Modeling Ch4 Emissions from Natural Wetlands on the Tibetan Plateau over the past 60 Years: Influence of Climate Change and Wetland Loss

The natural wetlands of the Tibetan Plateau (TP) are considered to be an important natural source of methane (CH 4) to the atmosphere. The long-term variation in CH 4 associated with climate change and wetland loss is still largely unknown. From 1950 to 2010, CH 4 emissions over the TP were analyzed using a model framework that integrates CH4MOD wetland , TOPMODEL, and TEM models. Our simulation revealed a total increase of 15% in CH 4 fluxes, from 6.1 g m ´2 year´1 to 7.0 g m ´2 year´1. This change was primarily induced by increases in temperature and precipitation. Although climate change has accelerated CH 4 fluxes, the total amount of regional CH 4 emissions decreased by approximately 20% (0.06 Tg—i.e., from 0.28 Tg in the 1950s to 0.22 Tg in the 2000s), due to the loss of 1.41 million ha of wetland. Spatially, both CH 4 fluxes and regional CH 4 emissions showed a decreasing trend from the southeast to the northwest of the study area. Lower CH 4 emissions occurred in the northwestern Plateau, while the highest emissions occurred in the eastern edge. Overall, our results highlighted the fact that wetland loss decreased the CH 4 emissions by approximately 20%, even though climate change has accelerated the overall CH 4 emission rates over the last six decades.

The atmospheric CH 4 concentration reached 1803.2 ppb in 2011, which was 150% greater than pre-1750 concentrations [9].The rate of CH 4 increase has been sustained over the past three decades, albeit with a temporary slowing to a near constant rate from 1999 to 2006 [10].Evidence suggests that the renewed increases in atmospheric CH 4 observed during 2007 and 2008 arose primarily from increased natural wetland emissions as a result of anomalously high temperatures in the Arctic and greater than average precipitation in the tropics [11,12].Compared with anthropogenic CH 4 sources, natural wetland sources are more variable, ranging from 115 Tg CH 4 year ´1 [13] to 237 Tg CH 4 year ´1 [14] at a global scale.
China has 304,849 km 2 of natural wetlands, accounting for 10% of the world's wetlands by area [15], and contributes 1.2%-3.2% to global wetland CH 4 emissions [16].Over one third of Chinese wetlands are situated on the Tibetan Plateau (TP) [17].Over the past 60 years, wetland loss has been reported on the TP.This wetland loss has been caused by global warming, leading to increased evaporation, subsequent increases in snow melting, and increased water outflow, as well as the draining and reclamation of the land as farmland [15].Large uncertainties exist in the estimation of wetland CH 4 emissions from the TP, ranging from 0.22 Tg year ´1 [17] to 1.25 Tg year ´1 [16].Most of the above estimations were based on the extrapolation of site-specific measurements of CH 4 fluxes to a regional scale [17][18][19][20].CH 4 emissions exhibit extreme spatial heterogeneity due to differences in climate, soil, topography, and vegetation throughout the TP.For example, the CH 4 emissions observed in the southeastern Plateau [21] were approximately five times greater than in the northwestern Plateau [17].Thus, the extrapolation approach may introduce uncertainties to regional estimations.
Compared with site-specific extrapolation methods, process-based models account for complexities in estimates of CH 4 emissions and are integrated with other processes, although site-specific parameters should be calibrated in order to give the model estimates regional reliability [22].In addition, process-based models can be used to estimate long-term historical CH 4 emissions, which were strongly influenced by climate change.For example, higher temperatures can increase the activity of methanogens and promote CH 4 production [23].Similarly, increased precipitation may result in a higher water table and subsequent acceleration in CH 4 flux [24,25].Moreover, alpine wetlands on the Tibetan Plateau are permafrost wetlands, which are quite sensitive to global climate change [26][27][28].However, less attention has been given to the evolution of regional CH 4 emissions from the TP in relation to climate change.
Recognizing the significance of climate change impacts on regional CH 4 budgets, this study focuses on quantifying the variation in CH 4 emissions from the TP via a process-based model.The objectives of this study are to estimate the change in regional CH 4 emissions from the wetlands on TP associated with climate change over the period from 1950 to 2010.

