Drivers of greenhouse gas emissions from standing dead trees in ghost forests

Coastal freshwater forested wetlands are rapidly transitioning from forest to marsh, leaving behind many standing dead trees (snags) in areas often called ‘ghost forests’. Snags can act as conduits for soil produced greenhouse gases (GHG) and can also be sources as they decompose. Thus, snags have the potential to contribute GHGs to the atmosphere, but emissions are not well understood. We assessed GHG emissions (carbon dioxide—CO2, methane—CH4, and nitrous oxide—N2O) from snags and soils in five ghost forests along a salinity gradient on the coast of North Carolina, USA. Mean (± SE) soil GHG fluxes (416 ± 44 mg CO2 m−2 h−1, 5.9 ± 1.9 mg CH4 m−2 h−1, and 0.1 ± 0.06 mg N2O m−2 h−1) were ~ 4 times greater than mean snag GHGs (116 ± 15 mg CO2 m−2 h−1, 0.3 ± 0.09 mg CH4 m−2 h−1, and 0.04 ± 0.009 mg N2O m−2 h−1). Hydrological conditions and salinity influenced soil GHG fluxes between the two field campaigns, but snags were less predictable and more variable. Snag and soil CO2/N2O fluxes were influenced by similar environmental parameters. The drivers for soil and snag CH4 however, were often not the same and at times oppositely correlated. Our results illustrate the need to further research into the drivers and importance of GHG emissions from snags, and the need to include tree stems into ecosystem GHG research.


Introduction
Coastal freshwater forested wetlands contribute significantly to the global carbon (C) cycle through C sequestration and long-term storage (Kirwan and Megonigal 2013), but sea-level rise and climate change are driving transitions from forest to marsh (Krauss et al. 2018). Freshwater forested wetlands are affected by changes in storm frequency and sea level rise, leading to the creation of ghost forests, areas where trees are replaced by marsh vegetation (Kirwan and Gedan 2019). Ghost forests are expected to expand as the climate changes, which could have important implications for local and regional greenhouse gas (GHG) budgets (carbon dioxide-CO 2 , methane-CH 4 , and nitrous oxide-N 2 O), by altering emissions of GHGs from soils and standing dead trees (snags). It is unclear how the increase in ghost forests could affect the emission of GHGs from coastal landscapes.
Recent studies have focused on GHG emissions from trees (live and dead), revealing their importance to the global GHG budget (Covey and Megonigal 2019). It is estimated that vegetation may represent up to 22% of annual global CH 4 flux, contributing as much as 32-143 Tg CH 4 year -1 (Carmichael et al. 2014). Standing dead mangrove trees have shown to exhibit higher CH 4 emissions than live mangrove trees (Jeffrey et al. 2019). Studies in tropical floodplains in the Amazon showed higher CH 4 emissions from live trees than peat soil surfaces nearby (Pangala et al. 2013). However, fewer studies have focused on snags, such as those found in ghost forests, which have the potential to transport soil produced gases as conduits, or produce them internally as wood decomposes (Carmichael et al. 2017;Jeffrey et al. 2019).
Production and emissions of GHGs from trees and soils in wetlands can be complex due to plant, microbial, and abiotic processes which can increase or decrease emissions to/from the atmosphere (Prendergast-Miller et al. 2011;Carmichael et al. 2014;Minick et al. 2019). Soil GHGs are affected by organic matter substrate availability, nutrient concentrations, soil redox conditions, pH, temperature, and salinity (Reddy and DeLaune 2008). Both CO 2 and CH 4 production in soils (and potentially snags) can be strongly affected by salinity, because sulfate (SO 4 2-) in seawater is a more energetically favorable electron acceptor, favoring microbial sulfate reduction over methanogenesis (Poffenbarger et al. 2011). The changes in availability of SO 4 2can increase CO 2 production (Weston et al. 2011) and decrease CH 4 production (Neubauer 2013). Processes driving soil GHG emissions can be further complicated in snags because decomposition of wood releases labile carbon for methanogens and nitrate for denitrification, potentially increasing production of CH 4 and N 2 O (Welch et al. 2019).
We assessed GHG emissions (CO 2 , CH 4 , and N 2 O) from snags and soils in five ghost forests experiencing saltwater intrusion, flooding, and high rates of sea level rise (0.45 cm y -1 , Kopp et al. 2015) on the coast of North Carolina, USA. We conducted field campaigns over two summers to answer the following questions: (1) what is the composition and magnitude of GHG emissions from snags and do they differ from nearby soils? (2) do snag and soil GHG fluxes vary by site?
(3) what influences snag GHG fluxes? (4) which GHGs contribute more to the radiative balance from soils and snags? We predicted that GHG emissions would be higher in soils than snags, and expected that emissions would differ among sites due to differences in salinity and hydrology. We explored various environmental factors that are known to affect GHG production.

