Global blue carbon losses from salt marshes exceed restoration gains

Abstract

Salt marsh soil organic carbon (SOC) is a key blue carbon pool affected by both disturbance and restoration; yet its long-term global dynamics remains poorly understood. Here we provide the global assessment of surface SOC changes in salt marshes from 2002 to 2019, combining multi-source remote sensing imagery with the machine learning calibrated by field observations. We find a net global SOC loss of 0.52 million tonnes, primarily driven by declines in North America and Oceania, which are only partially offset by gains in Asia and South America. The United States alone accounts for ~60% of the global loss, equivalent to 6.2 million tonnes of CO₂ if fully released. Losses are concentrated in mature salt marshes with large SOC storage, while gains occur primarily in newly formed salt marshes with relatively low SOC density. These patterns suggest that global restoration efforts are failing to keep pace with degradation. To avert irreversible climate and ecological damage, the protection of mature, carbon-rich salt marshes must become a core component of global climate strategies.

Introduction

Salt marshes are highly efficient carbon sinks and a key component of blue carbon ecosystems, sequestrating and storing carbon in coastal environments^1,2. The interplay of high sedimentation rates, high productivity, and anoxic respiration in salt marshes results in carbon burial rates approximately 40 times greater than those observed in terrestrial forests³. Despite covering less than 10% of the global land area, salt marshes contribute disproportionally to carbon storage, with an estimated capacity of approximately 1.84 Pg of carbon, accounting for approximately one-third of the global carbon stored in soil⁴. However, over the past few decades, climate change and intensified human activities have led to a net loss of more than 50% of the historical global salt marsh area^3,5,6. In response, the Blue Carbon Initiative (mitigating climate change through coastal conservation), which focuses on the protection and restoration of these ecosystems^2,7, has been proposed, and considerable recovery and ecosystem expansion have occurred as a result (~9700 km² globally⁸).

Changes in salt marsh areas lead to notable alterations in productivity and soil organic carbon (SOC)^5,9, which accounts for 49% to 98% of the total carbon stocks in salt marshes¹⁰. In newly formed (natural expansion) or restored salt marshes, where salt marsh vegetation has been recently established, SOC is accumulated in surface soils and gradually buried into deeper layers via sediment deposition^1,11. Conversely, in degraded or destroyed salt marshes, surface SOC is the first soil carbon pool to be impacted, with carbon lost eventually being released into the atmosphere or transported to the ocean via tidal inundation, tidal currents, and other hydrodynamic processes^1,12. Compared with SOC stored in deep layers, recent changes in surface layers (net changes incorporating carbon losses and gains) more accurately reflect the impact of restoration and protection efforts on net SOC dynamics¹³. Therefore, the SOC discussed in this study refers specifically to surface SOC.

The net outcome of surface SOC cycling in salt marshes is governed not only by changes in lateral marsh extent but also by variations in SOC density, defined as carbon stock per unit area within a specific soil depth, which could be influenced by vertical processes such as sediment accretion/erosion¹⁴. These two dimensions jointly determine overall SOC dynamics¹⁵. However, most previous studies only provide static snapshots (such as the salt marsh area or SOC storage for a certain year) and lack the temporal resolution required to capture SOC dynamics^6,8,16. In particular, global assessments of net SOC change, which reflect both losses and gains over time, remain limited¹⁷. Furthermore, substantial regional heterogeneity in ecological conditions and conservation policies^18,19 likely leads to divergent spatiotemporal SOC trajectories^17,20. This knowledge gap hampers our ability to evaluate the effectiveness of restoration efforts and the resilience of existing carbon stocks.

To address this, we presented the global, long-term assessment of surface SOC change in salt marshes, integrating spatial changes in salt marsh area with SOC density variation. By coupling multi-source remote sensing images with machine learning algorithms, we detected salt marsh gains and losses from 2002 to 2019 using a global salt marsh extent dataset⁸, and classified land-use and land-cover changes (LULCCs) via the Google Earth Engine (GEE) platform to determine the proximate causes of salt marsh transformations. For each dynamic pixel, we estimated SOC density using a remote sensing-based machine learning model trained on multitemporal Landsat surface reflectance (SR) imagery and calibrated against field-measured SOC samples. After that, the model, optimized through multi-algorithm testing, was deployed across all dynamic pixels using GEE. Finally, by overlaying predicted SOC density with mapped LULCCs, we quantified net SOC gains and losses between 2002–2019 globally. This integrated framework enabled spatially explicit, high-resolution tracking of SOC dynamics, and revealed the dual roles of land cover transitions and SOC density variation in shaping global salt marsh SOC trajectories. Our findings emphasize the urgency of protecting high-SOC-density marshes and provide a robust evidence to inform future conservation and restoration strategies under accelerating climate and anthropogenic pressures.

Results

Global patterns and trends in salt marsh soil organic carbon changes (2002–2019)

Our study revealed a net loss of approximately 0.52 million tonnes of surface SOC (0–20 cm) from global salt marshes between 2002 and 2019 (Fig. 1), whereas the changes in salt marsh distribution led to an increase in surface SOC of 0.61 million tonnes from 2011–2016. Geographically, distinct regional patterns emerged (Figs. 1a, b and 2), with North America and Oceania being the primary sources of surface SOC loss, whereas Asia and South America presented net surface SOC gains. Salt marshes in North America, particularly the USA, contributed the largest share to global surface SOC losses (~1.7 million tonnes; Figs. 1a, b and 2 and Supplementary Table 1), accounting for 60% of the global total. The greatest losses in USA ( ~ 0.97 million tonnes) occurred from 2005–2007 (Fig. 1f, Supplementary Fig. 1 and Supplementary Table 2). Oceania, despite being a region with notable SOC losses, experienced substantially lower SOC losses from salt marshes (~0.18 million tonnes) than North America did, primarily from 2002 to 2004. Australia was the dominant contributor to these losses, whereas New Zealand ranked among the top 10 SOC loss countries globally (Supplementary Table 2 and Supplementary Fig. 1).

Fig. 1: Temporal trends in salt marsh surface soil organic carbon (SOC) across different scales (2002–2019).

a Net surface SOC in countries with salt marsh changes, b surface SOC changes at the continental scale, c–f trends of surface SOC changes in major countries. The “Global Average” category in b includes six continents where salt marshes are distributed (middle to high latitudes): Africa, Asia, Europe, North America, Oceania, and South America, excluding Antarctica. Notable fluctuations in surface SOC occurred in North America, Asia, Europe, and Oceania, prompting the selection of four countries with large salt marsh areas: Australia (Oceania), China (Asia), Romania (Europe), and the United States of America (North America) to represent SOC trends in these regions.

