The study took place at Hinze Dam (28.05056° S, 153.28389° E) located near the Gold Coast, Australia (Fig. 1). The reservoir was built across the Nerang River in 1976, creating a large artificial lake with a maximum capacity of 310 730 megalitres. The reservoir wall height was raised in 1989, then again in 2011, increasing the water level by 15 m and doubling its capacity. The increase in water level in 2011 led to the death of trees around the lake perimeter, now covering ∼ 20 % of the reservoir surface area, creating the existing ghost forest. The western part of the reservoir is connected to the Nerang River and Waterfall Creek, while the eastern part is connected to Little Nerang Creek, with the Little Nerang Dam upstream. There is a small dam that reduces the amount of fresh organic material coming into the eastern side of the lake.

Two field campaigns were conducted in the winter (mid-August) and early summer (late November) of 2023 to capture seasonal variability in temperature and water level. All sampling was conducted from boats to minimize benthic disturbances within each plot. The three plots were located to the South, Mid and North of the reservoir western arm, which accounts for an inflow depositional gradient from South to North (Fig. 1). In the South, the Nerang River stream feeds the reservoir and deposits sediment and organic material. Ghost forest trees were sampled in belt transects perpendicular to the shoreline (Fig. 2a–c), which allowed us to measure trees along a water depth gradient. All trees sampled were in standing water during both field campaigns (Fig. 2d). The same trees were re-sampled in both seasons, keeping the stem height above the water surface consistent between campaigns, as the water table dropped.

A Van Veen grab was used to collect benthic sediments from six locations per site (Fig. 2e). The pH and redox potential of the sediments were measured in situ using a pre-calibrated multiprobe (Hach PHC101 and MTC301). During the first campaign, three soil samples were also taken at each site and analysed for total nitrogen, phosphorus and % C_org with samples oven dried at 40 °C and then homogenised before analysis. Total nitrogen was determined using a LECO TruMac CNS analyser. Soil samples for phosphorus determination were first treated with a 1:3 Nitric / HCl and then run on APHA 3125 ICPMS. Carbon content samples were pre-treated with acid to remove inorganic carbon before analyses on an Elemental Analyser (EA). During the second campaign, the depth of the organic-rich layer was estimated by gently tapping a 10 cm ø PVC pipe through the softer sediments until the hard clay layer was reached, and measuring the depth change. This was replicated at three locations per site. The sediment oxygen demand (SOD) was determined by adding 10 g of freshly collected sediment (Van Veen grab) into a 250 mL borosilicate glass bottle, filling it with surface water and noting the start and stop concentrations of dissolved oxygen at 0 and 48 h using a luminescent/optical dissolved oxygen sensor (Hach, LDO101).

To determine aquatic CH₄ concentrations, 150 mL water samples were collected into borosilicate bottles at six locations within each site along a water depth gradient. Surface (10 cm) and bottom water samples (30 cm above the benthic surface) were collected using a peristaltic pump, overflowing > 3 times the bottle volume before capping ensuring no headspace. The samples were initially kept on ice until later being treated with mercuric chloride (HgCl₂) at the end of the sampling day. Water temperature and dissolved oxygen levels were also measured at the same locations using a hand-held probe (Hach, LDO101). Additionally, a spatial transect collecting water samples in borosilicate bottles (as described above) at ∼ 1 km intervals along the reservoir was performed on each trip, spanning from the southern to northern plot (Winter n=16 and summer n=23 locations). The coordinates, windspeed and water depth were also noted at each sample location. All water samples in borosilicate bottles were analysed as followed: the dissolved partial pressure of pCH₄ was determined using the headspace method with a 90 mL water sample and 60 mL of ambient air added into a 150 mL syringe and shaken vigorously for 2 min (Jeffrey et al., 2019a). The headspace gas was then analysed using a CH₄ isotope cavity ringdown spectrometer (Picarro CRDS, G2201-i) and then corrected for the dilution with ambient air according to Eq. (1) (de la Paz et al., 2021): 1CH4,wsample=CH4pVwKVw+1RTVV where Vw is the volume of the water used, K is the solubility coefficient, R is the universal gas constant (8.205×10-5 m³ atm ⁻¹ K⁻¹ mol⁻¹), T is the air temperature (K), VV the volume of the headspace, CH_4 p is the CH₄ concentration (ppmV) measured with the SSIM CRDS.