Study sites
We examined five freshwater forested wetlands undergoing transition to marsh on the Albemarle Pamlico Peninsula (APP), North Carolina, USA (Fig. 1). The APP is a low-lying landform with the elevation of the majority of the peninsula within 1 m of mean sea level (Moorhead and Brinson 1995). The low elevation makes it more vulnerable to chronic disturbances from sea level rise, and saltwater intrusion (Moorhead and Brinson 1995;Poulter et al. 2006). Ghost forests in the APP have not been well classified, but a study by Ury et al. (2021 In press) estimated that ghost forests within the Alligator National Wildlife Refuge (total area: * 80,000 ha) take up approximately 3170 ha (* 4% of total area) and have been increasing since 2012 (Smart et al. 2020). Wetlands within the APP generally experience low salinity (2-11 mS/cm), but can be subject to acute and chronic saltwater intrusion from hurricane storm surge or lack of freshwater from rivers and drought (Ardón et al. 2013;Manda et al. 2014).
Soil and snag GHG fluxes were measured along a salinity gradient with freshwater wetlands located along the northern portion of the APP, and brackish marshes found on the southern portion (Fig. 1). The five wetlands we selected are currently managed by either the state or US Fish and Wildlife Service and are experiencing various levels of forested wetland retreat: Palmetto-Peartree Preserve (PPP- Fig. 1-1), Pocosin Lakes National Wildlife Refuge (PC- Fig. 1-2), Swanquarter National Wildlife Refuge (SQ- Fig. 1-3), Gull Rock State Game Lands (GR- Fig. 1-4), and Point Peter Rd within the Alligator River National Wildlife Refuge (PP- Fig. 1-5). Palmetto Peartree Preserve is located near the Albemarle Sound, which is less saline than the pamlico sound (Corbett et al. 2007). The salinity gradient from the largely freshwater Albemarle Sound (less than 6 ppt) to the more brackish pamlico sound ([ 12 ppt) is due to the number of inlets connecting the sounds to the ocean (Riggs et al. 1995). The vegetation from Gull Rock, Swanquarter, and Palmetto Peartree Preserve were further described by Taillie et al. (2019). The snags selected for all sites, were dead pine that had already lost most of the bark and few branches remaining ( Fig. 1), except Pocosin Lakes which included cypress in addition to pine.
The two field campaigns were during the growing season, the first from May to October, 2018, and the second from May to August, 2019. The mean tree diameter at breast height (DBH) was 27 cm ± 1.17 (± SE) and 28.4 cm ± 1.33 for trees sampled in 2018 and 2019, respectively. During the first field campaign, measurements included weather parameters, soil temperature and conductivity, and GHG fluxes from snag stems and soils (Table 1). During the second field campaign, measurements were similar to 2018 with additional sampling described below (Table 1).

