Introduction

Blue carbon ecosystems (saltmarshes, mangroves and seagrasses) can store 4–10 times more organic carbon in their soil than terrestrial forests (Donato et al. 2011; McLeod et al. 2011; Duarte et al. 2013; Macreadie et al. 2021), with mangroves and saltmarshes storing around 4 times more soil organic carbon per unit area than seagrasses (Siikamäki et al. 2013). Saltmarshes and mangroves are extremely valuable as efficient mechanisms of blue carbon sequestration and storage, as well as providing important ecosystem services such as habitat for juvenile fish and protection from coastal erosion (Himes-Cornell et al. 2018; Hagger et al. 2022). The total amount of organic carbon stored in a defined area is commonly referred to as a ‘stock’, which is composed of three main carbon reservoirs: aboveground biomass (live and dead), belowground biomass (live and dead) and the soil carbon (Howard et al. 2014). Most of the carbon in blue carbon ecosystems is in the soil carbon reservoir, which holds between 60 and 80% of the total organic carbon stock (Ouyang and Lee 2020).

Globally, blue carbon ecosystems are threatened by a number of factors such as land use change (e.g. coastal development or infrastructure for aquaculture), coastal squeeze (the prevention of habitat migration by physical barriers), pollution (e.g. wastewater treatment discharge) and habitat destruction (e.g. off-road vehicles; Alongi 2002; Atwood et al. 2017; Lovelock et al. 2017; Himes-Cornell et al. 2018). Current estimates put global tidal flat ecosystem loss at 16% in the last 40 years (Murray et al. 2019) and seagrass meadows at 19% since 1880 (Dunic et al. 2021). The rate of global mangrove deforestation has decreased since the early 2000s, yet regional and national losses remain concerning. Additionally, habitat fragmentation, the effect of which is not currently accounted for in habitat loss estimates, affects the functionality of mangrove ecosystems (Friess et al. 2019; Bryan-Brown et al. 2020). Carbon crediting frameworks such as ‘Reducing Emissions from Deforestation and forest Degradation plus’ (REDD +) and the Australian ‘Emissions Reduction Fund’ have recently been expanded for use in blue carbon ecosystems (Ahmed and Glaser 2016; Lovelock et al. 2022). These initiatives encourage the conservation and restoration of blue carbon ecosystems to aid in increasing organic carbon storage and avoiding emissions from habitat destruction (Ahmed and Glaser 2016). Without accurate estimates of soil organic carbon stocks from blue carbon ecosystems, coastal vegetated areas are at risk of misrepresentation in national carbon accounting and potential undervaluation in coastal land management.

Soil organic carbon stocks can be difficult to predict due to variability caused by the large number of environmental factors affecting carbon supply, capture and retention. Globally, soil stocks are influenced by climatic, hydrodynamic and geomorphic features (Doetterl et al. 2015; Osland et al. 2017; Atwood et al. 2020). However, on regional and local scales, these factors have less influence (Rogers et al. 2019; Ewers Lewis et al. 2020; Owers et al. 2020). Several potential drivers of soil organic carbon stock variability have been identified, including distance to the ocean, elevation, vegetation type and tidal inundation, but the effect that these drivers have at different spatial scales is not well understood (Young et al. 2021). Vegetation type can be classified at different spatial scales. For example, it may be classified using a broad ecosystem type (such as ‘mangrove’) or more specifically using the dominant species or taxonomic community. This results in different classification levels being used in different studies, which are often not directly comparable. Where species-level data on vegetation is available, it has been shown to provide accurate predictions of soil organic carbon stocks (Ewers Lewis et al. 2020).

Soil properties such as grain size have also been identified as important contributing factors to the capacity and longevity of carbon storage. Finer-grained soils have lower porosity, which reduces the rate of remineralisation of organic carbon (Saintilan and Rogers 2013; Kelleway et al. 2016; Mazarrasa et al. 2023). Tidal inundation is another local-scale driver of organic carbon storage, as increased soil saturation slows the decomposition rate of organic matter (Siikamäki et al. 2013; Lovelock et al. 2014; Owers et al. 2020). Site-specific assessments that include fine-scale drivers of carbon dynamics are therefore important for ensuring accurate site-level estimates of organic carbon storage (Gorham et al. 2021).

