Recent studies indicate that plants may be a previously overlooked but significant source of atmospheric CH4, though there is considerable disagreement on the mechanism of production. Our work sought to verify that woody deciduous trees grown under inundated conditions had the capacity for transporting CH4 from an anaerobic subsurface to the atmosphere and to consider if such a source could be important globally. Here, we report results from a greenhouse mesocosm study that indicate significant emissions of anaerobically produced CH4 transmitted to the atmosphere through broadleaf riparian tree species grown under flooded conditions. Using a leaf area normalized mean emission rate (0.7 ± 0.3 μg cm−2 hr−1), results were scaled globally for flooded forest regions and estimated to be 60 ± 20 Tg year−1, ∼10% of the global CH4 source. The carbon isotopic composition of CH4 emitted was found to be significantly enriched compared with expectations (δ13C ∼ −54‰) and provided an important isotopic constraint on the global source which coincides with the mean of the globally scaled greenhouse-based estimate.
 Though there has been significant research on emissions of CH4 from natural wetlands, early work concluded that the majority of emissions in natural wetland systems are mediated by aquatic macrophytes and through ebullition [Cicerone and Oremland, 1988]. Several more recent studies have indicated that woody tree systems could present a mechanism for transporting CH4 to the atmosphere from an anaerobic root zone under inundated conditions or, potentially, bypassing an aerobic oxidation layer that lies between deep roots and the atmosphere [Garnet et al., 2005; Megonigal and Guenther, 2008; Rusch and Rennenberg, 1998; Terazawa et al., 2007]. These tree emissions could particularly enhance CH4 flux in tropical regions that experience regular seasonal inundation.
 In this paper, we present a study of emissions of anaerobically produced CH4 from three deciduous riparian tree species and a global upscaling of results to determine the potential of tree emissions to impact the global CH4 budget.
2. Experimental Methods
 Three woody riparian tree species were grown in small mesocosms and studied in a research greenhouse: ash (Fraxinus latifolia), cottonwood (Populus trichocarpa), and willow (Salix fluviatillis). These experiments were conducted adjacent to ongoing mesocosm experiments with a rice cultivar reference (Oryza sativa L. ‘M-103’) and unplanted control plots. Plants were grown in triplicate in a sandy loam (71.8% sand, 25.6% silt, 2.6% clay and 3.23% organic matter, 1.6% C, 0.12% N, 0.01% S, 126 ppm P, 101 ppm K) contained in fiberglass tubs (61 × 48 × 36 cm). Soil organic content was enhanced through the addition of rice straw equivalent to 3 tons per hectare to stimulate below ground anaerobic methane production. All plants and controls were grown under fully inundated conditions except during one mid-season drainage.
 Static flux samples were drawn from translucent chambers enclosing each plant and its air-water interface approximately two times weekly between July and October primarily during morning hours (9–11 AM local time). Chambers were reinforced translucent polyethylene sheeting on a frame of PVC pipe (66 × 51 × 90 cm), with a gas sampling port and a 12 Volt battery powered fan that stirred the air inside the chamber during sampling. Tedlar bag branch enclosures were also used to confirm CH4 emissions through the tree biomass. Leaf area and above ground biomass were determined destructively at the end of the experiment.
 Over the course of the study, approximately 60 fluxes were measured from each species. Samples from inside the chambers were removed at 10 minute intervals for 30 minutes and CH4 concentrations were measured on a Agilent model 6890 gas chromatograph with a flame ionization detector (GC-FID) [Khalil et al., 1998]. Net CH4 flux is determined by linear regression of the change in concentration with time (ΔCH4/Δt) and expressed m−2 of water-atmosphere interface [Khalil et al., 1998]. Analyses were filtered for strong ebullition events and operator error, identified by their non-linearity (r2 < 0.9, ∼10% of data). Belowground porewater samples at 5, 10, 15, and 20 cm depths were collected bi-weekly and CH4 concentrations were measured by extracting samples in ultra-high purity N2, and analyzing them via GC-FID.
 During a sampling intensive period, flux (t = 30 min) and porewater samples were drawn once per week and measured for the carbon isotopic composition (δ13C) of CH4. The δ13C of CH4 was determined by continuous-flow gas chromatography-isotope ratio mass spectrometry on a Thermo Scientific Delta V Advantage IRMS using a method previously described [Rice et al., 2001]. Values of δ13C of CH4 were measured relative to a calibrated CO2 reference gas and are reported relative to the VPDB scale using the delta (δ) notation such that δ13C = [(13C/12Csample/13C/CVPDB) − 1] × 1000 as established by the International Atomic Energy Agency (IAEA) in Vienna, Austria [Coplen, 1995]. All samples were corrected for the ambient atmosphere in the greenhouse (∼2 ppm CH4), collected at initial chamber placement (t = 0 min).