Greenhouse gas analysis
During the 2018 field campaign, we selected 5-10 snags at each site that were deemed stable (e.g. intact and not at risk of being felled), representative of the area, and were within 10 m of each other (n = 39). For the 2019 field campaign, we selected the same 5 snags from the previous year if possible except in Gull Rock, which had 5 additional snags selected as part of another study (n = 30) (Carmichael et al. 2021 In Prep). We were unable to resample all the same trees from 2018 during 2019 because a few (n = 8) had fallen during the two hurricanes in 2018 (Florence-September 14, 2018 andMichael-October 11-12, 2018). Snag and soil GHG fluxes were measured via a closed dynamic chamber technique using a portable gas analyzer (Gasmet DX4040, Vantaa, Finland) recording gas concentrations every 20 s for 10-20 min (10 min for snags, and 20 min for soils due to the larger chamber volume). The Gasmet library contained generic references of alpha-and betapinene, which are terpenes released by pine trees and have shown to interfere with CH 4 concentrations using laser-based analyzers and can exerted a strong bias on FTIR-based instruments (Kohl et al. 2019). The spectra for 2,3-dimethylbutane was also included (only as an analytical help), however this component is not very likely to be present. From the snag chamber outlet, a 1 m gas tubing was connected to a drying agent (Drierite), and connected to Gasmet, then returned to tree chamber through a 1 m gas tubing connected to an additional outlet. The Gasmet flow rate was * 1.5 L min -1 with a cell sample size of 0.4 L. The Gasmet uses Fourier transformation infrared spectroscopy, and has a detection range of up to 30,000 ppm-CO 2 , 100 ppm-CH 4 , and 5 ppm-N 2 O. We used the closed static chamber technique with syringes when the Gasmet experienced battery or technical failures, sampling every 5 min over 40 min period (ESM 1). Approximately 15 mL of gas samples were stored in 12 mL pre-evacuated Exetainer vials (Labco-Lampeter, Wales, UK). Gases (CO 2 , CH 4 , and N 2 O) were analyzed using an Agilent gas analyzer (7890A GC system-Santa Clara, CA) equipped with a methanizer and two detectors, flame ionization detector (FID) and micro-electron capture detector (ECD), for CH 4 and N 2 O analysis respectively. The ranges for the GC were up to 10,000 ppm-CO 2 , 100 ppm-CH 4 , and 5 ppm-N 2 O.
Snag chambers were temporarily attached during measurements at various heights depending on optimal sealing conditions but were generally near * 60-70 cm from soil surface. Snag chambers were made of clear, flexible polycarbonate sheets (0.76 mm thick) modeled after Siegenthaler et al. (2016) and varied in size (Chamber 1: 45 9 25 9 2.5 cm & Chamber 2: 29 9 20.5 9 2.5 cm, length 9 width 9 depth respectively) based on snag diameters (ESM 1). Soil chambers (30 cm diameter) were modeled after  and placed 10 cm into the ground within 1-2 m from selected trees (n = 31 in 2018; n = 30 in 2019). GHG flux measurements were repeated every other month over the growing season for each field campaign (n = 2 per site per year). Both soil and snag chambers had temperature data loggers during gas sampling, measuring every minute inside the chamber using iButtons, which have an accuracy of ± 0.5°C (Maxim Integrated, San Jose, CA). Fluxes, expressed as mg m -2 h -1 , were determined using a linear and non-linear modelling approach called HMR after Hutchinson and Mosier (1981) and described in detail by Pedersen et al. (2010). The R package 'HMR' was used to calculate regressions and kept only significant (p \ 0.1) correlations (Pedersen et al. 2010). Snag stem GHG fluxes were calculated by estimating the stem surface area up to the chamber measurement, assuming cylindrical stem, and is expressed as mg stem -1 h -1 . CH 4 and N 2 O fluxes were also converted to CO 2 equivalents on a 100 year time horizon by multiplying by 45 and 270, respectively (Neubauer and Megonigal 2015).
In order to scale up GHG fluxes, snag and live tree density estimates were recorded within a 100 m 2 plot for each site (Table 2). Within the plot, snag and live tree DBH was measured for trees greater than 3 cm DBH. Basal area was calculated using the equation: p Â DBH=2 ð Þ 2 , with DBH converted to meters. Soil surface area was estimated by subtracting basal area of live trees and snags (Pangala et al. 2017). Scrubs/ shrubs basal area were not measured. Snag GHG CO 2(eq) equivalent fluxes were upscaled by calculating the surface area up to 1.4 m, assuming the stem was cylindrical, multiplied the median GHG flux rate by To measure soil solutes and porewater GHGs, porewater sippers were also installed in February 2019 at each site near three selected snags at two depths 15 cm (n = 15) and 30 cm (n = 15) (ESM 1). Porewater sippers were constructed from plastic tubing attached to an air stone (4 cm long 9 2 cm diameter). The air stone was housed inside a 2.54 cm diameter PVC, which was slotted near measuring depth and wrapped in a filter sock to prevent silt and sediment  (Wilson et al. 2018). A 60-mL syringe was used to create a vacuum to extract soil solution.
Porewater was drawn after soil GHG sampling and measured for GHG concentrations, NO 3 -N, NH 4 -N, Cl -, PO 4 -P, and SO 4 2-. To measure GHG concentrations, approximately 6-7 mL of soil solution was inserted into a 12 mL pre-evacuated Exetainer vial and was later filled with ultra-pure nitrogen gas. GHG concentrations were measured using headspace equilibrations techniques after adjusting to room temperature (Helton et al. 2014).
To measure porewater solutes, an additional 30-40 mL of soil solution was extracted from the sippers and filtered through 0.7 lm GFF filter for porewater solutes and stored in scintillation vials. Exetainer samples were stored upside down and kept cool until analysis, and scintillation vials were kept cool during transport and frozen until analysis. NH 4 -N and PO 4 -P were measured on a Seal Analytical AA3 Segmented Flow Analyzer, while SO 4 2-, Cl -, and NO 3 -N were analyzed on a Metrohm 930 Flex Ion Chromatograph using chemical suppression and conductivity detection. The measured values on both analyzers were calibrated using standards and were within 10% of expected values.

Soil geochemistry
For each site, four sets of soil half-cores were collected on June 2019 using a Russian Peat corer (5 cm diameter 50 cm length; AMS Inc. American Falls, ID). Cores were sectioned in the field into increments of 5 cm and placed in airtight sampling bags (Whirl-Pak, Madison, WI), and kept cool until lab analysis. One core was used for bulk density and organic matter measurements, the other three were used for water extractable solutes after sieving through a 2 mm sieve. Water extractable solutes (NO 3 -N, NH 4 -N, Cl -, PO 4 -P, and SO 4 2-) and pH were determined for each depth following methods from Helton et al. (2019). The soil subsample for bulk density was weighed and dried at 60C, and re-weighed to determine soil water content, and bulk density. The dry soil was then placed in muffle furnace at 550°C for 4 h to measure soil organic matter by loss on ignition (LOI). Soil volumetric water content was calculated by using the following equation: wet massÀ dry mass g ð Þ dry mass g ð Þ Â bulk density g cm À3 ð Þ density of water g cm À3 ð Þ .