Model Framework
The model framework utilizes a biogeophysical, process-based model called CH4MOD wetland , which was developed for modeling CH 4 emissions from natural wetlands [29].The model adopted the hypothesis of the CH4MOD model [30,31], developed to simulate CH 4 emissions from rice paddies, with modifications based on the supply of methanogenic substrates in natural wetlands, which differ significantly from that of rice paddies.In CH4MOD wetland , methane production rates are calculated by the availability of methanogenic substrates and the parameterized influences of environmental factors-e.g., soil temperature, soil texture, and soil redox potential.The methanogenic substrates are derived from the root exudation of wetland plants and the decomposition of plant litter and soil organic matter.CH 4 transportation occurs via diffusion, ebullition, and plant transportation.Oxidation occurs when CH 4 diffuses to the atmosphere or is transported through the plant aerenchyma.Model inputs include daily soil temperature, water table depth, the annual above-ground net primary productivity (ANPP), and soil texture.The outputs are daily and annual CH 4 production and emissions.More details about CH4MOD wetland are well-documented in previous studies [29,32].
In previous studies [29,32], we calibrated the model parameters based on the observation of CH 4 emissions from wetlands on the Sanjiang Plain of northeast China.The main parameters that should be calibrated included vegetation index (VI), the fraction of CH 4 oxidized during plant-mediated transport (P ox ), and the fraction of plant mediated transport available (T veg ).In this study, we used the same calibration values for the main parameters as described in our previous studies.Table 1 shows the main input and parameter values at the study sites.In order to obtain the ANPP on a regional scale, we used the outputs of the TEM model.TEM is a process-based ecosystem model that simulates the biogeochemical cycles of C and N between terrestrial ecosystems and the atmosphere [38,39].This model has been widely used to investigate regional and global NPP (e.g., [40][41][42]).The TEM model also outputs soil temperature data, which is used as an input for CH4MOD wetland .
Regional water table depth was obtained from the TOPMODEL.This is a popular method used to simulate regional water table depth in natural wetlands.This method is based on the topographic wetness index (TWI), with ki " ln pα i {tanβ i q representing the spatially distributed water table depth for a 1 km sub-grid within a grid of 0.5 ˝, where α i is the contributing area upslope from point i, and tanβ i is the local surface slope at that point.The central equation of TOPMODEL is: where zi is the local water table depth in a 1 km pixel, z is the average water table depth in a 0.5 ˝grid, m is the scaling parameter, ki is the local topographic wetness index (TWI) in the 1 km pixel, and λ is the average of ki over the 0.5 ˝grid cell.The value of z is calculated by the soil moisture content in a 0.5 ˝grid cell.More details about this method are given in previous studies [43][44][45][46][47].

Data Sources
The data sources included site-specific observations for model validation, the gridded input data sets for developing the model framework, and the wetland area.We first validated the CH4MOD wetland at two wetland sites to test its ability to simulate CH 4 emissions from the TP; one site was located on the Zoige Plateau (32 ˝47 1 N, 102 ˝32 1 E; 3470 m above sea level), and the other was located on the Haibei alpine wetland (37 ˝29 1 N, 101 ˝12 1 E, 3250 m above sea level) (Figure 1).Site-specific observations were obtained from the literature.
The Zoige Plateau has the largest peatland in China.The dominant plant species are Carex meyeriana (CME) and Carex muliensis (CMU).Methane emissions at this site were measured from May to September of 2001 using static chambers and gas chromatography techniques [48].Synchronous measurements of the climate and water table depth were also taken during the experiment.More details about these measurements were described in previously published work [49].
At the site located in the Haibei alpine wetland, in the northeast part of the Qinghai-TP (Figure 1), the primary species of vegetation are Carex allivescers (CAL) and Hippuris vulgaris (HVU).Methane flux was measured by the same method of as on the Zoige Plateau every two weeks from early July to mid-September 2002.Synchronous measurements of the climate and water table depth, as well as the plant biomass, were also taken during the experiment.More details about these measurements can be found in previously published work [35].
Atmosphere 2016, 7, x 4 of 17 the plant biomass, were also taken during the experiment.More details about these measurements can be found in previously published work [35].The gridded input data sets for the model framework are related to climate, soil, vegetation, and hydrology.For the TEM model, climate datasets from 1950 to 2010 were developed using the latest monthly air temperature, precipitation, vapor pressure, and cloudiness datasets from the Climatic Research Unit (CRU TS 3.10) of the University of East Anglia in the United Kingdom [50].The soil texture data, used to assign texture-specific parameters to each grid cell in the TEM model, were derived from the soil map of the Food and Agriculture Organization of the United Nations [51].The IGBP vegetation map was referenced to specify the vegetation parameters for TEM.The map was derived from the IGBP Data and Information System (DIS) DISCover Database [52,53].
For the TOPMODEL, monthly soil moisture data from 1950 to 2010 were obtained from [54].The topographic wetness index data were obtained from the HYDRO1k Elevation Derivative Database, which was developed by the U.S. Geological Survey Earth Resources, Observation and Science (EROS) Center [55].
For CH4MODwetland, the monthly soil temperature data and the ANPP were obtained from the TEM model.The monthly water table depth data were obtained from the TOPMODEL.We used linear interpolation to develop the daily soil temperature and daily water table depth from 1950 to 2010.The soil sand fraction data were obtained from the Food and Agriculture Organization of the United Nations [51].The soil organic carbon content and the reference bulk density in wetland soils were from the Harmonized World Soil Database (HWSD) [56].
In this study, we used the definition of wetlands given by the US National Research Council (NRC) [57].For the purposes of this work, a wetland was defined as "an ecosystem that depends on constant or recurrent, shallow inundation or saturation at or near the surface of the substrate.The minimum essential characteristics of a wetland are recurrent, sustained inundation or saturation at or near the surface and the presence of physical, chemical, and biological features reflective of recurrent, sustained inundation or saturation.Common diagnostic features of wetlands are hydric soils and hydrophytic vegetation.These features will be present except where specific physiochemical, biotic, or anthropogenic factors have removed them or prevented their development".This definition is considered to be a relatively "narrow definition" of wetlands, as it does not include lakes and rivers.We used the remote sensing data of Niu, who developed gridded wetland maps for 1978, 1990, 2000, and 2008 with a resolution of 1 km × 1 km [15].The initial gridded wetland map for 1950 was estimated based on remote sensing data from 1978 [15] and the census The gridded input data sets for the model framework are related to climate, soil, vegetation, and hydrology.For the TEM model, climate datasets from 1950 to 2010 were developed using the latest monthly air temperature, precipitation, vapor pressure, and cloudiness datasets from the Climatic Research Unit (CRU TS 3.10) of the University of East Anglia in the United Kingdom [50].The soil texture data, used to assign texture-specific parameters to each grid cell in the TEM model, were derived from the soil map of the Food and Agriculture Organization of the United Nations [51].The IGBP vegetation map was referenced to specify the vegetation parameters for TEM.The map was derived from the IGBP Data and Information System (DIS) DISCover Database [52,53].
For the TOPMODEL, monthly soil moisture data from 1950 to 2010 were obtained from [54].The topographic wetness index data were obtained from the HYDRO1k Elevation Derivative Database, which was developed by the U.S. Geological Survey Earth Resources, Observation and Science (EROS) Center [55].
For CH4MOD wetland , the monthly soil temperature data and the ANPP were obtained from the TEM model.The monthly water table depth data were obtained from the TOPMODEL.We used linear interpolation to develop the daily soil temperature and daily water table depth from 1950 to 2010.The soil sand fraction data were obtained from the Food and Agriculture Organization of the United Nations [51].The soil organic carbon content and the reference bulk density in wetland soils were from the Harmonized World Soil Database (HWSD) [56].
In this study, we used the definition of wetlands given by the US National Research Council (NRC) [57].For the purposes of this work, a wetland was defined as "an ecosystem that depends on constant or recurrent, shallow inundation or saturation at or near the surface of the substrate.The minimum essential characteristics of a wetland are recurrent, sustained inundation or saturation at or near the surface and the presence of physical, chemical, and biological features reflective of recurrent, sustained inundation or saturation.Common diagnostic features of wetlands are hydric soils and hydrophytic vegetation.These features will be present except where specific physiochemical, biotic, or anthropogenic factors have removed them or prevented their development".This definition is considered to be a relatively "narrow definition" of wetlands, as it does not include lakes and rivers.We used the remote sensing data of Niu, who developed gridded wetland maps for 1978, 1990, 2000, and 2008 with a resolution of 1 km ˆ1 km [15].The initial gridded wetland map for 1950 was estimated based on remote sensing data from 1978 [15] and the census data [58].The wetland maps for other years were developed using the linear interpolation between the existing wetland between consecutive years.