With an increasing focus on blue carbon ecosystem restoration and conservation projects at specific coastal wetland sites, information on the effect of local-scale carbon dynamics may be invaluable for providing accurate soil organic carbon storage estimates. Robust methods for estimating and predicting soil organic carbon stocks in blue carbon ecosystems remains a challenge to the implementation of restoration projects, which are often driven in large part by the potential for carbon accumulation and crediting (Wylie et al. 2016; Vanderklift et al. 2019). Using fine-scale drivers to tailor stock estimates to specific ecosystems and locations will improve opportunities for restoration and conservation efforts (Macreadie et al. 2021).

In this study, we investigated the drivers of fine-scale spatial variability in soil organic carbon stocks across multiple sites in semi-arid South Australia. Our main aims were to (1) sample the under-represented semi-arid blue carbon habitats to establish soil organic carbon stock estimates for the region and (2) identify the primary drivers of fine-scale variability in soil organic carbon stocks using a modelling approach that indicates the effects of environmental factors on soil organic carbon stocks.

Materials and Methods

We sampled soil organic carbon stocks to a depth of 50 cm in supratidal saltmarsh, intertidal saltmarsh and mangroves. These data were used to train and evaluate a model of the environmental drivers of soil organic carbon stocks using various fine-scale environmental data collected in situ or collated from existing spatial datasets.

Study Setting

Multiple soil cores were sampled in 2020/2021 from within seven sites along the coast of the Eyre Peninsula, South Australia (Australia): three sites on the east coast and four sites on the west coast (Fig. 1, Table 1). All sample sites are classified as sheltered tidal flats or lagoons (Harris et al. 2002; DEW 2007). All sampling locations consisted of coastal wetlands characterised by either single species intertidal mangrove forests (inundated daily, Avicennia marina), intertidal saltmarshes (inundated daily, dominated by Salicornia quinqueflora and Tecticornia arbuscula) or supratidal saltmarshes (inundated infrequently, dominated by Tecticornia halocnemoides, Tecticornia indica, Maireana oppositifolia, Frankenia sp. and Disphyma crassifolium).

Fig. 1
figure 1

Inset map shows location of study area in Australia; main map shows location of sampling sites along the Eyre Peninsula coast, South Australia. Black circles denote sample locations; Mount Young (1), Franklin Harbour (2), Tumby Bay (3), Acraman Creek (4), Cape Missiessy (5), Nadia Landing (6) and Davenport Creek (7). Coloured circles denote tide gauge locations from East to West: Whyalla (orange), Middle Bank (pink), Port Lincoln (green) and Thevenard (blue)

Full size image
Table 1 Number of soil cores per vegetation type sampled at each site. Mangroves were not sampled at Nadia Landing or Acraman Creek as they were not present along the transects
Full size table

Vegetation Survey Data

Pre-existing coastal vegetation monitoring plots and coastal elevation profiles, set up by the South Australian state government Department for Environment and Water, were surveyed and sampled at each of the seven study sites. The elevation profiles varied in length from 800 to 3300 m, crossing the wetland ecosystems at each site, perpendicular to the coast. The vegetation monitoring plots were 30 × 30 m and were positioned along each elevation profile within large representative patches of different blue carbon vegetation types (supratidal saltmarsh, intertidal saltmarsh and mangrove). Plot locations were chosen by the state government team when establishing the monitoring profiles and were based on representative patches of the specific vegetation communities present. To minimise disturbance to the monitored vegetation plots, we collected core samples just outside the perimeter of plots, but within the representative vegetation patch and at a similar elevation. Inside each vegetation monitoring plot, vegetation type and environmental data were collected, including the dominant species, plant cover/abundance and soil type.

Soil Core Collection and Analysis

The sampling was designed to increase the spatial coverage of blue carbon soil samples in South Australia and broadly enable comparison across sites. At each of the sample locations, three replicate cores were taken close to each vegetation monitoring plot (30 × 30 m). Cores were collected using a stainless-steel Russian D sampler or a steel Jarret soil auger (depending on soil type). Cores were taken at least 5 m apart from each other. Additional core samples were taken at sites (outside of vegetation monitoring plots) where a target vegetation type was not represented in an existing plot. When this occurred, all of the same vegetation and soil type survey data were collected as described above for the plot monitoring. Of the 74 sampled soil cores, five were removed from this study, three samples were too rocky to perform carbon analysis on and two did not reach a depth of 50 cm. In total, 24 cores were taken from 9 supratidal saltmarsh plots, 28 cores from 10 intertidal saltmarsh plots and 17 cores from 6 mangrove plots.