Fig. 2: Surface soil organic carbon (SOC) changes in regions with salt marsh losses and gains from 2002–2019.

Surface SOC dynamics are shown for regions where salt marshes either degraded or expanded during the study period (Aggregated per 1° grid cell). Surface SOC losses are represented in regions where salt marshes were lost or degraded, whereas surface SOC gains correspond to regions where salt marshes were restored or expanded. Positive values (surface SOC gains) are indicated by the blue colorbar, while negative values (surface SOC losses) are represented by the red colorbar.

Conversely, Asia gained a total of 0.94 million tonnes of surface SOC, with an increase from 0.07 million tonnes of SOC between 2002 and 2004 to 0.22 million tonnes of SOC during 2017–2019 (Fig. 1b and Supplementary Table 1). China played a pivotal role in this SOC gain, 0.73 million tonnes of SOC in salt marshes accumulated with net SOC gains across all periods, followed by South Korea (0.11 million tonnes) and Bangladesh (0.97 million tonnes, Supplementary Table 2). However, salt marshes in Japan recorded a net SOC loss of 0.07 million tonnes, ranking fourth globally in terms of SOC losses, with persistent declines over time (Supplementary Table 2). Despite having the largest SOC losses in Asia, Japan had smaller losses than the USA and Australia. In contrast, South America presented continuous SOC gains, totaling 0.18 million tonnes (Fig. 1b and Supplementary Table 1). Salt marshes in Brazil, Uruguay, and Argentina were the primary contributors to these gains, with 89,120, 47,870, and 33,050 tonnes of surface SOC gains, respectively (Supplementary Table 2); these regions included the most widely distributed salt marshes in South America.

Salt marshes in Europe presented alternating surface SOC gains and losses, with a net gain of 0.33 million tonnes from 2002–2016, but losses of 0.12 million tonnes from 2017–2019 (Supplementary Table 1). Russia, which had the largest area of salt marshes in Europe, maintained a balance between surface SOC gains and losses. Salt marshes in Romania recorded the highest surface SOC gains among European countries and ranked second globally. However, most other European nations, including Sweden, Germany, Latvia, Morocco, and Denmark, experienced net SOC losses from salt marshes. In Africa, salt marshes experienced net surface SOC losses from 2002 to 2010, followed by gains from 2011 to 2019. Total SOC gains (36,090 tonnes) notably exceeded losses (7962 tonnes) in salt marshes (Fig. 1b and Supplementary Table 1), resulting in a net surface SOC gain in Africa.

Land use change drives surface soil organic carbon losses in global salt marshes

The dynamics of global surface SOC and salt marsh areas were accompanied by multiple LULCCs, including transitions to and from built-up areas, mudflats, aquaculture areas, mangroves, agricultural lands, and seawater intrusion/shoreline accretion areas. In this study, we found that the losses in global salt marsh areas were due primarily to shifts to mudflats (37%) and aquaculture areas (25%), whereas the gains in salt marsh areas resulted mainly from the conversion of aquaculture areas (35%) and mudflats (27%). The most significant losses in salt marsh areas occurred between 2002 and 2007, with 27% converted to agricultural lands (Fig. 3c). From 2008 to 2016, the reduction was smaller, with a focus on mudflats, aquaculture areas, and agricultural lands. Notably, between 2014 and 2016, salt marsh areas markedly increased. From 2011 to 2013, the increase was due mainly to the conversion of mudflats, whereas in other periods, it was due primarily to the conversion of aquaculture areas. After 2017, the reduction in salt marsh area intensified, primarily due to conversion to mudflats and aquaculture areas.

Fig. 3: Land use and land cover changes in salt marshes and related surface soil organic carbon (SOC) dynamics.

a Salt marshes lost to other land use types. b Salt marshes gained from other land use types. c SOC changes across land use types from 2002 to 2019. d Proportion of salt marshes lost to different land use types in each continent. e Proportion of salt marshes gained from different land use types in each continent. Lines in a and b indicate transition between salt marshes and other land-use types, with line width representing the proportional area of each transition. Black boxes in c indicate net SOC dynamics: a downward box with a negative value indicates that SOC losses from the conversion of salt marshes to a particular land use type exceeded the SOC gains from converting that land use type to salt marsh, contributing to a net SOC loss during this period.

Specifically, Oceania and Africa experienced conversions of mudflats and agricultural lands, whereas North America underwent changes (both salt marshes losses and gains) involving mudflats and aquaculture areas (Fig. 3d, e and Supplementary Fig. 2). In South America, mudflats and seawater intrusion/shoreline accretion areas played an important role, with more than 80% of LULCCs occurring between these areas and salt marshes (Fig. 3d, e). In Europe and Asia, the conversion of built-up and seawater intrusion/shoreline accretion areas into salt marshes outpaced the reverse trend (Fig. 3d, e and Supplementary Fig. 2), suggesting both inland and seaward migration, influenced primarily by coastal accretion and sediment deposition.

Conversely, the transformation of aquaculture areas into salt marshes contributed the most to net SOC gains globally, particularly between 2008 and 2010, and again from 2017 to 2019 (Supplementary Fig. 3). This transformation occurred mainly in Asia and Europe (Supplementary Fig. 2), largely driven by wetland restoration efforts, including the reconversion of reclaimed lands back to salt marshes. Moreover, the conversion of salt marshes to agricultural lands represented the second largest source of SOC loss (Supplementary Fig. 4), which was especially pronounced between 2005 and 2007, during which more than 0.2 million tonnes SOC was lost. After 2017, SOC losses were caused predominantly by the conversion of salt marshes to aquaculture areas. Seawater intrusion and shoreline accretion occurred during each 2-year period, causing both SOC losses and gains depending on whether sea-level rise outpaced sediment deposition or vice versa. From 2005–2007, significant SOC gains were observed due to the seaward expansion of salt marshes, whereas in recent years, SOC losses increased because of the degradation of salt marshes near the coastline caused by seawater intrusion. Transformations between salt marshes and mangroves or built-up areas caused relatively few SOC changes. Although the conversion of salt marshes to mangroves initially led to surface SOC losses, new soil carbon pools formed in mangroves over time, acting as long-term carbon sinks.

