The research team, led by graduate researcher Titilayo Tajudeen alongside co-authors Marcelo Ardon, Mirela Tulbure and Katherine Martin, focused on the Albemarle-Pamlico Peninsula, one of the largest expanses of coastal forest on the Atlantic seaboard and a region exceptionally vulnerable to rising seas. By training artificial intelligence models on decades of satellite imagery, the researchers reconstructed a 36-year history of land cover change, distinguishing healthy forest from marshland, shrub-dominated thickets and ghost forest. The results reveal a landscape under mounting pressure from saltwater intrusion, saturated soils and the compounding effects of extreme weather events.

The headline numbers tell a story of acceleration. Between 1985 and 2010, some 16,968 hectares of forest were converted into marsh, ghost forest or shrubland. Between 2010 and 2021, that figure climbed to 23,876 hectares, one and a half times the loss recorded over the preceding 25 years, despite covering less than half the time span. The transition into ghost forest, where salt-poisoned trees die but remain standing, accelerated even more dramatically. Ghost forest extent grew by 7,561 hectares between 2010 and 2021, a rate two and a half times faster than the 3,087 hectares gained between 1985 and 2010. In other words, the most visible and haunting symptom of coastal degradation is intensifying faster than overall forest loss itself.

At the heart of this transformation lies sea level rise, which the study identified as the chief driver of forest loss. As relative sea level climbs along the North Carolina coast, salty water pushes into freshwater wetlands and low-elevation upland forests, elevating soil salinity and waterlogging the root systems of salt-intolerant trees. The abstract accompanying the paper describes how healthy coastal forests that depend on freshwater are transitioning into landscapes dominated by dead or dying trees, which are in turn flanked by salt-tolerant shrubs and grasses before eventually giving way to marshes or open water. This progression, from forest to ghost forest to marsh, represents a one-way ecological ratchet in which each stage is harder to reverse than the last.

Geography matters as much as chemistry. The researchers found that the most heavily affected areas were concentrated within one kilometer of the coast, where proximity to channels exposes the land most directly to saltwater intrusion and storm-driven surges. Salinity and the increasing rate of relative sea level rise emerged as the key environmental drivers of the observed conversions, alongside the simple fact of being close to tidal waters. These findings provide a spatial template for identifying which remaining forest tracts are most vulnerable, information the authors say can form the basis for targeted conservation strategies before more land crosses the threshold of no return.

Yet sea level rise does not act alone. The study highlights how a sequence of extreme events compounded the slow creep of the ocean. The region endured a severe drought from 2007 to 2011, followed by Hurricane Irene in 2011, a one-two punch that stressed forests already contending with encroaching salt. Tajudeen noted that while these events occurred long before the end of the study period, some of the affected areas simply never recovered. Even lands that enjoy formal protection were not spared, as the combination of extreme events and rising seas pushed them into new ecological states, including ghost forest. The lesson for coastal managers is sobering: protected status cannot shield an ecosystem from a changing climate, and disturbance events can permanently tip vulnerable forests over the edge.

The methodological machinery behind these findings is as noteworthy as the results themselves. The researchers harnessed two complementary satellite imagery systems, Landsat 8 and Sentinel-2, and trained a convolutional neural network, a class of artificial intelligence model particularly adept at processing the grid-like structure of satellite raster data. By feeding the network multispectral, bi-seasonal, topographical and phenological metrics derived from the imagery, the team enabled it to classify which pixels represented intact forest and which had been converted to marsh, ghost forest or shrub. The study found that incorporating phenology and topographical indices significantly enhanced the separability of the ghost forest class from all other vegetation types, a critical technical achievement given how visually similar dying forests can appear from orbit.

Each satellite platform brought distinct strengths to the analysis. Sentinel-2 imagery offers a resolution of 10 meters, sharper and more detailed than the 30-meter resolution of Landsat 8, and in a head-to-head comparison for the year 2021, when both datasets were available, the higher-resolution Sentinel-2 data outperformed Landsat, achieving an F1 score of 96.3 against Landsat’s 93.4. But Landsat holds an irreplaceable advantage in its archive, which stretches back decades and allowed the researchers to extend their analysis all the way to 1985. The team therefore relied on Landsat for the long-term reconstruction while using the comparison to validate the accuracy of their approach, demonstrating that deep learning models can deliver reliable land cover classification across both modern and legacy satellite records.

