Extensive hydrodynamic model validation
To provide a comprehensive validation of the modeled surge-tide-wave dynamics during Sandy, this study made use of all available field data from an extensive network of coastal water level gauges and wave buoys in the study region, as well as hundreds of HWMs and many rapid deployment surge sensors (SI Fig. S1). The good agreements between simulated and observed water levels and waves were quantified in terms of the root-mean-square error (RMSE) and correlation coefficient (CORR).
Storm tide
Available data included those from hundreds of permanent and temporary surge sensors from NOAA (SI Fig. S2), Hudson River Environmental Conditions Observing System (HRECOS), and USGS. Statistics (SI Table S1) showed excellent agreement between the time series of simulated and measured storm tide data. The averaged RMSE at USGS temporary sites was 0.20 m and the averaged CORR was 0.94. CH3D-SSMS successfully reproduced the surge and tides with high confidence in not only the open ocean but also the complex estuarine system during Sandy (SI Fig. S3) with a maximum coastal storm tide at NY Bight of 3.36 m and NJ coast of 3.27 m.
High water marks and inundation
A total of 526 out of 653 independent HWM locations were located within the study region, most of which clustered in NJ and NY coastal areas. 83.7% (440 out of 526) of HWMs were captured by the CH3D-SSMS model grid and compared with the surveyed values. The model results had 0.33 m RMSE and 0.87 CORR (SI Fig. S4). When only “good” and “excellent” HWMs (rated by USGS) were used, the RMSE dropped to 0.30 m, and CORR increased to 0.90. The noticeable data disagreements (Fig. SI 4) were caused by the inconsistency of surveyed data.
Wave
During Sandy, four wave-buoys within the domain recorded the wave data: two at the apex of NY Bight, and another two in Long Island Sound (LIS). The significant wave height (H_{sig}) and peak wave period (T_{p}) simulated by the Simulating Waves Nearshore (SWAN) model in the CH3D-SSMS were compared with measured data and Wave Watch III (WW3) operational run results (SI Fig. S5)25,26. SWAN more accurately captured the evolution of (H_{sig}) in NY Bight and LIS. Maximum significant wave height over land was as high as 2.17 m at NY Bight and 2.60 m along NJ coast.
Impact of wetland on surge and wave during super storm sandy
To estimate the value of the coastal wetlands in reducing flood, we calculated the following four metrics for inundation: the Average Inundation Height ((AIH)), the Maximum Inundation Height ((MIH)), the Total Inundation Area ((TIA)), and the Total Inundation Volume ((TIV)) with and without wetlands. The (TIA) and (TIV) are defined as
$$ TIA = iint_{{{text{Landward}};{text{area}}}} {dxdy} , $$
(1)
$$ TIV = iint_{{{text{Landward}};{text{area}}}} [H_{max} left( {x,y} right) – H_{0} left( {x,y} right)]dxdy, $$
(2)
where (H_{max} left( {x,y} right)) and (H_{0} left( {x,y} right)) are the maximum water level and the land elevation at land cells (left( {x,y} right)), respectively11,12. The wave analysis was carried out by calculating the average wave height ((AWH)), the maximum wave height ((MWH)), and the total wave energy ((TWE)) which is defined as
$$ TWE = iint_{{{text{Landward}};{text{area}}}} [frac{1}{8}rho_{w} g{ }left( {H_{{rms,max{ }}} } right)^{2} ]dxdy, $$
(3)
where (H_{rms max}) is the maximum root-mean-square wave height over the flooded land. The wave energy, instead of wave height, is more directly related to the wave-induced structure loss.
We define the relative inundation reduction ((RIR)) as the difference in (TIV) (value without wetland minus value with wetland), divided by the (TIV) with wetland. The relative wave reduction ((RWR)) is defined accordingly using the (TWE). The inundation and wave analysis were carried out at the regional level (SI Table S2) and zip-code resolution (Figs. 1, S6, S7).
Zip-code resolution wetland’s effect on (TIV) and (TWE) during Sandy in 2012. Map showing zip-code resolution avoidance in (A) (TIV) and (B) (TWE) during Sandy without wetlands, as a percentage of those for the with-wetland scenario. Dark red values show zip-code with the most wetland benefit while dark blue areas have the least wetland benefit. Negative values indicate that the presence of wetland would increase (TIV)/(TWE) and positive values indicate that wetland would lower (TIV)/(TWE). The map is produced using ESRI ArcGIS Pro 2.7 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview).
If the wetlands were absent during Sandy, the (RIR) of the entire model domain would have been 4% and the (RWR) would have been 19%. Breaking the entire model domain into 6 regions (Table S2): South NJ (SNJ), Central NJ (CNJ), North NJ (NNJ), CT, Long Island (LI), and Mainland NY (MNY), it was found that the (RIR) during sandy was low (< 10%) for all regions, while the (RWR) was moderate for CNJ and SNJ (10–25%) where the wetland cover was high, and low for the other regions. Although CT ranked 3rd in wetland cover, its (RWR) ranked 4th during Sandy due to Sandy’s landfall location and the sheltering of waves provided by LI which ranked 3rd.
