Drivers of CO2 along a mangrove-seagrass transect in a tropical bay: delayed groundwater seepage and seagrass uptake

Water-to-air carbon dioxide fluxes from tropical coastal waters are an important but understudied component of the marine carbon budget. Here, we investigate drivers of carbon dioxide partial pressure (pCO2) in a relatively pristine mangrove-seagrass embayment on a tropical island (Bali, Indonesia). Observations were performed over eight underway seasonal surveys and a fixed location time series for 55 hours. There was a large spatial variability of pCO2 across the continuum of mangrove forests, seagrass meadows and the coastal ocean. Overall, the embayment waters surrounded by mangroves released CO2 to the atmosphere with a net flux rate of 18.1 ± 5.8 mmol m d. Seagrass beds produced an overall CO2 net flux rate of 2.5 ± 3.4 mmol m d, although 2 out of 8 surveys revealed a sink of CO2 in the seagrass area. The mouth of the bay where coral calcification occurs was a minor source of CO2 (0.3 ± 0.4 mmol m d). The overall average CO2 flux to the atmosphere along the transect was 9.8 ± 6.0 mmol m d, or 3.6 x 10 mol d CO2 when upscaled to the entire embayment area. There were no clear seasonal patterns in contrast to better studied temperate systems. pCO2 significantly correlated with antecedent rainfall and the natural groundwater tracer radon (Rn) during each survey. We suggest that the CO2 source in the mangrove dominated upper bay was associated with delayed groundwater inputs, and a shifting CO2 source-sink in the lower bay was driven by the uptake of CO2 by seagrass and mixing with oceanic waters. This differs from modified landscapes where potential uptake of CO2 is weakened due to the degradation of seagrass beds, or emissions are increased due to drainage of coastal wetlands.

Digital reproduction on this site is provided to CIFOR staff and other researchers who visit this site for research consultation and scholarly purposes. Further distribution and/or any further use of the works from this site is strictly forbidden without the permission of the Continental Shelf Research journal.
You may download, copy and distribute this manuscript for non-commercial purposes. Your license is limited by the following restrictions: 1.
The integrity of the work and identification of the author, copyright owner and publisher must be preserved in any copy.

2.
You must attribute this manuscript in the following format: This is a manuscript version of an article by Macklin

Introduction
The large net primary production of the world's coastal embayments are exported to coastal waters (Robertson et al., 1992) primarily through the interplay of tidal dynamics and seasonal river discharge. With a global area of ~45000 km 2 , intertidal areas of temperate embayments predominantly comprise of salt marsh habitats (Greenberg et al., 2006) occupying low-lying topographic zones (Scott et al., 2014) and high distributions of seagrass beds in subtidal zones (Short et al., 2007). In contrast, tropical coastal embayments typically consist of a continuum of fringing coral reefs, seagrass beds and mangrove forests (Torres-Pulliza et al., 2013). Near-shore tropical mangrove forests are the most carbon-rich forests on earth, storing and sequestering globally significant amounts of carbon in their soils (Donato et al., 2011).
Occupying only 0.02% of global surface area, mangrove forests are responsible for approximately 11% of the total terrestrial organic carbon delivery to oceans (Jennerjahn & Ittekkot, 2002;Sippo et al., 2017). Mangroves are tightly connected with their adjacent habitats (Signa et al., 2017) and support marine biodiversity, regulate water quality and protect tropical coastlines against storms (Ganguly et al., 2017).
Tropical seagrass beds are located shoreward of coral reefs and seaward of mangrove forests in areas with high light availability and favourable water quality (Guannel et al., 2016) and have been reported to be largely net autotrophic (i.e. a net atmospheric CO 2 sink) (Duarte & Cebrian, 1996). Coastal geomorphology is recognised as being important in seagrass abundance, distribution and diversity, as these habitats usually exist near fringing reefs in protected, shallow coastal lagoons (Torres-Pulliza et al., 2013). Combined, mangrove forests and seagrass beds play a major role in biological connectivity of coastal embayments, acting as coastal buffers by filtering sediment and nutrient loads to adjacent coral reefs (Hemminga & Duarte, 2000).
