Economic damages from Hurricane Sandy attributable to sea level rise caused by anthropogenic climate change

The study addresses how much of Hurricane Sandy’s more than $60 billion in damages can be attributed specifically to sea level rise caused by anthropogenic climate change, independent of any influence on the storm’s meteorology. While attribution of cyclones themselves remains challenging, sea level rise is robustly linked to human-induced warming and raises the baseline for all coastal floods. The authors aim to quantify the fraction of Sandy’s damages attributable to the anthropogenic component of sea level rise (ASLR), excluding non-climate-driven anthropogenic factors such as land-water storage changes and local land subsidence. The work focuses on New York, New Jersey, and Connecticut, where the majority of impacts occurred, and seeks to establish a lower bound by isolating the sea-level contribution without considering potential climate-change effects on Sandy’s track or intensity.

Prior research provides increasing confidence in attributing certain extreme events (temperatures, precipitation, drought) to anthropogenic climate change, with cyclone attribution less mature. Numerous studies have documented increases in coastal flood frequency, peak heights, and damages due to rising sea levels, typically by linearly adding sea level rise to water levels without dynamic modeling of nonlinear interactions and without isolating the anthropogenic climate contribution from other factors (e.g., vertical land motion). Studies on Sandy found no clear evidence that climate change made the storm’s intensity, size, or track more likely, and projections suggest future changes in cyclone characteristics may have smaller effects on extreme sea levels in the New York Bight than sea level rise itself. The paper builds on budget-based and semi-empirical sea level attribution literature, including Kopp et al. (2016) showing anthropogenic influence on 20th-century GMSL, and on work regarding regional sea level fingerprints and ocean dynamics, as well as discussions of mid-Atlantic sea level hotspots linked to internal variability.

The analysis proceeds in three main components: attribution of anthropogenic sea level rise (ASLR), hydrodynamic flood simulation with bias correction, and damage modeling with attribution of damages. 1) Sea-level-rise attribution - Budget-based approach: Construct global and New York sea-level-rise budgets for 1900–2012 from literature-based components: Greenland and Antarctic ice-sheet mass loss, global glacier melt, thermal expansion, and land-water-storage (LWS) changes. Apply GRD fingerprints to localize components to New York and include a regional ocean-dynamics term. Develop low, central, and high attribution scenarios by assigning literature-based attributable fractions to each component (including 0–100% ranges where uncertainty is high). Compare totals with observations for consistency. - Semi-empirical approach: Update Kopp et al. (2016) semi-empirical GMSL model relating rate of sea-level change to deviations in global mean surface temperature. Generate historical sea level using observed HadCRUT4 temperatures and counterfactual sea levels under three temperature scenarios: stable (500–1800 CE mean), cooling (reversion to preindustrial cooling trend), and CMIP5-based natural-forcing-only (paired with historical full-forcing). For each scenario, sample parameter posteriors (1000 draws) and temperature uncertainties to create large ensembles of paired historical and counterfactual sea-level curves. Derive ASLR as the difference between paired curves. Downscale global semi-empirical ASLR to New York using a 0.91 ratio from the budget-based central scenario. - Total ensemble: Pool semi-empirical results across scenarios and equally weight them with central budget-based distributions by extensive random resampling to form integrated global and New York ASLR ensembles, from which summary statistics are drawn. 2) Flood modeling - Use the U.S. Army Corps of Engineers CSTORM-MS system with ADCIRC (2D depth-averaged hydrodynamic model) coupled to STWAVE (nearshore waves) and WAM (deepwater waves). Force with a meteorological reanalysis for Sandy from OceanWeather, Inc. Include riverine inflows. - Historical run uses 2012 mean sea level offset of +10 cm relative to the 1983–2001 tidal epoch at The Battery, NYC. Counterfactual runs perturb mean sea level by −4, −8, −10, −12, −14, −16, −20, and −24 cm relative to the historical run to span ASLR uncertainty. 3) Model validation and spatial bias correction - Compare modeled maximum water elevations to 456 observed high water marks and gauge data (excluding poor-quality and wave-active V zones). Initial bias ~+1.2 cm with RMSE 35.5 cm and spatially correlated errors. Apply spatially coherent bias correction to remove correlation, yielding mean bias −0.1 cm and RMSE 22.3 cm. Apply the same correction to counterfactual runs. 4) Damage and exposure analysis - Generate flood depth maps by subtracting lidar DEM land elevations (~5 m resolution) from bias-corrected water levels, masking isolated depressions via connectivity checks. Compute exposure of land, population, housing units, and integrate water volume over land at Census-block level (2010 U.S. Census), assuming uniform within-block densities and excluding wetlands for population/housing. - Estimate property damages using FEMA’s HAZUS-MH depth-damage methodology across 33 building classes with depreciated replacement costs and depth-damage curves. For each Census block, compute mean inundation depth over flooded fraction, derive per-class damages, and scale by flooded area fraction. This approach targets relative differences across scenarios rather than absolute losses. - Attribute damages to ASLR by comparing modeled damages under actual sea level versus counterfactual lower sea levels corresponding to ASLR percentiles. Compute attributable damage fractions and apply them to reported state-level repair, response, and restoration costs (NY, NJ, CT) to obtain dollar estimates. Distribute state totals to counties proportional to modeled damages. - Evaluate simplified proxies: (a) hydrodynamic simulation + damage modeling without bias correction; (b) flood volume as a damage proxy without damage modeling. Compare relative errors against the full, bias-corrected approach. Data, code, and reproducibility details are provided, including public availability of semi-empirical ensembles, hydrodynamic outputs, and block-level exposure and damage estimates.