Beyond its regional significance, the study offers a replicable framework for tracking coastal ecosystem degradation worldwide. Ghost forests are appearing along shorelines from the Chesapeake Bay to the Gulf Coast and beyond, wherever rising seas meet flat, forested terrain. Quantifying the dynamics and pathways of forest-to-marsh conversion, as this research does, is vital for understanding the progression of degradation and for forecasting future change. By pinpointing the regions most exposed to salinity, proximity to channels and accelerating sea level rise, the North Carolina work gives conservation planners a way to prioritize interventions, whether protecting migration corridors for marshes, managing hydrology to limit saltwater intrusion, or accepting conversion in some areas while defending others.

What the bleached trunks of the Albemarle-Pamlico Peninsula signal, ultimately, is a coastline in motion. Forests that took centuries to establish are being converted to marsh and open water within a few human generations, and the study’s evidence that loss may still be speeding up suggests the coming decades will bring further, faster change. The ghost forest is more than a haunting photograph; it is a measurable, accelerating indicator of how rising seas are redrawing the boundary between land and water, and thanks to the marriage of long-term satellite archives and modern machine learning, scientists can now watch that boundary shift year by year, hectare by hectare.

Subject of Research: Accelerating coastal forest loss and ghost forest formation driven by sea level rise on North Carolina's Albemarle-Pamlico Peninsula

Article Title: Forests are dying along North Carolina’s coast – and it’s speeding up

Article References: Forests are dying along North Carolina’s coast – and it’s speeding up. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: ghost forests, sea level rise, coastal forests, North Carolina, saltwater intrusion, remote sensing, convolutional neural networks, Landsat, Sentinel-2, marsh conversion, Albemarle-Pamlico Peninsula, PLOS One

The concept at the heart of the study is known as escape habitat planning. As climate change pushes ecosystems beyond their historical ranges, species must either adapt in place or move to survive. For coastal wetlands, movement means migrating landward, a process that requires both vertical space for sediment accumulation and lateral space for wetlands to form on adjacent uplands. When seawalls, roads, and subdivisions block that retreat, habitats become squeezed between rising water and immovable infrastructure, a phenomenon ecologists call coastal squeeze. In the southeastern United States, projections suggest up to two-thirds of future wetland migration will come at the expense of freshwater wetlands, while the remainder will consume already fragmented uplands, many of which harbor endemic species found nowhere else.

Florida presents a uniquely difficult case. The state’s 1,931-kilometer coastline is home to more than two-thirds of its 23.4 million residents, and its flat topography and porous limestone bedrock make it exceptionally vulnerable not only to inundation but to saltwater intrusion into the aquifers that supply most of its freshwater. Projections indicate that up to 2 meters of sea-level rise could flood more than 7,770 square kilometers of the state, with the southern Gulf Coast and the Everglades bearing the brunt. Coastal forests are already converting to ghost forests, landscapes of dead trees killed by inundation and salt, while nearly half of Florida’s island-dependent species face major extinction risks. Yet some systems retain mobility: salt marshes can transgress landward, and mangroves are expanding northward as freeze events become rarer.

The research team built and compared two complementary models. The first, the Florida Ecological Greenways Network Coastal-to-Inland Connectivity Model, is deliberately simple. Using Esri’s Cost Distance tool and the state’s Cooperative Land Cover dataset, it treats all undeveloped, ecologically suitable land as equally permeable, assigning a uniform resistance value of one, and then calculates continuous pathways along which coastal natural communities could theoretically migrate inland. The study area was delineated using NOAA’s 3-meter sea-level rise model with an additional two-mile buffer. This data-efficient approach is highly transferable to other regions and is intended to support strategic land acquisition decisions with minimal input requirements.

The second model, the Integrated Coastal Connectivity Model, is far richer in data. Built with Omniscape, a circuit-theory-based connectivity tool implemented in the Julia programming language, it combines expert-ranked land cover classifications with high-resolution flood hazard maps from JBA Risk Management, topographic diversity data derived from satellite elevation products, and sea-level rise projections from two established marsh migration models. On the Gulf Coast, the team used the Sea Level Affecting Marshes Model, or SLAMM, a tool originally developed in the mid-1980s with funding from the U.S. Environmental Protection Agency that simulates inundation, erosion, sediment accretion, overwash, saturation, and salinity shifts. Along the Atlantic coast, where comprehensive SLAMM results were unavailable, they substituted the NOAA Marsh Migration Model, which combines land cover data, digital elevation models, and tidal datum surfaces to project marsh persistence, migration, or loss.