Furthermore, a sensitivity analysis accounting for partial surface SOC losses following land-use conversion indicates that, although the absolute magnitude of SOC losses was reduced, the spatial distribution of SOC loss hotspots remained largely unchanged across scenarios (Supplementary Fig. 5). Across regions, the partial release scenario consistently preserved the relative ranking of SOC loss intensity, with North America and parts of Europe remaining dominant loss hotspots, indicating that regional contrasts were robust to assumptions about post-conversion SOC (Supplementary Fig. 5). Temporally, differences between full and partial SOC release scenarios remained consistent across all periods, indicating that interannual variability in SOC loss was primarily driven by the timing and type of LULCCs rather than uncertainty in post-conversion SOC losses magnitude (Supplementary Fig. 5).

SOC density impacts surface SOC losses and gains in global salt marshes

While SOC density is not a direct causal driver of surface SOC losses, it serves as a robust proxy for salt marsh age and the extent of long-term SOC storage. High-density areas typically represent mature salt marshes that have accumulated substantial carbon over decadal to centennial timescales. Consequently, these are more susceptible to substantial carbon losses when subjected to disturbance or conversion. Our results (Fig. 4) indicated that the surface SOC density of newly formed salt marshes (salt marsh gains) globally ranged from 700 to 26,400 g/m² (Fig. 4a and Supplementary Fig. 6), which was much lower than that of mature salt marshes (salt marsh losses, Fig. 4b and Supplementary Fig. 7). Our observations showed high SOC densities concentrated along Russia’s eastern coastline and Canada’s Hudson Bay. Notably, newly formed salt marshes along the USA and European coasts exhibited relatively uniform SOC densities (3000–11,000 g/m²), whereas those in China averaged below 3000 g/m² (except during 2008–2010). A comparison of the patterns of SOC densities in these regions revealed that the surface SOC density in North America was substantially greater than that in Asia. Moreover, the SOC densities in the areas of both losses and gains in Asian salt marshes were similar, whereas in North America, the SOC density was markedly greater in areas with salt marsh losses than in areas with salt marsh gains, resulting in considerable SOC losses. In North America, particularly along the USA East Coast, the loss of high-SOC-density salt marshes contributed to large surface SOC losses. Moreover, in Russia, the SOC density of degraded salt marshes closely resembled that of newly formed salt marshes, leading to smaller net changes in SOC accumulation (Fig. 2). However, between 2014 and 2019, the loss of salt marshes with high surface SOC density led to net SOC loss in Russia. In China, salt marsh areas expanded, particularly after 2005, but degraded salt marshes tended to have lower SOC density than newly restored areas did. Consequently, changes in China’s salt marshes have contributed mainly to SOC gains.

Fig. 4: Surface soil organic carbon (SOC) density distributions in regions with salt marsh gains and losses.

a Surface SOC gains are represented in pixels where salt marshes were restored or expanded, whereas b surface SOC losses correspond to pixels where salt marshes were lost or degraded. Color classes (blue, yellow, orange, red) represent relatively low, moderate, and high SOC density levels, respectively, defined based on the empirical distribution of SOC density values to enhance visualization and contrast. These thresholds reflect relative differences within the dataset rather than absolute or standardized SOC density classifications.

Discussion

Drivers affecting surface soil organic carbon changes in global salt marshes

This study provides a global assessment of spatial and temporal changes in salt marsh SOC density, offering improved estimates of SOC dynamics in systems undergoing major transitions due to restoration, degradation, or LULCCs. The surface SOC changes exhibited patterns similar to those of the changes in the salt marsh area (Fig. 1b and Supplementary Fig. 8). Global salt marsh areas declined from 2002–2007 but expanded from 2008–2019, with a net gain of ~150 km² from 2014–2016. However, considerable losses in salt marsh area in North America (exceeding 150 km² from 2005–2007) led to substantial SOC declines. Considerable SOC loss in the USA was correlated with the highest rates of salt marsh losses from 2005 to 2009 (−0.35% to −0.46%)^6,21, making it one of the most prominent hotspots of SOC losses from salt marshes globally, consistent with previous studies^6,22. Salt marshes in Asia have acted as carbon sinks over the past two decades. However, Oceania’s gradual losses in the salt marsh area and Europe/Africa’s relative stability contrasted with consistent expansions in the salt marsh area in Asia and South America, which, while beneficial, were insufficient to offset global SOC losses (Fig. 2). The tight coupling between salt marsh area dynamics and SOC fluxes reinforces the necessity of integrating habitat preservation with carbon stock management, while urging future research to disentangle climate-driven vs. human-induced drivers of salt marsh loss and recovery.

In this study, we observed that the most significant LULCCs occurred between salt marshes and aquaculture areas or mudflats (Fig. 3a, b), which aligns with previous regional studies^10,23,24,25, underscoring the critical role of human activities in influencing changes in salt marsh areas and their subsequent effects on SOC dynamics. Despite efforts to control land reclamation, particularly after 2010, aquaculture, an essential economic activity for coastal communities, continued to expand in Europe and Asia. Aquaculture production was concentrated in a few countries, mainly in Asia and Europe, including China, Bangladesh, South Korea, and Norway, accounting for 89.8% of the global aquaculture output²⁶. Moreover, LULCCs exhibited geographic variation, with distinct patterns across different continents and countries. Notably, China, which experienced the greatest SOC gains in salt marshes, has relied heavily on government interventions for salt marsh restoration. Initiatives such as the “Returning farmlands to wetlands” program²⁴ have facilitated the conversion of built-up areas into salt marshes. While large-scale reclamation prior to 2000 led to dramatic salt marsh losses, since 2008, government policies, including the Ecological Redline and Ecological Islands⁷, have curbed reclamation activities and introduced measures to protect and restore these ecosystems. These efforts, along with spontaneous reforestation in Europe²⁷, have substantially supported salt marsh restoration and global SOC gains. These findings highlight the combined influence of human interventions and natural processes in shaping salt marsh dynamics, with region-specific restoration efforts playing a crucial role in supporting SOC storage.