Position and elevation data for each soil core location were collected using a Leica GNS RTK GPS system, which regularly produces locational accuracies in the sub-centimetre range (vertically and horizontally). The base station was a Leica Geodetic System 1200 GNSS unit, with the corrections being transmitted with a Satel PDL radio. The roving unit was a Leica Viva GS10 with a receiving radio. Soil cores were taken to a depth of 50 cm and subdivided in the field into depth increments: 0–5 cm, 5–10 cm, 10–20 cm, 20–30 cm, 30–40 cm, 40–50 cm. Samples were sealed in plastic bags and kept cool until processing and lab analyses that took place at Eurofins APAL Agricultural Laboratory. No sieving was undertaken as this is not standard practice in blue carbon ecosystems (Fest et al. 2022). Samples were air-dried at 40 °C, finely ground with a puck mill and homogenised. Sub-samples (200 mg) were then taken for analyses. A sub-sample was used to measure moisture content (oven dried to 105 °C), and these data were subsequently used in combination with sample volume to calculate dry bulk density. Organic carbon is calculated as the difference between total carbon and inorganic carbon. All forms of carbon, organic matter, biochar and carbonate (inorganic) were measured together using the dry combustion method. Total inorganic carbon (%) was measured on a separate sub-sample using a classical acid-based titration method with a Metrohm 855 Robotic Titration System (Rayment and Lyons 2010). Total inorganic carbon (%) was then subtracted from total carbon (%) to give total organic carbon (%). Organic carbon values were reported as percentage of the 105 °C dry sample weight. Soil organic carbon stock (t/ha) was calculated using the total organic carbon percentage and dry bulk density for each depth increment and summed across the depth of the core as per Howard et al. (2014). Eurofins APAL Agricultural Laboratory changed their instrumentation between sample analysis resulting in total carbon being measured on an Elementar Vario Macro C/N analyser for the east coast samples (sites 1–3 in Fig. 1) and a Leco CN828 C/N Furnace for the west coast samples (sites 4–7 in Fig. 1).

Environmental Data

A range of environmental variables were measured in situ or generated from existing datasets and used as predictors to model soil organic blue carbon stocks. The variables were chosen based on their hypothesised relationship to soil organic blue carbon stocks (Table 2). A description of each environmental variable and the data used to generate it is provided below.

Table 2 Hypothesised relationships between variables and carbon stocks, and sample sizes/ranges of values sampled for each continuous variable
Full size table

The vegetation survey data on all species present at each core sampling location were used to classify the sampled ‘vegetation type’. Intertidal saltmarshes were dominated by Salicornia quinqueflora and Tecticornia arbuscula; supratidal saltmarshes by Tecticornia halocnemoides, Tecticornia indica, Maireana oppositifolia, Frankenia sp. and Disphyma crassifolium; and mangroves by Avicennia marina.

Soil type was based on a visual assessment of core samples and was defined as peat (an accumulation of partially decayed vegetation or organic matter), loam (a fertile soil of clay and sand containing humus), sand (large grained particles, loose, visible shells) and clay (a stiff, sticky fine-grained earth that can be moulded when wet). Each depth increment was classified as one or a combination of soil types, and the most dominant soil type across each core was used to classify soil type for the whole core.

Distances from core sampling locations to the mean high water spring tide level and nearest tidal creek were measured in metres using ArcGIS Pro. Distance to the mean high water spring tide level was measured using a polygon layer of the mean high water spring tide level based on recent LiDAR DEM and tidal range data (Sarira and Clarke 2019). A polyline was manually drawn from each sample GPS location to the nearest point on the high tide polygon. Tidal creeks were identified using aerial imagery from multiple sources (ESRI World imagery 2022; Google Maps satellite imagery 2022) to ensure all creek locations were recorded (as some are only visible in the imagery under certain tidal conditions). Polylines were then drawn from core sample locations to creek edges.

