Both CH4 and N2O fluxes from agro-ecosystems are highly variable in space and time, affected by ecological drivers (e.g., climate, vegetation, and anthropogenic activity), soil environmental factors (e.g., temperature, moisture, pH, redox potential, and substrate concentration gradients), and biochemical or geochemical reactions [Li, 2000, 2001; Li et al., 2004]. Process-based models are used to quantify trace gas fluxes driven by the local climate, soil, vegetation, and management conditions at the site scale. GIS databases provide spatially differentiated information of climate, soil, vegetation, and management to drive the model runs across the region. To quantify the impacts of the water management change on C sequestration and CH4 and N2O emissions from ∼30 million hectares of rice paddies in China, we integrated a process-based model, DNDC, with a GIS database of paddy area, soil properties, paddy management (fertilizer use, water management, crop residue management, planting and harvest dates), and daily weather data.
2.1. DNDC Model
 DNDC was originally developed for predicting carbon sequestration and trace gas emissions for non-flooded agricultural lands, simulating the fundamental processes controlling the interactions among ecological drivers, soil environmental factors, and relevant biochemical or geochemical reactions, which collectively determine the rates of trace gas production and consumption in agricultural ecosystems [Li et al., 1992, 1994, 1996]. Details of management (e.g., crop rotation, tillage, fertilization, manure amendment, irrigation, weeding, and grazing) have been parameterized and linked to the various biogeochemical processes (e.g., crop growth, litter production, soil water infiltration, decomposition, nitrification, denitrification) embedded in DNDC. To enable DNDC to simulate C and N biogeochemical cycling in paddy rice ecosystems, we modified the model by adding a series of anaerobic processes. The paddy-rice version of DNDC has been described and tested in recent manuscripts [Li et al., 2002; Cai et al., 2003; Li et al., 2004], and is summarized briefly here.
 Paddy soil is characterized by the frequent changes between saturated and unsaturated conditions driven by water management. During these changes in soil water content, the soil redox potential (i.e., Eh) is subject to substantial changes between +600 and −300 mV. One of the key processes controlling CH4 and N2O production/consumption in paddy soils is soil Eh dynamics; CH4 or N2O are produced or consumed under certain Eh conditions (−300 to −150 mV for CH4, and 200 to 500 mV for N2O), so the two gases are produced during different stages of the varying soil redox potential.
 To quantify Eh dynamics and its impacts on N2O and CH4 production, DNDC combines the Nernst equation, a basic thermodynamic formula defining soil Eh based on concentrations of the existing oxidants and reductants in the soil liquid phase [Stumm and Morgan, 1981], with the Michaelis-Menten equation, a widely applied formula describing kinetics of microbial growth with dual nutrients [Paul and Clark, 1989]. The Nernst and the Michaelis-Menten equations can be linked in model calculations since the two equations share a common factor, oxidant concentration. A simple kinetic scheme was adopted in DNDC to define the effective anaerobic volumetric fraction of a soil, based on the Eh value as calculated with the Nernst equation, using the concentration of the dominant active oxidant (in order of descending Eh: O2, NO3−, Mn4+, Fe3+, SO42−, CO2). Each oxidant is assigned an Eh range (Table 1). When a soil layer's Eh value is in a particular oxidant's range, the layer is divided into a high Eh zone (Eh = range upper value) and a low Eh zone (Eh = range lower value), as a linear function of where within the range the Eh value is. DNDC allocates substrates (e.g., DOC, NO3−, NH4+, CH4) to reductive reactions (e.g., denitrification, methanogenesis) and oxidative reactions (e.g., nitrification, methanotrophy) based on relative fractional volumes of the oxidizing and reducing zones, and the potential oxidation and reduction reactions are determined by Eh (and pH) [Stumm and Morgan, 1981]. When a soil is flooded, oxygen diffusion into the soil is severely restricted, and ongoing decomposition lowers the oxygen concentration, reducing the soil Eh and causing the low Eh volume fraction to increase. As the oxygen is depleted, the low Eh volume fraction reaches its maximum. At this moment, a new oxidant (i.e., NO3−) will become the dominant species in the soil, and a new, lower Eh volume fraction will begin to swell, driven by NO3− depletion. By tracking the formation and deflation of a series of Eh volume fractions driven by depletions of O2, NO3−, Mn4+, Fe3+, and SO42−, consecutively, DNDC estimates soil Eh dynamics as well as rates of reductive/oxidative reactions, which produce and consume CH4 or N2O in the soil. This links the soil water regime to trace gas emissions for rice paddy ecosystems, and DNDC predicts daily CH4 and N2O fluxes from rice fields through the growing and fallow seasons, as they remain flooded or shift between flooded and drained.
