Evacuation zoning and sheltering
The NYC Office of Emergency Management (OEM) developed and released six Hurricane Evacuation Zones in 2013 as part of the City’s Hurricane Sandy After-Action report (Fig. 1b). The NYC evacuation zone system was primarily based on (i) the risk of coastal flood resulting from storm surge; (ii) the geography of the City’s low-lying neighborhoods; and (iii) the accessibility of the neighborhoods by roads and bridges (NYC Mayor’s Office, 2012). Zone 1 encompasses the City’s coastline and low-lying areas which are most likely to be inundated with storm surge in the event of a west-northwest bearing Category 1 hurricane. Zones 2–6 cover additional neighborhoods that are subject to the flood impact of more extreme storms (a Category 2 hurricane and above with larger sizes). The higher the zone number, the lower the flood risk. Table 1 summarizes the areal and population coverage of NYC Hurricane Evacuation Zones. All of the Evacuation Zones occupy almost half (~ 49%) of the city area and include over 3 million New Yorkers accounting for ~ 39% of the total population derived from the 2010 NYC census. Particularly, ~ 41% (18.9K) of the elderly (people over 75 years old) who are always disproportionately affected by coastal flooding, are estimated to reside within the boundaries of Evacuation Zones 1 through 6. Despite the lower population densities compared to the city’s average, such vulnerable people (including the elderly) are highly concentrated in NYC evacuation zones (8270 people/km2).
We delineated the borders of coastal flood evacuation zones for Shanghai, following the general principles of NYC Hurricane Evacuation Zoning and the regulations of coastal flood management in China. We begin by identifying coastal flood risks with various return periods and sea level rises (SLRs) utilizing a 2D flood inundation model (see Methods for details and Supplementary Fig. 1), and we then use the flood maps generated to define and classify vulnerable communities and associated accessible neighborhoods into six evacuation zones (Fig. 1a). Zone 1 includes the city’s low-lying and poorly protected waterfront areas that are susceptible to a present 100-year flood event, and Zones 2–4 further extend to the coastal and Huangpu River floodplains depending on the severity of longer return period events (200-, 500-, 1000-year) under current conditions. Considering the local effect of relative SLRs in Shanghai, Zones 5 and 6 are at risk from future high-end scenarios (i.e. 1000-year floods under 2030 and 2050 sea levels), respectively. As shown in Table 1, approximately ~ 37% (over 8.4 million) of the city’s population live within the zones of 2794.62 km2, equivalent to ~ 41% of the land area of Shanghai. Meanwhile, the percentage coverage of the evacuation zones for the elderly people (75+) is found to be relatively high (~ 46%). Perhaps most notable is that the spatial distribution of the residents is heavily uneven across the six evacuation zones, with Zone 6 and 1 being the top two most densely populated parts.
Both cities provide a number of evacuation shelters to meet the basic needs of evacuees during a coastal flood or other emergency. NYC’s shelter system consists of 60 evacuation centers which are generally well positioned outside of the evacuated area, and is designed to accommodate up to 600,000 people if fully activated for all evacuation zones (Fig. 1b). The provision of NYC emergency sheltering is considered to be sufficient because among residents who evacuated during Hurricane Sandy, 78% stayed with friends or relatives, and only 2% sought shelter (NYC Mayor’s Office, 2012). In Shanghai, there are a total of 91 emergency shelters scattered throughout the city, including 74 indoor facilities (e.g. schools and stadiums) and 17 outdoor places such as open spaces. Of those indoor shelters suitable for flood emergency (Fig. 1a), ~ 34% (25) fall within the six evacuation zones and the remaining ~ 66% (49) will be functional with a full capacity of over 230,000 people, representing less than 3% of resident population in all evacuation zones. It is noted that the residually limited capacity of operational shelters significantly mismatches the potentially huge demand of flood evacuation, which is usually organized by local government in China rather than predominantly by individual evacuees in the U.S. Such a great demand-capacity mismatch may impose an excessive burden on the city’s emergency evacuation services, leading to important imbalances in accessibility among population groups.
