Accounting for tropical cyclones more than doubles the global population exposed to low-probability coastal flooding

Coastal storm surges generated by tropical cyclones (TCs) and extratropical cyclones (ETCs) can cause extreme sea levels, particularly when coinciding with high tides. With roughly 100 million people living below current high tide lines, accurately estimating storm tide return periods (RPs) is critical for hazard and risk assessments and for designing coastal protection. Existing global RP datasets have known inaccuracies in TC-prone regions due to coarse meteorological forcing that under-resolves TC intensity and limited record lengths that poorly sample rare TC events. While ERA5 improves surge modeling for many events, its several-decade length remains insufficient for robust estimation of high RPs in TC regions. The research question addressed is how to produce robust global storm tide RPs that fully account for low-probability TCs and to quantify the impact on global exposure to coastal flooding when TCs are comprehensively included.

Prior global studies used reanalyses (e.g., ERA-based) to estimate storm tide RPs and flood risk but underperform in TC regions due to under-resolved TC intensity and short records. Regional approaches have extended records using synthetic TC tracks to empirically estimate low exceedance probabilities without relying on unstable extreme value extrapolations, with applications in Australia, the United States (including New York and Tampa), and other locales. New global synthetic TC datasets (e.g., STORM and others) now enable estimation of up to 10,000-year RP storm tides with higher statistical confidence. However, a fully global storm tide RP dataset that combines synthetic TCs with ETC-driven surges had not been produced prior to this study.

The study develops COAST-RP, a global dataset of storm tide return periods combining TC- and ETC-induced surge levels with tides and employing an empirical extreme value analysis. - Hydrodynamic model: GTSMv3.0 (global depth-averaged Delft3D FM), variable mesh (~25 km deep ocean to 2.5 km along coasts; 1.25 km in Europe). Storm surges forced by 10-m winds (U10) and mean sea level pressure. Tides simulated using 60 astronomical constituents including self-attraction and loading; 19-year tidal simulation (1999–2017) to include the 18.6-year nodal cycle. Output at 23,226 coastal locations (~25 km spacing); TC computations at 10,641 TC-prone locations. - TC surge modeling: Synthetic TC tracks from STORM (derived from IBTRACS 1980–2017) representing 10,000 years; simulate 3,000 years (sufficient for RP1000). Filtering: only TCs within 750 km of land and track segments within 1000 km of coast; multiple TCs placed and simulated simultaneously with a minimum 2000 km separation using a new placing algorithm. Wind/pressure fields generated using the Holland parametric model on a polar grid (36 radials, 375 arcs, 750 km radius); applied standard conversions and drag formulation (Garratt; Cd capped at 2.5×10^-3) and parameters (e.g., SWRF 0.85, 1-min to 10-min conversion 0.915, inflow angle ramping to 25°). Output: 10-minute surge time series for 3,000 TC-years. - Bias correction for TC surges: Compare STORM-based RPs against IBTRACS-forced simulations (2710 observed TCs, 1980–2017). Using bootstrapped 90% confidence intervals (599 repetitions), identify locations where observed RP(2–19y) averages fall outside STORM’s 5th–95th percentile. Apply multiplicative correction using the ratio of STORM’s 5th or 95th percentile to IBTRACS average. Notable corrections: NE China (~0.7), N Australia (~1.4), N South America (~0.6). - ETC surge modeling: Use ERA5 (0.25°, hourly) for 1980–2017 from prior work; remove TC-induced surge periods to avoid double counting by excluding times within 250 km (or 5×Rmax if Rmax>50 km) and up to 48 h after TC track end per IBTRACS. Apply constant Charnock parameter 0.04189. Output: 38 years of 10-minute ETC surge time series. - Extreme value analysis and tide combination: For TCs, identify independent surge peaks (≥3 days apart); for each, combine with 72-h tidal time series randomly sampled from the TC genesis month (from 19 tidal years), repeat 1000 times and retain maxima to build empirical exceedance. For ETC, combine full-year ETC surge with year-long tides (randomly phase-shifted by up to 30 days) and extract independent storm tide peaks (≥72 h apart) above a threshold; repeat 1000 times to emulate ~38,000 years. Compute empirical exceedance probabilities using Weibull plotting positions and derive RPs. Combine TC and ETC storm tide RPs probabilistically: for a storm tide level x, RP(x) = 1 / (1/RPTC + 1/RPETC). Where no TC influence exists, RP = RPETC. - Validation: Compare COAST-RP against regional studies in Australia and the U.S.; report RMSE, bias, and rank correlation; examine crossover return periods (TC vs ETC dominance) at selected sites. - Inundation and exposure modeling: Static inundation with water level attenuation (0.5 m km^-1) on MERIT DEM (30-arcsecond, ~1 km). Inundate connected cells below attenuated water level. Exposed population from GPWv4 (2020; UN-adjusted), with sensitivity test using WorldPop (resulting in ~10% lower exposure). No coastal protection included.