The transformation of salt marshes to various land-use types resulted in distinct patterns of surface SOC losses and gains (Supplementary Figs. 2 and 3). The most notable SOC losses occurred when salt marshes were converted into mudflats, a trend that has persisted over the past two decades. While the conversion of salt marshes to mudflats represents a substantial loss of SOC and coastal protection services, it does not constitute a complete ecological loss. Intertidal mudflats can provide important ecosystem functions, including habitat for benthic invertebrates, foraging grounds for migratory shorebirds, and support for coastal food webs^28,29. Nevertheless, compared with vegetated salt marshes, mudflats generally store substantially less SOC and offer reduced capacity for long-term SOC burial and wave energy dissipation. From a climate mitigation and coastal protection perspective, the widespread conversion of salt marshes to mudflats therefore represents a net loss of key ecosystem services, despite the persistence of certain ecological functions. Besides, although restoration and natural expansion are not explicitly separated in our quantitative analysis and are jointly classified as salt marsh gains, the underlying pathways of salt marshes establishment likely differ among LULCC types. In practice, salt marsh gains occurring on former built-up, agricultural lands, or aquaculture areas are predominantly associated with human-led restoration activities. In contrast, salt marsh gains detected on mudflats or open seawater are more likely driven by natural expansion processes, facilitated by sediment accretion, vegetation restoration, and relative sea-level rise. Due to the lack of globally consistent information on management interventions, these processes cannot be robustly distinguished at the global scale using remote sensing alone. Nevertheless, recognizing their different ecological contexts provides important insight into the heterogeneity of SOC accumulation trajectories within newly established salt marshes.

Spatial variations of surface soil organic carbon changes in global salt marshes and their drivers

North America represents the global hotspot for surface SOC losses in salt marshes, whereas East Asia emerges as the primary hotspot for SOC gains (Figs. 2 and 4), driven by distinct regional environmental and anthropogenic factors. In North America, particularly in the USA, a sharp decline in SOC from 2005–2007 was driven primarily by the conversion of salt marshes to mudflats following major hurricane events, including Hurricanes Rita (June 2005), Katrina (August 2005), and Wilma (October 2005). Although salt marshes effectively dissipate wave energy and storm surges³⁰, hurricanes significantly impact these ecosystems through the introduction of freshwater and deposition of large amounts of sediment, which causes extensive salt marsh erosion and the conversion of marshes to open water³¹. Hurricanes such as Katrina and Rita caused substantial wetland losses (including salt marshes), affecting thousands of kilometers of coastline, with storm surges exceeding 3 m³². Such disturbances are known to often result in the uprooting of marsh vegetation and the disruption of low-salinity marshes, ultimately threatening the ecological integrity and sustainability of salt marsh ecosystems³³. Furthermore, hurricanes caused extensive coastal erosion in regions such as Louisiana and Maryland, exacerbating the damage to salt marshes³⁴.

Similar to hurricanes, storm-induced LULCCs can substantially affect salt marsh area and SOC losses, for example, moderate storms with a return period of ~2.5 months caused the greatest salt marsh deterioration in Plum Island Sound³⁵. However, they are highly context dependent and vary with storm intensity, tidal range, and regional sediment dynamics^35,36. In microtidal systems and during high-magnitude events such as hurricanes, storms can overwhelm the wave-attenuation capacity of salt marshes, leading to rapid lateral erosion, vegetation uprooting, and conversion of vegetated marshes to unvegetated intertidal flats or open water. By contrast, in mesotidal to macrotidal systems, and particularly during less extreme storm events, storms may act as an important source of suspended sediment to marsh platforms, enhancing vertical accretion and potentially increasing salt marsh resilience to relative sea-level rise³⁷. Moreover, recent evidence suggests that in meso- and macrotidal environments, day-to-day hydrodynamic conditions can exert a stronger control on lateral marsh erosion than episodic storm events themselves³⁸. Furthermore, the growing global demand for seafood has led to the continued encroachment of aquaculture upon salt marshes, particularly in developing countries in Asia, further contributing to SOC losses³⁹. Balancing economic development and ecological conservation remains an ongoing challenge that warrants further investigation⁴⁰.

In addition to LULCCs, SOC density (carbon stock per unit area) is also a critical driver of salt marsh surface SOC dynamics. Higher SOC density signifies stronger carbon sequestration capacity, promoting long-term carbon storage, while lower values may indicate degradation or SOC release. Consequently, SOC density directly reflects an ecosystem’s net SOC balance, making it a key indicator for evaluating the ability of salt marshes to sequester carbon over time^41,42. Despite a slight global increase in salt marsh areas between 2008–2010 and 2017–2019, surface SOC experienced a net loss. This counterintuitive outcome reveals that the gains from salt marsh expansions were negated by LULCCs, such as converting carbon-rich salt marshes into low-SOC land uses and by declining SOC density driven by both lateral and vertical erosion processes. These include wave- and current-induced edge erosion, storm- and surge-related sediment removal, and surface soil erosion associated with reduced sediment supply, together with reduced vertical accretion and organic matter decomposition, which collectively contributed to net carbon losses⁶. The surface SOC densities in the degraded salt marshes were similar to those in the newly formed salt marshes, but the regional distributions differed slightly (Supplementary Fig. 7). Our study identified North America as a high-SOC-density region, whereas Asia is relatively low in SOC density. This regional disparity may reflect differences in the restoration age, sediment supply, or hydrological stressors, such as altered tidal regimes, increased storm surge frequency and intensity, freshwater diversion and channelization, and reduced sediment delivery associated with river regulation.

Building on the factors influencing surface SOC density, it is evident that various climatic, environmental, biotic, and societal factors play significant roles. In regions undergoing changes in salt marsh areas, surface SOC density is notably influenced by temperature, precipitation, and tidal forces (Supplementary Fig. 9). In North America, temperature and precipitation were the primary climatic factors affecting SOC density, whereas in Asia, sediment trapping and precipitation played more significant roles (Supplementary Figs. 10–13). Maximum temperature and SOC density were strongly negatively correlated. Salt marshes were predominantly found in mid- to high-latitude areas with cooler climates⁴³. Elevated temperatures inhibit salt marsh growth⁴⁴, reduce overall productivity⁴⁵, and increase respiration, which in turn decreases surface SOC by lowering net productivity and aboveground biomass⁴⁶. Additionally, high temperatures accelerate microbial activities, leading to faster SOC decomposition and the release of stored carbon into the atmosphere⁴⁷. This may explain why regions with high SOC density, such as Russia and Canada, are situated in cooler, high-latitude zones.

In contrast, sediment trapping in Asia has led to lateral carbon fluxes that have reduced SOC density, particularly along China’s coasts⁴⁸. High concentrations of suspended matter significantly influence SOC density in coastal salt marshes¹⁰. Furthermore, human-driven disturbances such as pollution from industrial emissions, agricultural runoff, and modifications to waterways (e.g., dam construction and canal digging) have disrupted sediment and water flow, adversely affecting SOC density in these regions⁴⁹. Another notable factor in this context is the invasion of Spartina alterniflora, which has further influenced SOC density in China’s salt marshes^50,51. Spartina alterniflora has a greater carbon sequestration capacity than native species do, and its widespread invasion increased the SOC density in salt marshes^52,53. However, due to its detrimental effects on native biodiversity, the Chinese government has been actively removing Spartina alterniflora since 2014⁵⁴. As the area of Spartina alterniflora has decreased, SOC density in the salt marshes has also been reduced.