Statistics
Due to the lack of normality in the data, we used an aligned rank transformation non-parametric factorial analysis to test for significant differences in GHG fluxes between years, sites, and their interaction effect for both snag and soil fluxes (Wobbrock et al. 2011). Kruskal-Wallis was used to test significant differences in porewater chemistry across sites and between CO 2(eq) equivalent gases. Dunn's multiple pairwise comparisons were then used for post-hoc analysis when Kruskal-Wallis showed significant differences, with a Bonferroni p-value adjustment (p \ 0.05). A one-way ANOVA was used to test significant differences in soil water extractable solutes (NH 4 -N, PO 4 -P, Cl -, NO 3 -N, and SO 4 2-), and soil properties (pH, dry bulk density, organic matter, %C, and %N) across sites after log transformations showed normality. Soil extractable solutes and properties were averaged from Water level data at PC was not logged from Jun 15 to Aug 1 due to logger malfunction the top 15 cm (5, 10, and 15 cm) depths for each soil core (n = 3 per site). Tukey HSD was then used for post-hoc analysis when ANOVA was significant (p \ 0.05), although we also considered marginally significant differences (p \ 0.1). Spearman rank correlations were used to explore relationships between snag and soil GHG fluxes, and potential environmental drivers (weather, soil GHG fluxes, porewater GHG concentrations, porewater solutes, and water parameters). Porewater nutrient data was averaged by site and month combining 15 cm and 30 cm soil solution. All analyses were conducted in R (version R 3.6.1) using stats and PMCMR packages (Pohlert 2014; R Core Team 2020).

Site characteristics
During the first field campaign in 2018, our sites received on average 1.5 times the rainfall (896 mm) compared to the 2019 sampling campaign (666 mm, which is similar to the long-term median ± SE of 672 mm ± 22, ESM 2). Mean temperature (± SE) during 2018 field campaign was 26.4 ± 0.2°C, and 25.4 ± 0.24°C during 2019 field campaign. Water level referenced to the ground varied from approximately -13 cm at Gull Rock and Point Peter, to 4.7 cm Pocosin Lakes in 2019 (Fig. 2). Mean specific conductivity from groundwater wells ranged from * 3 mS cm in Palmetto Peartree and Point Peter, to * 10 mS/cm in Swanquarter and Gull Rock. Specific conductivity remained relatively constant within each site, except Swanquarter which increased from 4.56 to 14.3 mS/cm throughout the growing season (Fig. 2).
Mean soil extractable solutes differed across sites for all solutes and pH, except for NO 3 -N, which was low across all sites (Table 3). Soil Cland SO 4 2concentrations were 15-26 times higher in Swanquarter than concentrations in Point Peter (Table 3). Soil pH was also significantly higher in Swanquarter (1.5 greater) than all other sites (Table 3). Soil NH 4 -N showed a similar trend, with lower concentrations in Point Peter and Palmetto Peartree (* 0.08 mg/L), and the highest in Swanquarter (0.71 mg/L, Table 3). Soil PO 4 -P was lower in Gull Rock, Point Peter, and Palmetto Peartree (* 25 lg/L), compared to Swanquarter and Pocosin Lakes (* 70 lg/L, Table 3). Soil bulk density was highest in Palmetto Peartree Preserve (1.05 g/cm 3 ), and lowest in Swanquarter (0.15 g/cm 3 ). Soil organic matter and %C was lowest in Palmetto Peartree Preserve (* 7%) compared to all other sites (Table 4).

Porewater dynamics
Porewater chemistry and GHG concentrations varied across all sites (Fig. 3). Porewater CO 2 concentrations ranged from 2.64 to 90.57 mM (Fig. 3). Point Peter had the lowest concentration of porewater CO 2 (12.39 mM), while Palmetto Peartree and Swanquarter had the highest (44 and 32 mM respectively) and most variable concentrations (Fig. 3a). Porewater CH 4 concentrations ranged from 0.06 to 566 lM and were different among sites (Fig. 3b). Porewater N 2 O concentrations ranged from 0 to 2.36 lM but were not significantly different among sites (Fig. 3c).  Porewater NH 4 -N ranged from 0.2 to 4.3 mg/L across all sites. Swanquarter and Point Peter had significantly lower NH 4 -N (0.2-0.4 mg/L), and Gull Rock (2.5 mg/L) had the highest (Fig. 3e). Porewater NO 3 -N was low (\ 0.09 mg/L) in all sites (Fig. 3f). PO 4 -P concentrations were significantly lower in Swanquarter and Gull Rock (0.004-0.006 mg/L) compared to Pocosin Lakes having the highest concentrations (0.07 mg/L). Porewater SO 4 2was also significantly different among sites, with higher concentrations in Swanquarter and Gull Rock (99-266 mg/L), compared to Pocosin Lakes (9.5 mg/L), Point Peter (5.3 mg/L), and Palmetto Peartree (3.6 mg/L). Similar to sulfate, chloride concentrations were also significantly higher in Gull Rock and Swanquarter (1969 and 4943 mg/L respectively), and much lower in Point Peter, (163 mg/L; Fig. 3h).