To quantify ebullition at each site, ten floating chambers were established along a perpendicular transect from the shoreline within the ghost forest (n=30) (Fig. 2a–c and f). The chamber diameter was 41 and 12.5 cm height, with a total volume of 12.9 L and surface area of 0.132 m². Gas within the headspace of each chamber was collected every ∼ 24 h. Each chamber had a 30 mL syringe pre-attached to the chamber outlet, which was extracted carefully to minimize disturbance to the chamber. The 30 mL gas sample was then transferred into duplicate evacuated 12 mL exetainer vials. The gas samples were later analysed for CH₄ concentration using the CRDS and a small sample induction module (Picarro SSIM, A0314).

The ghost forest tree stem CH₄ fluxes were measured by attaching semi-rigid foam-lined airtight chambers onto the tree stems (Siegenthaler et al., 2016) (Fig. 2g). Each tree stem CH₄ flux was measured at three different heights between 10 and 220 cm above the water level to account for variability of fluxes along stem height (Pangala et al., 2013). Chambers that were attached above algae, bark or over cracks were noted (Fig. S1 in the Supplement). To prevent potential chamber leaks due to small cracks, fissures or rough bark surfaces, white pottery clay was applied to seal these areas (Jeffrey et al., 2020b). Individual gas fluxes were measured for 3 to 7 min, with longer measurements required for lower fluxing tree stem surfaces. Diffusive CH₄ fluxes were measured adjacent to ghost forest trees by placing a smaller 28 cm diameter circular floating chamber onto the water surface. The tree stem and diffusive flux chambers were connected to a portable cavity ring-down spectrometer CO₂ and CH₄ analyser (CRDS Picarro, G4301) and CO₂ and CH₄ flux rates were measured in situ. Here we only present the CH₄ data, the CO₂ results are available in the Supplement (Fig. S4–S6) and dataset (Dittmann, 2025). The CRDS precision was factory calibrated ± 0.3 ppb with a lower detection limit of 0.9 ppb. The stem circumference and flux chamber height above the water level were noted for each sample tree, along with the water depth.

Ghost forest internal tree stem gas concentrations were collected from 11 trees spread across each site, similar to Carmichael and Smith (2016). Multiple 12 cm deep (13 mm internal diameter) holes were drilled horizontally into each tree stem at lower stem heights, using a cordless electric drill, with the hole immediately sealed using a sterilized rubber septum stopper. Gases were allowed to accumulate in the stem cavity for 24 h, at which time ∼ 12 mL of stem gas were removed and injected into a pre-evacuated exetainer, with the precise volume of gas sample noted. The CH₄ concentration was determined using the CRDS SSIM (as described above).

The stem and diffusive CH₄ flux (s) rate was processed using a modified gas flux R script “GasFluxes” (Fuss et al., 2020). The results of the linear regression for each flux measurement (Fig. S2) were added into the following Eq. (2) (Jeffrey et al., 2019b): 2F=sVRTairAt where s is the regression slope in ppm s⁻¹ for each flux measurement, V is the chamber volume (m³) (which includes the CRDS volume, chamber and tubing volume), A is the chamber surface area (m²) and t is the conversion factor from s to d, and to µmol of CH₄. Linear regression of CH₄ fluxes of r2>0.7 were used in the calculation and linear regression with lower r2 were analysed and used in the calculations when a clear increase was visible. In lower CH₄ fluxing trees, the CO₂ flux linear regression r2 values were used as a proxy for assessing air-tight chamber seals, as all tree stems emitted CO₂. The ebullition CH₄ flux rates E were calculated according to Eq. (3): 3E=CH4end-CH4startTd⋅VRTairA⋅t-Fw where Td is the duration of the deployment time, CH_4 end is final headspace concentration (ppm), CH_4 start is assumed 1.9 ppm and Fw is the average diffusive CH₄ flux at each site (as determined in Eq. (1) using portable GHG analyser). As our chambers (Sect. 2.5) measured total flux (i.e. diffusive + ebullitive), the average diffusive flux was subtracted from each ebullition chamber to estimate the ebullitive flux, similar to prior studies using the same approach (Hoffmann et al., 2017; Sø et al., 2024).

To upscale the axial tree stem CH₄ fluxes to individual tree emissions, the diameter at breast height (DBH in cm) was used and trees were assumed as a cylinder with no branching. To conservatively estimate total stem gas fluxes, we only up-scaled to a height of 2.5 m above the water level (even though trees were in excess of 30 m in height). This was the highest tree stem location we measured, so we did not extrapolate beyond this point. The upscaled tree stem CH₄ fluxes were used in comparison between sites and in total reservoir upscaling. The tree was divided into three radial bands and calculated as follows (Eq. 4) similar to Jeffrey et al. (2023): 4Ft=∫02.5(c⋅h⋅F) where Ft is the flux per tree up to 2.5 m (µmol d⁻¹), c is the tree circumference (m), F is the in situ measured gas flux rate for that height (µmol m⁻² d⁻¹) and h is the height of each band (m). The height per band was the same as the height of the chamber plus half the distance to the next chamber, or the distance to either end of 2.5 m.