Model Extrapolation to the TP
We established gridded (1 km ˆ1 km) and geo-referenced time-series input data sets of climate and soil data to drive the model described above and make spatiotemporal estimates of CH 4 fluxes from the wetlands on the TP.The gridded wetland maps, the topographic wetness index data and water table depths were created with a resolution of 1 km ˆ1 km.For the data set with coarse resolution (0.5 ˝ˆ0.5 ˝), we interpolated variables-e.g., climate, soil texture, and vegetation-over 1 km ˆ1 km using the nearest neighbor approach.We ran the model in each grid for the grid level CH 4 fluxes.The total CH 4 emissions from the inland and coastal wetlands in each grid cell were calculated as the product of the CH 4 fluxes and the gridded wetland area.

Model Validation
In this study, we used independent measurements of CH 4 fluxes to validate the model prior to extrapolation to the TP. Figure 2 shows the simulated and observed seasonal variations in CH 4 emissions from the Zoige and Haibei sites.In general, the model simulation was similar to the seasonal changes in CH 4 emissions from both sites (Figure 2a-d).However, there were some discrepancies between the simulated and observed CH 4 fluxes.For example, the model did not capture the higher CH 4 emissions from the Zoige CME site during June and September 2001 (Figure 2a).A systematic positive discrepancy between modeled and observed CH 4 emissions from the Zoige CMU site also occurred during the period from June to August 2001 (Figure 2b).For the Haibei alpine wetlands, the model overestimated CH 4 fluxes in early July and underestimated CH 4 fluxes in late July at both the CAL and HVU sites (Figure 2c,d).A regression of computed versus observed CH 4 emissions resulted in an R 2 of 0.53, with a slope of 0.59 and an intercept of 1.6 mg m ´2 h ´1 (n = 82, p < 0.001, Figure 3).
Atmosphere 2016, 7, x 5 of 17 data [58].The wetland maps for other years were developed using the linear interpolation between the existing wetland between consecutive years.

Model Extrapolation to the TP
We established gridded (1 km × 1 km) and geo-referenced time-series input data sets of climate and soil data to drive the model described above and make spatiotemporal estimates of CH4 fluxes from the wetlands on the TP.The gridded wetland maps, the topographic wetness index data and water table depths were created with a resolution of 1 km × 1 km.For the data set with coarse resolution (0.5° × 0.5°), we interpolated variables-e.g., climate, soil texture, and vegetation-over 1 km × 1 km using the nearest neighbor approach.We ran the model in each grid for the grid level CH4 fluxes.The total CH4 emissions from the inland and coastal wetlands in each grid cell were calculated as the product of the CH4 fluxes and the gridded wetland area.

