Assessing population exposure to coastal flooding due to sea level rise

The study addresses how different definitions of the spatial zone of coastal flooding exposure under sea-level rise lead to divergent estimates of the number of people affected and the implied timing of impacts. Prior estimates vary widely due to the choice of spatial zone (e.g., high-tide line, 100-year floodplain, LECZ) and its implicit temporal horizon, as well as methodological and data differences. The authors aim to: (1) compare exposure estimates across the three most commonly used zones; (2) integrate spatial and temporal dimensions via an Expected Annual Exposure (EAE) framework that captures the annual probability of flooding; and (3) apply this integrated approach to the coastal United States to produce more interpretable exposure metrics for adaptation planning.

A systematic review identified 46 studies estimating populations affected by SLR using varied spatial zones ranging from mean sea level to the LECZ. The three most common zones are: (i) specified levels of SLR (future high-tide line; n=20), (ii) coastal floodplains defined by return-period floods such as the 100-year floodplain (n=17), and (iii) the LECZ (<10 m above sea level; n=11). While 43% of studies used more than one zone, few examined all three. Prior work shows these zones imply different temporal horizons—from near-term nuisance flooding to long-term permanent inundation—and different hazard sets (e.g., salinization, storm surge). The literature has rarely integrated across zones to represent a continuum of flood exposure probabilities, and EAE has been relatively new and seldom combined with future population projections.

Overview: The authors combine a small-area demographic projection model with a flood risk probability model to compute population exposure under three common spatial zones (MHHW/future high-tide line, 100-year floodplain, LECZ) and to estimate Expected Annual Exposure (EAE) from 2000–2100 under RCPs 2.6, 4.5, and 8.5 and all five SSPs, with primary projections shown for SSP2-RCP4.5. Small-area demographic projections: Using a proportional fitting algorithm, the authors create spatiotemporally consistent Census Block Groups (CBGs) for 1940–2010 and project 2010–2100 via mixed linear/exponential methods. Inputs include Year Structure Built, group quarters counts (GQ), and persons-per-household (PPHU) from the 2013–2018 ACS, plus historical county housing unit counts. Population P_ijt for block group i, county j, time t equals H_ijt*PPHU_ijt + GQ_ijt, with H_ijt estimated historically by proportional allocation of county housing to block groups and projected forward to 2100. Projections are controlled to SSPs to align with national scenarios; validations indicate good fit in prior work. Elevation modeling and inundation surfaces: Airborne lidar-derived DEMs (NOAA), supplemented by USGS topobathymetric DEMs for Louisiana and the National Elevation Dataset, are referenced to NAVD88 and converted to MHHW via VDatum. A refined “bathtub” approach generates inundation surfaces from 0–10 m above MHHW at 0.25 m increments, noting each pixel’s minimum inundation threshold. Levee data are incorporated (FEMA/USACE), and connected components analysis removes isolated regions to enforce ocean connectivity. Pixels identified as isolated at lower thresholds are adjusted upward to the connected-index threshold to ensure consistent connectivity in subsequent analyses. SLR projections and flood probability: Probabilistic, site-specific SLR projections (Kopp et al. 2014) with local vertical land motion are sampled (10,000 Monte Carlo draws per tide gauge per year) for RCPs 2.6/4.5/8.5. Extreme water levels are characterized by fitting generalized Pareto distributions to NOAA tide station records (≥30 years through 2013), assuming stationary surge statistics. For each pixel, the nearest tide station’s return level curve is applied in a bathtub sense. The annual probability of exceedance of elevation E in year y is estimated by averaging over the SLR samples: P(H ≥ E | y) = (1/10000) Σ_j P(H + SLR_j(y) ≥ E | baseline). Probabilities are computed by decade from 2000–2100 for elevations 0–10 m (0.1 m increments) and stored in lookup tables to build spatial probability rasters for each RCP and year. Exposure computation: For each CBG, population exposure is estimated by overlaying inundation surfaces with per-pixel population density (dry land only per National Wetlands Inventory). Spatial zones are defined as: (a) Specified SLR level (MHHW/future high-tide line): pixels below SLR(y); (b) LECZ: pixels below 10 m + SLR(y); (c) 100-year floodplain: pixels where annual exceedance probability >1% (i.e., P(H ≥ E | y) ≥ 0.01; implemented as thresholding the probability surface to capture areas with return period ≤100 years). EAE is defined as the expected number of people below the maximum local storm surge height in a given year, computed by integrating per-pixel probabilities P(H ≥ E(lat,lon) | y) times per-pixel population density and summing within CBGs. Assumptions and computational notes: The bathtub approach does not model waves, rainfall, or hydrodynamics; wave attenuation and time-to-inundation are not represented, which may lead to overestimation in some contexts. Tide station spacing was previously found adequate for EAE in the US, but estimates near the 1% threshold may be sensitive. Adaptation measures (e.g., defenses, retreat) are not modeled; exposure is interpreted as potential absent adaptation.

