potential while also degrading soil C [7][8][9] . Additionally, SLR can lead to large C losses within the coastal zone by driving ecosystem transgression (for example, forest retreat, which prompts substantial aboveground biomass loss 10,11 and/or by driving erosion of C-rich sediments when exposed along open-ocean coasts 12,13 ). Thus, coastal landscapes facing the combined threats of SLR and erosion risk a blue carbon stock that is both diminished and more fleeting.

Barrier-island beach and dune systems protect the C-rich sediments of backbarrier marsh from wave erosion along many coasts globally and can supply sediments to fringing backbarrier marsh during high-energy events [14][15][16] , processes that support lateral and vertical resilience to SLR, respectively. However, this supportive function of barrier islands is jeopardized by SLR, which, compounded with intensifying coastal storms and sediment deprivation, forces oceanside barrier shorelines to transgress (through island narrowing via erosion and/or wholesale landward migration) at accelerating rates 17,18 . Soil C stocks previously protected by barrier islands are eventually exposed and subjected to high-energy, open-ocean processes, possibly shifting transgressive barrier-island systems from C sinks to C sources 12 .

Across the coastal landscape, the magnitude of the net C sink depends on the balance 19 between C loss due to erosion or drowning, and C accumulation in ecosystems migrating and/or accreting apace with SLR 2,6,11,20 . However, these landscape-scale C budgets typically focus on the evolution only of vegetated ecosystems, and assume shallow depths of erosion, as is common in protected environments. In contrast, wave action along open-ocean shorefaces can rework sediments well below mean sea level, exposing to erosion not only surficial salt marsh peat, but also far deeper sedimentary deposits. Failure to account for these processes may lead to large overestimates of C storage in coastal ecosystems.

Here, we combine geospatial data of barrier island retreat rates, organic carbon (OC) accumulation rates within backbarrier marsh soils and seagrass and lagoon sediments, and the OC content of eroding sedimentary facies to develop a regional-scale OC budget for the rapidly transgressing Virginia Atlantic coast (USA). Sedimentologic and geochemical analyses of 10 new sediment cores (each 3-19 m long) together with additional published stratigraphic data were used to determine facies-specific thicknesses, OC densities, and OC erosion rates (Fig. 1; equation 1). We find that buried lagoon sediments associated with unvegetated environments contribute the vast majority (> 80%) of OC eroded on the beach and shoreface of transgressing barrier islands. Moreover, we find that erosion of these deep deposits leads to rates of OC loss that exceed annual OC accumulation summed across the entire backbarrier environment, despite the well-known capacity of blue carbon ecosystems to sequester OC.

The largely undeveloped and rapidly transgressing Virginia Barrier Islands (VBI) are located in the mid-Atlantic SLR hotspot 21 and generally characterized by either wholesale landward migration or rotation of formerly progradational islands 22 (Fig. 1a). Stratigraphic and OC analyses reveal that those islands which are migrating landward are characterized by thin (< 2 m thick) sandy beach and dune deposits 22 perched atop discontinuous, thin (~0.9 m) marsh peat and thick (~6.6 m) lagoon deposits (Fig. 1c). In contrast, former backbarrier peats associated with historically progradational islands (Parramore, Hog) were long-ago eroded as those islands migrated to their landward-most positions, leaving only thinner (0.75-6.25 m) remnant lagoon deposits preserved under relatively thick (~4.5 m) barrier sands 22 . Averaged across the seven migrating islands, the beachface-exposed marsh is 0.9 m thick (ranging from 0.6 [Smith] to 1.3 m [Assawoman]) and characterized by a relatively homogenous mixture of marsh roots and silt-or clay-dominant minerogenic sediment with an average OC density of 26.8 kg OC m -3 (ranging from 23.3 [Smith] to 31.5 kg OC m -3 [Cobb]; Fig. 1c; Table 1). In contrast, lagoon deposits consist of a complex set of facies ranging from clay to medium sand, predominantly very dark greenish grey in color, with frequent shell fragments. Across all ten islands, average lagoon deposit thickness is 6.0 m (varying between 3.5 [Parramore] to 8.5 m [Wreck]) and average lagoon OC density is 7.6 kg OC m -3 (ranging from 5.3 [Smith] to 10.1 kg OC m -3 [Cobb]; Fig. 1c). Sandy units interbedded within lagoon complexes average 0.8 m of very fine to very coarse sand (ranging from 0.0 [Assawoman, Cobb, Myrtle] to 1.6 m [Metompkin and Cedar]). We estimate that 38.8 km 2 of backbarrier marsh was buried and re-exposed by island migration along the island chain from northern Assawoman to southern Smith between 1870 and 2017 C.E., at a system-wide rate averaging 0.26 km 2 per year.