 For purposes of global upscaling of results, inundated forests were identified using the Global Land Cover 2000 product with a spatial resolution of ∼1 km [Bartholomé et al., 2002]. Land cover classification included regularly and permanently flooded forests. We assumed that tropical riparian forests were flooded annually for five months which is consistent with the duration of the wet season and measurements of tropical inundation using satellite techniques [Eva et al., 2002]. We assumed that mangroves and coastal forests were flooded permanently as inundation here is primarily determined by sea level [Eva et al., 2002]. Tundra shrubland was considered waterlogged over the northern hemisphere summer. The canopy leaf area was derived from the Collection 5 MODIS Leaf Area Index (LAI) product at 1 km resolution [Yang et al., 2006]. The monthly averages from 2000 to 2001 were used. Emissions were calculated for each grid cell i and month m by:
where F is the mean methane flux, LA is the leaf area, D is the number of daylight hours, and I is either 0 or 1, depending on whether the pixel is inundated during the current month. Monthly emissions were summed to annual.
3. Results and Discussion
 All three riparian tree species showed average flux of CH4 greater than control, with mean emission rates of 2.6, 1.5, and 3.2 mg m−2 hr−1 for ash, cottonwood, willow and 0.9 mg m−2 hr−1 for control plots (Figure 1a). All tree plots produced lower fluxes than rice which averaged 6.6 mg m−2 hr−1. The effect of daily temperature variations was determined to have only a small impact on CH4 flux (temperature range 19–36°C, r2 = 0.14) and no clear seasonal behavior in CH4 flux was observed in tree fluxes [Khalil et al., 1998].
 The flux distributions were fit using gamma distributions and the method of moments was applied to assess the variability of each species [Rice, 2007]. Emissions from ash, cottonwood, and willow were higher than control plots at high levels of significance (p-value < 0.01, two-sided t-test). For purposes of calculating CH4 flux through the trees, the control plot distribution was subtracted from the aggregate tree distribution. Resulting data were then normalized to tree leaf area and the average tree CH4 emission rate was calculated to be 0.7 ± 0.3 μg cm−2 hr−1. This estimate is higher, but within collective error, than recent estimates of ∼0.5 μg cm−2 hr−1 from Bald Cypress (Taxodium distichum) [Garnet et al., 2005].
 Tedlar bag branch enclosures revealed significant emission in all species, but concentrations in the bags were non-linear on short time scales and resulting quantitative flux estimates using this approach were problematic. The origin of this response is unknown, but its rapidity is suggestive of stomatal control of CH4 conductance [Farquhar and Sharkey, 1982].
 The δ13C of emitted CH4 from chamber enclosure and bag samples are shown in Figure 1b. Rice and control plots had δ13C values of −62 ± 3‰ and −59 ± 2‰ respectively for emitted CH4 which is characteristic of rice agriculture and anaerobic wetland environments [Tyler et al., 1997]. All three tree species emitted CH4 enriched in 13C relative to rice and control plots with δ13C values of −54 ± 5‰ from ash, −54 ± 3‰ from cottonwood, and −52 ± 3‰ from willow. Though differences between tree species were not found, the difference between δ13C from aggregate tree data (−54 ± 4‰) and rice plots (and controls) is significant at high levels of confidence (p-value < 0.01).
 The ∼8‰ δ13C difference between rice and tree emitted CH4 can result from differences in CH4 production, gas transport, or oxidation. Though recent photosynthates can be an important source of carbon for methanogenesis, all plant species used in this work were C3 in photosynthetic pathway and likely have similar isotopic signatures in their organic matter (δ13C ∼ −25‰) [King and Reeburgh, 2002]. Active rhizodeposition of organic acids fixed via the enzyme PEP-carboxylase have been observed in some systems, including anaerobic root zones [Johnson et al., 1996; Jones, 1998]. The enriched δ13C signature observed could indicate this form of root exudates driving methanogenesis in the rhizosphere. Differences in carbon isotope fractionation factors with different methanogenic community structure could also potentially explain the observed 8‰ difference between rice and tree emitted CH4 [Chidthaisong et al., 2002].
 Alternatively, isotopic fractionation occurs in plants due to diffusive and effusive transport processes which are mass dependent, the magnitude of which depends largely on pore size [Chanton, 2005]. If responsible, the observed 8‰ shift in δ13C would suggest that CH4 transport in tree tissue is more convective in nature, which is at odds with measured fluxes which indicate a slower turnover time. CH4 oxidation in the rhizosphere, upper layers of soil, and potentially on the surfaces of the plant stems and leaves will enrich the isotopic composition of emitted CH4 due to a significant kinetic isotope effect (k12C/k13C ∼ 1.025) in aerobic bacterial CH4 oxidation [Raghoebarsing et al., 2005; Tyler et al., 1994]. Under this mechanism, measurements of δ13C above and below ground indicate that ∼20% of CH4 was oxidized in rice plots whereas 50–70% was oxidized in tree plots [Tyler et al., 1997]. Finally, it is noteworthy that the average δ13C of −54‰ emitted by trees is close to the δ13C of CH4 produced in aerobic environments of recent chamber studies involving whole C3 plants (−52‰) [Keppler et al., 2006]. Thus, in future field studies it may not be possible to distinguish between aerobic and anaerobic mechanisms of CH4 production based on the δ13C of emitted CH4 alone.