One of the study’s most technically interesting choices involves how water itself was parameterized. The researchers tested scenarios in which open marine areas were assigned extreme resistance values and found that both extremes distorted the results. Assigning a value of ten, intended to represent high permeability, inadvertently created artificial islands of low resistance surrounded by higher-resistance matrices, while a value of zero blocked movement entirely. They settled on an intermediate, neutral value of six, which better captured continuous ecological flow along the coastline. Flood data was processed using fuzzy logic, with a gamma overlay combining weighted fuzzy sums and products to highlight areas that flood repeatedly across multiple return periods, from 20-year to 1,500-year events, rather than treating each flood scenario independently.

The results reveal both convergence and striking divergence between the two approaches. The simpler cost-distance model identified its most extensive migration corridors in the Big Bend and Panhandle regions, where natural land cover remains intact, and highlighted the Suwannee River basin and Apalachicola lowlands as critical connectivity zones. The integrated model showed that under a 1-meter rise, many low-lying Gulf Coast wetlands would be inundated, shrinking corridors to localized pockets, and under 2 meters, connectivity would collapse further across southwest Florida and the Big Bend. Agreement between the two models fell as low as 59 percent in some regions, and adding topographic diversity to the resistance surface reduced agreement by an additional 5 to 10 percent, a finding the authors did not fully anticipate.

That divergence may be the study’s most consequential insight. Florida is often perceived as uniformly flat, but its subtle microtopography, relict dune ridges, karst features such as sinkholes and springs, and slight elevation gradients, has shaped its biodiversity for millennia. Over successive cycles of rising and falling seas, populations isolated on upland ridges evolved into endemic species through allopatric speciation. The integrated model’s sensitivity to topographic diversity suggests these modest features act as stepping stones and microrefugia, offering drier soils, thermal variation, and resistance to inundation. Conserving geophysically complex areas slows the velocity of climate change exposure for species and provides microclimates that buffer against extremes, meaning that even centimeter-scale terrain variation matters when the entire landscape sits barely above the tide line.

The practical payoff comes in the form of land acquisition priorities. The team overlaid their connectivity outputs on the Florida Forever Board of Trustees Projects layer, which catalogs 362 proposed conservation land purchases not yet protected, and ranked roughly 19,000 individual parcels based on corridor overlap, watershed priority, and elevation. Agreement between the two models increased steadily with ranking quality, from 29.45 percent in the lowest 30 percent of parcels to 63.60 percent in the top 10 percent. In other words, while the models disagree about lower-priority areas, they converge substantially on the highest-value conservation targets, and many of those top-scoring parcels already align with existing state acquisition goals. This convergence gives planners a defensible basis for tiered decision-making, where areas of model agreement can trigger immediate action and divergent areas warrant monitoring or adaptive management.

The authors are careful to note the limits of their framework. The models do not account for ocean warming, freeze-driven shifts between mangrove and marsh dominance, groundwater levels, sediment supply, or the dispersal capacities of individual species, and they should be read as broad-scale screening tools rather than precise predictions. Future work could incorporate focal species, soils data, and higher-resolution elevation analysis. Still, the study demonstrates that integrating climate exposure and geomorphological variation into connectivity analysis can bridge traditional coarse-filter modeling with advanced spatial techniques, producing a transferable template for any coastal region facing the same existential question: not whether the sea will rise, but whether the ecosystems that depend on dry land will have somewhere left to go.

Subject of Research: Modeling coastal-to-inland ecological connectivity and wetland migration corridors under sea-level rise in Florida

Article Title: Sea level rise escape planning for Florida’s coastal ecosystems

Article References: Sea level rise escape planning for Florida’s coastal ecosystems. (n.d.). https://doi.org/10.1007/s44353-025-00064-w

Image Credits: AI Generated

DOI: 10.1007/s44353-025-00064-w

Keywords: sea-level rise, ecological connectivity, coastal wetlands, marsh migration, Florida, SLAMM, Omniscape, conservation planning, topographic diversity, land acquisition, climate adaptation, ghost forests