Indonesia, lying between latitudes 6 ºN and 11 ºS, has a coastline of more than 95,180 km, the second longest coastline in the world (Spalding et al., 1997) and 2.9 Mha of mangrove cover, larger than any continent on earth (Atwood et al., 2017). With such an extent and high carbon stocks, Indonesia's mangrove forests store on average 3.14 PgC (Murdiyarso et al., 2015). However, in three decades , Indonesia has lost 40% of its mangroves, mainly as a result of aquaculture development (Giri et al., 2011). This has resulted in potential global annual emissions of 0.07 to 0.21 Pg CO 2 (Murdiyarso et al., 2015). Seagrass beds cover an estimated 30,000 km 2 of Indonesian coastline (Green & Short, 2003), and combined with mangrove forests, account for approximately 3.4 Pg C (~17%) of the global blue carbon reservoir (Alongi et al., 2016).
Mangrove-seagrass connectivity research has usually centred on the exchange of dissolved organic carbon (DOC) and particulate organic carbon (POC) Hemminga et al., 1994;Maher et al., 2013;Müller et al., 2015). Little is known about CO 2 interactions between near-mangrove forest surrounding waters (described as mangrove forest water from here on) and adjacent seagrass beds. Stable isotope studies show that seagrasses close to mangroves have a more depleted δ 13 C value than those further away (Bouillon, Connolly, et al., 2008;Hemminga et al., 1994), suggesting that seagrasses are fixing DIC sourced from mangrove respiration.
Mangrove groundwater and porewater exchange can be an important source of carbon to coastal waters (Bouillon et al., 2007;Maher et al., 2013;Maher et al., 2017;Sadat-Noori et al., 2016). A recent literature review demonstrates that groundwater fluxes in mangroves can be a major component of tropical coastal carbon budgets with fluxes on the same order of magnitude as rivers (Chen et al., 2018). Since mangrove forests usually coexist with seagrass beds and coral reefs (Fourqurean et al., 1992), and carbon exchange along this continuum supports cross-productivity (Unsworth et al., 2008), understanding the relationship between groundwater seepage, carbon dynamics and ecosystem connectivity in transition zones is important.
Here, we investigate the drivers of pCO 2 dynamics along a mangrove-seagrass transect in Bali, Indonesia. We performed coupled, automated seasonal pCO 2 and radon ( 222 Rn; a natural groundwater tracer) investigations to assess whether CO 2 is derived from groundwater or porewater pathways. We investigate temporal and spatial scales of pCO 2 dynamics, hydrological drivers such as groundwater seepage, delayed antecedent rainfall, and interplay along the mangrove-seagrass continuum in a non-impacted embayment.

Area description
Gilimanuk Bay, a 3.7 km 2 coastal embayment, is located in Jembrana Regency on the northwest coast of Bali, Indonesia. Including two small islands, Kalong Island and Burung Island, the area contains some of Bali's most pristine mangrove forests (Thoha, 2007) ( Figure   1). Oceanic upwelling and tidal exchange from the deep Java Strait supply nutrients to Gilimanuk Bay (Ningsih et al., 2013;Siswanto, 2008). The average depth of the embayment is ~2m, with intertidal zones in the upper embayment unnavigable at low tides (<0.5m; shallow and deep sea waters (3,520 ha) (Utama, 2015). The regional geology includes alluvial deposits and Prapat Agung Formation which consists of limestone, calcareous sandstone and marts (Purbo-Hadiwidjojo, 1971). Soils are hydromorphic alluvial in the nearshore zone and grey-brown alluvial surrounding the embayment. Dominant mangrove species in Gilimanuk Bay include Rhizophora apiculata, Excoecaria agallocha, and Ceriops tagal (Marbawa et al., 2015). Twelve of the world's sixty known species of seagrass are present in West Bali National Park (Purnomo et al., 2017). Dominant seagrass species include Cymodocea rotundata, Halophila ovalis and Enhalus acoroides (Purnomo et al., 2017;Zulkarnaen et al., 2014).

Approach and methods
To investigate hydrologic variations, groundwater seepage and CO 2 fluxes, we combined eight underway spatial surveys (S1-S8) with a detailed 55 hour fixed location time series (TS) in Gilimanuk Bay between 20 th November 2015 and 15 th November 2017 (see Table 1). The average tidal range throughout the surveys was 1.3m. As S1 was conducted at the end of a subsequent drought period and followed by a period of prolonged rainfall, following surveys (S2-S8) were intended to replicate rainfall events and possible groundwater relationships. .
During the underway surveys, we measured high resolution spatial variations at 10 minute intervals while location was tracked and logged by a Garmin GPS72 continuously. Each survey commenced at high tide beginning at the ocean mouth and ending upstream at the mangrove forest, in a small research vessel travelling between 4 and 6 km/hr, while the time series was conducted at the ocean mouth ( Figure 1).