 Porewater CH4 concentrations were found to be significantly lower in rice plots (mean 760 μg/L) than in either tree (mean 1570 μg/L) or control (mean 4960 μg/L) plots (Figure 2a) because of enhanced CH4 transport from the rhizosphere to the atmosphere. In fact, average belowground CH4 concentration was found to be inversely related to aboveground flux (r2 = 0.94), supporting the assertion that transport was a controlling mechanism of belowground CH4 concentrations. The isotopic composition of belowground CH4 was found to be highly variable (δ13C −63 to −48‰, Figure 2b) with no clear differences between species. The absence of a large δ13C difference between rice and tree species belowground tends to favor the oxidative hypothesis for explaining the 8‰ difference in δ13C of emitted CH4. However, given the variability in belowground δ13C and CH4 concentration, more measurements will be necessary to verify this result.
 To determine if tree emissions have the potential to impact the global CH4 budget, emissions were scaled using mean leaf area emission rate (0.7 ± 0.3 μg cm−2 hr−1) across all broadleaf tree species in flooded environments from the Global Land Cover 2000 data set. The leaf area of the vegetation canopy from MODIS LAI in these regions during times of inundation was then used to scale the measured emissions. With this technique, global CH4 emissions were estimated at 60 ± 20 Tg year−1. Delineated spatially (Figure 3), the majority of CH4 emissions (40 Tg) were in the tropical Amazon region of South America, the African Congo, and Indonesia. Significant emissions were also found in the northern mid to high-latitude regions of Eurasia (20 Tg). We note that these estimates assume belowground CH4 concentrations similar to those in the greenhouse experiments, concentrations that may be higher than in natural settings by the amendment of organic matter [Khalil et al., 2008].
 If this source of CH4 is significant on the global scale as our bottom up approach suggests, it will have important implications for the pre-anthropogenic CH4 budget. Recently, preindustrial atmospheric CH4 was observed to be unexpectedly 13C enriched based on an ice core record during the period 0–1300 AD (δ13C ∼ −47.5‰), which may have resulted from enhanced biomass burning and a higher than previously considered geological CH4 source [Etiope et al., 2008; Ferretti et al., 2005]. This compares with observations from ∼1700 AD that are depleted in 13C relative to modern CH4 (δ13C ∼ −49‰). Because trees appear to emit CH4 enriched in δ13C by ∼8‰ compared with conventional wetland sources, this biogenic source can provide an alternative mechanism to shift the δ13C of biogenic CH4. To better constrain the tree source we employed a box model of the atmosphere that includes categories for biogenic, biomass burning, and geological sources for the preindustrial period 0–1700 AD (Table 1). The atmospheric kinetic isotope effect was estimated at −6 ± 1‰ [(k13C/k12C − 1) × 1000] and assumed to be constant [Lassey et al., 2007]. With CH4 sources kept to lower estimates and source δ13C values from the literature, we estimate the maximum global tree source strength to be 60 Tg (Table 1) [Houweling et al., 2000; Quay et al., 1999]. This global constraint coincides with the mean of bottom-up estimates of 60 ± 20 Tg year−1.
Table 1. Global Emissions Strengths and Isotopic Compositions Used in the Box Model
 These results suggest that woody trees could present a sizeable global source of CH4 to the budget, plausibly as large as 60 Tg yr−1, and may help explain observed tropical enhancements in atmospheric CH4 without a large aerobic plant source. Though we have confirmed the potential of such a source in our greenhouse mesocosm study and identified key regions that may represent significant sources of tree CH4 emissions, field studies will be needed to confirm the magnitude and spatial distribution of this CH4 source. Laboratory, greenhouse, and field studies across a wide variety of tree species (particularly tropical species) and grown in differing soil and water conditions would also be particularly useful to better characterize the variability in CH4 emission rates. There are also several unanswered mechanistic issues including the pathway of conductance through the plant tissue to the atmosphere and the reason for the difference in δ13C of emitted CH4 between woody trees and herbaceous aquatic macrophytes.
 We thank the members of the Global Change Research program at Portland State University for their helpful discussions. This research was supported by the Office of Science (BER), U.S. Department of Energy, grant DE-FG02-08ER64515.