Although marsh peat is widely recognized for its large blue carbon stores 1 , we find that lagoon facies thickness is the single largest driver of shoreline-normalized OC erosion rates (equation [1]), accounting for 85% of variability (P < 0.001; Fig. 2a). Shoreline-change rate accounts for approximately half of the variability in OC erosion rates (R 2 = 0.52; P = 0.02; Fig. 2b). In contrast, neither the rate of long-term average marsh exposure (P = 0.14; Fig. 2c) nor marsh or lagoon OC densities (Supplementary Fig. 1) have a significant effect on OC erosion rates.

Applying new multi-decadal and island-specific shoreline-change rates, marsh-exposure rates, and island shoreline lengths to equation (1) (Supplementary Tables 123), we find that beach/shoreface OC erosion has accelerated over shorter time periods (Fig. 3), reaching an annual average rate of 42.9 ± 10.0 Gg OC yr -1 between 1994 and 2017. This is more than 125% greater than the average annual OC accumulation for the entire VBI backbarrier-including OC accumulated in marsh, seagrass, and lagoon soil/sediment-over a similar time period (33.8 ± 6.0 Gg OC yr -1 ;1984-2020 C.E.) [23][24][25] (Fig. 3; Supplementary Table 4).

Carbon budgets that cross traditional ecosystem boundaries are crucial for establishing the degree to which coastal landscapes can mitigate climate change through C sequestration 11 .

Recent studies demonstrate that ecosystem transitions associated with SLR (for example, conversion of forest to marsh or of marsh to open water) lead to shifts in magnitudes and loci of C burial and C loss 2,6,11,20 . However, such landscape-scale C budgets typically focus on vegetated ecosystems and include C loss due to marsh submergence or erosion only to a depth of 1 m 20,26,27 . Thus, widely-used protocols for assessing vulnerability of C stocks often overlook sediment C accumulation in unvegetated systems as well as C loss due to deeper erosion of nonvegetated facies. Here, by extending the landscape C budget to include sites of sediment/soil OC accumulation and erosion that traditionally have been ignored, we find that backbarrier lagoon and tidal-flat sediments contribute >80% of the total annual OC eroded in the VBI system (Fig. 3). Thus, incorporating these sediments into OC flux estimates not only magnifies the OC erosion term in our budget but also challenges previous understandings of the role deep, unvegetated sediments play in the coastal OC sink. Organic C capture in vegetated ecosystems has been the paradigm of coastal OC research since the term 'blue carbon' was first coined in the early 2000s 1,7,28 . However, emerging evidence demonstrates that non-vegetated and subtidal coastal environments can contain substantial OC stocks 29,30 , fed by the deposition of particulate matter (for example, organic matter from nearby erosion of vegetated systems or from productivity within the overlying water column) and in situ microphytobenthic productivity 31,32 , as has been shown for the VBI lagoons 33 . We find that, despite hosting OC densities that are approximately one-third of that of the marsh (Fig. 1c), the thickness of lagoon deposits is a more important driver of OC erosion fluxes than factors that commonly garner more attention, such as marsh OC density, marsh thickness, or marsh erosion rate (Figs. 2a and 2c; Supplementary Fig. 1). This aligns with emerging evidence that unvegetated coastal areas are important components of the coastal OC budget, and can, depending on their areal extent and thickness, account for more total OC storage than vegetated areas within the same landscape. In fact, we find that just the average annual erosion of lagoon OC (33.4 ± 9.8 Gg OC yr -1 ; 1994-2017 C.E. could negate the OC accumulated annually in the entire backbarrier averaged over a similar time period (33.8 ± 6.0 Gg OC yr -1 ; 1984-2020 C.E.) (Fig. 3).

The disproportionately high rates of OC burial in coastal ecosystems 1 leave large pools of OC subject to destabilization following rapid SLR and commensurate wetland drowning, forest dieback, and/or enhanced erosion 2,7,10 . Previous work by ref. 12 considered an additional consequence of SLR on OC storage (that is, transgression of barrier islands) and found that erosion of outcropping salt marsh along barrier-island beach and shorefaces can flip the system from a C sink to a C source. Likewise, our quantification of the most recent (1994-2017 C.E.) rate of annual OC erosion along the VBI shoreface is approximately 1.3 times the rate of OC accumulation across the entire VBI backbarrier over a similar time period [23][24][25] (Fig. 3; Supplementary Table 4). Including only marsh soil OC in these budgets would erroneously suggest that the VBI remains a strong sink for OC, netting an average 11.5 Gg OC yr -1 over the past two decades (Fig. 3; Supplementary Information). Like other landscape-scale carbon budgets 6,12,20,23 , our work assumes that eroded carbon represents a source of carbon to the atmosphere or to non-coastal ecosystems. However, fully classifying the VBI chain as a net OC source would require tracking the fate of this shoreface-eroded OC, which may include remineralization, offshore burial, or possibly transport and redistribution to the backbarrier through tidal inlets. Nevertheless, the imbalance we measure between annual rates of backbarrier OC accumulation and shoreface OC erosion implies that, at the very least, barrier-island transgression results in a coastal OC sink that is far more tenuous than commonly assumed.