Model Validation
In this study, we used independent measurements of CH4 fluxes to validate the model prior to extrapolation to the TP. Figure 2 shows the simulated and observed seasonal variations in CH4 emissions from the Zoige and Haibei sites.In general, the model simulation was similar to the seasonal changes in CH4 emissions from both sites (Figure 2a-d).However, there were some discrepancies between the simulated and observed CH4 fluxes.For example, the model did not capture the higher CH4 emissions from the Zoige CME site during June and September 2001 (Figure 2a).A systematic positive discrepancy between modeled and observed CH4 emissions from the Zoige CMU site also occurred during the period from June to August 2001 (Figure 2b).For the Haibei alpine wetlands, the model overestimated CH4 fluxes in early July and underestimated CH4 fluxes in late July at both the CAL and HVU sites (Figure 2c,d).A regression of computed versus observed CH4 emissions resulted in an R 2 of 0.53, with a slope of 0.59 and an intercept of 1.6 mg m −2 h −1 (n = 82, p < 0.001, Figure 3).The total amount of seasonal CH4 emissions at the two sites was determined as the summation of the daily values.The absence of CH4 emission measurements between consecutive days (Figure 2a-d) was linearly interpolated.The simulated seasonal CH4 emissions were approximately 8% higher than the observed values for the Zoige wetland, but were approximately 21% lower than the observed value for the Haibei wetland (Table 2).The model simulated the CH4 emissions from Zoige and Haibei wetlands with a model efficiency of 0.52.However, there was some bias between the simulated and observed values, with a RMSE value of 47.4% and a RMD value of −4.3%.

Temporal Variation in CH4 Emissions on the TP
Temporal variation in CH4 fluxes and the total amount of regional CH4 emissions are shown in Figure 4.The decadal mean area-weighted CH4 fluxes increased significantly from the 1950s to the 2000s (Figure 4a).The decadal mean CH4 flux was 6.1 g m −2 yr −1 and increased to 7.0 g m −2 yr −1 , with a total increase of 15% (Figure 4a).During the first 30 years of the modeled data, the decadal mean CH4 fluxes showed a decreasing trend.The lowest CH4 fluxes occurred in the 1960s, with a value of 5.8 g m −2 yr −1 (Figure 4a).The rate of increase in CH4 fluxes increased significantly from the 1980s, with a maximum value in the 2000s (Figure 4a).
The temporal variation in CH4 fluxes was strongly influenced by climate change during the 60 years examined.The increase in decadal mean air temperature from −1.4 °C in the 1950s to −0.51 °C in the 2000s promoted CH4 fluxes (Figure 4c).The lowest CH4 fluxes occurred in the 1960s (Figure 4a), which corresponded to the lowest temperatures (Figure 4c) and precipitation (Figure 4d).The highest temperatures (Figure 4c) and precipitation (Figure 4d) resulted in the high CH4 fluxes of the 2000s (Figure 4a).A significant positive correlation was found between the annual mean CH4 flux and temperature (Figure 5a), as well as between CH4 flux and annual precipitation (Figure 5b).This result suggests that a warmer, wetter climate accelerated CH4 fluxes from the wetlands on the TP over the past 60 years.The total amount of seasonal CH 4 emissions at the two sites was determined as the summation of the daily values.The absence of CH 4 emission measurements between consecutive days (Figure 2a-d) was linearly interpolated.The simulated seasonal CH 4 emissions were approximately 8% higher than the observed values for the Zoige wetland, but were approximately 21% lower than the observed value for the Haibei wetland (Table 2).The model simulated the CH 4 emissions from Zoige and Haibei wetlands with a model efficiency of 0.52.However, there was some bias between the simulated and observed values, with a RMSE value of 47.4% and a RMD value of ´4.3%.

Temporal Variation in CH 4 Emissions on the TP
Temporal variation in CH 4 fluxes and the total amount of regional CH 4 emissions are shown in Figure 4.The decadal mean area-weighted CH 4 fluxes increased significantly from the 1950s to the 2000s (Figure 4a).The decadal mean CH 4 flux was 6.1 g m ´2 year ´1 and increased to 7.0 g m ´2 year ´1, with a total increase of 15% (Figure 4a).During the first 30 years of the modeled data, the decadal mean CH 4 fluxes showed a decreasing trend.The lowest CH 4 fluxes occurred in the 1960s, with a value of 5.8 g m ´2 year ´1 (Figure 4a).The rate of increase in CH 4 fluxes increased significantly from the 1980s, with a maximum value in the 2000s (Figure 4a).
The temporal variation in CH 4 fluxes was strongly influenced by climate change during the 60 years examined.The increase in decadal mean air temperature from ´1.4 ˝C in the 1950s to ´0.51 ˝C in the 2000s promoted CH 4 fluxes (Figure 4c).The lowest CH 4 fluxes occurred in the 1960s (Figure 4a), which corresponded to the lowest temperatures (Figure 4c) and precipitation (Figure 4d).The highest temperatures (Figure 4c) and precipitation (Figure 4d) resulted in the high CH 4 fluxes of the 2000s (Figure 4a).A significant positive correlation was found between the annual mean CH 4 flux and temperature (Figure 5a), as well as between CH 4 flux and annual precipitation (Figure 5b).This result suggests that a warmer, wetter climate accelerated CH 4 fluxes from the wetlands on the TP over the past 60 years.
Atmosphere 2016, 7, x 7 of 17  Although CH4 fluxes showed an increasing trend (Figure 4a), CH4 emissions decreased on a regional scale during the past 60 years (Figure 4b).This decrease was driven by wetland loss.A total of 4.75 M ha of wetland existed in the 1950s, but that area decreased to 3.34 M ha in the 2000s (Figure 4b).As a result, the total amount of CH4 emissions decreased from 0.28 Tg in the 1950s to 0.22 Tg in the 2000s, a decrease of approximately 21% (Figure 4b).