GHG fluxes
Overall GHG fluxes were * 4 times higher in soils (mean ± SE GHG fluxes: 416 ± 44 mg CO 2 m -2 h -1 , 5.9 ± 1.9 mg CH 4 m -2 h -1 , and 0.1 ± 0.06 mg N 2 O m -2 h -1 ) than snags (mean ± SE GHG fluxes: 116 ± 15 mg CO 2 m -2 h -1 , 0.3 ± 0.09 mg CH 4 m -2 h -1 , and 0.04 ± 0.009 mg N 2 O m -2 h -1 ) for both years. When scaling snag fluxes per stem surface (mean stem GHG fluxes: 97 ± 12 mg CO 2 stem -1 h -1 , 0.21 ± 0.08 mg CH 4 stem -1 h -1 , 0.02 ± 0.005 mg N 2 O stem -1 h -1 ), soils were * 5 times higher. Both soils and snags were sources and sinks of all GHGs. There were a total of 68 and 81 tree chamber measurements, and 36 and 77 soil chamber measurements in 2018 and 2019, respectively. The number of measurements that met our flux validity criteria for both years combined varied for each GHG with CO 2 having higher percentages for both snag (88%) and soil (88%) measurements, followed by CH 4 (55%-snag, 69%-soil), and N 2 O (55%-snag, 67%-soil). Snag CO 2 fluxes ranged from -297 to 1107 mg CO 2 m -2 h -1 (n = 131), while soil CO 2 fluxes ranged from -369 to 2347 mg CO 2 m -2 h -1 (n = 100). Snag CH 4 emissions ranged from -3.2 to 3.1 mg CH 4 m -2 h -1 (n = 83), while soil CH 4 emissions ranged from -20 to 114 mg CH 4 m -2 h -1 (n = 78). Snag N 2 O fluxes ranged from -0.13 to 0.45 mg N 2 O m -2 h -1 (n = 83), and soil N 2 O fluxes ranging from -1.3 to 2.5 mg N 2 O m -2 h -1 (n = 76). When considering the snag density and upscaling CO 2(eq) GHGs for snag stems and soil area (without tree basal area) within a 100 m 2 plot at each site, we estimated snag GHGs contribute an additional 200-1200 mg CO 2(eq) hr -1 ( Table 6). The scaling of GHGs are conservative estimates due to lack of precise soil surface area. Snag CO 2 fluxes were not significantly different between years (2018-wet & 2019-dry) nor among sites and did not have an interaction effect (Fig. 4a). Snag CO 2 increased 3 and 10 times for Pocosin Lakes and Palmetto Peartree, respectively, while Point Peter and Swanquarter remained similar across years (Table 5). Soil CO 2 fluxes were different between years (p = 0.0005), and among sites (p = 0.05, Fig. 4b). Soil CO 2 fluxes were approximately 2 times higher in 2019 (502 mg CO 2 m -2 h -1 ) than in 2018 (240 mg CO 2 m -2 h -1 ) and were 2 times higher in Palmetto Peartree (532 mg CO 2 m -2 h -1 ) than Gull Rock (232 mg CO 2 m -2 h -1 ). There were no significant differences in snag CH 4 fluxes among sites, between years, or their interaction effect (Fig. 4c). Snag CH 4 fluxes in Gull Rock (0.30 mg CH 4 m -2 h) were higher than Palmetto Peartree (0.07 mg CH 4 m -2 h -1 ), while all other sites slightly increased from 2018 to 2019 (Fig. 4c). Snag stem CH 4 fluxes in Point Peter (2018) and Palmetto Peartree (2019) were sinks instead of sources unlike all other sites (Table 5). The interaction between year and sites was significant for soil CH 4 fluxes (p = 0.0003), although soil CH 4 fluxes differed between years (p = 0.01) but not among sites (p = 0.19, Fig. 4d). There was twice as much more variation in soil CH 4 fluxes in 2018 for Swanquarter, Gull Rock and Point Peter, compared to 2019, unlike Palmetto Peartree and Pocosin Lakes which had the least variation in 2018 and slightly increased from 2018 to 2019 (Fig. 4d). Snag stem N 2 O fluxes were highly variable both years (Fig. 4e) and marginally significant among sites (p = 0.07), and between years (p = 0.09). Gull Rock was the only site that decreased in snag stem N 2 O fluxes, while all other sites slightly increased from 2018 to 2019 (Table 5; Fig. 4e). Soil N 2 O fluxes were not different between years, but were among sites (p = 0.001), and had interaction effects (p = 0.006, Fig. 2f). Soil N 2 O fluxes were higher and more variable (* 3 times more variable) in 2018 for Gull Rock and decreased in variability in 2019 (Fig. 4f). Pocosin Lakes was the only site that was converted from an N 2 O sink in 2018 (-0.0012 mg N 2 O m -2 h -1 ) to a source in 2019 (0.008 mg N 2 O m -2 h -1 ), while Point Peter remained a N 2 O sink both years but was reduced from -1.3 to -0.02 mg N 2 O m -2 h -1 from 2018 to 2019 (Fig. 4f).
Radiative forcing CO 2 fluxes remained the dominant radiative balance contributor for snag fluxes both years (10 and 120 times higher than CH 4 and N 2 O respectively), but differed by year for soil fluxes. In 2018, soil CH 4 ÁCO 2(eq) fluxes were within the same range as soil CO 2 fluxes, and N 2 OÁCO 2(eq) fluxes were significantly less (Fig. 5). Soil CO 2 fluxes in 2019 were the dominant contributor followed by CH 4 , and much less for N 2 O fluxes (Fig. 5).