Table 1. DNDC Oxidation-Reduction Scheme
|Dominant Oxidant||Eh Range, mV||Dominant Reactions Simulated|
|O2||650–500||aerobic decomposition, CH4 oxidation, nitrification|
|NO3−, NO2−, NO, N2O||500–200||denitrification sequence|
 This new DNDC model has been tested against several methane flux data sets from wetland rice sites in the United States, Italy, China, Thailand, and Japan [Li et al., 2002; Cai et al., 2003]. For sites in East Asia, simulated seasonal CH4 emissions from paddy soils were in good agreement with field studies (r2 = 0.96, regression slope = 1.1, n = 23, range in fluxes 9 to 725 kg CH4-C ha−1 season−1) [Cai et al., 2003]. There were fewer sites with N2O flux data. If the site soil characteristics and crop and water management were well described, simulated fluxes were similar to observation (Figure 1) [Zheng et al., 1997]. However, for the three cases where the management details were not so well known, there were discrepancies between model and field, though there was overlap in the range in fluxes, 0.6 to 2.0 kg N2O-N ha−1 season−1 for the field and 0.4 to 5.7 kg N2O-N ha−1 season−1 for the model [Cai et al., 2003]. The results from the tests indicate that DNDC is capable of estimating the seasonal magnitudes of CH4 and N2O fluxes from paddy sites, although discrepancies exist for about 20% of the tested cases.
Figure 1. Comparison between observed and DNDC-modeled CH4 and N2O fluxes from a paddy rice field applied with mid-season drainage in Wu County, Jiangsu Province, China, in 1995. DNDC captured the episodes of CH4 emission depressions and N2O emission increases during the soil drying time periods by tracking the soil oxygen diffusion, CH4 oxidation, labile organic matter decomposition, and stimulated nitrification and denitrification fueled by the increased ammonium and nitrate production due to the conversions of soil anaerobic to aerobic conditions driven by the mid-season drainage. (Field data were adopted from Zheng et al. ).
Download figure to PowerPoint
2.2. GIS Database
 The model domain contained 30 million ha of rice paddies in China [Frolking et al., 2002]. There were 11 different crop rotations, including single-rice, double-rice, rice-winter wheat, rice-rapeseed, rice-rice-vegetable, etc. The area occupied by each rotation in each county was quantified by combining a county-scale statistical database of crop-sown areas with a Landsat TM-derived land-cover map for all of mainland China [Frolking et al., 2002]. The majority of rice production occurs in southern China, and particularly along the Yangtze River (Figure 2 [International Rice Research Institute (IRRI), 2004]). Daily weather data (maximum and minimum air temperatures, precipitation) for 1990 from 610 weather stations across China were acquired from the National Center for Atmospheric Research (http://dss.ucar.edu/index.html). Station data were assigned to each county on a nearest neighbor basis. Maximum and minimum values of soil texture, pH, bulk density, and organic carbon content were derived for each county from digitization of national soil maps [Institute of Soil Science, 1986] and other information [National Soil Survey Office of China, 1997]. Model simulations could then be conducted for each county by choosing soil parameter values spanning the observed range. Soil Mn, Fe, and sulfate contents were set to average values for Chinese paddy soils: Mn = 30 mg kg−1 soil, Fe = 80 mg kg−1 soil, and sulfate = 220 mg kg−1 soil [Li, 1992]. General data on fertilizer use, tillage, planting and harvest dates, crop residue management, and crop varieties were taken from Central Radio and Television School of Agriculture (CRTSA) , Huang et al. , Cui et al. , Liu and Mu , and Beijing Agricultural University . Shen  reported that based on national statistics, an average of 30% of total crop residue (leaves + stems + roots) was returned to the soil, which we adopted for all fields. Manure production was based on animal and human populations from the county database assembled by the Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences, Beijing, using standard manure production rates [Intergovernmental Panel on Climate Change (IPCC), 1997], and field application rates of 20% for animal manure and 10% for human manure.
Figure 2. Mean annual provincial rough rice production for 1991–2000 (IRRI 2004) in millions of tonnes per year. No provincial data were reported for Tibet (Xizang), Qinghai, Xinjiang, Gansu, Shanxi, Inner Mongolia, and Beijing; 1991–2000 total mean annual rough rice production for these seven provinces was 740,000 t yr−1. Taiwan was excluded from our entire analysis because of a lack of crop rotation data.