Implications for global salt marsh protection and restoration

Expanding on the variability in SOC density, it is clear that global salt marshes exhibited notable spatial heterogeneity due to the combined influences of environmental and biotic factors, climate, and human activities¹⁶. High-latitude marshes consistently maintained greater SOC densities than mid- and low-latitude systems, largely due to temperature-mediated suppression of organic matter decomposition. Conversely, intense precipitation and tidal inundation promoted lateral carbon export, reducing surface SOC accumulation in dynamic coastal zones. Current global carbon balance models, however, oversimplify this complexity by assuming uniform SOC densities, a methodological limitation that neglects demonstrated latitudinal trends, disturbance-driven variability, and biome-specific carbon dynamics^16,55. To address this gap, we have employed a machine learning framework integrating multi-spectral remote sensing and field-calibrated SOC data, enabling spatially explicit quantification of SOC densities and their temporal trajectories. This approach helps clarify the regional contributions to global SOC dynamics in salt marshes, providing a more accurate understanding of how changes in the salt marsh area impact the overall SOC balance. Nevertheless, our estimates of SOC densities were lower than previous global averages^6,16, likely due to our consideration of the spatial heterogeneity of SOC density. For instance, full restoration of the SOC storage potential of salt marshes may take up to 100 years⁵⁶, newly formed or recently restored salt marshes can exhibit relatively high short-term SOC accumulation rates driven by rapid sediment deposition and initial organic matter inputs. However, these early-stage systems typically maintain lower SOC densities than mature salt marshes because soil development, organic matter stabilization, and vertical carbon accumulated into deeper layers remain incomplete. As a result, estimates based on average SOC densities from mature marshes may overestimate SOC gains in newly formed systems while simultaneously underestimating SOC losses from degraded or destroyed marshes, thereby contributing to the uncertainties reported in previous global assessments^6,57.

In conclusion, our analysis demonstrates that North America experienced the greatest loss of salt marsh areas, which, coupled with its relatively high SOC density, played a major role in the global decline of SOC. While salt marshes in Asia underwent slight expansion, the newly formed salt marshes generally had lower SOC densities, leading to insufficient gains to counterbalance the losses from the high-SOC-density marshes in North America. Consequently, global salt marshes remained in a net state of SOC loss. However, while North America, particularly the United States, accounted for the largest share of global SOC loss in absolute terms, this pattern partly reflects its large extent of salt marshes, which constitute a major proportion of global salt marsh area²². When normalized by salt marsh extent, relative SOC losses in regions such as Oceania and Europe are comparable or even higher, indicating greater proportional vulnerability. Thus, the dominant contribution of North America to global SOC loss represents both its extensive salt marsh coverage and the degradation of long-established, high-SOC-density ecosystems, rather than an unusually high loss rate per unit area. As such, protecting and restoring salt marshes with high-SOC densities worldwide, as well as preventing their conversion to low-SOC land uses, are essential strategies for enhancing carbon storage and mitigating climate change.

Countries with high-SOC-density salt marshes that are rapidly degrading, especially the United States, Canada, and Australia, require urgent large-scale conservation measures, such as hurricane resilience strategies including sediment replenishment and shoreline stabilization. In contrast, Asian nations like China and Bangladesh, where aquaculture expansion poses a threat, should incorporate carbon stock assessments into coastal zoning to protect high-SOC ecosystems from short-term economic pressures. Future restoration must exceed merely increasing marsh areas and instead focus on enhancing SOC density through nature-based solutions, such as selecting sites with optimal sediment supply, preserving mature marshes, and controlling invasive species. Given that full SOC recovery in restored marshes can take up to a century, long-term policy commitments and international funding are essential to align restoration efforts with climate mitigation goals⁵⁸. Ultimately, protecting carbon-rich marshes and implementing targeted restoration strategies are essential to rebalancing carbon dynamics and reinforcing the role of these ecosystems as robust natural climate solutions⁵⁹. This interpretation is further supported by our analysis of salt marsh dynamics within protected areas, where direct human disturbance is minimized. As shown in Supplementary Fig. 14, surface SOC accumulation within major protected areas in China, including the Chongming Dongtan and Yellow River Delta National Nature Reserves, exhibited a consistent increasing trend over time, despite limited opportunities for active restoration. Notably, more than half of the national-scale SOC gains in salt marshes during the study period occurred within officially protected areas, underscoring the critical role of protection in sustaining and gradually enhancing blue carbon potentials. At the same time, these findings highlight the urgent need for long-term, site-based field monitoring in newly formed and restored salt marshes to directly quantify SOC accumulation trajectories, validate remote sensing-based estimates, and resolve uncertainties in post-restoration carbon recovery rates.

Methods

SOC estimation from remote sensing data and machine learning

We utilized multi-source satellite imagery to calculate the surface SOC density of salt marshes over time and at a global scale, enabling the rapid estimation of SOC. The methodology for SOC estimation involves the following components:

• (a) Satellite data: we used the global salt marsh change dataset by Murray et al.⁸ as the base map. Landsat-5/7/8 optical SR images, with a 30 m resolution, were obtained from GEE for the period from 2002 to 2019, aligning with the temporal range of the salt marsh change dataset to ensure consistency in analysis. Optical information was extracted from the pixels where salt marsh changes occurred (after excluding points with missing spectral information, a total of 1,276,219 pixels were identified for salt marsh losses, and 1,414,029 pixels were identified for salt marsh gains). For each period (2002–2004, 2005–2007, 2008–2010, 2011–2013, 2014–2016, 2017–2019), we used the annual average spectral values from the year when the change began or ended. For example, for pixels representing salt marsh losses from 2002 to 2004, the 2002 annual average spectral information was extracted, whereas for pixels representing gains from 2002 to 2004, the 2004 annual average was used.

