More meteorological events that drive compound coastal flooding are projected under climate change

A large share of the global population inhabits low-lying coastal zones exposed to flooding from both elevated sea levels and intense precipitation or river discharge. When these two hazards occur together or in close succession, impacts can be larger than from either hazard alone, as shown by recent compound flooding events in Australia, Italy, Ireland, and the UK. Compound flooding can occur when high meteorological tides (storm surge and wave setup) impede drainage of inland runoff, or when precipitation coincides with elevated sea levels, amplifying coastal inundation. While present-day compound flood hazard has been studied in various regions, global assessments have been limited by observational coverage of sea level and by a lack of comprehensive treatment of future changes in the meteorological drivers and their dependence. Sea-level rise will raise baseline water levels, but climate change will also modify precipitation extremes, meteorological tides, and their interplay. This study investigates, at the global scale, how the concurrence probability of extreme meteorological tide and precipitation events may change under high greenhouse gas emissions, identifies the physical drivers and seasonality of present-day co-occurrence, quantifies projected changes by late century, attributes these changes to individual drivers and their dependence, and assesses associated uncertainties.

Prior research has demonstrated increased risks from compound flooding when storm surge coincides with heavy rainfall, with case studies and regional to continental analyses for the United States, Europe, and Australia. Observational limitations have historically constrained global coastal analyses, but advances in ocean modeling now provide continuous global sea level time series that, together with precipitation or discharge data, enable broader assessments. Studies highlight that storm surges exacerbate flood levels in many river deltas, and that sea-level rise will increase extreme sea levels and compound flood hazard. Extreme precipitation is projected to intensify with warming due to higher atmospheric moisture, while storm characteristics affecting meteorological tides may change regionally. Previous European-focused work shows increased probability of compound flooding under anthropogenic climate change, but a comprehensive global assessment of the meteorological drivers’ concurrence and its future evolution has been lacking. This study builds on copula-based multivariate extreme value techniques used in prior compound hazard analyses and leverages cyclone tracking literature to relate co-occurrence to storm tracks.

Data and variables: Meteorological tide was defined as storm surge plus 0.2 times significant wave height (Hs), derived from ocean model simulations: storm surges from D-FLOW Flexible Mesh (forced by 6-hourly winds and pressure) and waves from Wavewatch III (forced by 6-hourly winds). Sea-level data are available approximately every 100 km along the global coastline. For the historical period, ocean models were forced by ERA-Interim reanalysis; for projections they were driven by six CMIP5 GCMs under RCP8.5 (ACCESS1-0, ACCESS1-3, GFDL-ESM2M, GFDL-ESM2G, CSIRO-Mk3-6-0, EC-EARTH). GFDL-ESM2G was excluded for the Black Sea and Red Sea due to surge model instabilities. Tropical cyclone effects were enhanced in the reanalysis-based dataset by dynamic downscaling for surges and satellite-based corrections for waves, but such refinement was not feasible for CMIP5-based simulations. Precipitation time series were taken from ERA-Interim and the CMIP5 models at the coastal grid point nearest each sea-level location and aggregated over 3-day windows to better capture rainfall-driven runoff relevant to small-to-medium coastal catchments and pluvial flooding. High-latitude regions with snowfall require cautious interpretation because snow does not directly translate to immediate flooding. The analysis does not aim to represent long-river estuaries (catchments ≥ 5–10 × 10^3 km^2), where inland processes dominate.

Definition of extremes and joint return periods: Extremes of meteorological tide and aggregated precipitation were defined at levels that occur on average once per year in each model’s present climate. Bivariate dependence between the two variables was modeled using parametric copulas fit to pairs where both variables exceeded high thresholds (generally the 95th percentiles), with marginal tails modeled by Generalized Pareto Distributions. Joint return periods focused on the AND-type definition at the 99.7th percentiles of each variable, yielding the expected waiting time for concurrent exceedance events. Copula families considered included Gaussian, t, Clayton, Gumbel, Frank, Joe, BB1, BB6, BB7, and BB8; the best family was selected using the Akaike Information Criterion. Fits were performed via maximum likelihood (R packages VineCopula, ismev, eva), and goodness-of-fit was evaluated using Cramér–von Mises criteria for both copulas and marginals.

