Sampling sites were chosen in Laamu Atoll (also known as Haddhunmathi) in the Southern Central Maldives, specifically the islands of L.Hithadhoo, L.Maabaidhoo and L.Gaadhoo (Figure 1). These three sites consisted of sandy areas adjacent to residential housing development and were characterized by mangroves on the high ground and seagrasses close to the shoreline. Other sites within the atoll have similar characteristics but these were chosen given previous cooperation with the local communities that facilitated sampling and ease of access. Gaadhoo was chosen specifically because its mangrove ecosystem is enclosed in the center of the island, whereas the other systems are open and intertidal. Mangrove habitats consisted of Red Rhizophora mucronata, yellow Ceriops tagal, black Lumnitzera racemosa mangroves, and ironwood Pemphis acidula. Seagrass habitats were made up of diverse species assemblages including Thalassia hemprichii, Thalassodendron ciliatum, Halodule pinofilia, Syringodium isoetifolium, and Cymodocea rotundata.

In 2019, 18 sediment cores were collected from both mangrove (total of 9; Figure 1) and seagrass (total of 9; Figure 1) ecosystems across all chosen sites, to quantify the soil carbon parameters for the first time. At each site, we collected a minimum of three cores from each of the two ecosystems, at a minimum of 1 m apart, parallel to the coastline. At each of the sampling points, a coring device made from PVC pipe (100 cm length, 5 cm internal diameter) was hammered into the sediment to a depth of 1 m below the sediment surface or until hard bedrock sediments were reached. Core compaction, estimated by measuring the distance between the sediment surface within the core and the outer sediment surface, was recorded in the field for all PVC cores before dislodging from the seafloor. Once dislodged, the sediments were transported to the lab and extruded in situ.

To stabilize sediments for transport and storage, the cores were filled with foam material and sealed with tape. While we strived for 1 m cores (and this was achievable at some locations – e.g. 1134 cm at Gaadhoo Island seagrass meadows), there were many instances where we could not achieve such depths. Figure 2 shows photos of extruded cores, illustrating the varying lengths of achievable cores (in this case cores of ~40-60 cm are shown) and sediment types. Therefore, we extruded and divided sediments into the following sections that were common to all cores to enable between-site comparisons: 0- 2, 10-12 and 28-30 from the surface of the core. Plant belowground biomass was not separated from sediment due to its interwoven nature. Dry samples were sent to the UHH Analytical Laboratory, University of Hawai’i at Hilo, to estimate the organic matter content via Loss on ignition (LOI) method. LOI is a measure of the mass of a sample lost when heated to high temperatures of 450°C for 4–8 hours (Heiri et al., 2001).

Subsequently, we used Kauffman and Donato (2012) and Fourqurean et al. (2012b) to establish the relationship between LOI% and organic carbon (OC%) for mangroves and seagrass samples. While LOI is known to drift from a direct relationship with percent organic carbon, this drift mainly occurs when LOI values are high [> 20% (Santisteban et al., 2004)], which was not the case for any of our samples. LOI can also produce more error for organic carbon measurements where soils contain calcium carbonate, but this error is usually < 0.25% (Li et al., 2020). Since this study was exploratory in nature, the use of LOI to obtain organic carbon (OC%) was deemed appropriate. Soil carbon density and carbon stock were calculated according to (Howard et al., 2014). The 28-30 cm section was used to estimate the carbon parameters for the remainder of the sediment core (20-50 cm). Visual analysis of the core showed that upper layers were more variable than lower, while lower layers tended to be homogenous; therefore, the lowest layer of 28-30 cm was used to extrapolate estimates for the deeper layers, and data from deeper parts of the core (> 30cm) were not used.

Due to methodological issues that arose during sample processing, soil bulk densities from Laamu atoll could not be used for its intended purpose. Since this data is necessary to calculate carbon stocks, and no such data is available for the Maldives, instead we have used surrogate values from cores obtained in the same fashion from the next atoll directly south from Laamu, Huvadhu Atoll. Preliminary analysis from seagrass sediments around H.Fares-Maathodaa and H.Hoandedhdhoo islands provided organic carbon values (Helber et al., unpublished data) that were within the same range of the current study. As sediment organic carbon was the most important variable in models to successfully predict Dry Bulk Density (DBD) of soils (Katuwal et al., 2020; Alemu I et al., 2022) and raw DBD data are rarely provided in other seagrass studies [see Piñeiro-Juncal et al. (2022)], we feel confident to use the DBD values obtained from H.Fares-Maathodaa and H.Hoandedhdhoo in our study. Dry bulk density values from mangroves were not available from this atoll, so instead mean values from Sri Lankan sites (the nearest available data) were used (Perera and Amarasinghe, 2019).

