Climate-driven increase of natural wetland methane emissions offset by human-induced wetland reduction in China over the past three decades

Both anthropogenic activities and climate change can affect the biogeochemical processes of natural wetland methanogenesis. Quantifying possible impacts of changing climate and wetland area on wetland methane (CH4) emissions in China is important for improving our knowledge on CH4 budgets locally and globally. However, their respective and combined effects are uncertain. We incorporated changes in wetland area derived from remote sensing into a dynamic CH4 model to quantify the human and climate change induced contributions to natural wetland CH4 emissions in China over the past three decades. Here we found that human-induced wetland loss contributed 34.3% to the CH4 emissions reduction (0.92 TgCH4), and climate change contributed 20.4% to the CH4 emissions increase (0.31 TgCH4), suggesting that decreasing CH4 emissions due to human-induced wetland reductions has offset the increasing climate-driven CH4 emissions. With climate change only, temperature was a dominant controlling factor for wetland CH4 emissions in the northeast (high latitude) and Qinghai-Tibet Plateau (high altitude) regions, whereas precipitation had a considerable influence in relative arid north China. The inevitable uncertainties caused by the asynchronous for different regions or periods due to inter-annual or seasonal variations among remote sensing images should be considered in the wetland CH4 emissions estimation.


Simulation performance
To evaluate and separate the effects of wetland area dynamics and climate change on China's wetland CH4 emissions (between 1978 and 2013)，simulations were driven with different composition of historical climate data and remote sensing based wetland distribution data (Supplementary Table S1).
The 1:1,000,000 China soil dataset was used to generate the initial soil carbon content, the fractions of sand, clay, and silt, and the soil pH for each cell.  Table S2) and most of the PFT phenological and physiological parameters were adopted from the C3 grass PFT in the original model 3 . The definition of inundation stress effects on gross primary productivity (GPP) of the added PFT in wetlands followed the assumption made by Wania et al. (2009) 4 , that sphagnum and C3 graminoids photosynthesis will increase or decrease when water table rises or drops. The wetland PFT would be kept fixed over wetland regions during simulation periods.
For each simulation, a 300-year spin-up procedure, running with multi-year (between 1960 and 2000) averaged historical meteorological data, was set up and allowed the ecosystem carbon pools to reach a relative equilibrium state. For reaching soil carbon equilibrium, the model has an internal speed-up process during the soil spin-up period. It allows the model to run up to 40 times of additional soil carbon cycling during one normal simulation day, which means a 300-year soil carbon spin-up has up period, 1951-1978, 1979-1990, 1991-2000, and 2001-2013 (Supplementary Table S1 Table S1). In each simulation, the wetland distribution was kept unchanged by using the wetland map of 1978, 1990, 2000, or 2008. For different simulations, only result slices of particular years (1978, 1990, 2000, 2008, and 2010-2013) were extracted for analysis (Supplementary  Tropical broadleaf drought-deciduous trees 3 Warm-temperate broadleaf evergreen trees 4 Temperate conifer evergreen trees 5 Temperate broadleaf cold-deciduous trees 6 Boreal conifer evergreen trees 7 Boreal broadleaf cold-deciduous trees 8 Boreal conifer cold-deciduous trees 9 Evergreen shrubs 10 Cold-deciduous shrubs 11 C4 grasses 12 C3 grasses 13 Wetland vegetation

Model evaluation at site level
The wetland CH4 emission modeling performance of TRIPLEX-GHG was evaluated using global field measurements from a previous study 3,5 . Observed data, which were collected from more than 10 studies across China, are used to evaluate the model. The results indicated that the TRIPLEX-GHG

