The Federated States of Micronesia (FSM) comprises four island states approximately 4,000 kilometers (km) southwest of Hawaii, just north of the equator. Pohnpei is the largest island state in the FSM at 346 square kilometers (km²) and is home to about 35,000 people. The rugged topography rises steeply from the coast to the nearly 800-meter (m) summit of a dormant volcano. Pohnpei is ringed by coral reefs and 57 km² of mangroves in the coastal lowlands. The extensive mangroves provide habitat to several endangered species such as the Pohnpei lorikeet (Trichoglossus rubiginosus) and fruit bats (Pteropus molossinus). Mangroves are important for human communities throughout the FSM; the fish, crabs, and firewood harvested from mangroves provide the equivalent of more than half of the annual household income (Naylor and others, 2002).

Pohnpei is characterized by a tropical rainforest climate. One of the wettest places on Earth, Pohnpei receives an average of 4,800 mm of rainfall each year and has no dry season (Lander and Khosrowpanah, 2004). The El Niño–Southern Oscillation (ENSO) is responsible for the majority of the interannual variation in rainfall, with drier conditions in post El Niño years (Lander and Khosrowpanah, 2004). ENSO is also a strong driver of sea-level variation, and changes of ±30 centimeters (cm) are possible (Lander and Khosrowpanah, 2004).

We developed a new process-based soil cohort model for understanding the vulnerability of mangroves to environmental change, especially sea-level rise (fig. 2). The model simulates changes in land-surface elevation in response to potential increases in sea level by estimating sediment accretion from inorganic inputs and biomass inputs from common mangrove species growing around Pohnpei (appendix 1, eq. 1). Mangrove biomass is estimated by fractional basal area, from which aboveground and belowground inputs are estimated using allometric relations and shoot:root ratios for each species, with additional belowground dynamics estimated from soil cores. Estimates of mass and density for each pool were used to calculate cohort volumetric contributions to depth by summation, and the total normalized by area to estimate land-surface elevation relative to mean sea level. The model uses estimates of elevation change relative to sea level to project changes in mangrove species’ distribution around Pohnpei based on the optimal elevation ranges of various mangrove species.

The probability of species presence was determined from elevation relative to mean sea level (fig. 2). Fractional basal area for each species is held in living and standing dead pools. A negative change in fractional basal area, which results when surface elevation moves outside the optimal elevation range for a given species, adds fractional basal area to the standing dead pool. Both living and standing dead cover are converted to aboveground biomass using species-specific allometric equations. Standing dead biomass falls at a constant rate and is moved to the fallen dead pool where it undergoes decomposition. Litterfall and root turnover contribute to the labile (LOM) and refractory (ROM) soil organic pools, which are subject to loss by decomposition at differential rates. Rates of organic matter decomposition are tied to inundation period and slow with increasing depth in the soil column. Root density distribution by depth is determined by an exponential decay function, whereas mineral sediments are deposited in the surface cohort.

Each cohort represents both the age and depth of the soil and contains the organic and inorganic soil and root pools. A total of 80 age-depth cohorts were used in this model, transferring material to deeper cohorts at differential rates (for example, Rybczyk and others, 1998). The variable transfer rates were set such that there was higher temporal resolution in the upper soil layers and lower resolution in the deeper layers, where material accumulated (transfer rates of 1, 2, 10, and 20 years/cohort, 20 cohorts in each). Using a constant number of cohorts was important for computational efficiency during multi-century model runs, where old, deep material changes very slowly.

We selected seven representative sites on Pohnpei, Federated States of Micronesia, for modeling based on the availability of long-term accretion data derived from lead–210 dating (R. MacKenzie, U.S. Forest Service, written commun., August 15, 2018; fig. 4), which is used for dating materials up to about a century old (Appleby and Oldfield, 1978). Thirty 1-m long cores were used to calibrate the model. Each core was sectioned into 2-cm intervals and analyzed for dry bulk density and organic matter percent through loss-on-ignition. Average accretion, mineral accumulation rate, and organic accumulation rates for each site were used to calibrate the model, assuming a 75-year horizon (table 4). The soil-core data were also used to calculate the parameters in the mixing model (eq. 21) for estimating bulk density (k₁ = 0.145, k₂ = 0.908, k₃ = –0.029; root mean squared error, RMSE = 0.072; fig. 5); we assumed a maximum bulk density limit of 1.2 g/cm³. We used an analysis of variance to test for site and hydrologic zone differences in soil-core elevation, accretion rate, and the organic and mineral accumulation rates.