- Inclusion of tropical cyclones more than doubles estimated global exposure to low-probability coastal flooding: 77.8 million people exposed to ETC-only RP1000 floods increases to 191.6 million when TCs are included (about 1.0% vs 2.5% of global population). - The COAST-RP-based RP1000 flood map yields a 31% higher global exposure estimate compared to an earlier dataset (Aqueduct/GTSR and IBTRACS-based TCs), indicating prior underestimation due to incomplete TC representation. - Spatial patterns: RP1000 storm tide levels exceed 5.0 m at ~3% of output locations, notably over wide/shallow continental shelves. TC dominance at high RPs in the Bay of Bengal and Gulf of Mexico; large tidal range and ETC surges dominate parts of Europe and Canada. Removing TC influence from ERA5 reduces ETC RP25 surges by up to ~1.0 m in eastern China, northern Australia, and U.S. Gulf and east coasts; relative reductions exceed 50% on many tropical islands. - Crossover RPs (where TC storm tides exceed ETC): e.g., New York transitions near RP45 (vs ~RP60 in a regional study), Cairns transitions near RP391; low (<10 years) along parts of Vietnam, SE China, N Philippines, and some Pacific/Caribbean islands. - Country exposure at RP1000 (absolute share of global exposure): Bangladesh (~24%), India (~19%), China (~14%), Vietnam (~7%). Without TCs, largest relative contributors shift to China (17%), Netherlands (12%), Vietnam (10%), India (9%). TC contribution to national exposure exceeds 90% in Puerto Rico, Belize, Cuba, Mexico, and Bangladesh. Top countries by absolute exposure: Bangladesh (46.3M; 91% TC), India (37.0M; 80% TC), China (25.9M; 49% TC), Vietnam (13.6M; 46% TC), Netherlands (9.4M; 0% TC), etc. By relative exposure, Netherlands (55.9%; 0% TC), Bahamas (43.2%; 62% TC), Bangladesh (28.4%; 91% TC), Myanmar (15.7%; 84% TC), etc. - Validation: Australia (600 locations) RP1000 RMSE 0.60 m, mean relative bias +2.9%, Spearman’s rho 0.94. U.S. (23 locations, TC-only comparison) RP1000 RMSE 0.90 m, bias +5.8%, rho 0.76; all COAST-RP RP1000 values fall within reported 5th–95th percentiles. Differences in some regions attributed to model resolution and tide contributions. - TC vs ETC RP curve behavior: TC storm tide RP curves generally show steeper tails than ETC, emphasizing TC importance at high return periods even where TCs are relatively rare.

By constructing a global, empirically derived storm tide RP dataset that combines long synthetic TC surges with ETC surges and realistic tides, the study addresses the core challenge of under-sampling low-probability TC events in prior global models. The results demonstrate that TCs substantially elevate high-RP storm tide levels and associated exposure, particularly in densely populated deltaic regions and along coasts with wide continental shelves. The identification of crossover RPs where TCs surpass ETCs clarifies where and at what design levels coastal management must explicitly consider TC hazards. Validation against regional studies shows good agreement, supporting the robustness of the methodology. The substantial increase (31%) in globally exposed population for RP1000 relative to prior global products underscores the policy relevance: flood risk has been materially underestimated in many TC-prone regions, with implications for insurance, infrastructure design standards, and adaptation planning (e.g., U.S. 100-year flood zoning along the northeast).

The study introduces COAST-RP, a global storm tide return period dataset that, for the first time at global scale, fully incorporates low-probability tropical cyclones by leveraging synthetic TC tracks and empirical extreme value analysis. It shows that accounting for TCs more than doubles global exposure to RP1000 coastal flooding (from ~78M to ~192M people) and increases global RP1000 exposure estimates by 31% compared to earlier datasets. The framework provides improved inputs for broad-scale flood risk assessments and adaptation planning. Future work should extend this approach to future climate scenarios including sea-level rise and projected changes in TC and ETC characteristics, and enhance process representation (e.g., wave setup, tide–surge interaction), wind field modeling for TCs, and reduce biases in synthetic TC generation.

- Synthetic TC dataset (STORM) is based on statistical resampling within 5°×5° boxes and does not enforce all physical constraints on track evolution; can over/under-represent TC intensity and tracks locally (e.g., NE China, N Australia, N South America), necessitating bias correction. - TC wind fields use the Holland parametric model assuming near-symmetric structure with limited asymmetry representation; may underestimate surges away from the eyewall and miss structural complexities. - Process simplifications: tides and surges modeled separately and combined probabilistically; no explicit tide–surge interaction, wave setup, coastally trapped waves, or non-wind-driven mean sea level variability included. These omissions can cause local over- or underestimation (e.g., tide–surge interaction often reduces extremes; wave setup can increase nearshore levels). - GTSM is 2D barotropic; mean sea level variability (steric and circulation-driven) up to ~10 cm not included. - Inundation modeling is static with a uniform attenuation factor (0.5 m km^-1), does not include defenses, fluvial backwater, or full hydrodynamics; may overestimate exposure where strong protection exists, and underestimate inland propagation in some deltas/estuaries. - Population datasets and DEM uncertainties affect exposure estimates; sensitivity to population product noted (~10% lower using WorldPop).