Blue C storage dynamics have traditionally been considered a negative climate feedback, whereby SLR drives enhanced soil OC accumulation in coastal ecosystems like salt marshes 2- 4,6,9,34 . For the VBI, we find that an increase in the rate of island transgression by only 1 m yr -1 intensifies OC erosion by approximately 73 kg OC m -1 yr -1 (Fig. 2b). Thus, our results confuscate the current understanding of coastal OC processes by suggesting that dynamics along open-ocean coasts can constitute a positive climate feedback. Given newly uncovered multidecadal lags in barrier response to SLR 18 , our findings suggest that OC erosion along migrating barrier islands will continue to accelerate as island movement equilibrates to modern (and even faster, future) rates of SLR. Narrowly focusing on OC gains and losses within the top meter of vegetated environments underestimates the OC potentially eroded from deeper and unvegetated ecosystems, especially within dynamic coastal systems. Therefore, landscape-scale OC budgets based on the evolution of shallow, vegetated environments may obscure the potential for coastal landscapes to switch from net C sinks to C sources, a threshold which the VBI may already have crossed. Regardless of the magnitudes and sites of OC accumulation and erosion, our findings demonstrate that, for systems in which barrier islands are free to move landward, blue carbon stored in wetland and thick lagoon sediments is largely ephemeral.

The Virginia Barrier Islands (VBI) comprise a 110-km-long chain of 12 mixed-energy islands backed by salt marsh and shallow lagoons along the US Mid-Atlantic Coast (Fig. 1a).

The absence of artificial shoreline stabilization along all but Wallops Island allows most to erode and/or migrate landward in response to storms and SLR, which they do at an average rate of 4.35 m yr -1 (1851-2017) 18 . Excluding net-progradational Fisherman's Island (located at the southern longshore depocenter at the mouth of Chesapeake Bay), individual island shorelines transgress at rates between 3.1 m yr -1 (Cobb) and 7.5 m yr -1 (Ship Shoal) 18 (Fig. 1a). This process exposes expansive marsh deposits along the seaward side of many of these islands (Fig. 1b) and, visible at very low tide, lagoon deposits along the marsh periphery or directly under barrier sands.

Nine vibracores (each 3-9 m long) and one GeoProbe core (19 m long) collected from across seven islands (Fig. 1a) were analyzed for organic-matter (OM) content via loss-onignition (LOI) and grain size, and a subset for total organic carbon (TOC) content (Supplementary Information). We apply the resulting marsh-and lagoon-specific conversion factors (Supplementary Fig. 2) to approximate OC content based on OM values for all downcore samples.

Contact between the marsh and lagoon unit, as well as the base of the Holocene barriersystem were determined according to sediment texture, mineralogy, and OM content, in keeping with the unit descriptions of ref. 35,36 . We estimated OC erosion rates (g OC yr -1 ) associated with loss of both marsh and lagoon deposits for each island as: where, following ref. 12 , we apply island-average OC densities, OCx (g OC m -3 ), to the islandaverage thicknesses, Tx (m), of the marsh and lagoon units based on new and published cores 35,[37][38][39][40] (Fig. 1c; Table 1; Supplementary Table 5). Unlike ref. 12 , however, we account for lagoon sediment OC in our erosion terms, quantifying a maximum blue carbon loss term for erosion of the entire Holocene unit. Except where replaced by inlet fills, lagoon deposits ubiquitously underlie both transgressive and progradational islands within the VBI chain 35,36,39,41 .

Thus, lagoon sediment volume loss is approximated by multiplying the shoreline length, Lshoreline (m) (Supplementary Table 1), by the island-specific shoreline-change rate, SCR (m yr -1 ) (Supplementary Table 2). In contrast, beach/shoreface marsh erosion is confined to discontinuous portions of migrating islands. Following ref. 42 , we used the earliest-mapped backbarrier marsh extent and overlaid successive island positions up to 2017 C.E. to calculate a time-averaged annual marsh exposure rate due to island transgression, ERmarsh (m 2 yr -1 ) (Supplementary Table 3). We used Digital Shoreline Analysis System (DSAS) 43 to calculate shoreline positions at 50-m spaced transects along the length of the VBI to calculate both longterm (1870-2017) and short-term (1870-1942; 1942-1994; 1994-2017) shoreline-change rates, SCR. System-wide rates are valued as the sum of component islands.