• (b) In situ soil organic carbon data: the SOC dataset integrates 90 in situ field samples collected across China’s coastal wetlands and 123 sites compiled from peer-reviewed studies (2002–2019), providing global coverage across major salt marsh biomes⁶⁰ (Supplementary Fig. 15 and Supplementary Data 1). Field sampling was conducted from 2019 to 2022, primarily in Chongming Dongtan (China) and Ximen Island (China), targeting dominant species such as Suaeda glauca, Phragmites australis, and Spartina alterniflora. Soil cores (20 cm depth) were systematically collected to measure key properties: soil organic carbon content (SOCC) via elemental analysis, and soil bulk density (SBD) using fixed-volume cores. Additionally, samples from peer-reviewed studies (2002–2019) were curated, standardized to the 0–20 cm depth range with confirmed global positioning system (GPS) coordinates (latitude/longitude) and sampling dates. Metadata for these samples included SOCC, SBD, vegetation type, and sediment texture, where available. The combined dataset spans North America, Europe, Asia, and Oceania, encompassing nine dominant plant families commonly reported in global salt marsh ecosystems, including Poaceae, Cyperaceae, Asteraceae, Alismataceae, Araceae, Amaranthaceae, Typhaceae, Juncaceae, and Chenopodiaceae. The spatial distribution of SOC sampling sites was designed to align with the global pattern of salt marsh change during the study period (2002–2019), rather than to provide a global distribution of SOC stocks. Because the objective of this study is to quantify SOC gains and losses associated with salt marsh expansion and degradation, sampling points were intentionally concentrated in regions experiencing the most pronounced net changes in salt marshes extent. Specifically, North America and China together account for approximately 70% of the documented global net salt marsh area change over the study period (Supplementary Table 3), and these two regions collectively comprise ~83% of the SOC sampling sites used for modeling (Supplementary Fig. 16 and Table 3). This proportional alignment between sampling density and the magnitude of salt marsh change enhances the robustness of SOC density estimates in regions that dominate global SOC gains and losses. Sampling sites outside North America and China were included to capture broader ecological and climatic variability across salt marsh systems and to constrain model behavior beyond the primary change hotspots. The surface SOC density (20 cm) of these sampling sites were calculated using Eq. (1).

  $${{{\rm{SOC\; density}}}}={{{\rm{SBD}}}}\times {{{\rm{SOCC}}}}\times {{{\rm{ST}}}}\times 100$$

  (1)

  where SOC density is the soil organic carbon density (g C/m²) at a certain depth for each sampling point, SBD is the soil bulk density (g/cm³), SOCC is the soil organic carbon content (%), and ST is the soil thickness (e.g., 20 cm in this study).

• (c) Machine learning models: to predict surface SOC density from remote sensing data, we implemented machine learning models using key spectral values and derived indices, namely normalized difference vegetation index (NDVI)⁶¹, enhanced vegetation index (EVI)⁶², modified normalized difference water index (mNDWI)⁶³, and land surface water index (LSWI)⁶⁴, as input variables for training (Supplementary Table 4). The models utilized were random forest (RF), decision tree (DT), and support vector machine (SVM), trained on field-sampled SOC data¹⁶, alongside optical bands (Landsat spectral bands B2–B7) and indices derived as part of our dataset (Supplementary Fig. 17). To ensure spatiotemporal alignment between the in situ SOC data with Landsat imagery, we selected cloud-free scenes within 15 days of the sampling dates and extracted spectral indices (NDVI, EVI, mNDWI and LSWI) from 30 × 30 m pixels centered on the sampling coordinates⁶⁵. The machine learning models were parameterized as follows: RF with 500 trees (min_samples_leaf = 5, max_depth = 15), DT with cost-complexity pruning (ɑ = 0.01, max_depth = 10), and SVM with a radial basis function kernel (γ = 0.1, C = 10). Before training, feature normalization using min-max scaling was applied, followed by model optimization with early stopping (patience = 10 epochs) based on validation loss. The dataset was randomly split into 70% for training and 30% for validation, adhering to the standard procedure for remote sensing inversion models, which have demonstrated high accuracy in SOC density prediction^66,67. All steps, including parameter selection, model training, and validation, were conducted using PyCharm Community Edition 2023.1.1 to ensure reproducibility and consistency throughout the analysis. This workflow was designed to maximize the predictive performance of these algorithms for global SOC density mapping.

• (d) Model accuracy verification: we rigorously evaluated model performance by comparing predicted surface SOC densities against independent field observations using three standardized metrics: the coefficient of determination (R²), root mean square error (RMSE), and Pearson correlation coefficient (r). These metrics provide complementary assessments of predictive accuracy and robustness, and are commonly adopted in large-scale SOC mapping studies^3,68. To ensure an optimal balance between model complexity and interpretability, we systematically tested multiple combinations of satellite-derived spectral bands and indices. The final model configuration incorporated the B3, B4, SWIR, mNDWI, and the EVI as predictors (Supplementary Fig. 17) within an RF framework (500 trees, maximum tree depth = 15). This variable set was selected based on its ability to capture vegetation productivity, surface moisture, and sediment-water interactions that jointly regulate surface SOC conditions in salt marshes. Importantly, the robustness of this covariate selection was further supported by spatial cross-validation and pixel-to-sample distance diagnostics (k-NN distance-based model evaluation) as shown in Supplementary Figs. 18 and 19, which indicate that predictive performance remains stable across regions with varying sampling density. Together, these results demonstrate that a parsimonious set of carefully selected spectral predictors can achieve reliable global estimates of surface SOC while maintaining computational efficiency and minimizing uncertainty propagation from heterogeneous auxiliary datasets. Despite using a smaller number of covariates than previous environment-driven SOC models⁶⁸, the final RF model exhibited strong and stable performance, with validation results of R² = 0.54 and r = 0.74 (p < 0.001) (Supplementary Fig. 20). These values are consistent with, and in some cases exceed, reported benchmarks for SOC estimation using remote sensing-based approaches¹⁶. Comparative testing further showed that the RF model outperformed alternative algorithms, including decision tree (R² = 0.35) and support vector machine models (R² = 0.26), particularly in capturing nonlinear relationships between vegetation structure, surface moisture, and SOC across tidal gradients.

• (e) Model application: the calibrated RF model was systematically applied to all identified salt marsh change pixels, as identified through the combination of Landsat SR data and the global salt marsh change dataset (described in component a). This approach enabled us to estimate the spatial distribution of surface SOC density across regions of salt marsh change and to quantify SOC dynamics over time. By applying the model to these changing pixels, we were able to map temporal variations in SOC density and then assess the impact of land use and environmental changes on carbon sequestration capacity in salt marshes.

Salt marsh maps and changes in the salt marsh area

In this study, the term “salt marshes” is used consistently to refer to vegetated saline tidal wetlands that are periodically inundated by seawater, in accordance with widely adopted global tidal wetland classifications^8,69. We integrated the data on the changes in salt marshes to analyze LULCCs trends and identify the underlying driving factors through remote sensing reclassification and geostatistical analysis. First, we accessed the global database⁸ showing the changes in the intertidal distribution in the GEE and extracted location data for areas in which salt marshes exhibited changes from 2002 to 2019. We then obtained remote sensing optical information for areas with losses or gains during the same periods.