Water-level records from the long-term gage at the airport on Pohnpei were used to calculate mean sea level (MSL), mean high water (MHW) and mean low water (MLW) for 1983–2001, the most recent National Oceanic and Atomspheric Administration (NOAA) epoch. Pressure transducer water-level dataloggers (Solinst^® Edge Jr.) were also deployed at three study sites (Enipoas, Enipein, Sapwalap), but only the loggers at Sapwalap and Enipein were recovered. The loggers were deployed just above the soil surface and recorded data in 12-minute intervals, from which the frequency and duration of high tides that inundated the mangrove forest (from July 7, 2016, to March 13, 2017) were determined. The data from the loggers and the airport gage were aligned, and MSL was estimated for each logger site by the MHW difference (appendix 1, eq. 22). For sites without water-level data, the MSL was assigned according to the nearest site with data (table 5).

Elevation surveys were completed in February 2017 and April 2019 (Irwin and others, 2021). Temporary benchmarks were established outside the mangrove forest using a Trimble global navigation satellite system base station. We then used differential leveling to establish the elevation from the benchmark with the surface elevation table (SET) and soil-core locations at each study site. At each turning point along the leveling transect, the presence of each mangrove species within a 10-meter (m) radius was recorded. We used the species-elevation data from 152 plots to derive probability distribution functions for each species, which in turn control the growth rate in the population model (fig. 6). We used the soil-core elevation to calibrate a site-specific mineral-accumulation function.

Digital elevation models (DEMs) were created for each site using transects of differential leveling data and the elevation profile from seaward to upland edge. We digitized the shore and upland edge of the mangrove forest with high resolution (0.3 m) WorldView 2 satellite image, and Euclidean distances from each edge were calculated as rasters. An elevation-profile transect was generated using a subset of data from Enipoas, Enipein, Palikir, and Pwudoi (fig. 4). Elevation profiles of the transect were generated using a general additive model (GAM) spline and the relationship between elevation and distance from the seaward edge (fig. 7). We used the relative width of the mangrove forest to extrapolate the elevation profile GAM model across the study site (appendix 1, eq. 23).

For each site, we used an optimizing function to determine the maximum width, W_m, value that resulted in a DEM with the lowest RMSE when compared with the elevation points for that site, constraining W_m between 600 and 950 m. Model accuracy varied by site (table 5); the average RMSE for all the sites was 0.26 m.

We used a multi-step calibration scheme to estimate several important parameters while preserving site differences (table 6). We allowed three parameters to change for site-specific model calibrations. First, the mineral accumulation rate was calibrated to the data from the soil cores. Second, carrying capacity, here defined as fractional basal area, was calculated as the 95th percentile from field surveys in each region. The litter deposition rate was set to 2 percent of aboveground biomass, a value that produced approximately 8,800 kilograms per hectare (kg/ha (R. Hauff, unpublished data, in Gleason and Ewel, 2002). Finally, we calibrated the labile decomposition coefficient for inundation (k_l) to minimize the difference in modeled and observed elevation after 100 years in a model simulation that used the historic sea-level rise rate (3.75 mm/year) and initial elevations estimated from soil-core accretion rates. Modeled values of all roots, living roots (fine and large), and ratio of dead:live roots were compared with values from Cormier and others (2015); for all sites, these values were within the range of literature values (appendix 2). The proportion of structural roots was calibrated such that the living fine and large root mass pools were less than the value provided in Cormier and others (2015); we assumed structural roots were not incorporated in the soil cores of that study. The parameters that control root collapse (l_h, r_h,d_r, d_l) were constrained by limiting the elevation loss of a dead forest to a maximum of 10 mm/year, the rate of soil collapse observed in a harvest mangrove stand (Krauss and others, 2010). The model only considers elevation and does not directly consider the potential effects of salinity or macronutrients on organic deposition or decomposition; we assumed that site-specific values of carrying capacity (table 6) and the organic accumulation rate reflect these factors. A 400-year model spin-up period, using sea-level rise rates equal to the mean soil-core accretion rates and three initial land-surface elevations (5, 15, 35 cm relative to MSL), was used to assess model accuracy by comparing modeled and observed vertical accretion, mineral accumulation, and organic matter accumulation rates (appendix 2). By using the mean accretion rates to calibrate the model, we are assuming that these historic (75 year) accretion rates represent a forest in steady state with its environment.

At Sapwalap and Enipoas, we used a long-term (1999–2019) SET record to independently validate the model. At each site, there are three replicate SETs deployed in fringe, riverine, and interior hydrologic zones. Using the differential elevation data, the absolute elevation of each pin measurement was calculated relative to the Earth Gravitational Model of 2008 (EGM2008) datum, and the elevation in 1999 was determined. We used a 400-year model spin-up period to establish the mean initial soil profile and mangrove community at the mean elevation of each group of SETs. We used mean annual water level from the Pohnpei Harbor gage to calculate the annual change in sea level that was used to drive the model. We then compared the average modeled and observed accretion rates for the 20-year period (1999–2019).

