Many freshwater forested wetlands along the southeastern U.S. coastline are rapidly transitioning from forest to marsh or open water, due to climate change related disturbances, such as saltwater intrusion and increasing flooding frequency. These changes in wetland state are considered a regime shift, and the timing and trajectory of change are not well understood. Recent studies have found early warning signals (EWS) of regime shifts in other ecosystems, but it is unclear if these can be detected for coastal wetlands.

In this study, we examined the ability to detect EWS of regime shifts in coastal wetlands within the Albemarle Pamlico peninsula, North Carolina, U.S.A. We used 35 years of the Landsat record to examine trends and variance of normalized difference vegetation index (NDVI) time series for selected areas known to have undergone regime shifts.

We found that NDVI time series trends combined with changes in standard deviation of NDVI allowed us to identify four scenarios of change for coastal wetlands: 1) unstable transitioning; 2) gradual transition (declining); 3) unstable re-vegetated (recovering); and 4) stable vegetated. At the landscape scale within the Albemarle Pamlico peninsula, we found that approximately 114,294 ha (40%) of natural wetlands are considered stable, while 77,732 ha (27%) are areas are re-vegetating following a disturbance (mostly fires). We also found that 39,828 ha (14%) experienced a regime shift with an abrupt change, while 24,092 ha (8.5%) are forests gradually shifting to marshes.

Syntheses and applications: Our results suggest that the transition from a forest to a marsh can occur both rapidly and slowly, and remote sensing of NDVI time series can help identify ecosystem trajectories. Remote sensing provides the ability to measure and monitor the resilience of ecosystems, identify trajectories of change, and opens windows of opportunity for intervention, and prioritization of conservation/restoration of coastlines, all of which will become more important in the face of climate change and sea level rise.

We used Google Earth Engine (GEE) computing platform for processing and extraction of NDVI time series data due to its multi-petabyte catalog of satellite imagery and geospatial datasets (Google Earth Engine, 2019). Landsat Surface Reflectance Tier 1 (TM, ETM+, and OLI) data was used to assess the selected transitioned sites (Fig. 3). Sensor differences between Landsat 7 and 8 images were harmonized using transformation function following Roy et al. (2016). Images were processed, masking out clouds and cloud shadows using QA band. For Landsat 5 and 7, the ‘sr_atmos_opacity’ band was used to find and mask hazy pixels, and for Landsat 8, the ‘sr_aerosol’ band was used to remove pixels labelled as high probability of haze. NDVI values were then computed using the red and near-infrared bands.  Raster NDVI composites were generated for each year from 1985 to 2020 selecting the maximum NDVI per pixel during the growing season (May 1 to August 31). It is important to note however that the number of images per satellite available varied each year and per pixel (Fig. S1).