With the extracted optical information, we selected spectral information from each band and calculated various characteristic indices as classification features. A random forest classifier was trained on the GEE using these selected samples and features⁷⁰. Finally, we reclassified the changed pixels to determine their land use and land cover types and validated the classification accuracy through cross-validation with field data. A 10-fold cross-validation method in GEE demonstrated an overall accuracy exceeding 70% (Supplementary Table 5), with a general increasing trend over time due to improved image quality. These results ensure reliable SOC estimation and analysis of LULCCs in this study.

Estimation of SOC gains/losses and analysis of the driving force

Estimation of SOC gains and losses

The surface SOC gains and losses from changes in salt marshes were derived from the remote sensing imagery and LULCCs. The methodology involves the following assumptions: (1) the conversion of salt marshes to built-up areas, aquaculture areas, agricultural lands, mudflats, or seawater results in the complete destruction of these areas, leading to the eventual release of all surface SOC⁷¹. (2) Transformations to mangroves preserve the original SOC⁷², whereas changes from mangroves to salt marshes do not contribute to carbon gains.

The cumulative SOC of other land use and land cover types represented the SOC losses and gains from salt marsh losses or gains in a given period and region. We summed the SOC losses and gains for each degree of latitude, with SOC losses defined as negative and gains defined as positive. The region-level SOC gains and losses were calculated via Eq. (2):

$${{\mathrm{SOC}}}={10}^{-6} {\sum }_{i=0}^{n}{{{\mathrm{SOC\; density}}}}_{i}\times 900\times \left\{\begin{array}{l}1,\; {{\mathrm{loss\; pixel}}}\\ -1,\; {{\mathrm{gain\; pixel}}}\end{array}\right.\,$$

(2)

where SOC is the soil organic carbon (tonnes; negative values indicate SOC loss, and positive values indicate SOC gain); SOC density_i is the soil organic carbon density for pixel i (g/m²); n is the number of pixels in the target region; and 900 is the area for each Landsat pixel (30 × 30 m).

To address uncertainty in the fate of surface SOC following salt marsh conversion, we also conducted a scenario-based sensitivity analysis comparing a full-release assumption with a LULCC-specific partial-release framework. The full-release scenario assumes complete SOC loss upon conversion, providing an upper-bound estimate. In the partial-release scenario, conversion-specific SOC loss fractions were applied based on published empirical and mechanistic studies^73,74: 40–70% for agricultural lands, 30–80% for aquaculture areas, 100% for built-up areas, and 0–30% for mudflats and seawater, reflecting differences in drainage, sediment disturbance, and burial potential. Global SOC losses were recalculated for each time period using these fractions to derive uncertainty bounds and to assess the sensitivity of net SOC balance estimates to assumptions about post-conversion SOC pathways.

Analysis of the factors driving SOC density changes

To evaluate the drivers of SOC density changes in salt marshes, we began with a cluster analysis based on coastal hydrobasins⁷⁵ to categorize salt marsh pixels undergoing change into distinct clusters (Supplementary Figs. 21 and 22). This classification considered regional population, environmental, and climatic factors. Using insights from a comprehensive literature review (Supplementary Figs. 23 and 24), we systematically identified key variables influencing salt marsh distributions and carbon sequestration processes. The review was conducted following a structured workflow, including keyword-based searches in the Web of Science database (covering 1946–2023), screening of peer-reviewed studies reporting salt marsh extent, sediment accretion, carbon density, carbon accumulation rates and potential driving factors, and rigorous filtering to retain studies with clear geographic locations, vegetation information, and soil depth descriptions. Based on the frequency of occurrence in the literature and their demonstrated mechanistic relevance to salt marsh dynamics, we identified key variables affecting the distribution of salt marshes and their carbon sequestration capacities, which are population pressure (DMSP and VIIRS nighttime light data⁷⁶), typhoon frequency (IBTrACS)⁷⁷, sea-level rise⁷⁸, sediment trapping⁷⁹, tidal patterns (FES2014⁸⁰), temperature, and precipitation (WorldClim)⁸¹. Average values of these variables during the observed period of change were calculated for each hydrobasin cluster. We then employed a two-level mixed-effects model to quantify the impacts of these factors on SOC density changes. The Level 1 model captured within-cluster dynamics and is expressed as:

$$\Delta {{{\mathrm{SOC}}}}_{{{\rm{ij}}}}={\upbeta }_{{{\rm{0j}}}}+{\upbeta }_{{{\rm{1j}}}}({{{\mathrm{PopPressure}}}}_{{{\rm{ij}}}})+{\upbeta }_{{{\rm{2j}}}}({{{\mathrm{TyphoonFreq}}}}_{{{\rm{ij}}}})+{\upbeta }_{{{\rm{3j}}}}({{{\mathrm{SLR}}}}_{{{\rm{ij}}}}) \\+{\upbeta }_{{{\rm{4j}}}}({{{\mathrm{SedimentTrapp}}}}_{{{\rm{ij}}}})+{\upbeta }_{{{\rm{5j}}}}({{{\mathrm{Tide}}}}_{{{\rm{ij}}}})+{\upbeta }_{{{\rm{6j}}}}({{{\mathrm{Temp}}}}_{{{\rm{ij}}}})+{\upbeta }_{{{\rm{7j}}}}({{{\mathrm{Precip}}}}_{{{\rm{ij}}}})+{e}_{{{\rm{ij}}}}$$

(3)

where ∆SOC_ij is the change in soil organic carbon density (g/m²) for observation i within hydrobasin j; PopPressure_ij is the change (e.g., hfp2004–hfp2002) in annual average human footprint (the synthesized effect of multiple human activities including built environment, population density, nighttime lights, croplands, pastures, roads, railways, navigable waterways) within hydrobasin j; TyphoonFreq_ij is the number of tropical cyclone occurrences (points along their paths) within hydrobasin j; SLR_ij is coastal sea level rise trend (2002–2021) for each hydrobasin j estimated by using the value of the nearest altimetry stations; SedimentTrapp_ij is the average sediment trapping index score within hydrobasin j; Tide_ij is the mean tidal amplitude within each hydrobasin using the principal lunar semi-diurnal or M2 tidal amplitude as this is this most dominant tidal constituent; Temp_ij is the average value of minimum temperature of coldest month; Precip_ij is the average value of precipitation of driest quarter (both temperature and precipitation are extracted from the standard WorldClim Bioclimatic variables for WorldClim version 2).