Download figure to PowerPoint
 Changing water management of China's rice paddies is not well documented. Li et al.  constructed a simple scenario of the evolution of paddy water management in China from 2000, as paddies went from near 100% continuous flooding in 1980 [Shen et al., 1998] to ∼80% midseason drainage by 2000 (Qingmu Chen, Chinese Academy of Agricultural Sciences, personal communication, 2002). In this study we wish to quantify the potential impacts of alternative water management practices, so we designed two water management scenarios: continuous flooding (CF) and midseason drainage (MSD) for 100% of the rice paddies in China.
 The county was chosen as the basic spatial unit for GIS database construction since most of the statistical cropland data was county-based. Meteorological data, soil properties, and agricultural management data were obtained from ground-based sources. Since each county is regarded to be uniform during a single model simulation, uncertainty estimates related to the inherent heterogeneities of many input parameters within the county must be generated during the scaling-up process. With ∼2500 counties, most with rice crops and many with several crop rotations, a set of annual simulations with a single parameter set required about 10,000 model runs, so Monte Carlo analysis (randomly adjusting all parameters thousands of times to get a statistical distribution of outputs) was computationally prohibitive.
 Instead, sensitivity tests were conducted to prioritize the environmental factors regarding their effects on CH4 or N2O emissions [Li et al., 2004]. Among the tested factors, including soil properties, temperature, and precipitation, the most sensitive factors for CH4 and N2O emissions were soil texture and soil organic carbon (SOC) content, respectively. Varying soil texture and SOC over the ranges reported in the county-scale database produced a range of CH4 and N2O emissions. The range in fluxes generated by this “most sensitive factor” method was compared to the distribution of fluxes generated by a Monte Carlo simulation for three paddy sites (Yunnan Province, China; Suphan Buri Province, Thailand; and California, United States); 70–98% of annual CH4 fluxes and 61–88% of annual N2O fluxes generated in 5000 Monte Carlo simulations for each site fell within the range generated by varying the most sensitive factor [Li et al., 2004].
 Applying the “most sensitive factor” method, we ran DNDC for each rice rotation in each county twice, once with low SOC, low pH, and high clay content, and once with high SOC, high pH, and low clay content, to produce a range in CH4 and N2O emissions wide enough to represent likely variations in actual fluxes caused by the heterogeneous nature of soil properties.
2.4. Emissions Calculations
 Of 11 possible paddy rotations mapped across China [Frolking et al., 2002], a single county typically had 2–4 paddy rotations. For each paddy crop rotation in each county, we simulated annual change in SOC (ΔSOC), and annual CH4 and N2O emissions for the range of soil conditions (as discussed in section 2.3). We consider ΔSOC as the negative of the site's net annual CO2 flux. Methane flux is another pathway for carbon out of wetland soil, but the substrates for methane production are primarily plant-derived C (e.g., root exudation, deposition, and respiration CO2) [Watanabe et al., 1999; Lu et al., 2002; King et al., 2002], which are not included in the simulation's SOC budget, so methane flux would only be a small fraction of simulated ΔSOC. We then summed the low and high emissions from all of the rice cropping systems in the county, based on each rotation's fraction of total cropland area, to get a range in total CO2, CH4, and N2O emissions for the county. We calculated a change in flux due to changing water management from continuous flooding to mid-season draining as the different between the mean MSD flux and the mean CF flux, where the mean flux is the average of the high and low fluxes.
 The climate impact of these flux changes can be compared by converting them to a common basis of CO2-equivalent fluxes using the global warming potential (GWP) methodology [Ramaswamy et al., 2001]. GWP values for CH4 and N2O depend on the time horizon chosen. Methane, with a much shorter atmospheric adjustment time than N2O or CO2, has very different 20-year, 100-year, and 500-year GWP values [Ramaswamy et al., 2001]. A CO2-equivalent flux for a 1-kg CH4 or N2O flux is the mass of CO2 emissions that would generate the same integrated radiative forcing, over the chosen time horizon, as a 1-kg emission CH4 or N2O. (Note that GWP calculations are done with molecular masses, not C or N masses.) For example, let ΔCO2, ΔCH4, and ΔN2O be the differences between the mean carbon dioxide, methane, and nitrous oxide fluxes (MSD minus CF) in kg C or N ha−1 yr−1. Then the 20-year time horizon CO2-equivalent flux, in kg CO2 ha−1 yr−1, is calculated as
where the numerical fractions in the second line are the ratios of molecular weights (CO2, CH4, and N2O) to elemental weights (C and N); equals 62 kg CO2-eq kg−1 CH4 and equals 275 kg CO2-eq kg−1 N2O; equals 23 kg CO2-eq kg−1 CH4 and equals 296 kg CO2-eq kg−1 N2O; equals 7 kg CO2-eq kg−1 CH4, and equals 156 kg CO2-eq kg−1 N2O [Ramaswamy et al., 2001].