The density of ghost forest trees (trees ha⁻¹) was estimated from duplicate 100 m² plots in the Northern site and four 100 m² plots in the South and Mid sites. Tree heights were measured during May 2025 using a laser range finder (Nikon, Forestry Pro). The overall ecosystem CH₄ flux from trees for each site was calculated using Eq. (5): 5Ftol=xFt⋅d⋅c where Ftol is the total tree flux per ha (mol d⁻¹), xFt is the average tree flux per site, d is the tree density (trees ha⁻¹) and c is the conversion factor from µmol d⁻¹ to mol d⁻¹. To upscale the diffusive and ebullition flux inside the ghost forest area, the sum of the tree basal surface areas was subtracted from the total aquatic surface area. The remaining aquatic surface area was multiplied by the average areal diffusive and ebullition CH₄ flux rates per site. (Note: tree basal areas only accounted for 1.6±0.4,1.6±0.5 and 2.5±1.0 % of the ghost forested area within South, Mid and North sites, respectively).

The total surface area of the reservoir, the ghost forested areas and the shallow zones around the edges (8 m water depth) were estimated using ArcGIS Pro (version 3.4.0), using a bathymetry map provided by Seqwater. Ebullition was assumed to only occur in all areas < 8 m in depth (Bastviken et al., 2004). For the open water areas of the reservoir (not containing trees), the diffusive CH₄ flux was estimated using spatially collected dissolved CH₄ concentrations (from the transect grab samples above) and the CH₄ flux rate numerically modelled using k estimated from windspeed (Wannikhof, 1992). We separated the reservoir surface area into three zones, the southern zone (with the inflow), the middle zone (spanning most of the open water reservoir including the eastern part) and the northern zone (which combines the far north and eastern embayment). The average flux per site for each of the three CH₄ pathways (trees, diffusive and ebullition) for each trip was used in the calculation. The sum of the three pathways represents the total zone emissions.

For better comparison we normalized the tree CH₄ flux relative to the water surface area (m²) (Amaral et al., 2025). The average tree stem CH₄ flux per site were converted to a flat surface equivalent (Fsurface), tree heights were upscaled to 2.5 m and only here we also upscaled to 10 m to compare to a less conservative approach. Equation (6): 6Fsurface=d100⋅(c⋅h)⋅xFt

All reported errors represent standard errors of the mean. All analyses were undertaken using R (4.3.2), figures and graphs were modified using Gimp (v. 2.10.36). The Shapiro-Wilk normality test and Q-Q visualisation were used to test for normality. For normally distributed data, t-tests were applied to compared differences between groups. Otherwise, Wilcoxon rank-sum tests were used when data did not conform to normality. Differences between stem height measurements and between fluxes from the three plots were tested using Kruskal-Wallis One Way Analysis of Variance on Ranks, followed by the Dunn post hoc test for pairwise comparisons. Relationships between water depth and aquatic gas concentrations were assessed using linear models, where regression lines were fitted to log-transformed water depth.

Reservoir water capacity declined during the two campaigns from 86 % in winter to 80 % in summer, resulting in a water level drop of ∼ 1.5 m (Fig. 3). This was among the lowest levels since the dam wall was raised in 2011. Average water temperatures increased from 17.2±0.1 °C in winter to 24.6±0.7 °C during summer. Water temperature was similar between the top and bottom waters inside the ghost forest areas indicating no thermal stratification. The average maximum and minimum air temperature (including 14 d prior and after the fieldwork campaign) increased from 23.9 and 5.4 °C in winter, to 29.2 and 16.0 °C in summer. During the winter campaign, the reservoir received 1.5 mm of rain, while during the summer campaign 29 mm of precipitation occurred on the first two days and 4 mm over the rest of the week.

Dissolved oxygen (DO%) in the water column of the reservoir was lower overall during winter than in summer (Table 1). The biggest change was observed at the southern site, where bottom DO increased from 60 % to 90 % saturation between the two campaigns. The DO concentrations were 4 %, 3 % and 11 % lower at the bottom of the water column compared to surface, in the North, Mid and South sites, respectively. During the summer campaign, the difference was 11 %, 21 % and 27 % in the North, Mid and South sites, respectively. Surface water pH increased between campaigns, from 7.3 to 7.6, 7.3 to 7.9 and 7.2 to 8.3 for the North, Mid and South sites, respectively.