A Li-820 CO 2 detector and a radon-in-air monitor (RAD7, Durridge) were deployed to measure pCO 2 and 222 Rn concentrations at approximately 1m depth. The Li-820 and the RAD7 were connected with a closed-air-loop to a shower head gas exchange (Dulaiova et al., 2005;Santos et al., 2012) using the methodology of Santos et al. (2012) and references therein. A Li-820 (calibrated before underway surveys with 0, 400 and 10 000 ppm spans) was checked for accuracy alongside a calibrated Li-820 which was deployed in the 55 hr time series at the end of the study campaign. The RAD 7 was, pre-calibrated by the manufacturer (Durridge) and is expected to hold calibration for at least one year. Mole fraction measurements provided by the Li-820 CO 2 detector were later calculated to pCO 2 according to the recommendations of Pierrot et al. (2009). Atmospheric pCO 2 was assumed to be constant at 400 µatm. A Hydrolab DS-5 water quality sonde was calibrated before each survey and deployed to measure temperature, salinity and dissolved oxygen at 10 min intervals. We used a calibrated handheld YSI EcoSence EC300 meter and a Hatch 40D LDO Sensor for field measurements every 10 to 15 minutes, which were comparable to the Hydrolab DS-5 measurements. CO 2 fluxes at the water-air interface were then calculated as function of air-water CO 2 gradient (ΔpCO 2 ), temperature and salinity dependant solubility (k 0 (Weiss, 1974)) and gas transfer velocity (k) according to: where k is the CO 2 gas transfer velocity, K 0 is the solubility of CO 2 (Weiss, 1974) and ΔpCO 2 is the difference between sea and air (pCO 2sea − pCO 2air ).
Positive CO 2 flux values signify a CO 2 exchange from water-to-air (CO 2 source) while negative CO 2 flux values signify an exchange from air-to-water (CO 2 sink). Estimations of CO 2 fluxes were calculated by using 10 minute sampling times for both underway and time series measured pCO 2 data and average wind speeds from 28 days prior to each survey and the time series (Table 1). Rainfall and wind data were sourced from Banyuwangi Weather Station (8.21700°S, 114.38300°E; average atmospheric pressure=1021.5 hPa (1.0 atm); 10 m above sea level) located 7 km from the study site (www.dataonline.bmkg.go.id).

Seasonal spatial surveys
Water temperature was lowest at the ocean mouth increasing towards the shallow mangrove forest water endmember (max=34.7 ºC; Table 2) throughout the eight underway surveys. The lowest salinity was observed in the mangrove forest water endmember (26.2; Survey 1) however overall average salinity ranges were relatively close to seawater (31.5 -33.0) reflecting the characteristics of an ocean dominated embayment. Dissolved oxygen (DO) was slightly undersaturated at the ocean entrance and increased significantly in the seagrass beds in 5 out of 8 surveys (reaching 166.3 %; Survey 5, Figure 3), while the lowest DO was observed in the mangrove forest water area (≤ 60% in 5 surveys). There was minimal tidal variation on survey days (mean= ~1.3m although Survey 6 tidal range was ~2m.  Stronger correlations between 222 Rn and pCO 2 than between dissolved DO and pCO 2 ( Figure   4) in 7 out of 8 surveys suggest that the seepage of groundwater, not pelagic respiration and photosynthesis, drove pCO 2 supersaturation within the embayment, particularly in the mangrove forest water (Figure 4). pCO 2 was mostly above atmospheric equilibrium (>400 µatm) with up to a five-fold increase in the mangrove forest water (max = 2101µatm). On average, pCO 2 in the mangrove forest water was ~four-fold that of the overall embayment ( Figure 3). In contrast, the seagrass beds were undersaturated in 30% of surveys suggesting that, seagrass was fixing mangrove water-derived CO 2 from the upper embayment.  Figure 5) between rainfall and radon, CO 2 , and salinity were used to interpret the lag time between rainfall, groundwater discharge, and subsequent response in groundwater-derived pCO 2 or radon in Gilimanuk Bay. In spite of the relatively dry conditions during sampling, cumulative antecedent rainfall 17 to 63 days prior to surveys had significant correlations to pCO 2 (p≤0.05) with peak correlations (R 2 = 0.81; p<0.002) when 29 days of cumulative rainfall was used. Significant correlations between pCO 2 and cumulative antecedent rainfall was observed 15 and 84 days prior to our survey in the mangrove waters (p<0.05) with the strongest correlation at 48 days cumulative antecedent rainfall (R 2 = 0.90; Figure 5). Correlations in the seagrass area were significant between 29 and 104 days cumulative antecedent rainfall (p<0.05) with the strongest correlation for 73 days of cumulative antecedent rainfall (R 2 = 0.77). This wide range of significant lagged correlations shown may imply that these correlations may not necessarily represent a causation in the seagrass area. Seagrasses and the embayment mouth only showed significant salinity correlations up to 11 days and 4 days, suggesting direct rainfall is the major influence on local salinity. Interestingly, groundwater seepage (as traced by 222 Rn) mimicked pCO 2 in the mangrove forest water, but was decoupled in the highly productive seagrass ( Figure 5).