Cyclone diagnostics and seasonality: Extratropical cyclone tracks were identified using the TRACK algorithm applied to 850 hPa relative vorticity from ERA-Interim. Tropical cyclone tracks were obtained from IBTrACS. Track densities were computed with spherical kernel density estimators. The peak month and the length of the season during which 90% of concurrent extremes occur were derived to characterize seasonality and the role of tropical and extratropical cyclones.

Time periods and change metrics: Present-day concurrence was assessed from ERA-Interim (1980–2014). Future changes were computed by comparing 2070–2099 to a baseline of 1970–2004 using the six CMIP5 models under RCP8.5. Changes in joint return period were expressed as percentage change ΔT = 100 × (Tfut − Tpres)/Tpres and translated to probability changes via ΔP(%) = −100 × ΔT(%) / (ΔT(%) + 100). A change was deemed robust where the multi-model median change lay outside the ERA-Interim-based 95% natural variability range and at least five of six models agreed on the sign.

Natural variability estimation: Present-day natural variability of joint return periods was estimated via bootstrap resampling of interannual variability. For each location, 700 surrogate bivariate time series of equal length were constructed by randomly sampling calendar years (preserving autocorrelation), and 95% confidence intervals were taken from the 2.5th–97.5th percentiles of the resulting return periods.

Attribution of drivers and uncertainty partitioning: To attribute changes to (1) precipitation marginal changes, (2) meteorological tide marginal changes, and (3) changes in dependence structure, three partial-change experiments were constructed by transforming variables to impose future marginals with present dependence (cases 1–2) or future dependence with present marginals (case 3), and recomputing joint return periods. The total change is not the sum of partial changes due to nonlinearity. Uncertainty contributions from each driver were quantified using symmetrized percentage changes in return periods across the six models; intermodel spread (difference between second highest and second lowest values) was computed for each driver, aggregated by IPCC regions, and normalized to obtain relative contributions to total uncertainty.

Assumptions and exclusions: Locations above the Polar Circle affected by ice processes were excluded. Interactions among sea-level components (surge, waves, astronomical tide, and mean sea level) were assumed independent, an accepted approximation at the global scale given current modeling capacity. Analyses used quantiles, so no bias correction was applied.

• Present-day concurrence: The global median joint return period for concurrent extreme precipitation and meteorological tide is 17 years (vs 365 years under independence), implying strong dependence and roughly 20-fold higher likelihood of co-occurrence than if independent. Concurrence is highest in regions with frequent tropical or extratropical cyclones (e.g., United States, eastern Central America, Madagascar, Europe, northern Africa, northern/eastern Australia, India, northern Southeast Asia, China, Japan), with joint extremes every 4–8 years. Northern Hemisphere coasts show higher concurrence (median 15 years) than the Southern Hemisphere (23 years). Low concurrence typically aligns with differing seasons of extremes between inland and coastal drivers in many tropical regions.
• Seasonality and drivers: Concurrent extremes peak during the tropical cyclone season in the tropics and in autumn–winter at midlatitudes when extratropical cyclone activity is highest. The longest season for concurrent extremes occurs along the eastern US coast due to both TCs and ETCs occurring in different seasons.
• Future changes (RCP8.5, 2070–2099 vs 1970–2004): The concurrence probability increases along about 60% of the global coastline. The global median change in joint return period is −20%, corresponding to a 26% increase in probability of concurrent extremes by 2100. Above 40°N, joint events become on average 2.6 times more frequent (median ΔT = −61%). Regions with strong increases include northern North America, northern Europe, northern Mediterranean, Russia, Japan, Korea, China, Bangladesh, and around Cameroon; also parts of northwestern South America, southern Chile, northern Australia, Gulf of Carpentaria, and New Zealand. Significant decreases occur in smaller regions: northwestern Africa, southern Spain, western South Africa, eastern Madagascar, southwestern Australia, and Central Chile. Many tropical/subtropical areas show high model disagreement.
• Attribution of change: Changes in precipitation extremes dominate the increase in concurrence (global median ΔT_prec = −25%; increasing concurrence along 83% of coasts) and contribute 77% of the global signal. Changes in meteorological tides have a weaker and regionally mixed effect (global median ΔT_met.tide = +7% implying reduced concurrence along 61% of coasts), contributing 20% of the global signal, with patterns consistent with projected storm track changes (e.g., increased tides in northern Europe and the Baltic; decreases in the Mediterranean). Changes in dependence have small net global effect (ΔT_dep ≈ −1%; about balanced coast fractions increasing vs decreasing) and contribute 3% of the global signal, but can be locally important.
• Uncertainty: The largest contributor to projection uncertainty is the future dependence between precipitation and meteorological tide (≈55% of total uncertainty globally). Uncertainty from precipitation and meteorological tide changes contributes ≈24% and ≈21%, respectively. Regions with strong tropical cyclone influence (e.g., Central America) show high dependence-related disagreement.
• Physical interpretation: Thermodynamic moistening generally intensifies precipitation extremes, increasing concurrence, while circulation changes modulate this effect regionally. Projected storm track shifts help explain regional patterns in meteorological tide changes (e.g., increases in northern Europe and western Canada; decreases in the Mediterranean and parts of the subtropics).