Benthic sediment organic carbon varied between sites, it was lowest at the South site (5.42 %) and highest at the Mid site (9.32 %) (Table 2). Phosphorus concentrations were the highest at the South site (963.3 mg kg⁻¹), more than double the other sites. The average uncorrected redox potential at the southern site ranged from -120±12 to -113±10 mV during the two campaigns, respectively. For the North site the average uncorrected redox potential shifted from -90±12 to -53±6 mV during summer. The Mid site had a larger variance in its uncorrected redox values during both campaigns, ranging from -121 to -85 mV on one shore to -17 to -1 mV on the opposite shore, inside the embayment. Similarly, during summer the redox potential ranged from -132 to -123 mV on one side to -62 and 53 mV on the other side of the bay. The sediments in the South site had the highest average sediment oxygen demand (SOD) with 0.11±0.001 mg O₂ mL⁻¹ d⁻¹ followed by the Mid site 0.08±0.017 mg O₂ mL⁻¹ d⁻¹ and the North site 0.07±0.009 mg O₂ mL⁻¹ d⁻¹.

Measured tree DBH ranged from 17.0 to 37.5 cm, 19.7 to 60.0 cm and 18.0 to 41.8 cm for the North, Mid and Southern site, respectively. The average tree DBH per site was 30.2±3.6 cm, 28.7±3.5 and 23.5±2.5 cm for the North, Mid and Southern site, respectively. The ghost forest tree density was 2200, 1350 and 1975 trees ha⁻¹ for the North, Mid and South sites, respectively. The tree basal areas covered between 1.5 % and 2.5 % of the water surfaces in the ghost forest areas. Tree height ranged from 10.0 to 28.8 m above the water surface and averaged 22.0±2.7 m.

Dissolved CH₄ concentrations in the surface water declined at our sample locations as the water depth increased during both winter (R2=0.28, p<0.005) and summer (R2=0.43, p<0.001) (Fig. 4). Aquatic CH₄ concentrations were highest during the summer campaign.

There was considerable spatial and temporal heterogeneity in the ebullition CH₄ fluxes between and within sites. In winter, the average ebullition fluxes were 2253±688, 437±104 and 2270±819 µmol m⁻² d⁻¹ for the North, Mid and South sites, respectively (Fig. 5, Table 3). During the summer, average ebullition CH₄ fluxes were 1103±339, 1677±427 and 6286±1048 µmol m⁻² d⁻¹ for the North, Mid and South sites, respectively. There was a significant increase in ebullition flux between the winter and summer campaigns for the Mid (p<0.05) and South site (p<0.005). The decreased ebullition flux at the North site between winter and summer was not significant. For the winter campaign, the ebullition fluxes showed significant difference between the Mid and North site (p<0.05), as well as the Mid and South site (p<0.05). During the summer campaign we found significant differences in the ebullition fluxes between the South and North site (p<0.005), as well as South and Mid site (p<0.001).

Diffusive CH₄ flux rates differed between the three sites within the reservoir. During winter, the highest average diffusive CH₄ fluxes were measured at the North site (264±25 µmol m⁻² d⁻¹), followed by the South site (131±41 µmol m⁻² d⁻¹) and lowest at the Mid (69.5±7.0 µmol m⁻² d⁻¹) site (Fig. 6, Table 3). The North site showed significantly higher fluxes compared to the Mid site (p<0.001) and to the South site (p<0.005). During summer, the North had the highest average diffusive flux (306±25 µmol m⁻² d⁻¹) and Mid and South sites had similar fluxes (218±51 and 218±77 µmol m⁻² d⁻¹). Significant differences were only observed between the North and South site (p<0.01). Only the Mid site showed a significant increase in diffusive CH₄ fluxes between campaigns (p<0.001).

All trees at all stem heights (total n=192) emitted CH₄ during both campaigns. There was a high variability in the amount of CH₄ emitted per tree at all three sites. The CH₄ stem flux measurements ranged from 1.66 to 1584 µmol m⁻² d⁻¹ and 2.43 to 8614 µmol m⁻² d⁻¹ for the winter and summer campaigns, respectively (Fig. 6, Table 3).