Spatial Variation in CH4 Emissions from the TP
Spatial variation in CH4 emissions from the TP are shown in Figure 6.The highest CH4 fluxes occurred at the eastern edge of the Plateau, with peak fluxes as high as 40 g m −2 yr −1 (Figure 6a,b).CH4 fluxes showed a decreasing trend from the southeast to the northwest, with the lowest CH4 fluxes of 5 g m −2 yr −1 occurring in the northwestern Plateau (Figure 6a,b).Compared with the early 1950s, a   Although CH4 fluxes showed an increasing trend (Figure 4a), CH4 emissions decreased on a regional scale during the past 60 years (Figure 4b).This decrease was driven by wetland loss.A total of 4.75 M ha of wetland existed in the 1950s, but that area decreased to 3.34 M ha in the 2000s (Figure 4b).As a result, the total amount of CH4 emissions decreased from 0.28 Tg in the 1950s to 0.22 Tg in the 2000s, a decrease of approximately 21% (Figure 4b).

Spatial Variation in CH4 Emissions from the TP
Spatial variation in CH4 emissions from the TP are shown in Figure 6.The highest CH4 fluxes occurred at the eastern edge of the Plateau, with peak fluxes as high as 40 g m −2 yr −1 (Figure 6a,b).CH4 fluxes showed a decreasing trend from the southeast to the northwest, with the lowest CH4 fluxes of 5 g m −2 yr −1 occurring in the northwestern Plateau (Figure 6a,b).Compared with the early 1950s, a Although CH 4 fluxes showed an increasing trend (Figure 4a), CH 4 emissions decreased on a regional during the past 60 years (Figure 4b).This decrease was driven by wetland loss.A total of 4.75 M ha of wetland existed in the 1950s, but that area decreased to 3.34 M ha in the 2000s (Figure 4b).As a result, the total amount of CH 4 emissions decreased from 0.28 Tg in the 1950s to 0.22 Tg in the 2000s, a decrease of approximately 21% (Figure 4b).

Spatial Variation in CH 4 Emissions from the TP
Spatial variation in CH 4 emissions from the TP are shown in Figure 6.The highest CH 4 fluxes occurred at the eastern edge of the Plateau, with peak fluxes as high as 40 g m ´2 year ´1 (Figure 6a,b).CH 4 fluxes showed a decreasing trend from the southeast to the northwest, with the lowest CH 4 fluxes of 5 g m ´2 year ´1 occurring in the northwestern Plateau (Figure 6a,b).Compared with the early 1950s, a widespread enhancement of 0-2 g m ´2 year ´1 became apparent in the 2000s from the wetlands of the TP (Figure 6c).In some wetlands of the central Plateau, this enhancement was as high as 2-5 g m ´2 year ´1 (Figure 6c).Decreases in CH 4 fluxes were shown at the northeastern edge of the Plateau, with values of 0-2 g m ´2 year ´1 (Figure 6c).
widespread enhancement of 0-2 g m yr became apparent in the 2000s from the wetlands of the TP (Figure 6c).In some wetlands of the central Plateau, this enhancement was as high as 2-5 g m −2 yr −1 (Figure 6c).Decreases in CH4 fluxes were shown at the northeastern edge of the Plateau, with values of 0-2 g m −2 yr −1 (Figure 6c).
The grid-level regional emissions showed similar spatial patterns to that of CH4 fluxes (Figure 6d,e).The highest emissions occurred at the eastern edge of the Plateau, with values higher than 8 Gg yr −1 .The northwestern Plateau had the lowest emissions at 2 Gg yr −1 (Figure 6d,e).Unlike the CH4 fluxes (Figure 6c), a widespread decrease of approximately 0.5 Gg occurred in the 2000s compared with the 1950s (Figure 6f).The greatest decrease occurred in the eastern and northeastern Plateau, with changes as high as approximately 1.5 Gg yr −1 (Figure 6f).

Uncertainties in Model Validation
Model validation is important prior to extrapolation to a regional scale.A validation should use independent observed data that is not also used to calibrate the model [59].Previous studies demonstrated a better model performance than that shown here when simulating CH4 emissions from the Haibei wetland in the TP-e.g., Jin's study [60].However, as reported by Jin, the same dataset was used to both calibrate and validate the model.Here, we calibrated the model based on CH4 observations from the Sanjiang Plain obtained from previous studies [29,32].The parameters were unchanged for the validation of CH4 emissions from the TP.The model obtained estimates using different datasets for calibration and validation purposes, usually resulting in more accurate greenhouse gas flux predictions.
The uncertainties in model validation were induced in part by the variable water table depth, which is one of the environmental factors most sensitive to CH4 emissions [24,61].In the Zoige wetland, water table depth generally changed quickly and presented as patches from July to August 2001 [49].The rapid changes were difficult to capture, which may have resulted in discrepancies between the observed and simulated CH4 fluxes (Figure 2a,b).In addition, a negative bias occurred between the simulated and observed seasonal CH4 emissions from the Haibei wetland (Table 2), which may be explained in part by the sparse observations from this area.The proportions of observed CH4 fluxes extrapolated for the seasonal fluxes from the Haibei and Zoige wetlands were The grid-level regional emissions showed similar spatial patterns to that of CH 4 fluxes (Figure 6d,e).The highest emissions occurred at the eastern edge of the Plateau, with values higher than 8 Gg year ´1.The northwestern Plateau had the lowest emissions at 2 Gg year ´1 (Figure 6d,e).Unlike the CH 4 fluxes (Figure 6c), a widespread decrease of approximately 0.5 Gg occurred in the 2000s compared with the 1950s (Figure 6f).The greatest decrease occurred in the eastern and northeastern Plateau, with changes as high as approximately 1.5 Gg year ´1 (Figure 6f).