Table 1
Areal and population (percentage) coverage of coastal flood evacuation zones in Shanghai and NYC
Evacuation
Zone
|
Shanghai
|
NYC
|
Area (km2)
|
Population
|
Elderly (75+)
|
Area (km2)
|
Population
|
Elderly (75+)
|
Zone 1
|
289.17 (4.27%)
|
1055066 (4.58%)
|
66167 (5.79%)
|
123.69 (15.81%)
|
516241 (6.31%)
|
36329 (7.87%)
|
Zone 2
|
237.34 (3.51%)
|
330400 (1.44%)
|
22902 (2.00%)
|
79.07 (1.01%)
|
429171 (5.25%)
|
30175 (6.54%)
|
Zone 3
|
438.23 (6.47%)
|
844650 (3.67%)
|
54299 (4.75%)
|
36.31 (4.64%)
|
348325 (4.26%)
|
19939 (4.32%)
|
Zone 4
|
503.53 (7.44%)
|
1623265 (7.05%)
|
102439 (8.96%)
|
45.00 (5.75%)
|
534995 (6.54%)
|
31876 (6.90%)
|
Zone 5
|
669.92 (9.90%)
|
1934135 (8.40%)
|
121175 (10.60%)
|
49.51 (6.33%)
|
593086 (7.25%)
|
32153 (6.96%)
|
Zone 6
|
656.42 (9.70%)
|
2657732 (11.55%)
|
154900 (13.55%)
|
50.32 (6.43%)
|
753007 (9.21%)
|
38621 (8.37%)
|
Zone 1–6
|
2794.62 (41.28%)
|
8445248 (36.69%)
|
521882 (45.66%)
|
383.91 (49.07%)
|
3174825 (38.84%)
|
189093 (40.96%)
|
Evacuation routing and planning
There exists worldwide evidence that aged and infirm populations are the prime victims of the destructive flooding, the botched evacuation, and the inadequate shelter (Coates, 1999; Rappaport, 2000; Kates et al., 2006; Diakakis and Deligiannakis, 2017). For example, approximately ~ 71% of the flood victims associated with Hurricane Katrina (in 2005) were over 60 years old, and ~ 47% of those were older than 75, attributing to a progressively increasing (with age) physical inability to flee from danger (Jonkman et al., 2009). Therefore, the aged who are least able to evacuate to safe places outside of the impact areas during a flood emergency should be given the top priority, regardless of disparities in disaster response behaviors. We assume that in advance of a predicted storm, an organized, phased evacuation operation will be implemented with a mandatory evacuation order, providing transportations (e.g. buses) to orderly evacuate vulnerable population from zones nearest to the hazard before others who are in less danger. Here, we present a system-level network analysis to enhance city-scale coordination of evacuee transfer and make efficient use of shelter capacities, exclusively focusing on the elderly (75+) that are most dependent on mass-transit emergency evacuation before a major coastal flooding. Specifically, we conduct a multi-objective multi-phase origin-destination (O-D) estimation for city-wide pre-event evacuation, based on a maximal covering location-allocation model. Then we apply a time-dependent routing algorithm to identify critical emergency corridors and to minimize average evacuation times for each city (see Methods for details).