This suggests that in-situ productivity was driving pCO 2 rather than groundwater seepage in the seagrasses. In the ocean mouth, 222 Rn had significant correlations with the widest range of delayed rainfall (29-268 days; p<0.05) also implying that these correlations may not necessarily represent a causation in this case. Overall, the lagged correlations imply that delayed groundwater seepage following seasonal rainfall plays a significant role in CO 2 concentration and distribution in this embayment. CO 2 fluxes revealed shifts from a strong source to the atmosphere in the mangrove forest water, an intermittent sink in the seagrass beds to equilibrium or a weak source at the ocean mouth (Table 3). The total mangrove forest area covered ~30 % of the embayment transect and seagrass beds covered ~40 %. Measurements suggested that the release of CO 2 to the atmosphere occurred in mangrove forest water (total average CO 2 flux= 18.1 ± 5.8 mmol m -2 d -1 ) was much greater than in the seagrass beds where the CO 2 flux was 2.5 ± 3.4 mmol m -2 d -1 . Seagrass beds were a sink of CO 2 in 2 out of the 8 surveys ( Figure 3). Average fluxes along the transect were 9.8 ± 6.0 mmol m -2 d -1 , a source of CO 2 (Table 3). No clear seasonal patterns were observed.  Raymond and Cole (2001) presented a mid-range flux value, which is likely more representative of the study site, which does have some tidal flow, but much lower current velocities than the macrotidal Scheldt. Table 4 also includes the oceanic wind speed parameterisations of Wanninkhof and McGillis (1999) and Wanninkhof (2014) for reference. Table 4. Survey (1-8) and time series, average windspeeds 28 days prior (AWS) and CO 2 flux calculations using four author's transfer velocity parameterizations for equations 2-5: Wanninkhof and McGillis (1999), Raymond and Cole (2001), (Borges et al., 2004) and Wanninkhof (2014)

Time Series at the ocean entrance
A 55 hr time series was conducted at the embayment entrance from 16:25, 31 August 2017 to 00:30, 3 September 2017 to investigate aquatic CO 2 exchanges between the embayment and the ocean (Figure 6). There was no precipitation during observations and 11.0 mm of rain was recorded in the month preceding the time series observations. Water temperatures ranged from 25.1 to 27.7 ºC and were lowest between midnight and early morning. Salinity ranged from 34.9 to 35.2. The higher salinity measurements on the outgoing tide may be a result of evaporation within this shallow embayment (data not shown). DO ranged from 82.5 % to 104.3 %. In spite of some data gaps, DO appeared to have diel trends reflecting photosynthesis during the day and respiration at night (Figure 7). The pCO 2 trends did not follow DO. Indeed, pCO 2 had stronger correlations to 222 Rn than DO during the day ( Figure   7) implying the groundwater CO 2 source was stronger than the photosynthesis sink. 222 Rn followed a tidal cycle and correlated with pCO 2 during both incoming and outgoing tides (R 2 =0.53 & R 2 =0.22, respectively; data not shown). The highest 222 Rn, salinity and CO 2 values were observed during the two lowest tides (overall tidal range = 1.5m) which is consistent with groundwater-derived inputs that are well known to occur at low tide (Atkins et al., 2013;Call et al., 2015;McMahon & Santos, 2017;Santos et al., 2009). It is difficult to explain the increase in pCO 2 from ~440 to 500 µatm in the last 4 hours of the time series since both radon (groundwater proxy) and DO (respiration proxy) had no similar changes.
pCO 2 was above atmospheric equilibrium throughout with the highest pCO 2 observations at low tides and lowest at high tides. Low pCO 2 at the embayment mouth are consistent with the survey observations of CO 2 outgassing and/or uptake by the seagrass beds (Table 4; R&C01: 1.4 ± 0.4 mmol m -2 d -1 ).

