TC climatology
Figure 2 shows the CDF of synthetic TC properties when TCs are at their closest distance to the reference point of Jamaica Bay. CDFs are estimated based on a generalized extreme value distribution fitted to the data. We perform a t-test to quantify the significance of changes over time in TC properties. We consider changes to be statistically significant when the p value is equal or smaller than 0.05. Averaged over all models, the TC translation speed, VT, would significantly change from the historical time period to the mid- and late-twenty-first century (p values of 0.010 and 0.019, respectively). We find that the average probability (averaged over all models) of a TC, when at its closest distance to Jamaica Bay, to move with a VT < 5 m s−1 increases by 0.9% and 5.4% in the mid- and late-twenty-first-century periods, respectively. The average probability that a TC moves with a VT > 15 m s−1 decreases by 2.2% and 7.2%, respectively. These findings are consistent with the storm characteristics in the basin-scale synthetic TC datasets used by Marsooli et al. (2019), which similarly showed an increase in the number of slow-moving TCs in the future climate in the western North Atlantic basin. While there exists limited discussion on how TC translation speed would change in the future, using global TC “best-track” data, Kossin (2018) found that the global TC translation speed has reduced by 10% over the time period of 1949–2016.
Projections from all models agree that the radius of maximum wind speed, Rmax, would reduce over time. The t-test indicates statistically significant reduction in Rmax, with an averaged p value of 0.036 for the mid-twenty-first century and about zero for the late-twenty-first century. Averaged over all models, the probability that a TC has a Rmax < 40 km and Rmax > 60 km changes by, respectively, 2.2% and − 4.7% by the mid-twenty-first century and 9.8% and − 9.8% by the late-twenty-first century.
Synthetic tracks show an increase in the probability that a TC with a hurricane intensity can reach the region of study by the end of twenty-first century (the region of study covers an area within 200 km from Jamaica Bay) (Supplementary Fig. S4). All models except MRI5 agree that the intensity of TCs (represented here by the maximum wind speed, Vmax) would significantly increase by the end of twenty-first century. Averaged over all models, the probability of a TC to have a Vmax > 33 m s−1 (hurricane strength) increases by 1.7% (2.3% if MRI5 excluded) by the mid-twenty-first century and 5.9% (7.9%) by the late-twenty-first century. This projected increase in TC intensity in the future climate is consistent with most other projections (Knutson et al. 2019). The projected increase in Vmax is also consistent with the projected decrease in Rmax, given their physical relationship.
Differences (averaged over all models) between future and historical synthetic TC track densities (see Supplementary Fig. S5) indicate an eastward shift in storm tracks by the end of twenty-first century. This is consistent with Garner et al. (2017) who found an offshore shift of storm tracks at the latitude of NYC.
There are some discrepancies among projections from each individual model. For example, projections from GFDL5, HADGEM5, MPI5, and MRI5 show, respectively, a change of − 15.1%, − 4.5%, − 1.2%, and − 8.0% in the probability that a given TC has a translation speed of greater than 15 m s−1 from the historical period to the late-twenty-first-century period. We performed a bootstrap analysis to assure that our TC track datasets are large enough for making stable conclusions. The analysis showed that the calculated percentage changes in TC parameters agree with the distribution of percentage changes obtained across the bootstrap samples, assuring that our datasets are sufficiently large. For instance, based on the original GFDL5 dataset, the probability of a TC to have a translation speed of smaller than 5 m s−1 would increase by 4.59% from the historical period to the mid-twenty-first century period, which agrees with the distribution of changes calculated based on the bootstrap samples which shows a mean increase of 4.55% and a standard deviation of 0.62%. Therefore, discrepancies among projections from the different climate models can be due to systematic differences in the models such as the resolution, initial conditions, and emphasized physical processes.
Supplementary Table S1 summarizes the annual TC frequency, i.e., the number of TCs (passing within 200 km from Jamaica Bay) in any given year within a specific time period. The annual frequency in 1980–2000 is about 0.39 based on NCEP-based and most model-based datasets. The GFDL5 model shows a smaller frequency for the historical period. However, compared to other models, the GFDL5 model projects a substantial increase (up to 377%) in the future TC frequencies. Projections from HADGEM5 and MPI5 indicate a moderate increase in the TC frequency while projections from MRI5 show only a subtle change.