To generate spatial results, we used the model to simulate a range of initial elevations and interpolated the results of fractional species cover across the 10-m resolution DEM in 10-year intervals. Again, we used a 400-year model spin-up to develop the forest and soil profile, holding the initial elevation constant and the rate of sea-level rise to zero. The sea-level rise scenarios were simulated after the spin-up period.

We used four future sea-level rise scenarios to assess the relative vulnerability of mangrove forests around Pohnpei. Relative to 2020, the representative concentration pathway (RCP) scenarios 4.5 and 8.5 from the IPCC project 37 and 67 cm of sea-level rise, respectively, by 2100 (Church and others, 2013; fig. 8). To illustrate model sensitivity, we included a moderate scenario of 52 cm by 2100. The AR5 IPCC report did not consider potential instability of Antarctic ice sheets, so we also included a more extreme scenario by adding 50 cm to the RCP 8.5 scenario, resulting in a 117 cm sea-level increase by 2100 (fig. 7). In addition to sea-level rise, we included a constant 0.074 mm/year rate of subsidence, as estimated from deep soil-core data by J. Ellison (University of Tasmania, written commun., July 10, 2019).

Wetlands developing under larger tidal ranges tend to demonstrate increased resilience to sea-level rise through a greater capacity to build soil-surface elevations vertically (Kirwan and others, 2010), and also, they often have more time before undergoing submergence due to sea-level rise, especially if positioned at a distance inland (Lovelock and others, 2015). The El Niño–Southern Oscillation (ENSO) exerts control over interannual variation in sea level across the equatorial Pacific, such that El Niño periods lower sea levels around Pohnpei and La Niña events drive water levels higher. We fit a second-order polynomial to the 45-year record of mean sea level from Pohnpei to deduce the local trend in sea-level rise. We then fit a Weibull distribution to the residuals from the regression, which represent the recent historic influence of ENSO, storms, and long-period lunar nodal harmonics. We ran 100 Monte Carlo simulations for each sea-level rise scenario and calculated the annual average elevation and species cover area for those at greater than 15-percent cover.

As with any model, we were required to make several simplifying assumptions and omit important processes because of a lack of local information. For this effort, we focused on the above and belowground processes that control elevation and mangrove response to changes in flooding. Salinity can be a dominant driver of community composition and productivity that we chose to omit owing to a lack of field data and specific knowledge of species responses. Similarly, the macronutrients nitrogen and phosphorus are known to play an important role in limiting mangrove growth in the Caribbean (Castañeda-Moya and others, 2013); however, we left nutrients out of our model to reduce model complexity and because we lacked calibration data. To some degree, the effects of salinity and nutrients are both captured in the elevation distributions for each species, and at a regional scale, by basal area distributions derived from forest inventory plots.

By calibrating the model with soil-core accumulation rates, we are assuming that the relationships between basal area and rates of productivity and decomposition over the last 75 years are representative of future conditions. Human development and climate change could both cause changes in the environment that affect mangrove productivity, however. For example, changes in freshwater delivery or sediment loading could have a strong effect on the health and resilience of Pohnpei mangroves. Additional scenarios of changes in sedimentation or productivity could be explored to understand the sensitivity of the projections to these assumptions. Furthermore, data that identify the source of the sediment in mangrove forests across each region are needed to understand what is ultimately driving relative resilience. For example, discharge and suspended sediment monitoring of major rivers would provide valuable information about variability of sediment inputs into mangrove forests.

One of the greatest sources of uncertainty in our projections of future mangrove area is related to elevation. Without a continuous DEM of the mangrove forest, we chose to spatially extrapolate the elevation leveling transects and made broad assumptions about the slope of the forest at each site. Like similar models of wetland soils (for example, WARMER, Swanson and others, 2014), the projections are sensitive to initial elevation and the results should be interpreted with this uncertainty in mind. Additionally, there was no information about nearshore bathymetry, so our model domain only includes the mangrove extent as of 2019. It is likely that mangroves could prograde into adjacent reef habitats in some areas; however, this process is not considered in the model.

Additionally, the site-specific mean sea-level datum elevations were highly uncertain. Comparisons of short-term water-level data at two locations to the long-term water-level gage at the airport indicated there are differences in MSL around the island, which we attributed to prevailing winds and currents. Given the larger area of mangroves on the leeward southwest side of the island, it is probable there are differences in MSL or wave energy that affect mangrove establishment. Water-level monitoring around the island could ascertain the temporal and spatial nature of these differences and improve projections of mangrove sea-level rise vulnerability.