Tidal inundation period is an estimate of the proportion of time that each core sampling site is inundated by the tide, based on their exact elevation and the tidal regime at the site. We used the precise RTK GPS position data and information from local tide gauges, sourced from Flinders Port Holdings, who maintain and calibrate tide gauges and manage tide data reporting for the State of South Australia. Data were sourced from four tide gauges (Whyalla, Middle Bank, Port Lincoln and Thevenard) for the 5 years prior to soil sampling (2015–2020) at 1-min intervals. Tide gauge data for each site was sourced from the closest tide gauge (Fig. 1). For each tide gauge, measurements were corrected to Australian height datum (AHD) using information supplied by Flinders Port Holdings. Frequency of inundation was calculated for each 5-cm increment in tide height AHD and then averaged annually, giving a frequency distribution for each tide height at each tide gauge. A frequency of inundation was then assigned to each sample location (using the precise RTK GPS elevation measurements); and we calculated the annual average percentage of time that a given elevation increment was inundated by the tide.

Modelling

We modelled the relationship between environmental variables and soil organic carbon stocks to a depth of 50 cm. A Bayesian linear mixed-effects approach was used to model organic carbon stocks as a function of vegetation type (reference level ‘mangrove’), soil type (reference level ‘clay peat’), distance to mean high water, distance to tidal creek and tidal inundation period, using the ‘brms’ R package (Bürkner 2021). A set of candidate models were created using the environmental variables listed in Table 2 (Supplementary Information Table S1). Interactions and parameter combinations were determined by their likely ecological significance based on the literature (Table 2). To avoid over-parameterising the model, no more than 6 parameters were considered per model (n = 69, Peduzzi et al. 1996). All numeric variables were checked for correlation and standardised by centring and scaling. Elevation and distance to tidal creek were not included in the same models due to strong positive collinearity (R = 0.76).

To account for non-independence between cores collected at the same sites, ‘site’ was used as a random effect in all models. Four sampling chains ran for 2000 iterations with a warm-up period of 1000 iterations, for total post-warm-up draws of 4000. Default priors of the ‘brms’ package were used, namely a Student’s t-distribution (3, 4.2, 2.5) for the intercept and unbiased priors for regression coefficients. We confirmed model convergence with visual checks and the Rhat statistic. Model selection was based on Bayes R2 score calculated using the bayes_R2 function, leave-one-out cross-validation using the loo function and mean prediction error calculated using the posterior_epred function, all from the ‘rstantools’ package (Gabry et al. 2022). While traditional R2 tends to increase as more variables are added to the model, Bayes R2 penalises the inclusion of additional variables that do not contribute significantly to the model’s predictive power (Gelman et al. 2018).

Results

The raw data on soil organic carbon stocks found that mangroves had the highest total average soil organic carbon stocks of 116 t/ha followed by intertidal saltmarsh 70 t/ha and supratidal saltmarsh 43 t/ha. The pattern of soil organic carbon stocks being highest in mangroves followed by intertidal and supratidal saltmarsh was found across all sites except Cape Missiessy, which had higher stocks in supratidal saltmarsh than intertidal saltmarsh (Table 3).

Table 3 Raw carbon stock data (t/ha) to 50 cm depth at each site by vegetation type. Standard deviation in parentheses
Full size table

The measured soil organic carbon stocks showed a trend of lower stocks at higher elevations, which interacts with vegetation type, as these were influenced by elevation (and tidal inundation; Table 2). Across the sampled sites, mangroves had a mean elevation of 0.44 m AHD (SD = 0.12), intertidal saltmarsh of 0.59 m AHD (SD = 0.16) and supratidal saltmarsh of 1.18 m AHD (SD = 0.38).

Modelling

Model selection was performed on a set of candidate models (Supplementary Information Table S1). The top-ranked model was selected based on having the lowest mean prediction error of 13.45 t/ha (SD = 10.03), lowest leave-one-out cross-validation score of 648.18 and lowest WAIC score of 646.71. This model included all variables (vegetation type, distance to high water, distance to tidal creek, soil texture and tidal inundation period) except elevation which was not included in the model due to collinearity with distance to tidal creek. We found no evidence of systematic patterns in the model prediction error across the categorical and continuous variables (Supplementary Information Fig. S1).

Variables in the model that had a significant effect on soil organic carbon stocks were distance to tidal creek, vegetation type and soil texture. Of the soil texture classes, ‘loam sand’ had the strongest negative relationship with soil organic carbon stocks, followed by ‘clay sand’ and ‘clay loam’. The highest soil organic carbon stocks were associated with the soil texture class ‘clay peat’. Intertidal saltmarshes had lower soil stocks compared to mangroves. Tidal inundation period, distance to mean high water, supratidal saltmarsh and ‘loam peat’ did not have significant effects on soil organic carbon stocks (Fig. 2). The model showed a negative relationship between soil organic carbon stocks and distance to tidal creek, indicating that organic carbon stocks decrease further away from tidal creeks (Fig. 3). The random effect in the model, site, overall had a significant effect on soil organic carbon stocks (Fig. 2), but the direction of the relationship differed for individual sites (Supplementary Information Fig. S3).