Flood return levels
Figure 3 shows the estimates of flood return levels at the representative site in the bay (40.60°N and 73.80°W), located near the bay head and the Inwood USGS gauge where the model accurately simulated historical flood levels (see Fig. 1), for the historical and future time periods. The very-likely range (5th–95th percentiles; i.e., 90% statistical confidence interval) is shown by the shaded area. Under the compound effects of SLR and TC climatology change, the flood level for a given return period would substantially increase from the historical period in the late twentieth century to the future periods in the twenty-first century. For example, while the 100-year weighted-average flood level in the historical period is 1.68 m (with a very-likely range of 1.63–1.75 m), it increases to 2.13 m (2.09–2.19 m) and 3.23 m (3.14–3.42 m) in the mid and late twenty-first century, respectively.
Flood return periods presented here are bias-corrected based on biases that are estimated by comparing NCEP-based projections for the historical period of 1980–2000 with model-based projections for the same historical period. Thus, it is assumed that NCEP-based projections realistically represent flood hazards for the historical period. One may perform a sanity check by comparing the NCEP-based flood return periods with return periods estimated based on water level observations at tide gauge stations. Because there are no records of water level observations in Jamaica Bay for the historical period of 1980–2000, we use water level observations at a nearby tide gauge station, i.e., The Battery NY station in the Upper Bay, operated by the National Oceanic and Atmospheric Administration (NOAA). The Battery station is located at the southern tip of Manhattan in NYC, about 15 km away from the Jamaica Bay’s inlet.
Figure 3 (circles in magenta) displays flood return periods estimated based on the observed TC-induced water levels at The Battery. No attempt was made to fit a return period curve to the observation-based estimates, given the low number of TC events between 1980 and 2000. The NCEP-based and observation-based return levels compare relatively well but show some discrepancies. We hypothesized that the discrepancies could be due to major geometric and bathymetric differences in Jamaica Bay and Upper Bay, resulting in different flood levels during a single storm. For example, the peak storm tide generated by Hurricane Sandy in 2012 in Jamaica Bay was up to 0.7 m smaller than that at The Battery station. However, based on our model outputs, the NCEP-based return levels at the Battery also show discrepancies with the observations. We investigated the effects of small observed water level sampling size by generating NCEP-based return period curves for small sample chunks from NCEP tracks, but the discrepancies persisted. Thus, the discrepancies exist mainly because, in synthetic modeling approach, the generated NCEP tracks are only statistical representations in the reanalysis climate environment of the historical tracks.
Weighted-average return period curves in Fig. 3 show that, under the effect of only TC climatology change, flood levels with a return period smaller than 100 year barely change. However, the effect of TC climatology change on low-probability flood levels (i.e., return periods greater than 100 year) is non-negligible, especially by the end of twenty-first century. To further illustrate, Fig. 4 displays contributions of SLR and TC climatology change to the flood level increase from the historical period to the future periods. We find that the effects of SLR dominate the effects of TC climatology change. However, TC climatology change becomes a major source of increases in low-probability, high-consequence flood levels, especially in the late-twenty-first-century period. For example, from the historical period of 1980–2000 to the late-twenty-first-century period of 2080–2100, while TC climatology change contributes to only 11% of the increase in 100-year flood level it contributes to 19%, 22%, and 31% of the increases in 500-, 1000-, and 10,000-year flood levels in Jamaica Bay. The large increase in these low-probability flood levels is likely due to the increase in the probability that slower and more intense TCs pass within 200 km from Jamaica Bay by the end of twenty-first century (Figs. 2 and S4).
Similar to the weighted-average projections, projections from each individual model (Fig. 3) suggest that the combined effects of SLR and TC climatology change result in a substantial increase in flood levels. Projections based on GFDL5 show the largest increase in the future flood levels whereas MRI5 projects the smallest change. Projections based on the GFDL5 and HADGEM5 model suggest that TC climatology change by the end of twenty-first century substantially increases flood levels associated with low probability but high consequence events (i.e., flood return periods greater than 100 year). Similar patterns but with a smaller magnitude can be observed from projections based on the MPI5 model. In contrast, projections based on the MRI5 model indicate that while the TC climatology change has negligible effects on flood levels due to high-probability events, it decreases flood levels associated with rare but devastating events in the late-twenty-first-century period. The reason for this projected decrease in flood levels is likely that TC projections based on MRI5 show negligible changes in TC intensity and frequency (Fig. 2 and Table S1) but a profound eastward shift in TC track densities.
