured by Grigal et al. , the model currently does not consider mortality. Soil C increases at the hollow site, but changes little in 1998 and even decreases in 1990 at the hummock site. The average soil C balance of the hollow and hummock sites is 281.4 kg C·ha−1·yr−1, which is comparable to the 180–280 kg C·ha−1·yr−1 estimated by Verry and Urban . They also found a soil C loss with water flow of 370 kg C·ha−1·yr−1. However, their estimate of soil microbial respiration (4710 kg C·ha−1·yr−1) was much higher than our simulated results. As a result, their estimate of litter input to soil would also be much higher, possibly due to tree mortality. The simulated NPP is 5434.9 and 5658.5 kg C·ha−1·yr−1 at the hollow and hummock sites, respectively; these numbers are comparable to the above ground NPP of 3700 kg C·ha−1 ·yr−1 measured by Grigal  and Grigal et al. .
Table 5. Simulated Annual Carbon Budget of the MEF-Fen Sitea
Unit: kg C ha−1 yr−1.
soil microbial respiration
soil C balance
soil microbial respiration
soil C balance
3.3.5. Sensitivity Analysis
 Sensitivity analysis reveals the effects of given factors on selected state variables. The three groups of factors are initial conditions, model parameters, and climate drivers, while the five selected state variables are water table (WT), NPP, soil microbial respiration (Rs), CH4 emissions, and NEP. WT is given as absolute change, whereas the others are expressed as changes in percentage (Table 6).
The parameters are changed from the baseline data of BLP-Fen in Table 3, either by 10% or by a specified quantity. The selected response variables are the annual average water table (ΔWT), net primary production (ΔNPP), soil microbial respiration (ΔRs), net ecosystem productivity (ΔNEP), and methane emissions (ΔCH4).
Water table is expressed as the annual average, while the others are expressed as the annual totals.
Changes of climate drivers are for each day based on daily climate data.
 The five state variables respond differently to different input variables. WT and NPP are sensitive to only a few model parameters and environmental variables, whereas Rs, CH4, and NEP respond strongly to many (Table 6). One simple reason is that calculating C fluxes like CH4 and NEP involves many components of Wetland-DNDC. WT is primarily influenced by hydrological parameters (e.g., the critical levels of outflow; D1 and D2 in (5)) and by climate variables (e.g., precipitation, temperature). A decrease in either D1 or D2 by 10 cm could lower the annual average WT by about 10 cm (Table 6); this is because outflow increases linearly with WT as soon as WT rises above these boundary levels. Such high sensitivity of WT to D1 and D2 indicates that good long-term WT data are essential since D1 and D2 (together with a1 and a2,) are calibrated parameters (5). Furthermore, although an increase in precipitation may produce limited effects on WT, a drought may exert greater influences on WT (Table 6). NPP is sensitive to the maximum photosynthesis rate (Amax,g in equation (18)) and the initial biomass, similar to what was observed by Aber et al.  for woody plants. In addition, while NPP shows little response to a decrease in temperature, it may be reduced by 12.4% with a 2°C temperature increase, perhaps because of increasing autotrophic respiration. Rs is significantly affected by temperature because temperature exerts direct effects on microbial activity and on soil moisture conditions. The simulations show that the number of days when WT is above the surface decreases dramatically from 171 days under the baseline condition to 85 days with a temperature increase of 2°C. Rs is also sensitive to changes in D1 and D2 and initial soil organic C. CH4 responds strongly to temperature and D1 and D2, and, to a lesser extent, to precipitation and initial soil conditions (e.g., organic C, pH, and porosity). For NEP, the most critical factors are temperature, D1 and D2, Amax,g, plant biomass, and initial soil organic C.