Fig. 2
figure 2

Effect size plot of predictor variables and random effect (site) on soil organic carbon stocks. Parameter estimates are Bayesian posterior median values, 95% highest posterior density uncertainty (thin black lines) and 50% uncertainty intervals (thick black lines). Grey colouring indicates no significant effect (95% intervals overlap zero), green represents a significant positive effect and orange represents a significant negative effect on soil organic carbon stocks. The asterisk symbol (*) indicates significance. Intercept includes categorical variable reference categories (mangrove, clay peat)

Full size image
Fig. 3
figure 3

Modelled relationship of soil organic carbon stocks and distance to tidal creek. Grey shading indicates 95% confidence intervals

Full size image

Discussion

This study aimed to provide stock estimates for an under-sampled region and develop an understanding of the environmental drivers influencing fine-scale variability in soil organic carbon stocks across mangrove and saltmarsh ecosystems. We developed a Bayesian model of soil organic carbon stocks as a function of several environmental variables. The most important variables in the model were vegetation type, soil texture and distance to a tidal creek. The significance of site-based differences in the model also indicates that broader scale (regional) factors, such as geomorphology, hydrological inputs and oceanography, contribute to variability in soil organic carbon stocks.

Establishing Stock Estimates for an Under-sampled Region

This study presents the first comprehensive soil organic carbon stock dataset from the semi-arid regions of South Australia. Soil organic carbon stocks (to a depth of 50 cm) were found to be lower at the sampled sites on Eyre Peninsula than in previously sampled temperate areas of South Australia along the Adelaide metropolitan coast. Adelaide’s average mangrove, intertidal saltmarsh and supratidal saltmarsh stocks are 73.85 t/ha, 20.20 t/ha and 31.56 t/ha greater respectively than the average Eyre Peninsula stocks reported in this study (Table 3, Jones and Russell 2021, 2022). Including stock measurements from semi-arid regions will provide more accurate state-wide estimates (Schile et al. 2017; Adame et al. 2021).

When the broad category of ‘saltmarsh’ is used as a vegetation class, it encompasses a wide variety of species compositions and locations in the tidal frame. However, in semi-arid South Australia (and beyond), there are considerable differences in the species composition of supratidal and intertidal saltmarshes that may influence carbon storage. We accounted for this in our study, by separating intertidal and supratidal saltmarsh communities based on the dominant plant species, which enabled us to detect a significant difference in soil organic carbon stocks between mangrove and intertidal saltmarshes (Fig. 2). This difference would likely have been insignificant if supratidal saltmarshes had been grouped together with intertidal saltmarshes under a generic ‘saltmarsh’ category. Previous research supports this effect of species zonation and vegetation structure on soil organic carbon stocks (Hayes et al. 2017; Owers et al. 2020). In southern Australian coastal wetlands, species composition was shown to be the most reliable indicator of soil organic blue carbon stocks down to a depth of 30 cm in tidal marshes, mangroves and seagrass meadows (Ewers Lewis et al. 2020). In mangrove forests with multiple species, community composition has also been associated with differences in soil organic carbon stocks (Atwood et al. 2017), and seagrass meadows have been shown to have an 18-fold difference in soil carbon stocks between different dominant species (Lavery et al. 2013).

The vegetation types observed at our sample locations (supratidal saltmarsh, intertidal saltmarsh and mangrove) were compared to the state-wide saltmarsh and mangrove habitat mapping spatial layer (DEW 1997). If we had relied on intersecting our sampling coordinates with the state’s digitised mapping layer, rather than using in situ observations, we would have classified approximately half of our samples incorrectly. Most errors in classification occurred near the border between two vegetation types, making it difficult to determine if inaccuracies are due to migrating ecosystems, or are within the margin of error in the original spatial data that the mapped classifications are based on. However, some sites showed clear changes in vegetation compared to the maps, where mangroves have migrated inland and become dominant in areas that were previously intertidal saltmarsh, and supratidal saltmarshes have grown in previously supratidal flat bare areas. This highlights a need for up-to-date and accurate mapping of coastal vegetated ecosystems, especially if spatial data layers are to be used in predictive models of soil organic carbon storage in coastal areas, or site selection and prioritisation for coastal restoration and management activities. Depending on the quality and age of coastal vegetation maps, site-level data may be required if carbon abatement is a key restoration, conservation or management aim. Ewers Lewis et al. (2020) proposed that calculating stocks for saltmarsh as an ecosystem is sufficient in data-deficient areas. However, we propose that in semi-arid regions where ecosystem classification level, but not species level, spatial data are available, elevation could potentially be used to separate intertidal from supratidal saltmarshes and increase the accuracy of soil organic carbon stock estimates. Future research could identify appropriate elevation ranges to stratify saltmarshes and determine regions where this approach could be suitable.