Uncertainties in Model Validation
Model validation is important prior to extrapolation to a regional scale.A validation should use independent observed data that is not also used to calibrate the model [59].Previous studies demonstrated a better model performance than that shown here when simulating CH 4 emissions from the Haibei wetland in the TP-e.g., Jin's study [60].However, as reported by Jin, the same dataset was used to both calibrate and validate the model.Here, we calibrated the model based on CH 4 observations from the Sanjiang Plain obtained from previous studies [29,32].The parameters were unchanged for the validation of CH 4 emissions from the TP.The model obtained estimates using different datasets for calibration and validation purposes, usually resulting in more accurate greenhouse gas flux predictions.
The uncertainties in model validation were induced in part by the variable water table depth, which is one of the environmental factors most sensitive to CH 4 emissions [24,61].In the Zoige wetland, water table depth generally changed quickly and presented as patches from July to August 2001 [49].The rapid changes were difficult to capture, which may have resulted in discrepancies between the observed and simulated CH 4 fluxes (Figure 2a,b).In addition, a negative bias occurred between the simulated and observed seasonal CH 4 emissions from the Haibei wetland (Table 2), which may be explained in part by the sparse observations from this area.The proportions of observed CH 4 fluxes extrapolated for the seasonal fluxes from the Haibei and Zoige wetlands were 27% and 8%, respectively.
The linear interpolation used to fill gaps between consecutive observed values in order to calculate the seasonal CH 4 emissions may disregard non-linear variations between dates.

Feedback between Climate Change and CH 4 Emissions
We concluded that the increased CH 4 fluxes were primarily induced by increasing temperature and precipitation over the past 60 years.A previous study [62] simulated CH 4 fluxes from China from 1949 to 2008 and also determined that climate change was the main factor influencing CH 4 fluxes.Zhu reported a significant positive correlation between soil temperature and CH 4 emissions, as well as water table depth and CH 4 emissions [63].This finding is consistent with our study (Figure 5).Such a relationship may arise because soil temperature is controlled by air temperature, and water table depth is primarily determined by precipitation.
Other simulation studies also demonstrated that historical climate change could increase CH 4 fluxes in different wetlands.Jin simulated annual mean CH 4 emissions from potential wetland areas on TP increased gradually from 6.3 g m ´2 y ´1 in 1979 to 7.4 g m ´2 y ´1 in 2010, an increase of 17% [60].Similarly, we estimated an increase of 13% over the same period, from 6.2 g m ´2 y ´1 in 1979 to 7.0 g m ´2 y ´1 in 2010 (Figure 3).The influence of climate change on CH 4 emissions may differ by region.For example, in the Sanjiang Plain, the CH 4 increase induced by climate warming has been offset by the decrease in annual precipitation over the last 60 years [32].
In addition to the temperature and precipitation, the increase in the atmospheric CO 2 concentration may have also influenced CH 4 emissions over the past 60 years.CO 2 fertilization increased the primary production (NPP) of plants [64,65].This could stimulate CH 4 emissions, as they are the main source of methanogenic substrates [66,67].As simulated by the TEM model in this study, CO 2 fertilization resulted in an increase in the plant NPP, with a rate of a rate of 2.2 g m ´2 per decade (data not shown).This increase also promoted CH 4 fluxes during the past 60 years.
CH 4 is an important greenhouse gas, and its emission has created a positive feedback loop effect for climate warming.According to our study, wetland loss on TP reduced the CH 4 emissions by approximately 20% (Figure 4b), which may decrease the global warming potential (GWP).However, the wetland loss also results in a loss of soil organic carbon (SOC) (e.g., [68][69][70][71]), releasing additional CO 2 into the atmosphere.In addition, draining wetlands may increase N 2 O emissions [72,73].Future studies should investigate the concomitant changes in CH 4 , CO 2 , and N 2 O emissions, as well as the consequent GWP.4.3.Estimates of the Regional CH 4 Emissions from the TP Regional CH 4 emissions were estimated using several methods and exhibited a high degree of uncertainty (Table 3).This uncertainty resulted from both the observed or simulated CH 4 fluxes, as well as the estimated wetland area.The site-specific method extrapolated the site-specific CH 4 fluxes to the entire region, which may obscure spatial variations in CH 4 fluxes (Figure 5).The earliest estimates of CH 4 emissions for the TP were from Jin [18], who reported estimated CH 4 emissions ranging from 0.7 to 0.9 Tg year ´1.This estimation was based on observations from the Huashixia wetland.Estimates from both Ding [19] and Chen [16] were primarily based on measurements from Zoige at the eastern edge of the TP.However, the two studies produced different estimates, with Chen estimating CH 4 emissions at 1.25 Tg year ´1, while Ding estimated only 0.56 Tg year ´1.This difference arose primarily because Chen used an observation of CH 4 fluxes that was much higher than the observation of Ding for the 2000s.Recently, Wei [17] made a challenge estimation, which considered the wetland species on the TP.According to Wei, swamp meadows release much less CH 4 than a typical swamp.Previous estimates that extrapolated the CH 4 fluxes from a typical swamp may have significantly overestimated regional CH 4 emissions.
In this study, the simulated spatial variation in CH 4 fluxes (Figure 5a,b) was consistent with previous observations.In the Zoige Plateau, located at the eastern edge of the TP, CH 4 fluxes ranged from 5 to 40 g m ´2 year ´1 (Figure 5a,b), which corresponded to the range of measurements previously reported (1.3 to 44.9 g m ´2 year ´1) [21,49,75].The simulated CH 4 fluxes around Namucuo wetlands ranged from 0 to 10 g m ´2 year ´1 (Figure 5a,b), which was consistent with prior observations (0.6 to 12.6 g m ´2 year ´1) [17].The simulated CH 4 fluxes around Huashixia wetlands ranged from 0 to 10 g m ´2 year ´1 (Figure 5a,b), which also corresponded to prior observations (ranging from 2.4 to approximately 10.5 g m ´2 year ´1) [18].Xu and Tian [62] also simulated CH 4 fluxes ranging from 0 to 27 g m ´2 year ´1 from the TP.In contrast, the simulated range of Jin [60] was lower than that found in this study, with a range from 0 to 8 g m ´2 year ´1.
The estimated wetland area ranged from 3.2 M ha to 18.8 M ha (Table 3).The variation may have been due to the estimation method.Most previous studies used the survey wetland area (e.g., [18,19,74]).Jin [60] estimated an area of 13.4 M ha based on the soil wet extent [76].Our area estimate was based on remote sensing data [15] and was consistent with both Xu and Tian's study [62] and Zhang and Jiang's study [74].