Twelve seagrass sediment cores were sampled in November and December 2022 around Fares-Maathodaa and Hoandedhdhoo and sectioned into 2 cm slices until 20cm depth. From below 20 cm depth, 2 cm sediment slices were taken every 5 cm, and from 50 cm onwards every 10 cm. Sediment slices were transferred into ziplock bags and frozen at -20°C until further analysis. For the determination of Dry Bulk Density (DBD), sediment sections were oven-dried at 60°C until there were no further changes in dry weight and their final weight recorded.

Sediments often contain inorganic carbon that could lead mixing models to produce inaccurate results for δ¹³C values. As a result, sediments were acidified using an HCl bath until they stopped bubbling. Since this acidification can inadvertently impact δ¹⁵N values, a separate un-acidified sample would also be processed to obtain accurate δ¹⁵N values. One set of sediment samples from Gaadhoo had extremely high %C and %N indicative of low or no inorganic carbon, and so these were not acidified for either δ¹⁵N or δ¹³C value measurements.

Samples of sediment and possible sediment sources were processed at the UHH Analytical Laboratory at the University of Hawai’i. Stable isotopes of ¹⁵N and ¹³C were measured as well as %C and %N. Isotope ratios were measured relative to standards USGS 40 and 41 (atmospheric air for δ¹⁵N values and Vienna Pee Dee Belemnite for δ¹³C values. Elemental precision relative to the quality control samples of similar elemental composition run alongside samples (Buffalo river sediment for sediments and peach leaf tissue for plant tissues) with known stable isotope ratios was < 0.2‰ for both δ¹³C and δ¹⁵N values.

Provenance of sediment organic matter was determined using Bayesian stable isotope mixing models through package MixSIAR in R V 4.0.0 (Stock et al., 2018). A key assumption with these mixing models is that all sources need to differ (Phillips et al., 2014). As a result, sources that had standard deviations that overlapped were grouped a posteriori (Mangrove and Screw Pine at Gaadhoo). In all cases, the final number of sources was less than the recommended maximum of n + 1 where n is equal to the number of tracers (Parnell et al., 2010). Since the number of sources differed between locations, each island had a separate mixing model run. No trophic enrichment factors were used given there should be no fractionation of sediments, but standard deviations were set to account for elemental precision (± 0.2‰). Although there is some evidence that decomposition can impact stable isotope fractionation, these effects are unpredictable (Kelleway et al., 2022), and the effect of decomposition is generally < 3 ‰ so will not significantly change the mixing polygon. As a result, fractionation effects from decomposition were not included here. Concentration dependencies were used given the differences in %C and %N between some of the sources present (Phillips and Koch, 2002). Model runs were adjusted to either “long” or “very long” run lengths to meet Geweke and Gelman-Rubin diagnostic criteria (Stock et al., 2018).

Across all sites, loss on ignition results indicated sediments cores contained 7.75 ± 3.43% organic matter (mean ± SD), with mangrove cores containing higher concentrations (9.31 ± 3.51%) than seagrasses (6.68 ± 2.98%), and there was no significant difference between sites for mangroves (df = 2, F = 2.02, p > 0.05) or seagrasses (df = 2, F = 0.09, p > 0.05). Percent organic carbon from ashed sediments was 6.54 ± 12.90%OC, with the high standard deviation reflecting the variability across sites and sediment cores, whereby mangrove sites at Gaadhoo had very high %OC (33.56 ± 5.56). Excluding these sites at Gaadhoo with very high concentrations of organic carbon, sediments cores contained 0.73 ± 0.50%OC. Mass spectrometry indicated un-acidified sediment samples were 11.13 ± 0.65%C (mean ± SD), whereas acidified samples without inorganic carbon were 26.72 ± 6.42%C, suggesting roughly a quarter of organic matter was carbon.

Soil carbon density in the top 30 cm of the soil profile in mangrove and seagrasses ecosystems was on average 0.09 ± 0.01 g cm^-3 (± SE, n= 18) and 1.22 ± 0.03 g cm^-3 (± SE, n= 144, Figure 3), respectively. The average dry bulk density in the mangroves and seagrasses was 1.28 ± 0 g cm^-3 and 1.32 ± 0.03 g cm^-3, respectively (Figure 3). Organic carbon content was on average 2.7 times higher in the mangrove soils 6.81 ± 0.3% compared to seagrass soil at 2.52 ± 0.2% (Figure 3).