Discussion
Aquatic systems in Southeast Asia are recognised as significant sources of CO 2 to the atmosphere but remain poorly represented in global databases (Müller et al., 2015). The few studies available focus on tropical river-dominated estuaries which produce significant CO 2 fluxes (Borges & Abril, 2011;Müller et al., 2015). Global summaries of water-to-air CO 2 fluxes are generally confined to human impacted river-dominated estuarine systems (Cai, 2011). A recent study in a temperate autotrophic marine dominated system in Australia reported ten-fold lower and reversed CO 2 fluxes than the more studied river-dominated counterparts . Since there is a paucity of data from marine-dominated tropical coastal embayments, our investigation contributes to filling gaps in global estuarine CO 2 fluxes.
By categorizing CO 2 fluxes into three classes (mangrove forest water, seagrass beds and the ocean-dominated mouth), comparisons with temperate coastal water CO 2 fluxes could be made. For instance, mangrove forest water had CO 2 fluxes (18.1 ± 5.8 mmol m -2 d -1 ) similar to estuarine systems such as the York River estuary in the U.S.A. (17 mmol m −2 d −1 ) (Raymond et al., 2000) and the Pearl River in China (24 mmol m −2 d −1 ) (Yuan et al., 2011).
Overall, the mangrove forest water dominated surrounding water CO 2 fluxes and were within estimated flux ranges for many global coastal waters (Chen et al., 2013). A study by Ho et al. (2017) reported average CO 2 fluxes of 105 ± 9 and 99 ± 6 mmol m −2 d −1 at the end of the wet season, where pCO 2 values ranged from 1000 to 6200 µatm (present study ranges = 516 (S1; Nov.2015) to 1164 µatm (S6; Jan. 2017; Table 2). We report overall embayment CO 2 fluxes of 9.8 ± 6.0 mmol m -2 d -1 comparable to the lower range of the Everglades study, estimated global mangrove fluxes of 4.6 to 113.5 mmol m −2 d −1 (Borges et al., 2003) and revised global estimates of 56.8 ± 8.9 mmol m -2 d -1 (Rosentreter et al., 2018). This is most likely due to lack of river inputs and large uptake by seagrass beds in the mid-embayment.
Seasonality seems to play a greater role in temperate waters (Guo et al., 2009). In the Changjiang Estuary (China) seasonal ranges were 52.9 mmol m −2 d −1 (December) to 92.9 mmol m −2 d −1 (August) based on the parameterization of Raymond and Cole (2001) (Zhai et al., 2007). Seasonality was not observed throughout our study. Gilimanuk Bay CO 2 emissions ranged from 2.8 mmol m −2 d −1 (S1; October 2015) to 16.9 mmol m −2 d −1 (S2; January 2016; Table 4) with no clear changes over an annual temperature cycle. Autotrophic coastal systems have been reported as sinks of atmospheric derived CO 2 (Borges and Abril, 2011;Maher and Eyre, 2012). The seagrass-dominated area of Gilimanuk Bay alternated between a CO 2 source and sink. The more consistent CO 2 source in the near shore ocean was possibly due to coral calcification in Gilimanuk Bay's fringing coral reefs (Figure 8). Groundwater seepage, porewater exchange and rainfall events have been linked to delayed groundwater discharge as a source of CO 2 in subtropical systems (Ruiz-Halpern et al., 2015).
Rainfall, particularly flood events, rapidly transport carbon stored in upper soil profiles across the land-to-water interface (Atkins et al., 2013;Gatland et al., 2014;Jeffrey et al., 2016;Webb et al., 2016). Paquay et al. (2007) reported that water residence times and periodic floods drove the distribution of pCO 2 in Hilo Bay (Hawaii) with groundwater seepage suggested to explain a 3 day delay of elevated pCO 2 after a heavy rainfall event. In contrast to Hilo Bay, Gilimanuk Bay has no riverine input, making groundwater seepage a dominant source of CO 2 year-round (average R 2 =0.74) followed by temperature (average R 2 =0.65) and salinity (average R 2 =0.63; Figure 4).
Tidally-driven porewater exchange (tidal pumping) has been suggested to release carbon dioxide and nutrients to estuaries (Sadat-Noori et al., 2016) and intertidal flats (Bouillon, Connolly, et al., 2008;Santos et al., 2014). Large tidal amplitudes (spring tides) have been related to enhanced 222 Rn in coastal waters as observed off a sandy beach in Korea (Kim & Hwang, 2002) and in Florida (Santos et al., 2009) and mangrove forest water in Australia (Call et al., 2015). Radon traces any water in contact with sediments regardless of salinity.