 To further test the interactions between plant, soil, and hydrology, we conducted two simulation experiments. First, we arbitrarily controlled the C substrates for CH4 production. Simulated annual CH4 emissions decrease 49.4% when C substrates from plants are eliminated, while simulated CH4 emissions decrease 69.6% when C substrates from soil decomposition are excluded. Second, we ran the model with constant water table levels (i.e., 20, 10, and 0 cm above the surface, and 10, 20, and 30 cm below the surface). When the water table is maintained at 10 cm above surface, annual CH4 emissions are the highest, 89% of the baseline emissions. When water table is kept at 20 cm above surface, annual CH4 emissions decrease by about 30% compared to that observed for a water table 10 cm above the surface due to a decrease in soil decomposition and soil temperature. This suggests that a fluctuating water table may be more favorable for CH4 emissions than a constant water table. Because the effect of water table on NPP is small, change in NEP is mostly due to the effect of water table on soil heterotrophic respiration. NEP is 1205.7 kg C·ha−1·yr−1 with a water table 20 cm above the surface and −706.8 kg C·ha−1·yr−1 with a water table 30 cm below the surface. The baseline NEP is 396.7 kg C·ha−1·yr−1. These results and the above sensitivity analysis show that C dynamics and CH4 emissions respond differently to input factors and they also strongly depend on the interactions among thermal/hydrological conditions, plant growth, and soil C dynamics. It is therefore critical for models to integrate hydrology, vegetation, soil, and climate in predicting C exchange and CH4 emissions of wetland ecosystems.
 A biogeochemical model, Wetland-DNDC, was developed by integrating the complex processes of hydrology, soil biogeochemistry, and vegetation in wetland ecosystems. In comparison to its parent model (PnET-N-DNDC), Wetland-DNDC includes several important changes, which enable the new model to capture the specific features of wetland ecosystems. The major improvements include functions and algorithms for simulating water table dynamics, the effects of soil composition and hydrologic conditions on soil temperature, C fixation by mosses or other ground growth species, and the effects of anaerobic status on decomposition and CH4 production/oxidation. Wetland-DNDC was tested against data sets of observed water table dynamics, soil temperature, CH4 flux, CO2 exchange, and annual C budget at three wetland sites in North America. The modeled results are in agreement with the observations. Sensitivity analysis indicates that wetland C dynamics are sensitive to temperature, water outflow rate, initial soil organic C content, plant photosynthesis capacity, and initial biomass in the wetland ecosystems. NEP and CH4 fluxes are sensitive to a wide scope of input parameters of climate, soil, hydrology, and vegetation. The ecosystem C dynamics as well as the CH4 emissions simulated by Wetland-DNDC respond to changes in thermal and hydrological conditions in a complex manner. The results further confirm the necessity of utilizing process models to integrate many interactions among climate, hydrology, soil, and vegetation for predicting C dynamics in wetland ecosystems.
 Wetland-DNDC currently does not include distribution hydrological routines to handle water inflow and outflow for a given watershed due to the complexity in calculation and amount of spatially differentiated input data required (e.g., topography, soil etc.). Instead, we focused this model at the site scale, and empirically parameterized the inflow and outflow indices for individual wetland watersheds where multiple-year observations of water table dynamics were available. For applying the model at large scales, these hydrological indices can be generated based on the normalized ranges or variations in water table and the lateral water fluxes. In this case, uncertainty analysis must be conducted to assess the potential errors produced from the generalized hydrological parameters.
plant aerenchyma factor of layer l
plant specific aerenchyma factor, (cm2).
maximum photosynthesis rate of ground vegetation (kg C·kg C−1·h−1).
maximum photosynthesis rate of woody plant (nmol CO2·g−1·s−1).
efficient photosynthetic biomass of ground vegetation (kg C·ha−1).
heat capacity of a layer (J·cm−3·°C−1).
heat capacity of component i (J·cm−3·°C−1).