National level estimation and comparison
Early estimations of wetland CH4 emissions in China, i.e., 1.70 TgCH4 yr -1 8 and 2.20 TgCH4 yr -1 9 around 1990, were conducted using very simple approaches. Based on the observed emission rate at only one permafrost station located in the QTP, Jin et al. 10  lacked detailed temporal or spatial information on estimations 8,9 . In the study of Jin, et al. 10 , a uniform CH4 emission rate (as measured in the QTP) was assumed for all wetlands in the country, including coastal wetlands. In the studies of Ding, et al. 11 and Ding and Cai 12 , the methane emission during the unmeasured period was set to 15-23% of that measured in the growing season to calculate the annual CH4 emissions, and the estimated wetland area in the studies was approximately 9.4×10 4 km 2 , which is lower than that derived in our study. Using the assumption that CH4 emissions are proportional to the area of wetlands at a national scale, an updated annual CH4 emissions estimation of approximately 4.5 TgC was suggested in a follow-up study, based on a new evaluation of wetland area 16 . The CH4 emissions are much larger than those estimated in all other studies to date.
Although the estimated national CH4 emissions strongly agreed with those of similar studies, considerable differences exist in regional estimations. For example, Chen, et al. 14  emissions. However, CH4 released from wetlands could have notable diurnal variations, increasing in the morning, reaching a peak at noon, decreasing in the afternoon, and decreasing by approximately 50% to 80% under dark conditions at night 17 . Therefore, bias will be introduced in the estimation of annual CH4 emissions using the inventory method if diurnal, daily, and seasonal variations are not considered.

Relationship between climate (precipitation and temperature) and wetland CH 4 emissions
A long-term analysis (1951-2013) was conducted on the fixed wetlands (i.e., the areas that remained wetlands from 1978 to 2008) to investigate the effects of precipitation and temperature on wetland CH4 emissions. The correlation coefficient and the significance level between the CH4 emission rate and precipitation or temperature were calculated for each wetland grid cell, and only those grid cells with statistical significance and a determination coefficient greater than 0.3 are shown in Figure 3b, 3c.
For some northern areas, precipitation was significantly (P<0.05) and positively correlated with wetland CH4 emissions and was responsible for more than 30% of the variation in wetland CH4 emissions (R2≥0.3) (Fig. 3b). Over most wetland areas in the NE, QTP, and SCN regions, nonsignificant relationships were detected between precipitation and CH4 emissions.
However, wetland CH4 emissions were extremely and significantly (P<0.001) correlated to temperature in most areas, with an R 2 value greater than 0.3, particularly in the QTP and NE regions (Fig. 3c). For wetlands in the western part of the QTP, the temperature explained more than 60% of the CH4 emission variation. Combined positive effects of temperature and precipitation were detected in some areas in the northeast QTP and in the northwesternmost region of China (Fig. 3b, 3c).
The high spatial heterogeneity in wetland CH4 emissions was caused by the substantial spatial variations in the driving variables, including precipitation, temperature, soil temperature, soil moisture, soil text, soil pH and soil redox potential, which control CH4 production and consumption. The wetlands in South China showed the highest CH4 emission rates for two reasons. First, abundant precipitation would keep the water table at a relatively high level and maintain good aerobic conditions for methanogenesis. Second, high plant primary production and organic matter decomposition rates provide full substrate availability, which is also an important control over methane production 18 .
According to the highly significant positive correlation between CH4 emissions and temperature outlined in this study, temperature could be one of the most important controls over CH4 emissions in the high-latitude (NE) and high-altitude (QTP) areas. This link could partially explain why wetlands in the NE and QTP were more sensitive to climate warming than other regions.
In northern China, the significant positive correlation between CH4 emissions and precipitation indicated that precipitation had a considerable influence on wetland CH4 emissions. Precipitation became the limiting factor for the wetlands located in the driest area of China. In the water-limited areas, increased precipitation would increase water table position, reduce the oxic portion of the soil and decrease the oxidative loss of CH4 19 . In other regions, temperature is more important to wetland CH4 emissions than precipitation. For example, the CH4 emissions are significantly correlated with temperature but not precipitation in the Sanjiang Plains (northeastern China). In a study conducted in this area, Song, et al. 20 also found that CH4 exponentially increased with temperature but found no significant relationship with water depth (controlled by precipitation). Precipitation has been suggested to have a smaller influence on wetland CH4 emissions than temperature in China 13 and globally 19 . The bacteria that produce CH4 has been found to be more sensitive to temperature than other variables and thus a change in temperature may significantly control the emissions 19,21,22 .
Only the straightforward relationships between wetland CH4 emissions and precipitation or temperature are analyzed in this study. CH4 emissions from wetlands are also influenced by other factors, such as soil redox potential, pH, salinity, the quantity and quality of methanogen substrates, water depth, and topography 18, [23][24][25][26] . Further investigations should consider these factors, particularly in the areas where CH4 emissions had no significant relationship with either precipitation or temperature, for example, in the most northern part of Northeast China.