Model simulations show that, during Sandy, if the wetland in the entire region were completely made up of woody wetland, TIV and TWE would decrease by 9.1% and 12.4%, respectively, while increase by 1.2% and 2.2%, respectively, if the wetland were composed of marsh only. (RIR) and (RWR) for the all-marsh wetland are 6.5% and 19.1%, respectively, while those for the all-woody-wetland are 17.6% and 36.8%, respectively. These results demonstrate that the woody wetland is much more effective than the marsh in buffering flood and wave.
Benefit of wetland on surge and wave during a Black Swan (BS) storm
We consider a rare “Black Swan” (BS) storm (with a 0.0034 annual frequency vs. 0.0014 for Sandy27) which made landfall in Staten Island with a storm surge that surpassed the maximum surge during Sandy with maximum coastal values of 7.17 m at NY Bight and 4.51 m at NJ coast of (Fig. S8). On the other hand, maximum significant wave heights over land were not greater than those of Sandy: 1.99 m at the NY Bight and 2.12 m on the NJ coast, but average wave height was higher than during Sandy (Fig. S8). During the Black Swan storm, wetlands created the biggest (RIR) and (RWR). MNY and NNJ experienced significant (RIR), while the remaining regions experienced low (RIR). In contrast to Sandy, in which CT, MNY, and NNJ had less (RWR), for the Black Swan storm these regions experienced high (> 50%) (RWR), while LI. CNJ and SNJ had significant (~ 25–50%) (RWR). The regions with the highest (RIR) and (RWR) were closest to the storm landfall location.
Benefit of wetland on surge and wave for the 1% annual chance flood
The maximum 1% annual chance maximum flood elevation (Fig. S9) was 5.15 m for NY Bight while 5.34 m for the NJ coast. The 1% annual chance maximum wave height (Fig. S9) overland is 2.24 m at NY Bight and 1.90 m at NJ coast. All the regions had moderate (RIR) except for SNJ which had a significant (RIR) due to highest (25.4%) wetland cover (see Table S2) and NNJ experienced significant (RWR) while the remaining regions had moderate values. CT ranked 2nd in (RIR ) although it has less wetland cover than CNJ because the mostly woody wetlands in CT are more effective in buffering storm surge than the marshes in NJ and NY. On the other hand, CT ranked 3rd in (RWR) due to the blocking of offshore wave energy by LI. (RIR) and (RWR) are found to be functions of storm characteristics, wetland type, and cover, and local conditions. Fig. S10 shows the percent wetland cover, RIR (relative TIV reduction), and RWR (relative wave energy reduction) in six regions (New York, North New Jersey, Long Island, Connecticut, Central New Jersey, and South New Jersey) during 1% annual chance events. As the wetland cover increases from less than 5% (NY, NNJ, and LI) to more than 10% in CNJ and SNJ, RIR and RWR generally increase, showing the increasing role of wetland in reducing inundation and wave. Relative reduction in inundation and wave energy are modest, between 10 and 30%. NNJ has properties behind the relatively sparse marsh, followed by woody wetland which protects properties behind them. Connecticut has less wetland than Central Jersey, but the mostly woody wetland is more effective in reducing flood and wave.
Benefit of wetlands on reducing residential structure loss
The monetary loss of residential structures was estimated using the simulated inundation and wave results while employing damage functions from the United States Army Corps of Engineer (USACE) North Atlantic Comprehensive Coastal Study (NACCS) and was validated using the NFIP building loss payouts aggregated by zip-code21,24. Direct simulation of wave-induced damage requires understanding and calculation of wave loads on structures using a depth- and phase-resolving model, which is beyond the capability of the models used in this study. Therefore, in this paper, we did not directly simulate wave-induced damage, but are accounting for wave-induced change in total water level which results in increase in estimated damage based on depth-damage functions. Overall, 96 coastal zip-codes in the state of NJ were used to validate the estimated loss. The model showed a correlation coefficient (CORR = 0.69) between simulated structure losses and NFIP payouts (Fig. 2). In NJ, as of 2019, the total NFIP payout was $3.9 billion USD, in comparison to the estimated total structure loss of $3.6 billion USD (SI Table S3), with an absolute error of 7.7%. This good agreement, plus the good agreement between the simulated and observed surge and wave reported earlier, confirms the validity of our “dynamics-based” loss assessment.
Economic model validation at zip-code resolution. Simulated losses during Sandy in NJ using the USACE damage functions versus FEMA NFIP payouts. Results were aggregated by zip-code and the corresponding correlation coefficient is 0.69 (R2 = 0.47). Validation use transformed structure loss ((PL_{T})) instead of the structure loss ((PL)). The figure is produced using MATLAB R2020 (https://www.mathworks.com/).