C substrate for CH4 production in a soil layer (kg C·ha−1).
change rate of soil redox potential under saturated conditions (100 mv·day−1).
damping depth (cm).
critical depths for lateral outflow (cm).
depth of soil layer i (cm).
day length (h · day−1).
redox potential (mv).
water lost through evaporation from layer l (cm).
potential soil evaporation (cm).
potential evapotranspiration (cm).
effects of redox potential and soil moisture on decomposition.
volumetric fraction of component i in soil.
effects of growing degree days on amount of effective photosynthetic biomass.
fgL, fg,T, fg,W
effects of light, temperature, and water on photosynthesis of ground vegetation, respectively.
effects of soil moisture on evaporation and transpiration.
effects of redox potential on CH4 production.
effects of redox potential on CH4 oxidation.
effects of CH4 concentration on CH4 oxidation.
effects of pH on CH4 production.
effects of temperature on CH4 production.
effects of temperature on CH4 oxidation.
field capacity of soil layer l0 (cm3·cm−3).
net water input to layer l and l0, respectively, through infiltration, gravity drainage, and matric redistribution (cm).
the area of the cross-section of a typical fine root (0.0013 cm2).
gross photosynthesis of ground vegetation (kg C·ha−1).
thickness of layer l and l0, respectively (cm).
Julian date when solar altitude is the highest (200 for the Northern Hemisphere, 20 for the Southern Hemisphere).
a constant (5 μmol·L−1) for effects of CH4 concentration on CH4 oxidation (μ mol·L−1).
soil layer in which water table resides.
leaf area index (one side area) (m2·m−2).
CH4 content in a soil layer (kg C·ha−1).
CH4 concentration in a layer (μ mol·L−1).
CH4 content decrease through diffusion (kg C·ha−1).
CH4 emission through ebullition (kg C·ha−1).
CH4 emission through plant-mediated transport (kg C·ha−1).
CH4 oxidation rate of a soil layer (kg C·ha−1).
CH4 production in a soil layer (kg C·ha−1).
number of layers above water table level.
lateral outflow from the saturated zone (cm).
relative capacity of a plant for gas diffusion from its root system to the atmosphere.
plant interception of precipitation (cm).
fraction of CH4 oxidized during plant mediated transport (0.5).
porosity of soil layer l0 (cm3·cm−3).
squared correlation coefficient.
ratio of the area of a watershed to the area of a wetland in the watershed.
root length density of layer l (cm−2).
maximum root water uptake rate (0.003 cm2·day−1).
the surface-runoff fraction of precipitation in a watershed.
surface inflow (cm).
snowpack (cm water).
soil moisture of layer l and l0, respectively (cm3·cm−3).
soil temperature (°C).
top surface temperature on the current day and the day before, respectively (°C).
annual average air temperature (°C).
annual amplitude of air temperature (°C).
bottom layer soil temperature (°C).
daily mean air temperature (°C).
water lost through transpiration from layer l (cm).
potential plant transpiration (cm).
fraction of water filled pore space.
snowmelt (cm water).
soil moisture at wilting point of layer l (cm3·cm−3).
water table position in reference to soil surface (cm).
height of water table level above the bottom of layer l0 (cm).
amount of water required for a unit water table change (cm cm−1).
depth of the bottom soil layer (cm).
surface inflow relative to precipitation.
rate parameters for outflow.
thermal conductivity of a layer (W·cm−1·°C−1).
thermal conductivity of component i (W·cm−1 ·°C−1).
change of CH4 content in a soil layer (kg C·ha−1·day−1).
change of redox potential (mv·day−1).
change of soil moisture in layer l in the unsaturated zone (cm3·cm−3·day−1).
change of water table position (cm·day−1).
 The authors gratefully acknowledge the contributions of the source codes of PnET-N-DNDC provided by John Aber, Florien Stange, Klaus Butterback-Bahl, and Hans Papen. The authors are indebted to Nigel Roulet for supporting Yu Zhang to conduct his study in McGill University. The authors would like to thank the two anonymous reviewers for their comments and suggestions, and to Karen Bartlett for editing the paper. This study is funded by the USDA Forest Service Southern Global Change Program, U.S. multiagency TECO grant, NASA EOS-IDS grant, and the National Council on Air and Stream Improvement (NCASI).