Discussion
We found evidence to support our prediction regarding differences between snag and soil GHG fluxes. Our results showed that soil GHG fluxes were * 4 times higher in soils than snag GHG fluxes for all sites and years (Fig. 2). Our second prediction was that snag and soil GHG fluxes would differ by site, which we found to be true only for snag N 2 O fluxes, and all soil GHG fluxes (Fig. 4). We were surprised to see that snag GHG fluxes were not as significantly affected by differences in soil hydrologic conditions and salinity as soil GHG fluxes, and had much more variability both years (Fig. 4a, c, e). There were some differences in correlations between snag GHG fluxes and soil GHG fluxes when compared to environmental parameters. For example, soil CH 4 fluxes were strongly negatively correlated with porewater CO 2 , Cland SO 4 2-, while snag CH 4 fluxes were positively correlated with these same parameters (ESM 3). Environmental parameters that influence soil CH 4 production or inhibition, are not the same in snags and are often oppositely correlated. It is important to note however that our snag GHG flux estimates are conservative because they were not measured directly at the base of snags. Studies have shown that tree stem GHG flux rates have been known to decrease either linearly or Fig. 5 Median (± SE) CO 2 equivalent flux (CO 2 , CH 4 9 45, N 2 O 9 270) for soils and tree fluxes by year. Note different scales between soils and tree fluxes. Different letters indicate significant differences between CO 2 equivalent GHGs within respective year exponentially with increasing stem height sampling position (Pangala et al. 2017;Jeffrey et al. 2019). Lastly, regarding our last prediction about the contribution of snag CO 2(eq) , we found that CH 4 and N 2 OÁCO 2(eq) from snags do not contribute the same amount as CO 2 (Fig. 3). As expected CO 2 gases were the largest contributor in soils during the dry year (2019), but during the wet year (2018) the CH 4 contribution could be just as high as CO 2 (Fig. 5). Overall, we found that flooding conditions and soil characteristics can substantially affect soil GHGs, but the magnitude and direction of snag GHGs can be less predictable.