Average stem CH₄ fluxes per tree (upscaled to 2.5 m) were 510±90 and 1770±508 µmol per tree d⁻¹ for the winter and summer campaign, respectively. The tree CH₄ fluxes ranged from 26.8 to 2196 µmol per tree d⁻¹ during the winter campaign and from 5.34 to 14,944 µmol per tree d⁻¹ in summer.

Average tree stem CH₄ fluxes were the highest in the South site during both campaigns (Fig. 4), winter 617±162 and summer 3545±1145 µmol per tree d⁻¹ (Table 3), but only significantly different during the summer campaign (Mid-South p<0.01 and North-South p<0.005). The second highest tree fluxes were from the North site (453±133 µmol per tree d⁻¹) during the winter and the Mid site (1008±524 µmol per tree d⁻¹) during summer. Average tree stem CH₄ fluxes increased during the summer campaign, but were only significantly higher at the South site (p<0.005).

In winter, the ghost forest stem CH₄ fluxes decreased with stem height. The average flux rates were 318±55.8, 251±57.0 and 153±54 µmol m⁻² d⁻¹ for the lower, middle and upper measurements (Fig. 7), with the lower and upper stem height fluxes being significantly different (p<0.005). One tree at the South site showed the highest flux of the campaign (1507 µmol m⁻² d⁻¹), however, the flux chamber was located over a stem fissure.

In summer, no significant differences between stem heights were found (Fig. 7), however the highest average CH₄ flux observed was at the uppermost measurement (1224±409 µmol m⁻² d⁻¹), especially at the South site. The internal tree stem gas samples (n=14) contained an average CH₄ concentration of 51086±10126 ppm (or 5.1 % saturation), being four orders of magnitude higher than the atmosphere.

The surface water CH₄ concentrations decreased from south to north, during both campaigns (Fig. 8). Up scaling CH₄ fluxes to the whole reservoir showed that total CH₄ emissions tripled between the winter and summer campaigns, from 1418 to 4604 mol d⁻¹ (Table 4). Tree stem CH₄ emissions tripled between the two campaigns, from 207 to 628 mol d⁻¹ and contributed 14 % and 15 % in winter and summer emissions respectively. Overall, ebullition was the dominant CH₄ emission source (67 % in winter and 58 % in summer), while diffusive fluxes from the deeper open water contributed only 2 % and 14 % for winter and summer, with diffusive fluxes of the shallow ghost forested areas contributed 16 % and 14 % respectively.

At the flux per water surface area within the ghost forest, tree fluxes were between 56.4 and 428 µmol m⁻² surface d⁻¹, ebullition between 437 and 6286 µmol m⁻² surface d⁻¹ and diffusive flux between 69.5 and 306 µmol m⁻² surface d⁻¹. If tree stem emissions were upscaled to 10 m of stem height this increased their relative contribution at a per water surface area to between 225 and 1711 µmol m⁻² surface d⁻¹.

All trees emitted CH₄ during both campaigns and from all three measured stem heights, addressing research question one. We observed high variation between the fluxes of the different trees undergoing different stages of decomposition, and due to the influence of the stem fissures and bark remnants (Jeffrey et al., 2020b). Cracks result from the decomposition and drying of the dead trees (Oltean et al., 2007). All sampled trees died at around the same time in 2011 as the water levels rose due to the raising of the dam wall. Due to their state of decomposition, we could not identify the tree species, but different species can have different decomposition rates (Freschet et al., 2012; Kahl et al., 2017). Many of large trees were likely Eucalyptus sp. based on the species in the adjacent areas located above the dam maximum water level. Eucalyptus sp. are classified as hardwoods (FAO, 2002) and therefore are somewhat more resistant to decomposition. Previous studies have found high variance in tree CH₄ fluxes even of the same species, in living (Jeffrey et al., 2023) and dead trees (Warner et al., 2017; Kipping et al., 2022). Decomposition stage is known to influence tree stem CH₄ fluxes, with recently dead trees exhibiting both elevated internal CH₄ concentrations (Covey et al., 2016) and higher CH₄ fluxes (Sakabe et al., 2025). In this study, all trees died simultaneously following reservoir inundation, meaning decomposition stage is unlikely to introduce meaningful variability among individual flux measurements. However, this dynamic may be relevant in other reservoirs where water levels rise gradually, producing ghost forests of mixed decomposition stages.