- Anthropogenic sea level rise (ASLR) through 2012 is estimated at 10.5 cm (6.6–17.1) globally and 9.6 cm (5.6–15.6) in the New York area (total ensemble 50th, 5th–95th percentiles). - Approximately 13% (7.5–22.5%) of Hurricane Sandy’s tri-state damages are attributable to climate-mediated ASLR, equating to $8.1B ($4.7B–$14.0B) when applied to reported state costs (NY, NJ, CT). - State-level attributable damages using total ensemble ASLR: New York $4.2B ($2.4B–$6.7B), New Jersey $3.7B ($2.2B–$7.0B), Connecticut $0.18B ($0.10B–$0.30B). New York City alone: $1.5B ($0.9B–$2.5B) attributable. - Additional exposure due to ASLR: 70.6 thousand (40.4–131.0) people and 36.3 thousand (20.6–65.8) housing units in the tri-state (about 9.2% and 8.8% of those flooded, respectively). Increased flood depth across the entire flooded footprint increases volume and damages more than simple marginal spatial exposure alone. - Hydrodynamic model performance after spatial bias correction: mean bias −0.1 cm and RMSE 22.3 cm against 456 observations; baseline model RMSE was 35.5 cm with spatially correlated errors. - Simplified methods approximate full-model attribution reasonably: simulation+damage without bias correction yields −4.2% relative error; simulation-only using flood volume as proxy yields +9.9% relative error, compared to the fully bias-corrected, depth-damage approach. - Budget-based and semi-empirical ASLR estimates are broadly consistent; ensemble integrates both to provide a robust characterization. New York ASLR medians are ~87–92% of global medians across low/central/high attribution scenarios due to GRD fingerprints and component mix (e.g., reduced contribution from Greenland/glaciers locally).

The study directly addresses the research question by isolating the contribution of anthropogenic climate-driven sea level rise to Hurricane Sandy’s impacts, independent of potential climate influences on storm characteristics. By combining independent attribution methodologies (sea-level budget reconstructions and semi-empirical temperature-driven models) with high-resolution hydrodynamic flood simulations and standardized depth-damage assessment, the authors demonstrate that ASLR materially increased both the extent and severity of flooding and caused multi-billion-dollar additional damages. The agreement among approaches, the consistency with observations, and the quantified uncertainties support a robust attribution of approximately one-eighth of total damages to ASLR. The findings underscore that even absent changes in storm frequency or intensity, rising baseline sea level amplifies coastal flood impacts. The additional exposure of roughly 71,000 people and 36,000 housing units, together with the nonlinear depth-damage relationships, explains why damages scale more strongly than marginal increases in flooded area. The work also clarifies regional nuances: New York’s ASLR slightly below the global mean reflects component fingerprints and limited long-term ocean-dynamics contribution over the century, despite short-term mid-Atlantic hotspots linked to internal variability. The approach generalizes to other historical and future storms, enabling quantification of the sea-level-linked fraction of damages and contributing to improved estimates of the economic costs of climate change. A lower-bound framing (excluding any storm-track/intensity changes due to climate) suggests results may be conservative; alternative counterfactuals with cooling trends and inclusion of pre-1900 ASLR would increase attributed fractions. The analysis highlights that unaccounted indirect economic effects and any nonlinear macroeconomic responses could further elevate total climate-linked costs, though likely modestly for the small differences between actual and counterfactual floods explored here.

Human-driven sea level rise substantially increased the damages and exposure from Hurricane Sandy. Using multiple independent ASLR attribution methods integrated into an ensemble, coupled with validated hydrodynamic flood modeling and standardized damage assessment, the study attributes about 13% of tri-state damages—approximately $8.1B—to anthropogenic sea level rise, along with tens of thousands of additional people and housing units exposed. The methodology provides a transparent, generalizable framework to quantify the sea-level-linked share of impacts for other coastal storms, offering an important tool for climate impact accounting and for informing adaptation and risk management. Future research could incorporate indirect and long-term economic effects, refine local subsidence and ocean-dynamics components, explore higher-resolution building inventories and vulnerability functions, evaluate alternative counterfactual climate baselines (e.g., cooling), and extend attribution to include potential climate influences on storm characteristics.

- The analysis isolates sea level rise and does not evaluate potential climate-related changes to storm track, size, or intensity; thus, results represent a lower bound of climate influence on damages. - Potential mismatch between observed and modeled/budgeted New York sea-level totals may reflect unmodeled local land subsidence or short-term ocean-dynamics variability, possibly biasing ASLR and damages low. - Indirect and long-term economic impacts (e.g., sectoral disruptions, input-output effects) are not included; prior studies suggest these can be nonlinearly related to direct losses. - Damage modeling via HAZUS-MH emphasizes depth-damage relationships and does not fully capture non-depth mechanisms (debris, contamination, erosion, flow velocity) beyond adjusted curves in wave zones; building distributions within blocks are assumed uniform. - Hydrodynamic modeling uses a 2D depth-averaged formulation (ADCIRC 2D), which may not capture certain baroclinic processes; although spatial bias correction reduces errors, residual uncertainties remain. - The downscaling of semi-empirical global ASLR to New York using a fixed 0.91 factor relies on the budget-based relationship and slightly reduces methodological independence. - Counterfactual scenarios primarily assume stable temperatures; adopting a cooling counterfactual and including pre-1900 ASLR would likely increase attributed ASLR and damages.