Uncertainties and Future Needs
This study estimated regional CH 4 emissions from natural wetlands in TP over the past 60 years.However, uncertainties still persisted in the estimations due to an incomplete model structure, model inputs, and inaccuracies in the defined wetland area.
First, some physical and biogeochemical processes are still neglected in CH4MOD wetland .For example, the impact of nitrogen deposition on CH 4 production and oxidation was not considered in this model.Nitrogen deposition can regulate plant growth and microbial activities [77,78].Nitrogen addition stimulates plant growth [79], and litter with higher N levels decomposes faster [80], leading to increased CH 4 emissions.In addition, nitrate can decrease CH 4 production by increasing redox potentials [81], and ammonium usually inhibits CH 4 oxidation by competing for methane monooxygenase [82].It was previously reported that nitrogen deposition has increased during the past 60 years in China [83].The impact of nitrogen deposition on CH 4 emissions should be considered in CH4MOD wetland in the future to decrease the uncertainties in the long-term estimations.
Secondly, model inputs-especially the spatial variability in the water table depth-account for a large proportion of the uncertainty in regional estimations.TOPMODEL has been widely used to simulate the water table distribution of the natural wetlands.It has been validated for both site-specific water table seasonal variation (e.g., [44]) and the spatial variation on the regional and global scales.On the regional scale, areas where the water table is at or above the soil surface level can be interpreted to correspond to the surface water extent.Thus, the validation of the spatial distribution of the water table depth was usually compared to remotely sensed inundation datasets (such as GIEMS [76,84], or [85]) and wetland and land cover mapping products (including [86,87]).For example, Kleinen [45] and Melton [87] showed reasonable validation by comparing the monthly global distribution of the water table with remote sensing data [76,84] as well as the GLWD map [86].Similar validations conducted on the regional and global scale have also been reported in previous studies [88,89].However, the limited resolution would inevitably induce bias into the water table variations-especially on TP, where the natural wetland has a large degree of microscale topographic variation [17].More accurate descriptions of the hydrology process and higher-resolution datasets will be needed to reduce the error in the simulated water table depth.
Last but not least, popular methods for defining the extent of wetlands include using "Prescribed constant wetland extents" (as was done here) and the "Hydrological model" [90,91].The latter method uses a model to simulate the dynamical wetland extent.However, both methods may produce uncertainties [88].Improving the ability to obtain accurate data on the distribution and extent of wetlands should be a research priority in the future [63].