GHG fluxes
Precipitation had an important effect on soil GHG fluxes. The cumulative precipitation in 2018 was the third wettest growing season since 1979 (Climate Engine gridMET; Huntington et al. 2017), while the drier 2019 growing season was near the overall median (ESM 2). Because the water table was much lower throughout most of the growing season in 2019 (Fig. 2), soil CO 2 fluxes increased at all sites (Fig. 4b), soil CH 4 fluxes were reduced in variability, with the exception of Palmetto Peartree (Fig. 4d), and soil N 2 O fluxes decreased in three out of the five sites (Gull Rock, Point Peter, and Swanquarter) (Fig. 4). The increase in soil CO 2 fluxes could have been caused by increased carbon mineralization due to a lower water table, or due to higher plant root/microbial respiration. In a microcosm experiment using soils from a nearby site, CO 2 fluxes increased under intermittently flooded conditions, compared to flooded conditions (Helton et al. 2019). The decrease in water levels also caused a decrease in variability and increased the potential for CH 4 oxidation near soil surfaces for all sites except Palmetto Peartree and Pocosin Lakes. The decrease in soil N 2 O fluxes in Gull Rock and Swanquarter could have been caused by an increase in completed nitrification (decrease in denitrification), or due to a decrease in soil N due to increased herbaceous vegetation uptake given the drier conditions in 2019 (Chapin and Matson 2011).
Although soil CH 4 fluxes decreased during the drier year (2019), snag CH 4 fluxes slightly increased for Point Peter, and Palmetto Peartree, slightly decreased for Swanquarter and Pocosin Lakes, and remained similar for Gull Rock (Fig. 4c). The increase in snag CH 4 fluxes during the drier year could be caused by stem release of CH 4 produced in deeper saturated soils diffusing up snag. The drop in water levels could have induced CH 4 release through the open internal cavities of the snags, allowing gas to transport in trees by molecular diffusion, and concentration gradients (Megonigal et al. 2020). Transport through the snag allows CH 4 to bypass the oxidation that occurs in upper horizons of soils, but it is also possible that CH 4 is oxidized as it is transported up the snag stem, which would explain the decrease in snag CH 4 fluxes in some sites (Hornibrook et al. 1997). The contrasting response of snags and soils CH 4 fluxes to changes in precipitation illustrate the complexity in drivers of soil-plant-atmosphere interactions, especially in snags which appear to be more variable than soils.
Snag N 2 O fluxes also had significant increases from 2018 to 2019 (except at Gull Rock), which could have been related to the increases in soil N 2 O fluxes, although the correlation between snag and soil N 2 O fluxes was not significant. The difference between years highlights the importance of snag N 2 O fluxes under different hydrologic conditions. Previous studies have suggested a decrease in live tree stem CH 4 and N 2 O emissions and variability during drier conditions compared to wetter conditions (Barba et al. 2019b), but our study shows mixed responses, which demonstrates the complexities in GHG production from snags (Fig. 4a, c). During the wet year (2018), Point Peter was a major N 2 O sink in soils which can occur in high water content and low inorganic N availability (Neubauer and Megonigal 2015). During drier years, such as 2019, snag N 2 O gases need to be taken into account because of their high global warming potential.

Porewater dynamics
Porewater solutes showed distinct characteristics between sites (Fig. 3). Trees have been shown to act as conduits for the atmospheric flux of GHG from wetlands (although the source of origin may vary), therefore it is possible that the snags were emitting GHGs from deeper soils. The mechanisms of gas transport in snags, however, differs from live trees because there is no active water transport through the xylem and phloem (Megonigal et al. 2020). It is most likely that snag GHGs are mostly driven by molecular diffusion driven by concentration gradients from soil and porewater.
Swanquarter had higher snag CH 4 fluxes during the wetter year (2018) and was reduced during the drier year (2019). Out of all sites, Gull Rock had among the highest porewater NH 4 -N (Fig. 3) and the second to highest soil extractable NH 4 -N after Swanquarter (Table 3). The high amount of NH 4 -N in soil solution at Gull Rock and Swanquarter is most likely due to salinization causing displacement of NH 4 ? ions by marine cations (Ardón et al. 2013). Although Swanquarter was more saline and had higher concentrations of SO 4 2- (Table 3 & Fig. 4), Gull Rock was significantly higher in porewater NH 4 -N, most likely due its proximity to a waterfowl impoundment (Winton and Richardson 2017). Herbivorous birds contribute to the amount of N via digestion and excretion of consumed plants making N more soluble and mobile . High amounts of NH 4 -N can suppress CH 4 consumption by oxidizers, potentially increasing CH 4 concentrations, which could explain why snag CH 4 fluxes were high in Swanquarter and Gull Rock in 2018 (Morse et al. 2012;Ardón et al. 2018).