We define the structural loss reduction ((SLR)) as the structural loss without wetland minus the structural loss with wetland, and the relative structural loss reduction ((RSLR)) as the ratio between (SLR) and the loss with wetland. A state-level analysis of structure loss in NJ showed a (RSLR) of 8.5%, 26.0%, and 52.3% for Sandy, BS storm, and 1% chance flood/wave, respectively (Table S3). Analysis of losses due to flood and wave indicated that for Sandy and the 1% event, most of the loss came from flood, while most of the loss in the BS storm came from waves. Avoided wave-induced loss was comparable to the avoided flood-induced loss during the BS storm and the 1% event, but much higher during Sandy, suggesting that NJ wetlands are more effective in reducing wave-induced loss vs. flood-induced loss. Results from the zip-code scale analysis during Sandy (Fig. 3) showed less variability: for most north NJ, (RSLR) ranged from 10 to 100% except for those along the Hudson River that had small increased loss (negative avoided loss). Most zip-codes in SNJ had (RSLR) between 5% and more than 100%. (RSLR) during the BS storm (Fig. S11) was more notable: ~ 50–100% for north NJ, 0–100% for central NJ. The 1% flood event (Fig. 4) showed an average RSLR > 25% for NJ.
Percent structural loss reduction during Sandy. Map showing zip-code resolution avoided loss (difference in loss without wetlands and loss with wetlands) during Sandy, as a percentage of the loss of the with-wetland scenario. Dark red values show zip-code with the highest wetland benefit, while dark blue areas have the least benefit. Negative values indicate that the presence of wetland would increase structural losses and positive values indicate that wetland would lower the structural losses. The results are shown for the NJ coastal zip-codes affected by Sandy. This study shows that the percent avoided loss in this figure does not always represent the actual wetland value for loss reduction because areas with few structures and losses could give misleadingly high values of percent avoided loss, as shown in south NJ. The primary purpose is for comparison with a similar figure in NAR17. The map is produced using ESRI ArcGIS Pro 2.7 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview).
Effect of wetlands on structural losses over zip code scale during the 1% annual event. Map showing zip code resolution difference in losses if the wetlands were absent, as a percentage of the wetland present scenario. Dark red values show zip code with the highest benefit of having wetlands while dark blue areas show the least benefited area. Negative values indicate that the presence of wetland would increase structural losses and positive values indicate that wetland would lower the structural losses. The map is produced using ESRI ArcGIS Pro 2.7 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview).
The above results showed that the value of coastal wetlands for flood/wave protection varies significantly with the storm. While wetlands may be more effective in reducing wave loss in some storms but flood loss in other storms, they may be ineffective in extreme storms. The 1% annual chance flood and wave event, which resulted from an ensemble of many less extreme but more frequent storms28, provides a more reasonable integrated scenario for the loss analysis. This is similar to the preferred use of the 1% flood map, instead of the flood map associated with a single design storm, for assessing the flood risk in any coastal region.
As shown in Fig. 5, in zip-codes with larger wetland coverage area in SNJ, wetlands could only prevent < $25,000 annual loss (loss in the 1% event divided by 100), while wetlands in zip-codes with medium wetland coverage area in central NJ could prevent > $60,000 annual loss. On the other hand, wetlands in north NJ zip-codes with smaller wetland coverage area would increase the loss. The annual loss in most of the zip-codes is < $10,000 except in one zip-code where the annual loss is ~ $50,000.
Wetland annual avoided losses for all NJ zip-codes. Map showing zip-code resolution avoided losses. Annual avoided loss is calculated as the difference in losses for the without wetland and with wetland scenarios using the 1% annual chance flood and wave map and then dividing it by 100 years. Dark blue values show zip-code with the lowest wetland value/benefit, while red areas have the highest wetland value/benefit. Negative values indicate that the presence of wetlands would increase structural loss and positive values indicate that wetlands would lower structural loss. Negative values are relatively small while only increasing losses by ~ $10 thousand per year. The map is produced using ESRI ArcGIS Pro 2.7 (https://www.esri.com/en-us/arcgis/products/arcgis-pro/overview).
To explain the significant spatial variation in (SLR) by wetlands, we developed a generalized linear regression model using the normal distribution with the logit link function, which was fitted with data for all coastal zip-codes during three flood events and with and without wetlands. Percent wetland cover, percent at-risk structural value, and average wave crest height in the zip-code were used as predictors to estimate the percent structural loss (PSL = structural loss value divided by at-risk structural value) in the zip-code (Table 1). For Sandy, the regression model had a significant correlation (R2 = 0.75, p < 0.001). For the BS storm and the 1% annual chance flood, there was a strong correlation with R2 = 0.87 and R2 = 0.84, respectively, both with p < 0.001. Using the three storm scenarios, each with and without wetland, the regression fit (Fig. 6) had a significant correlation with R2 = 0.75 and p < 0.001.
Zip-code structural loss regression model. A generalized regression model using a normal distribution with a logit link function. Constructed using the Structure Value at Risk, Wetland Cover, and Average Wave Crest Height as predictors to estimate structural losses in zip-codes (R2 = 0.75, p < 0.001). Coupled with a Rapid Forecast and Mapping System33 being developed for the region, this regression model can be used to forecast potential structural loss during an approaching hurricane. It can also be used to predict the potential loss in any hypothetical hurricane, or to plan wetland restoration to reduce loss in any specific zip code. The figure is produced using MATLAB R2020 (https://www.mathworks.com/).
Source: Ecology - nature.com