The study addresses how climate change alters the meteorological co-occurrence of extreme precipitation and elevated coastal water levels that can trigger compound flooding. Present-day analysis reveals strong dependence between inland and coastal meteorological extremes, tightly linked to cyclone activity and seasonality. Future projections indicate that, beyond mean sea-level rise, the meteorological conditions conducive to compound flooding will become substantially more frequent globally—especially at latitudes above 40°N—primarily due to intensifying precipitation extremes. Regional circulation changes, including shifts in storm tracks and atmospheric stability, modulate both precipitation and meteorological tides, producing spatial heterogeneity with some subtropical regions experiencing reduced concurrence. The dominance of dependence-related uncertainty highlights the need to better understand and model how joint behavior of extremes may evolve, particularly in cyclone-prone regions. These findings imply that coastal risk assessments and adaptation planning must incorporate changing joint probabilities of inland and coastal drivers, not only mean SLR or marginal extremes, to avoid underestimating compound flood risk and to design robust emergency response and protection strategies.

This work provides the first global assessment of how climate change affects the meteorological concurrence of extreme precipitation and meteorological tides relevant to compound coastal flooding. Under a high-emissions scenario, the joint triggering conditions become about one-quarter more probable globally by 2100, and more than 2.5 times as frequent north of 40°N. Changes are driven mainly by intensifying precipitation extremes, with meteorological tide changes and evolving dependence playing secondary but regionally important roles. Uncertainty is dominated by the dependence between drivers. The results emphasize the need for adaptation strategies that integrate changing joint hazards alongside sea-level rise and total extreme sea level projections. Future research should include higher-resolution and large-ensemble simulations (to better resolve tropical cyclones and circulation changes), localized studies for identified hotspots, explicit treatment of interactions among sea-level components, and improved representation of dependence dynamics and catchment processes, especially for larger river systems and high-latitude regions.

• Model resolution and tropical cyclones: CMIP5-driven projections do not include the reanalysis-based tropical cyclone downscaling and wave corrections used for the historical dataset, potentially underrepresenting TC-driven meteorological tides in projections; tropical regions should be interpreted with caution.
• Sea-level component interactions: Interactions among surge, waves, astronomical tides, and mean sea-level rise are assumed independent, a necessary simplification at global scale.
• Data and thresholds: Analyses use percentile-based definitions and do not apply bias correction; while this mitigates absolute bias effects, it may not capture all process nuances. Joint return period estimates rely on high-threshold selections and copula choices.
• Geographic coverage: Polar regions affected by sea ice are excluded. High-latitude snowfall complicates the direct link to immediate flooding.
• Hydrological scope: The approach targets pluvial and small-to-medium coastal catchments; it does not represent compound flooding dynamics in long river estuaries where inland processes dominate.
• Natural variability and ensemble size: The six-model CMIP5 ensemble limits characterization of internal variability, especially for dependence; larger initial-condition ensembles would better separate anthropogenic signals from natural variability.