Scenario-based coastal flooding
Results presented in the previous section elucidated that the effects of SLR on future flood levels dominate the effects of TC climatology change. Here we use the high-resolution model to quantify effects of SLR on surface waves and the extent of flooding under scenarios described in Section 2.2. The SLR scenarios are simulated for two selected synthetic TCs shown in Supplementary Fig. S2. The return periods of the combined SLR and TC scenarios are estimated based on the climatology-hydrodynamic modeling discussed in the previous section. Under no SLR, the two selected TCs generate flood levels of 1.76 and 2.61 m, which are approximately equal to 125- and 1300-year flood levels in the historical time period of 1980–2000. Hereafter, we refer to these TC scenarios as the historical high- and low-probability TC scenarios. Under a SLR of 0.59 m (5% chance of exceedance in the mid-twenty-first century), for example, these TCs generate flood levels of 2.39 and 3.08 m, which are equal to 214- and 1167-year flood levels in the 2030–2050 time period. Under a SLR scenario with the same chance of exceedance but in the late-twenty-first century (1.54 m), the TCs generate flood levels of 3.13 and 3.97 m, which are equal to 74- and 515-year flood levels in the 2080–2100 time period.
Figures 5 and 6 illustrate the spatial extent of flooding induced by the selected TC and SLR scenarios. The extent of flooding is defined here as the area above the mean-higher-high-water level that is normally dry but becomes wet during a storm event. With a 0.0 m SLR (i.e., control runs), the extent of flooding covers, respectively, 3.55 km2 and 13.45 km2 of floodplains under the historical high- and low-probability TC scenarios (Table 1). The flooded areas are mainly located on the eastern side of the bay. For the same TCs, the extent of flooding dramatically increases as sea level rises. The extent of flooding induced by the high-probability TC scenario increases, respectively, 38%, 99%, and 210% under SLR scenarios of 0.19 m, 0.38 m, and 0.59 m (which have a chance of exceedance of 95%, 50%, and 5% in the mid-twenty-first century). The extent of flooding induced by the low-probability TC scenario increases, respectively, 28%, 67%, and 106% for the same SLR scenarios. Under SLR scenarios of 0.44 m, 0.96 m, and 1.54 m in the late-twenty-first-century period (which have a chance of exceedance of 95%, 50%, and 5% in the late-twenty-first century), the extent of flooding increases 130%, 477%, and 925% for the high-probability TC and 79%, 160%, and 359% for the low-probability TC scenario.
Table 1 Flooded area and extreme waves area induced by the high- and low-probability TC scenarios under selected SLR scenarios. The chance of SLR exceedance is based on SLR projections at the location of The Battery tide gauge station In addition to the extent of flooding, rising sea levels amplify wave hazards. As sea level rises, water depth increases which, in turn, allows larger waves to reach currently shallow areas in the bay. Figure 7 compares the peak significant wave height, Hs,max, calculated for the control runs (no SLR) and the 5% exceedance SLR scenarios. For the control runs, wave heights are larger in the deep regions of the bay including the bay’s inlet, shipping channels around the bay, and Grassy Bay (the easternmost region of the bay near the J.F. Kennedy International Airport). Smaller wave heights are calculated in shallow areas in the center of the bay. For the future SLR scenarios, larger wave heights are calculated in the center of the bay as deeper waters allow larger waves to be developed.
Under no SLR, the calculated Hs,max for the historical high- and low-probability TC scenarios is greater than 1 m in about, respectively, 11.40 km2 and 13.16 km2 of the study area (Table 1). Under SLR scenarios that have a chance of exceedance of 95% in the mid- and late-twenty-first century, i.e., SLR of 0.19 m and 0.44 m, the area with a Hs,max greater than 1 m is 12.37 km2 and 13.96 km2 for the high-probability TC scenario and 14.10 km2 and 15.54 km2 for the low-probability TC scenario, suggesting 8.5%, 22.5%, 7.1%, and 18.1% of increases compared to the no SLR scenarios (see Supplementary Fig. S6). Under SLR scenarios that have a chance of exceedance of 5% in the mid- and late-twenty-first century, i.e., SLR of 0.59 m and 1.54 m, the area of Hs,max greater than 1 m changes to 15.03 km2 (31.8% increase) and 23.26 km2 (104% increase) for the high-probability TC, and 16.62 km2 (26.3% increase) and 28.55 km2 (117% increase) for the low-probability TC scenario (Fig. 7).