Soil organic carbon data is often reported to a depth of 1 m in accordance with the requirements of some carbon accounting and crediting schemes (Howard et al. 2014). However, these data have often been extrapolated from shorter soil cores. Sampling to 1 m can be physically difficult to achieve in some coastal areas due to shell grit and calcrete layers, and as this study focused on linking soil organic carbon stocks to current/recent environmental conditions and existing vegetation communities, we focused sampling on the top 50 cm (Ewers Lewis et al. 2017). A recent review of sediment carbon sampling methods in mangroves by Fest et al. (2022) found that the average sampling depth was 51 cm, despite most published soil stock values being given to 1 m based on interpolation from mean values. We did not feel that extrapolating to 1 m was robust, as organic carbon percentage and bulk density vary significantly with depth (Sanderman et al. 2018, Fest et al. 2022, Walden et al. 2023). As blue carbon science is developing rapidly, groups like the Coastal Carbon Research Coordination Network (CCRCN) are working towards internationally standardising sampling protocols and data reporting. Our data reporting (providing all measured parameters for each depth increment from each core) is in line with current recommendations and will allow for collation and comparison with other datasets using specific depth horizons (Sanderman et al. 2018, Fest et al. 2022).

Effect of Fine-Scale Drivers on Soil Organic Carbon Stocks

Soil organic carbon stocks are driven by multiple interacting environmental variables, acting at different spatial scales. On a global scale, much of the variation in soil organic carbon stocks can broadly be attributed to climatic conditions (Jardine and Siikamäki 2014; Osland et al. 2017; Rovai et al. 2018). Temperature and rainfall have been shown to have a strong positive correlation with carbon stocks in global models, due to contributing to both an increase in species diversity, plant productivity and soil saturation, which acts to slow the decomposition of organic material (Jardine and Siikamäki 2014; Sanders et al. 2016; Feher et al. 2017). These environmental variables are useful predictors for global, national and regional patterns in soil organic carbon stocks, though at finer spatial scales they are unlikely to vary enough to be useful in explaining the variability of stocks (Owers et al. 2020).

Our results showed that the random effect in our model, ‘site’, had a significant effect on soil organic carbon stocks. This suggests that some broader scale factors, not explicitly included in our modelling (but captured by the random effect for site), are influencing soil organic carbon stocks. These differences could be driven by regional variability in coastal geomorphology, as the shape and evolution of a coastal landscape can explain hydrology, sediment type and potential sources and supply of organic carbon. Coastal geomorphologies can broadly be classified as deltas, tidal systems, lagoons and fjords, with tidal systems, such as the ones we sampled, having the highest soil organic carbon stocks due to increased sediment supply (Hayes et al. 2017; Rovai et al. 2018; Twilley et al. 2018).

Distance to tidal creek incorporates information about several important factors for soil organic carbon storage such as elevation and hydrology, making it a useful predictor which can be viewed as a proxy for multiple environmental factors. Our results show soil organic carbon stocks decreased as distance to tidal creek increased (Fig. 3), which is expected for several reasons. Tidal creeks supply particulate matter that contributes to sediment (and carbon) accumulation, as well as being an indication that the water table is closer to the surface creating a more anoxic environment, which slows the decomposition of organic material (Chmura et al. 2003; Chmura and Hung 2004; Owers et al. 2020). Owers et al. (2020) found that the interaction between elevation and distance to water was driving the distribution of vegetation in saline coastal wetlands. Our results show proximity to a tidal creek can indicate areas likely to have higher soil organic carbon stocks; however, varying soil types throughout the ecosystem can alter the magnitude of this effect.