Soil carbon density in the top 2 cm of the soil profile in mangrove and seagrasses ecosystems was on average 0.07 ± 0.01 SE (mg cm^-3, n = 6) and 1.01 ± 0.07 SE (mg cm^-3, n = 12, Figure 3), respectively. The 10-12 cm depth range recorded an average carbon density of 0.1 ± 0.01 SE (g cm^-3, n = 6) for mangroves and 1.16 ± 0.10 SE (g cm^-3, n = 12) for seagrasses. At depths of 28-30 cm and 29-31 cm, carbon densities were 0.09 ± 0.01 SE (g cm^-3, n = 6) and 1.59 ± 0.10 SE (g cm^-3, n = 10), respectively (Figure 2). The average dry bulk density was the highest in the bottom part of the soil profile at the 28-30 cm depth, with mangroves at 1.35 ± 0.09 SE (g cm^-3) and, at 29-31 cm depth, seagrasses at 1.59 ± 0.10 SE (g cm^-3). Organic carbon content was highest at 10-12 cm depth (7.55 ± 0.6%) under mangroves and at 0-2 cm depth under seagrasses (2.99 ± 0.6%).

The carbon stock recorded in the mangrove ecosystem soil was on average 393 ± 29 tonnes ha^-1 (± SE) and 167 ± 20 tonnes ha^-1 in the seagrass ecosystem (± SE, Figure 4).

A total of 38 plant samples and 55 sediment samples were analyzed. These were from a mix of species that were not always present across sites, including: Red Rhizophora mucronata, yellow Ceriops tagal, black Lumnitzera racemosa mangroves, and ironwood Pemphis acidula, all of which were classified as ‘mangrove’. Species of seagrass included Thalassia hemprichii, Thalassodendron ciliatum, Halodule pinofilia, Syringodium isoetifolium and Cymodocea rotundata. Sediment stable isotope values were generally bound within the polygon produced by the likely sediment sources (mangrove, screw pine and seagrass, Figure 5). Depending on the island sampled, sediment stable isotope values were either grouped closely to seagrass or mangrove/screw pine values (e.g. Gaadhoo), or more broadly distributed between the available sources (e.g. Maabaidhoo).

Mangroves consistently provided the smallest proportion of organic matter to sediments, with screw pine contributing slightly more (0.23 ± 0.12, mean ± SD; Figure 6). Across all islands, seagrasses contributed the most to sediment organic content, with the highest contribution from seagrasses ranging to 0.72 ± 0.05 (SD) in Maabaidoo and lowest mean contribution of 0.55 ± 0.10 (SD) in Gaadhoo. In some cases, some of the distributions of modelled contributions to sediments were broad for sources, ranging from 0.25 to 0.75 for mangroves in Gaadhoo, for example.

One of the key assumptions in Bayesian stable isotope mixing models is that all sources have been included in the mixing model (Phillips et al., 2014). This typically results in a mixing polygon that encompasses all the consumers within the model (Smith et al., 2013). While these requirements appeared to be met for Gaadhoo and Hithadhoo, in Maabaidhoo the stable isotope values of the sediments appeared high relative to the closest available source (Mangroves). This placed them at the outer limits of what is feasible given the mixing polygon. While not able to be confirmed, it is possible that this high δ¹⁵N value is driven by sewage or wastewater inputs that are known to have this effect on stable isotope values (Gaston et al., 2004; Page et al., 2023). The very high values could also indicate that, while we attempted to sample every likely source in the system, a sediment source with stable isotope values like those of screw pines in the other islands was not included in this model. This means that the Maabaidhoo model likely overestimates the contribution of mangroves to sediments.

We recorded differences in measurements of organic carbon content between mass spectrometry used for stable isotope analysis and more traditional ignition techniques. Mass spectrometry recorded higher percentages of organic carbon (25%) than ashed sediments (6.5%). Loss on ignition approaches are widely accepted as the industry standard, however, the mineral composition of sediments can result in measurement errors (Santisteban et al., 2004), and refining ignition temperatures can reduce error (Wang et al., 2011), but was not possible in this instance. Conversely, acid-treated samples can overestimate carbon content due to unreacted carbonate carbon (Kennedy et al., 2005). The higher variability of organic carbon measurements from mass spectrometry in our results suggest some unreacted carbonate carbon remained in some of the samples, meaning loss-on-ignition measurements were likely more accurate. Thus, while mass spectrometry is a more precise method, it may be more prone to measurement error from contamination. Future studies should investigate how to reduce this error so that carbon sediment assessments can benefit from the greater precision of mass spectrometry and interpret our results in the context of these uncertainties.