- Baseline and recent changes: In 2000, the US EAE was just over 600,000 people; about 150,000 lived below the high-tide line; and 2.4 million lived in the 100-year floodplain. By 2020, EAE increased ~60% to ~980,000; population below the high-tide line increased ~60% to ~240,000; and population in the 100-year floodplain rose ~45% to ~3.5 million, despite only ~25% growth in the total coastal population. - Projections under SSP2–RCP4.5 (no adaptation): Between 2020 and 2100, EAE is projected to rise ~325% to 4.1 million (2.3–6.4 M); population below the high-tide line increases ~435% to 1.2 million (0.3–5.1 M); and population in the 100-year floodplain increases ~160% to 9.0 million (3.4–22.3 M). Total population in 406 coastal counties grows ~40% (from ~133 M to ~190 M), indicating flood exposure grows much faster than population. - Uneven exposure: Exposure growth is heterogeneous across counties. In 2000, only two counties had >100,000 people in the 100-year floodplain; by 2100, nine counties are projected to have >100,000 people annually exposed (EAE) and 13 counties >100,000 in the 100-year floodplain. In all counties and zones, exposure grows faster than population. - Metric dependence: Counties similar on one metric (e.g., share below MHHW) can differ substantially on others (EAE, RL100, LECZ). For example, Currituck County, NC and Orange County, TX both have 100% in the LECZ but show different profiles across the other metrics; McIntosh County, GA versus Franklin County, FL exhibits similarity on several metrics but large differences on another. Such discrepancies have implications for targeted adaptation strategies. - Contemporary impacts: Nearly 1 million US coastal residents are presently exposed to an annual flood event, underscoring that SLR-related flooding is already occurring.

Relying on a single spatial zone (e.g., only high-tide line, only 100-year floodplain, or only LECZ) obscures the heterogeneity of SLR hazards and timing, potentially misleading adaptation decisions. Integrating across zones with EAE provides a directly interpretable measure of the population expected to be flooded in any given year, capturing the continuum from frequent nuisance tides to rare extremes. The analysis shows that exposure increases faster than coastal population growth and varies widely across places and metrics, meaning areas with similar permanent inundation risk may face very different annual flood risks or floodplain exposure over time. These insights help prioritize adaptation investments (protection, accommodation, or retreat) and planning horizons (near-term recurrent flooding vs. long-term inundation), and suggest the need for multiple-zone assessments to identify which segment of the population is changing most rapidly in exposure. The study also highlights equity considerations: adaptation strategies must account for social differences that shape vulnerability under similar exposure levels and consider impacts beyond the 100-year floodplain.

The paper demonstrates that: (1) the choice of spatial exposure zone strongly influences population exposure estimates and implied timing of SLR impacts; (2) SLR-driven exposure is increasing much faster than population growth in US coastal counties; and (3) nearly 1 million coastal residents are already exposed to annual flooding. By integrating across spatial zones, the EAE framework offers a unified, annually interpretable exposure metric that supports more nuanced, locally relevant adaptation planning. Future research should incorporate social heterogeneity and equity into exposure analyses, expand to include non-flood SLR hazards, and explore dynamic adaptation pathways to understand how protective measures, accommodation, and retreat alter exposure trajectories.

- Hydrodynamic simplifications: The bathtub model does not simulate waves, rainfall, drainage, or time-dependent flood propagation, likely overestimating exposure in some settings and especially affecting rare-event thresholds. - Stationarity of surge: Storm surge statistics are assumed stationary; potential changes in storm climatology are not modeled. - Tide-gauge interpolation: Applying tide-station return levels over space may introduce uncertainty; estimates near the 1% AEP threshold are particularly sensitive to local factors. - Adaptation excluded: No coastal adaptation (e.g., levees, seawalls, managed retreat) is modeled; estimates reflect exposure absent future protective measures or land-use change responses. - Hazard scope: Primary focus is on flooding; extended SLR impacts (e.g., saltwater intrusion, erosion, socioeconomic ripple effects) are not quantified. - Data constraints: Some elevation and probability inputs are licensed and not fully public; although this does not affect methodology, it may limit complete reproducibility by external users.