The Level 2 model captured between-cluster variability and is represented as Eqs. (4)–(7):

$${{\rm B}}_{{{\rm{0j}}}}={\gamma }_{00}+{\gamma }_{01}({{{\mathrm{PopPressure}}}}_{{{\rm{j}}}})+{\upmu }_{{{\rm{0j}}}}$$

(4)

$${{{{\rm{\beta }}}}}_{{{\rm{1j}}}}={\upgamma }_{10}+{\upmu }_{{{\rm{1j}}}}$$

(5)

$${{\rm B}}_{{{\rm{2j}}}}={\upgamma }_{20}+{\upgamma }_{21}({{{\mathrm{Latitude}}}}_{{{\rm{j}}}})$$

(6)

$${{{{\rm B}}}}_{{{\rm{3j}}}}={\upgamma }_{30}$$

(7)

where PopPressure_j is the change in annual average human footprint within hydrobasin j; Latitude_j is the centered latitude of hydrobasin j; μ_0j and μ_1j are Hydrobasin-specific random effects (μ_0j ~ N(0, τ₀₀)_, μ_1j ~ N(0, τ₁₁)).

Additionally, SOC losses and gains were calculated for each continent, with North America identified as a loss hotspot and Asia as a gain hotspot. By integrating cluster analysis with multilevel modeling, we assessed spatial patterns of SOC density dynamics, providing a global and regional understanding of the carbon balance in salt marshes. This framework highlights how environmental, climatic, and anthropogenic factors jointly drive SOC density changes, offering actionable insights for salt marsh restoration and carbon sequestration strategies. All analyses were performed using R4.3.3⁸².

Uncertainty and limitation analysis

Limitations remain regarding the SOC dynamics derived in this study. First, our estimates of SOC accumulation are based on bulk SOC density and therefore integrate both autochthonous carbon, produced by in situ vegetation and microbial processes, and allochthonous carbon^57,83, imported from adjacent ecosystems such as seagrass meadows, tidal flats, or upstream terrestrial sources^84,85. Consequently, the SOC gains and losses reported here reflect the total carbon retained or released from salt marsh soils, rather than the net additional sequestration achieved through local primary production. This distinction is important for interpreting the carbon balance of coastal systems. Expansion of salt marshes through either natural sediment accretion or restoration may trap and store organic matter transported from other environments, which doesn’t necessarily represent new atmospheric carbon uptake. Thus, degradation or conversion of mature salt marshes can lead to the remobilization of both locally produced and imported carbon^11,86, potentially amplifying greenhouse gas emissions. Although disentangling these contributions is beyond the scope of this study, our spatially explicit framework provides a foundation for integrating isotopic, biomarker, or sediment source-tracking approaches in future research to better distinguish autochthonous and allochthonous carbon fractions^60,87,88. Recognizing this compositional complexity is essential for accurately attributing carbon fluxes and assessing the true potential of blue carbon ecosystems in global climate mitigation strategies.

Second, the SOC density inversion did not account for aboveground biomass; it was assumed that aboveground biomass would eventually be incorporated into the soil and show a positive correlation. In addition, vegetation reflectivity in remote sensing images affected inversion accuracy, particularly when the density of the vegetation distribution was high. The use of data from various sources (field-, published-, and satellite-datasets) collected using different methods and with varying degrees of error may have impacted the model results. Moreover, SOC density varied with soil depth and elevation, which can cause discrepancies compared with existing estimates^16,89. Estimations based on spectral data were typically limited to surface soils, with most studies focusing on depths of less than 30 cm^90,91. Therefore, we quantified the surface SOC density (0–20 cm) in salt marshes, and this measurement can represent the SOC balance for salt marsh changes, particularly for newly formed or restored salt marshes, although it doesn’t account for the total SOC. Besides, we focused on surface SOC for ecological reasons. Surface soils represent the most dynamic and biologically active layer of salt marsh soil profiles, where vegetation growth, tidal inundation, and anthropogenic disturbance exert the strongest and most immediate influence on carbon cycling^92,93. This layer therefore provides the most sensitive indicator of short- to mid-term SOC changes driven by degradation, restoration, or accretion processes^90,94. Moreover, surface SOC plays a critical role in nutrient retention, primary productivity, and microbial metabolism, making it a functionally important proxy for ecosystem condition^95,96. We acknowledge that this relatively shallow depth doesn’t capture the full vertical extent of below-ground carbon storage, which can extend to meter-scale deposits in mature salt marshes. However, our study aims to resolve decadal-scale changes that primarily occur near the surface, where carbon is most vulnerable to erosion, oxidation, and land-use conversion. Focusing on this active layer provides a conservative yet policy-relevant estimate of SOC dynamics associated with LULC transitions. This approach is consistent with current carbon accounting frameworks that often rely on surface-based emission factors to estimate carbon fluxes from land-use change^96,97. Future research combining deep-core sampling, age-dating, and long-term monitoring will be essential to extend these assessments to subsurface carbon pools and evaluate their long-term stability.

Third, to calculate SOC losses and gains, we assumed idealized conditions in which land-cover transformations, except those involving mangroves (which also have a high capacity for SOC storage), either result in the release of original surface SOC or the uptake of an equivalent amount of SOC. In reality, SOC dynamics in coastal wetlands are governed by complex hydrodynamic processes, including tidal inundation, wave action, and tidal and residual currents, which regulate sediment transport, erosion-deposition balances, and organic matter redistribution in vertical directions. These processes can introduce substantial spatial and temporal variability into actual SOC losses and gains^1,41,98. Our estimation supported rapid, large-scale SOC balance accounting, aiding global policy formulation. For detailed regional accounting, incorporating biophysical-chemical models is essential for understanding SOC cycles at a mechanistic level. Furthermore, while the current framework focuses on surface SOC dynamics with training samples mostly from North America and East Asia (hotspots of salt marshes gain and loss), future studies could benefit from integrating expanded datasets such as Maxwell et al.⁶⁸ to refine model calibration, reduce possible regional uncertainty, and improve representation of under-sampled regions. Incorporating these datasets would likely strengthen confidence in regional estimates and further enhance the robustness of global SOC change assessments. Lastly, the accuracy of remote sensing depended on the quality of the images, including their spatial and temporal resolutions. Using Landsat images (30 m resolution), areas smaller than 900 m² may have mixed pixels, affecting classification and SOC estimation. The limited spectral bands and susceptibility to atmospheric conditions also impacted the final accuracy. Future studies could use synthetic aperture radar or hyperspectral imagery for better estimates^99,100.