During the winter campaign, CH₄ fluxes declined with stem height, consistent with gas diffusion observations from living wetland trees, which are ecosystems featuring a clear soil CH₄ source (Jeffrey et al., 2023; Jeffrey et al., 2020a; Pangala et al., 2017; Sjogersten et al., 2020). Previous studies focused on dead tree axial trends, found that CH₄ also decreased in stem concentration, and with increasing stem height (Carmichael and Smith, 2016; Carmichael et al., 2018; Jeffrey et al., 2020b). δ13C–CH₄ measurements in dead trees suggest that CH₄ may be oxidized while moving upwards within tree stems, suggesting a soil source (Martinez et al., 2022). This soil source of CH₄ is further supported by research that showed wood samples produced little CH₄ during anaerobic incubations (Martinez et al., 2022).

In our study, the dead trees were in several metres of standing freshwater, featuring low-oxgyen and high SOD soil conditions. It is therefore possible that the majority of CH₄ originates from the sediments (with the trees acting as passive gas conduits) and diffuses outwards with increasing stem height. Nevertheless, due to the different decomposition states, we cannot rule out that internal microbial production of CH₄ may contribute to some of the observed CH₄ flux. The opposite trend was detected during summer, with the highest CH₄ fluxes occurring at the highest stem height. The water level had dropped ∼ 1.5 m between the two campaigns, thus potentially changing the preferential pathway of CH₄ emissions higher up the tree stems. We acknowledge that further factors, like wood water content or changing internal CH₄ production and oxidation likely also played a role. These factors complicate our system and lead us to the conclusion, that our summer results do not necessarily point to a wood source.

In living trees, it has recently been shown that CH₄ (gas phase) can travel rapidly through the bark of some wetland species, separate to the transport within the transpiration stream (liquid phase) (Jeffrey et al., 2024). Because dead trees no longer have bark and lack an active transpiration stream, this suggests that diffusion through cracks and open spaces through the dead wood may have been a more important pathway for CH₄ egress (gas phase) to be released from the reservoir sediments. However, the wood below the water line was saturated and thus reduced CH₄ diffusion rates (dissolved phase), being several orders of magnitude slower than gas. This may have led to a lag time between CH₄ production changes within the soil, and stem CH₄ gas emissions above the waterline. Further research is required to confirm both the CH₄ source(s), the transport rates and preferential pathway(s) for ghost forests.

We found no correlation between surface water depth and tree stem CH₄ flux (Fig. S3). This was similar to riparian living tree studies in the Amazon (Gauci et al., 2022) and Australian wetland forests (Jeffrey et al., 2023) which found that as the below-ground water table rose, it promoted anaerobic conditions in the soil, leading to increased CH₄ production and, consequently, higher tree fluxes. However, once surface inundation occurred, further increases in surface water depth did not significantly alter soil conditions, and therefore tree CH₄ fluxes remained similar. For the flooded ghost forest, as stem CH₄ fluxes were decoupled from variation in surface water depth, this likely reflected the dominance of soil CH₄ production, stem wood properties and decomposition state, over hydrological controls.

Overall, there were three-fold higher CH₄ fluxes observed during the summer campaign, along with elevated dissolved CH₄ concentrations and ∼ 7 °C increase in surface water temperature, particularly in the south site. This temperature change likely increased microbial metabolism, including methanogenesis and, therefore resulted in higher rates of CH₄ production. This was further evidenced as ebullition rates also increased in both the Mid and South sites, during summer. Temperature has been shown to enhance CH₄ flux rates (Yvon-Durocher et al., 2014) and increased ebullition in summer across multiple lake systems (Sanches et al., 2019; Aben et al., 2017). Sediments from boreal and temperate lakes have shown that CH₄ production and release increase exponentially with rising temperature (Liikanen et al., 2002; Duc et al., 2010). Previous studies have found higher ebullition fluxes above 30 °C compared to below 20 °C (Xun et al., 2024). In contrast, other studies have found that ebullition has no significant positive correlation with water temperature (Grinham et al., 2018b), suggesting organic matter content and water level changes as more important drivers. Ebullition release can also be influenced by sudden atmospheric pressure changes and wind driven turbulence (Kellner et al., 2006; Tokida et al., 2007). Aside from seasonal variability, ebullition bubble release events are highly heterogeneous (Anthony and Anthony, 2013) and temporally variable (Linkhorst et al., 2020) making them a difficult CH₄ flux term to accurately constrain (Rosentreter et al., 2021).

The aquatic surface CH₄ concentration was lower in the deep reservoir central basin, compared to the shallower ghost forested embayments. Deeper water results in a longer residence time of CH₄ within the water column, allowing for greater oxidation of CH₄. Up to 80 % of CH₄ being produced in deep water sediments can be oxidized, while CH₄ produced in shallower zones undergoes less oxidation prior to being emitted to the atmosphere (Bastviken et al., 2008).