GHG correlations
The lack of correlation between snag and soil GHG fluxes suggests that snag GHG production is complex and may be indirectly affected by soil GHG production. The drivers that produce soil CH 4 and snag CH 4 fluxes were not the same, and at times oppositely correlated (ESM 3). It is possible that the GHGs from snags depend on production of gases in the soils and undergo various processes within the trunk including oxidation/consumption of CH 4 and N 2 O, which has been shown in other studies (Machacova et al. 2016;Carmichael et al. 2017;Jeffrey et al. 2021). This hypothesis is even more likely considering the snag chambers were placed 60-70 cm from the ground, thus are conservative. This also helps explain the disconnect between environmental parameters that influence soil CH 4 versus snag CH 4 (ESM 3). Snag CH 4 and N 2 O fluxes were negatively correlated with DBH, which was similar to a previous study (Pangala et al. 2017), although there may be different reasons. In the Amazonian study by Pangala et al (2017), young live tree stems emitted more CH 4 than mature live tree stems. In our study it is likely that smaller snag stems emit higher concentrations of CH 4 because there is less time for oxidation through the sapwood as opposed to larger diameter snags.
Some of the environmental parameters that influence soil CO 2 and N 2 O fluxes were similar to those of snag CO 2 and N 2 O fluxes. For example, CO 2 fluxes for soils and snags are both positively correlated with temperature, porewater CO 2 , and water level deviation, and both are negatively correlated with relative humidity. This suggests that faster decomposition occurs during warmer and drier conditions in both snags and soils (ESM 3). The positive correlation between soil CO 2 and NO 3 -N implies that available inorganic N may help produce CO 2 indirectly because of its role in the microbial community (Zhang et al. 2019). N 2 O production in soils and snags were also both positively correlated with NO 3 -N, SO 4 2-, and Cl -, and both negatively correlated with PO 4 -P. The inorganic N availability in soils and the periodic alterations in soil water content can trigger nitrification and denitrification processes and produce N 2 O in soils (Moldaschl et al. 2021). Snag N 2 O emissions could be formed within snag tissues, similarly to CH 4 emissions, by a different set of microorganisms that live in association with trees (Machacova et al. 2019). Paxillus involutus (Batsch) Fr. and Tylospora fibrillosa (Burt.) are two ectomycorrhizal fungi found to form symbiotic associations with pine tree roots in acidic temperate soils (Prendergast-Miller et al. 2011). They become highly competitive when inorganic N concentrations are high, thus producing N 2 O through nitrate reduction under low O 2 conditions (Prendergast-Miller et al. 2011). The soils at Gull Rock were conducive to these types of symbiotic fungi due to acidity (pH * 5) and high amounts of NH 4 -N. This could explain why soil and snag N 2 O fluxes were high at Gull Rock, and thus had higher snag N 2 O fluxes than all other sites especially during the wet year, 2018 (Fig. 4e).

Radiative forcing
Soil CO 2 fluxes were the dominant radiative balance contributor for both years, although during the wet year (2018) CH 4 fluxes contributed as much to radiative forcing as CO 2 (Fig. 5). For snags, CO 2 fluxes were the dominant contributor, although its contribution was slightly higher in 2019 (drier year). Although snag CH 4 and N 2 O fluxes did not contribute as much as CO 2 fluxes it is still important to take into account because ghost forests could collectively contribute a significant amount, as shown by the upscaling of these GHGs within a 100 m 2 plot (Table 6). This evidence also supports the need to estimate ghost forest GHG production, since there is no longer leaf canopy consuming CO 2 through photosynthesis, potentially making ghost forests major GHG sources.

Snag GHG fluxes in context
Our CO 2 flux rates from snags were slightly lower than other wetland regions, and 6-10 times lower than upland trees (Table 7). Snag CH 4 fluxes from our study were within the range of other wetland studies, except for Amazonian live tree species which were 1-2 orders of magnitude higher (Pangala et al. 2017). Live tree CH 4 fluxes in upland regions were generally much lower than wetlands. Our study also demonstrated the ability of some snags to uptake CH 4 , which was also reported in a previous study within the same region (Carmichael et al. 2017). There is limited information on tree stem N 2 O emissions (live and dead), but we were able to include N 2 O fluxes from snags within the same area (unpublished) and are also within range as our study (Carmichael et al. 2017). Very few tree studies in wetland environments report N 2 O fluxes, most likely due to the low magnitude and high variability of N 2 O emissions. Live tree stem N 2 O fluxes in uplands have been reported to be both sources and sinks, which is similar to what we found in our study for a few snags in Pocosin and Point Peter, although the majority of snags we measured were sources.

Conclusion
Although snag area over soils is small, snag GHGs collectively increased the total ecosystem CO 2(eq) by 25%. Hydrological conditions and salinity significantly influenced soil GHG fluxes between the two field campaigns, but snags were less predictable. The environmental parameters that influence the production of CH 4 in soils however were often oppositely correlated with snag CH 4 fluxes. The disconnect between drivers in soil CH 4 and snag CH 4 production may have been due to oxidation of CH 4 within snag stem trunks, further highlighting the complexities of snags vs live trees. Our results also show that soil CH 4 fluxes contributed as much radiative forcing as CO 2 to the radiative balance during wet years. Although snag CH 4 and N 2 O GHGs did not contribute as much as CO 2 they should still be taken into account given the cumulative amounts that they contribute. Salinity induced stress and increasing frequency of flooding events is projected for much of the coastal southeastern US (Sweet et al. 2018), which will lead to a decrease in primary productivity and an increase in the spatial extent of ghost forests (Krauss et al. 2009). This study demonstrates the potential snags have to emit GHGs to the atmosphere, but also how unpredictable they can be under various hydrological conditions and salinity levels. Therefore, more efforts should be done to include tree GHGs (both live and dead) in regional and global budgets. Soil GHG fluxes scaled up with live and dead tree basal area subtracted. Snag stem GHG fluxes scaled using DBH of snags measured and calculated cylindrical surface area up to 1.4 m. Median GHG fluxes for each site was used for scaling  Code availability The code used to process and analyzed data are available from the corresponding author at request.

Declarations
Conflict of interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.