However, our dataset cannot resolve whether tidally-driven porewater exchange or fresh groundwater discharge are the source of radon enrichments in the mangrove forest water.
Considering the small tidal amplitude, lack of well-defined tidal creeks in the mangroves, steep topography surrounding the upper embayment, and the significant correlations between radon and antecedent rainfall, we suggest that delayed fresh groundwater discharge (rather than tidal pumping) is more likely to be the radon and CO 2 source to the upper embayment. .
Groundwater discharge at the shoreline is well known to lag rainfall for several months.
Seasonal oscillation in groundwater level and inland recharge can explain large saline groundwater discharge several months after rainfall as observed in a Massachusetts aquifer (Michael et al., 2005). Unfortunately, no groundwater level data are available for the Gilimanuk Bay area to build on this hypothesis.
CO 2 derived from the mangrove forest in the upper embayment appears to be fixed by seagrass beds in the lower embayment, intermittently transforming the lower embayment into a net sink of CO 2 (Figure 1; Figure 3; Figure 8). However, there was higher pCO 2 and waterto-air pCO 2 fluxes in wetter periods in the seagrass beds ( Figure 9; Surveys 3, 6 & 8) which coincided with the highest antecedent rainfall (Table 3). Seagrasses, although known to be net autotrophic, are largely under-represented in global carbon budgets .
Additionally, there are large uncertainties in the extent of seagrass beds globally, resulting from a) many regions in Indonesia being known to support extensive seagrass beds but have not been surveyed and b) the continuing loss of seagrass bed areas as a result of well-reported anthropogenic degradation (Alongi et al., 2016;Duarte et al., 2010;Unsworth & Cullen, 2010). The ocean's blue carbon sinks (mangroves, saltmarshes and seagrasses) capture and store approximately 70% of the carbon perpetually stored in aquatic systems (Nellemann & Corcoran, 2009). In contrast to mangrove forests habitats, which occur only in warmer tropical and subtropical climates, seagrasses are also present in colder northern and southern latitudes (Orth et al., 2006). The importance of their combined presence with mangrove forests in tropical systems is noteworthy as both mangrove forests and seagrass beds stabilise sediment (Hogarth, 2015). About 50% of net primary production produced by seagrasses is buried within seagrass sediments (Kennedy et al., 2010), and ~ 8% of mangrove net primary production is retained in mangrove forest soils (Bouillon, Borges, et al., 2008).
Degradation or loss of mangrove forests and seagrass beds may cause the release of large stores of sedimentary carbon to both the atmosphere and the coastal ocean (Mcleod et al., 2011). Therefore, continued degradation may increase atmospheric CO 2 . Due to the high carbon stores in blue carbon systems, and in particular with the tropical regions such as Indonesia, prevention of ecosystem degradation is crucial for limiting the potential release of this carbon to the atmosphere as CO 2 (Tollefson, 2018). Strategies to counter the drainage of blue carbon ecosystems would be an effective measure in maintaining carbon in soils (Crooks et al., 2011). For example, construction or artificial canals draining coastal wetlands more than doubled CO 2 emissions from waterways on the highly urbanized Gold Coast, Australia (Macklin et al., 2014). Our investigation further demonstrates the role seagrass beds play in taking up dissolved inorganic carbon and preventing emissions from groundwater-derived CO 2 in nearby ecosystems.

Conclusion
Our investigations across a coral reef-seagrass-mangrove continuum revealed a CO 2 source in the mangrove dominated upper bay apparently associated with delayed groundwater inputs, and differing CO 2 dynamics in the lower bay driven by the uptake of CO 2 by seagrass. The bay mouth was a source of CO 2 possibly due to production of CO 2 during fringing coral reef calcification. The average CO 2 water-to-air flux along the transect was 9.8 ± 6.0 mmol m -2 d -1 . Antecedent rainfall and radon were the best predictors of CO 2 dynamics, with no clear seasonality observed, in contrast to better studied seasonal temperate systems. Potential changes in rainfall events due to climate change in Indonesia (Overpeck & Cole, 2007) as well as ecosystem degradation may alter aquatic CO 2 emissions. This study may assist when determining anthropogenic modification buffer zones in mangrove forest embayments. As pCO 2 data in tropical coastal embayments are scarce, more studies are needed to assess the role of these systems in the global carbon cycle.