Methane dynamics showed notable differences among the three investigated sites, suggesting site specific factors driving CH₄ fluxes. Although temperature is a key factor, it is not the only factor influencing CH₄ concentration and fluxes. The South site consistently showed the highest ebullition and tree stem fluxes between all three sites, confirming research question two. The South site is located at the inflow into the reservoir through the Nerang River, resulting in fresh organic matter being deposited into this section of the reservoir. Even though the organic carbon concentration in the sediment was similar between the three sites, the depth of the organic layer in the South site was ∼ 3-fold deeper (< 30 cm compared to > 10 cm), leading to greater substrate availability for methanogenesis, likely explaining the higher tree and ebullition CH₄ fluxes observed. Similar to our findings, a study at nearby Little Nerang Dam, Australia (upstream from the eastern inflow), found that the ebullition fluxes were driven by sediment organic matter content, especially at the inflow site (Grinham et al., 2018b). Other studies have reported organic matter availability influencing ebullition CH₄ fluxes (Casper et al., 2000; Sobek et al., 2012) and have shown higher ebullition at inflow sites in reservoirs (Tušer et al., 2017; Shi et al., 2025; DelSontro et al., 2011). A previous study at Hinze Dam, before the dam was raised, also found the highest fluxes in the south-west of the reservoir at the same inflow site (Sherman and Ford, 2011).

The lability of the carbon in the organic material and the microbial community in the sediments should also be considered. Studies have found that CH₄ production is closely linked to the degradability of the organic material (Praetzel et al., 2020; Grasset et al., 2018; Zhou et al., 2025; Berberich et al., 2020). Phosphorus is also an important nutrient for microbial growth. The South site sediment had the highest phosphorus concentrations, which has been linked to increased primary and higher CH₄ production (Bastviken et al., 2004). In addition, another proxy for microbial activity and organic matter lability is the sediment oxygen demand (SOD). During the summer SOD was highest at the South site, further confirming the assumption that there is more easily degradable organic material and higher microbial activity within the sediment. Our findings also show lower aquatic dissolved oxygen concentration at the bottom of the water column and the lowest dissolved oxygen concentration at the South site. Oxygen consumption and the resulting depletion in the water column is usually driven more by SOD in shallow aquatic ecosystems, than water column oxygen consumption (MacPherson et al., 2007; Caldwell and Doyle, 1994).

This is the first study measuring CH₄ fluxes from ghost forest trees inside a reservoir. Ghost forests can originate from diverse forest types and due to different causes. Due to this heterogeneity, ghost forest tree CH₄ fluxes are still relatively unexplored. Therefore, further measurements across different environments are needed to compare with our reservoir observations and better constrain the potential range of CH₄ emissions from ghost forest tree stems. Compared to those previous ghost forest tree CH₄ studies, our average CH₄ stem fluxes were within the reported range addressing research question one. However, it should be noted that each study considered a different amount of stem height measurements (Table 5). As CH₄ fluxes can change with stem height, the height and number of heights measured can strongly influence estimated fluxes. One tree in our study emitted the highest single flux from a dead tree, almost twice that of previous maxima (Table 5), highlighting potential for extreme heterogeneity. One reason our higher CH₄ fluxes may occur is because our study was located within a subtropical region, featuring higher CH₄ production and fluxes, compared to temperate climates. Although Jeffrey et al. (2019b) studied mangroves in tropical regions, seawater-derived sulphate in mangrove ecosystems likely lowered soil CH₄ production. Sulphate and iron reducers can outcompete methanogens for H₂ or acetate when sulphate and iron are available (Roden and Wetzel, 2002). Similarly, salinity and sulphate on coastal forests can influence soil CH₄ production (Martinez and Ardón, 2021) and reduce tree CH₄ fluxes. Our study showed that the site conditions in a freshwater reservoir ghost forest were favourable for high CH₄ production.

Internal stem CH₄ gas concentrations were similar to that of subtropical dead mangroves, but higher than the other studies. This may be attributed to the longer incubation period (24 h) used in our study, compared to others, which typically have incubation times under one hour. Additionally, our stem gas concentrations were collected during the summer campaign, when higher tree fluxes and potentially elevated stem gas concentrations were observed likely due to increased microbial activity. Importantly, we did not observe any uptake of CH₄ which had previously been observed in ghost forest trees (Martinez and Ardón, 2021). Previous studies measured net CH₄ uptake in 38 % and 20 % of their measurements, respectively (Carmichael et al., 2024; Carmichael et al., 2018). Carmichael et al. (2024) found sequencing based evidence for methanotroph communities inside dead trees, showing the potential oxidation capacity, whilst our results suggest that net CH₄ emissions exceeded any CH₄ oxidation.