Future Trends in CH 4 Emissions from Natural Wetlands on TP
The projected annual mean air temperature will increase by 17 ˝C and 3.9 ˝C under the RCP 4.5 and RCP 8.5 scenarios, respectively, which were designed in the IPCC fifth assessment report (AR5) [92,93] on TP by 2100 [94].The projected annual precipitation will also increase by 19%-22% and 37%-44% under the RCP 4.5 and RCP 8.5 scenarios, respectively, by the end of the 21st century [94].Climate change can influence the CH 4 fluxes and the wetland area.According to our results (Figure 5), increasing the temperature and precipitation would increase the CH 4 fluxes.In addition, future climate would change the wetland area.The rising temperature will increase evapotranspiration, thus decreasing the wetland area.However, this may be balanced by glacial retreat and more precipitation, increasing the water supply to wetlands [15].
The Chinese government has increasingly recognized the importance of wetland protection, particularly after joining the Ramsar Convention in 1992.Thus, the shrinkage and degradation of the wetland began to be reduced after 1990 (Figure 4b).According to the China National Wetland Conservation Action Plan (NWCP) by the Chinese government, 1.4 ˆ10 9 ha of wetland will be restored by 2030 [95].Thus, the NWCP will result in the expansion of wetlands on TP in the future, which may promote regional CH 4 emissions.

Conclusions
The temporal and spatial patterns of CH 4 emissions from natural wetlands on the TP from 1950 to 2010 were simulated using a model framework that integrated the CH4MOD wetland , TEM, and TOPMODEL models.Model validation at the site level indicated that the model provided a reasonable description of the observed CH 4 emissions from the TP.Simulation results showed that CH 4 fluxes increased by 15%, from 6.1 g m ´2 year ´1 in the 1950s to 7.0 g m ´2 year ´1 in the 2010s.This change in fluxes was primarily induced by increases in air temperature and precipitation.However, during the same period, CH 4 emissions were reduced by an estimated 0.06 Tg year ´1 in the TP wetlands, which was primarily due to extensive wetland loss from 4.75 million ha to 3.34 million ha.On a regional scale, CH 4 fluxes ranged from 0 to 40 g m ´2 year ´1.The lowest and highest CH 4 emissions occurred at the northwestern and eastern edges of the Plateau, respectively.To decrease the model uncertainty in estimates of regional CH 4 emissions, accurate simulations of CH 4 fluxes and estimates of wetland area are needed in the future.

Figure 1 .
Figure 1.The distribution of natural wetlands across the Tibetan Plateau (data from Niu et al., 2012 [15] ) and location of the study sites.

Figure 1 .
Figure 1.The distribution of natural wetlands across the Tibetan Plateau (data from Niu et al., 2012 [15]) and location of the study sites.

Figure 2 .
Figure 2. Simulated and observed seasonal variations of CH4 emissions and the observed air temperatures and water table depths.(a), (b), (c) and and (d) are the CH4 emissions from the Zoige CME site, the Zoige CMU site, the Haibei CAL site and the Haibei HVU site; (e), (f), (g) and (h) are the air temperatures and water table depths from the Zoige CME site, the Zoige CMU site, the Haibei CAL site and the Haibei HVU site.

Figure 2 .
Figure 2. Simulated and observed seasonal variations of CH 4 emissions and the observed air temperatures and water table depths.(a), (b), (c) and and (d) are the CH 4 emissions from the Zoige CME site, the Zoige CMU site, the Haibei CAL site and the Haibei HVU site; (e), (f), (g) and (h) are the air temperatures and water table depths from the Zoige CME site, the Zoige CMU site, the Haibei CAL site and the Haibei HVU site.

Figure 3 .
Figure 3.Comparison of observed and simulated CH4 fluxes

Figure 3 .
Figure 3.Comparison of observed and simulated CH 4 fluxes.

Figure 4 .
Figure 4. Temporal variations of (a) simulated decadal mean CH4 fluxes, (b) decadal mean wetland area and regional CH4, (c) decadal mean air temperature, (d) decadal mean precipitation.Boxplots show the median, average values, and interquartile range, with whiskers extending to the most extreme data point within 1.5 × (75% -25%) data range.Triangles represent the decadal mean wetland area.

Figure 5 .
Figure 5. Regression between annual mean CH4 fluxes and (a) air temperature and (b) annual precipitation.

Figure 4 .
Figure 4. Temporal variations of (a) simulated decadal mean CH 4 fluxes, (b) decadal mean wetland area and regional CH 4 , (c) decadal mean air temperature, (d) decadal mean precipitation.Boxplots show the median, average values, and interquartile range, with whiskers extending to the most extreme data point within 1.5 ˆ(75% -25%) data range.Triangles represent the decadal mean wetland area.

Figure 4 .
Figure 4. Temporal variations of (a) simulated decadal mean CH4 fluxes, (b) decadal mean wetland area and regional CH4, (c) decadal mean air temperature, (d) decadal mean precipitation.Boxplots show the median, average values, and interquartile range, with whiskers extending to the most extreme data point within 1.5 × (75% -25%) data range.Triangles represent the decadal mean wetland area.

Figure 5 .
Figure 5. Regression between annual mean CH4 fluxes and (a) air temperature and (b) annual precipitation.

Figure 5 .
Figure 5. Regression between annual mean CH 4 fluxes and (a) air temperature and (b) annual precipitation.

Table 1 .
Site-specific parameters and model inputs of CH4MOD wetland .
ˆModel parameters; * Model inputs; a For the Carex meyeriana site (CME); b For the Carex muliensis site (CMU); c For the Carex allivescers site (CAL); d For the Hippuris vulgaris site (HVU).

Table 2 .
Model performance on the study sites.

Table 2 .
Model performance on the study sites.

Table 3 .
Regional CH 4 estimations from wetlands in the Tibetan Plateau after 1990.
# Nm means not mentioned in the reference.