Soil type is indicative of properties that control the breakdown and retention of organic carbon in the soil matrix. Grain size dictates pore size which affects oxygen exchange in the soil and has been shown to be a key predictor of soil organic carbon (Kelleway et al. 2016). In our modelling, the soil type ‘loam sand’ was associated with the lowest carbon stocks followed by ‘clay sand’ and ‘clay loam’. These three soil types correlated with lower soil organic carbon stocks are the larger grained soils, associated with greater oxygen penetration and more aerobic conditions, resulting in increased decomposition of organic materials (Santín et al. 2008). The soil type ‘clay peat’, which has a high proportion of organic matter and was the finest soil grain size in our study, was associated with the highest soil organic carbon stocks. Greater soil organic carbon stocks are generally associated with finer soil texture and high fractions of clay in surface sediments (Zhou et al. 1994; Santín et al. 2008; Ward et al. 2021). Finer soil particles have a larger surface area for organic carbon to bind to, which affects soil organic carbon storage potential (Keil and Hedges 1993). The soil types found at our study sites were classified as the dominant soil type down the 50-cm deep soil cores, rather than surface soil texture. Future research could investigate whether surface soil texture is indicative of depth averaged soil type, and could therefore be used as a proxy, as it is easier and quicker to record in the field.

Vegetation type is an important local-scale predictor of soil organic carbon stocks in tidal wetlands. Mangroves and saltmarshes have different elevation and tidal ranges, with mangroves, on average, found at lower elevations with more frequent and sustained tidal inundation (Leong et al. 2018). The primary source of organic carbon generation in blue carbon ecosystems is the current in situ vegetation, as plant material from the system makes up the majority of organic carbon being incorporated into the soils (Owers et al. 2020). Blue carbon stocks are often calculated by obtaining a single estimate of the average carbon stock value for a particular ecosystem and multiplying it by the area of that ecosystem type (e.g. ‘mangrove’ or ‘saltmarsh’ (Howard et al. 2014). Though our results support the use of vegetation type as a predictor of soil organic carbon stocks, this approach fails to capture variability within broad vegetation categories such as ‘saltmarsh’ (as discussed above) and can lead to error in estimates of soil organic carbon stocks (Owers et al. 2016).

With growing interest in the potential for blue carbon ecosystems to be managed for greater carbon sequestration benefits and the protection of stored carbon, it is important that an accurate, cost-effective method of estimating site-based stocks can be implemented. Restoration and conservation efforts that aim to increase carbon sequestration in tidal wetlands require accurate estimates of soil organic carbon stocks at specific locations, so that additionality can be reliably measured (Duarte et al. 2013; Thomas 2014; Hilmi et al. 2021). Identifying sites with significant soil organic carbon stocks is also crucial to informing management decisions and avoiding potential carbon emissions from habitat degradation (Lovelock et al. 2017). In general, the finer-scale, site-based drivers of variability in soil organic blue carbon are poorly understood, with many studies focusing on global- or national-scale carbon predictions and accounting (Jardine and Siikamäki 2014; Atwood et al. 2017; Sanderman et al. 2018; Kelleway et al. 2020; Young et al. 2021; Chen and Lee 2022). Our results suggest that local-scale studies attempting to predict soil organic blue carbon need to include some regional-scale factors such as geomorphology and oceanography, in conjunction with finer-scale variables such as soil type, vegetation community, hydrology and tidal inundation. Understanding the fine-scale drivers of soil organic carbon variability in combination with broad-scale drivers will allow for more accurate local-scale predictions of stocks, which is crucial for accurate carbon accounting and the effective implementation of carbon crediting schemes (Vanderklift et al. 2019).

Conclusions

This study has determined vegetation type, soil type and distance to tidal creek as the primary drivers of fine-scale variability in soil organic carbon stocks and has contributed significantly to the soil organic carbon stock dataset for under-sampled, semi-arid blue carbon ecosystems. We established an important distinction between the contributions of intertidal and supratidal saltmarshes to soil organic carbon stocks and highlighted the need for future blue carbon estimation to consider these as distinct ecosystems. We have established that soil organic carbon stocks in Australian semi-arid regions are, on average, lower than those in temperate or tropical regions, suggesting that current state- and national-scale estimates may overestimate blue carbon stocks in semi-arid regions. The results can also support improvements in blue carbon abatement estimates more broadly, by building understanding around the considerable influence of site-based environmental variability on soil organic blue carbon stocks. The study should encourage greater consideration of blue carbon restoration site selection, which may lead to improved carbon abatement outcomes.