Our study reveals that reservoir ghost forest trees can emit a substantial amount of CH₄ to the atmosphere. Our subtropical ebullition rates were the largest contributor to overall reservoir CH₄ fluxes at 58 % and 67 %. Studies have found herbaceous (non-woody) plant-mediated diffusion can dominate shallow aquatic CH₄ fluxes, with ebullition only playing a minor role (Jeffrey et al., 2019a; Whiting and Chanton, 1992; Chanton et al., 1992; Bastviken et al., 2023). However, in small and shallower lakes in Europe, and ponds in Australia, ebullition was determined to be the primary CH₄ source to the atmosphere (Schmiedeskamp et al., 2021; Grinham et al., 2018a). Indeed, ebullitive fluxes have been estimated to account for 80 % of the total reservoir flux globally (Johnson et al., 2021), especially within (sub)tropical regions (Harrison et al., 2021). Despite our deployment strategy (and daily sampling) of ∼ 60 ebullition chambers, spanning three sites and various water depth gradients – our sampling approach is still seasonally limited, and it therefore remains challenging to draw a definitive conclusion. Greater temporal and spatial sampling approaches would reduce these uncertainties and better constrain this CH₄ flux term. Degassing downstream of the dam wall has been found to emit significant amount of CH₄ (Harrison et al., 2021). This is particularly important in hydropower dams. However, Hinze Dam is a drinking water reservoir and water outflow of surface water is restricted (none during 2023) resulting in minimal degassing. Although we could not quantify that pathway, we argue that it is not a significant contributor in our system.

Ghost forest tree stem CH₄ contributions of 15 % and 14 % during winter and summer, respectively, suggest a persistent, and substantial CH₄ source, addressing research question three. We acknowledge that extrapolating reservoir-wide emissions from only three sites should be considered with caution, particularly given the absence of data from the eastern part of the reservoir. Also, our approach to collect ebullition likely underestimates this term, as CH₄ saturation within the headspace of the floating chambers may limit the continued diffusion of the gas. However, tree stem upscaling was done conservatively to only 2.5 m of stem height, therefore ghost forest tree CH₄ flux contribution may well be underestimated. Future studies should therefore consider ghost forest tree stem emissions, when ghost forests are present, to improve the accuracy of reservoir greenhouse gas budgets. Furthermore, because we used the Mid site – which exhibited lower emissions compared to the other sites – to scale up for large areas of the reservoir, our approach likely results in a conservative estimate of ebullition and total emission. Previous estimation of CH₄ emissions of wood decay in a reservoir represented 26 %–45 % integrated over 100-year time scale (Abril et al., 2013). Here we measured the CH₄ flux emitted by the trees, including CH₄ originating from the sediments. Based on our findings, emissions based on total carbon biomass are underestimating overall emissions, and it is crucial to measure the actual CH₄ flux from the trees. Although this pathway will eventually cease as trees decompose and disappear, the high moisture content of the timber slows decay and likely prolongs emissions.

Comparing the three CH₄ pathways at a per water surface area, ghost forest trees (upscaled to 2.5 m) contribute up to 428 µmol m⁻² water-surface d⁻¹. This is greater than the average diffusive flux (306 µmol m⁻² surface d⁻¹), but ebullition (6286 µmol m⁻² surface d⁻¹) is still the main contributor on water-surface areal basis. Comparing the fluxes on a per surface area has its limits, due to the three dimensions of trees, but it can help show the differences and directly compare between the fluxes. When upscaled to 10 m stem height (or about half way to the canopy), the contribution of ghost forest tree fluxes would increase to 1711 µmol m⁻² surface d⁻¹, demonstrating the sensitivity of upscaling total flux estimates to other stem heights.

Pre-impoundment vegetation clearing is a recognized management strategy and can been implemented prior to dam construction. This harvesting of timber would prevent emissions of CH₄ from ghost forest trees. Dams are sometimes constructed in rugged, hilly terrain to maximize storage capacity, making large-scale timber removal logistically difficult and cost-prohibitive. Trees were cleared during the initial construction of the Hinze Dam, but once the reservoir was filled and the dam level increased, it was difficult to reach the trees at the then edges of the reservoir.