Uncertainties and sensitivity analyses of the impacts of wetland dynamics on CH 4 emissions
The remote sensing data based wetland dynamics of China from 1978 to 2008 in this study was derived from the research of Niu, et al. 27 . The mapping was verified and validated by manually interpretation, with help of other reference data such as digital elevation model (DEM) data, land use/cover data, and Google Earth information 27 . However, some uncertainties still exist. On one hand, it is a challenge to gather all remote sensing images covering the whole country at a specified time, particularly for early years and need to span the time window approximately 3-5 years to complete map over whole country 27 . The wetland dynamic patterns were actually described for a period (around a base year) other than a specific year 27 . On the other hand, the retrieved wetland dynamics based on remote sensing data used in this study could describe a general trend on the patterns of wetland distribution, but unfortunately, further intra-annual dynamics information could not be obtained 27 . Using multiple sources of remote sensing data (e.g. from active and passive sensors) and applying a correlation analysis between wetland change and environmental factors will certainly reduce the uncertainty and improve the accuracy of wetland mapping.
To investigate the impacts of the inter-and intra-annual dynamics of wetland area on wetland CH4 emissions, additional simulations were carried out using the most recent inundated area dataset of Surface WAter Microwave Product Series (SWAMPS, http://wetlands.jpl.nasa.gov) 28 . From the global dataset, the monthly wetland fraction for each 0.5º×0.5º grid was extracted for the whole China during

Wetlands mapping of China based on remotely sensed data
The wetland area dynamics, one of the input data for the model, was directly derived from the previous served as reference images, and the other period images (1978, 2000, and 2008) were corrected by using the image-to-image method with ENVI software. The georegistration error was constrained to within two pixels for CBERS-02B images and less than one pixel for ETM and MSS as measured by root-mean-square error 27,33 . Given the heterogeneity of China's landscape, manual interpretation was chosen over automation to map wetland vegetation and most of the classification was done by visual interpretation 27,33,34 .
Accuracy assessment was also made for the wetland mapping. For example, in the study of Gong et al. 32 , the uncertainty and interpretation errors of wetland area change between 1990 and 2000 were evaluated from two aspects: misregistration and misinterpretation. The conversion error matrix indicated that inland natural wetlands has the false alarm error between 1% and 2% 32 . Weighted by area, the total error in change area statistics due to misregistration-caused false alarm would be less than 2% 32 . The overall image interpretation errors for inland natural wetlands were estimated to be 1.3% 32 . By taking into account the misinterpretation of wetland classes, the overall error for the inland natural wetlands should be less than 5% 32 .
Wetland reconnaissance for the 2008 map was undertaken from July 2009 to September 2009 and more than 10,000 field photos were taken and 459 qualified samples (located at representative wetland regions including the Qinghai-Tibet Plateau, the Northeast Plain, the Yellow River delta, east and southeast of China) were collected to validate the quality of wetland mapping for the period of 2008 27 . The overall accuracy of wetland and nonwetland was 0.98 while the overall accuracy of wetland types was 0.70 and kappa coefficient is 0.63 27,33 . Both of producer accuracy and user accuracy for inland wetlands were 0.83 33 . After validation of the wetland map in 2008, the other three maps (circa type was constructed 33 . According to the change model of wetland types between different periods, combinations were identified as impossible wetland transformation, such as from coastal wetlands to inland wetlands, and were rechecked on the images, and the corresponding map was revised accordingly 27,33 . Auxiliary materials and literature related to wetland distribution were collected to validate regions we could not confirm during the visual interpretation. The Google Earth information was fully utilized for the visual interpretation and validation processes 27,33 .
Supplementary Figure S5. An example of natural wetland fraction distribution of Northeast China. The maps were generated with ArcGIS 10.2, http://www.esri.com/.