Figure 2 shows the spatial distribution of the generated O-D pairs for evacuee-shelter assignments in Shanghai and NYC. We observe distinct evacuation patterns for both cities associated with varying shelter capacities and demographic characteristics. As illustrated in Fig. 2a, the existing non-uniform distribution and capacity constraint in sheltering will cause a massive, chaotic evacuation in Shanghai. Only the elderly who live in Zones 1–3 and the vast majority of Zone 4 can be mobilized during a large-scale evacuation of Zones 1 through 6. In particular, due to a complete loss of shelter functions in Chongming island, evacuees who are forced to leave their home neighborhoods will mostly be transferred to mainland Shanghai where many operational facilities are located. Additionally, major destinations of Zone 4 evacuees from the city center, are two mega-shelters in the southwestern suburb, according to the residual capacities of the system. The supplementary information includes a description of the patterns of O-D pairs for zones 1 to 5, independently (Supplementary Fig. 2). Unlike Shanghai, shelter capacities are not a bottleneck of hurricane evacuation in NYC. All targeted people within Zones 1–6 can be distributed to the closest evacuation centers in an efficient, structured way. Even if several facilities (e.g. around Jamaica Bay and Manhattan island) are heavily occupied or operate at full capacity, the first tier of the city’s emergency shelters in close proximity to the edge of evacuation zones are still enough to accommodate the maximum estimated numbers of the elderly residents who need special assistance during a coastal storm (Fig. 2b). An emergency evacuation plan that implements such a demand-oriented strategy will result in a more balanced allocation of shelter facilities to vulnerable neighborhoods throughout the city.
In terms of evacuation routing and timing, Fig. 3 displays the estimates of the evacuee transfers and associated time costs between origin neighborhoods and shelter destinations in the transport systems of Shanghai and NYC. Consistent with the characteristics of the O-D patterns, we find clear differences in the spatio-temporal distributions of evacuees’ mobility for the two cities. Dozens of critical evacuation arteries can be readily identified from the vessel-like structure in Shanghai, according to the estimated number of transfers (Fig. 3a and Supplementary Fig. 3). For instance, cross-island emergency evacuation will lead to heavy use of multiple routes (i.e. a ferry route, a bridge + tunnel and several highways) from Chongming island to Mainland Shanghai, with the corresponding time costs mostly higher than 1.5 hours under baseline traffic condition. Although a great number of evacuees who will be transported from the city center to the periphery may overload critical roads, the evacuation times are found to be relatively low (0.5–1.5 hours) if appropriate traffic control measures are applied. The critical routes in NYC are less discernable than that of Shanghai, because the distribution of evacuee transfers generally matches well with the distribution of shelter capacities via the city’s high-density and well-developed road network (Fig. 3b). However, we still detect several important roads and bridges (e.g. Belt Parkway and Cross Bay Boulevard) with mass evacuation flow around Jamaica Bay, which connect zones with high risks (e.g. Rockaway Peninsula and Coney island) to places with high capacities. In addition, apart from Rockaway Peninsula, the travel times for NYC evacuation neighborhoods are markedly less than 20 minutes (Fig. 3d), whereas each trip of flood evacuations can take up to 3–4 hours in Shanghai under baseline traffic condition (Fig. 3c).
Strategic positioning of shelters
The selection, modification and/or construction of shelters play a crucial role in addressing the high demand-capacity gaps across the system and thus increase the effectiveness and fairness of city-scale emergency evacuation. As the NYC emergency plan (including evacuation centers) is well prepared and the evacuation arrangements has previously worked well during coastal flood emergencies such as Hurricane Sandy, only the shelter system in Shanghai should be supplied with additional resources in locations and quantities that are critical to improve the performance of coastal flood evacuation. Therefore, we design two alternative strategies for shelter deployment in the city. One strategy only requires minimum additional resources being focused on the combination of all existing shelters to meet the evacuation demands of the elderly within Zones 1 to 6, whereas the other provides strategic optimization of the shelter system by ensuring sufficient capacity in key locations to minimize the total evacuation time for each route. Our analysis uses a multi-coverage facility location model to select and deploy additional shelters outside of the city’s evacuation zones for both strategies, assuming that the capacities of these facilities range from 2000 to 30000 people based on the local design standard of emergency shelter. We also conduct a sensitivity analysis to evaluate the spatial and temporal performance of the model with varying shelters (see Methods for details).
Figure 4 presents the deployment of shelters and the resultant distribution of O-D pairs for both strategies. While only 22 additional shelters are employed across the city in the first strategy, 128 shelters are strategically deployed in the periphery of the city’s evacuation zones with the second optimization strategy, including 8 existing shelters, 114 candidate shelters, and 6 potential mega-shelters which need to be newly built on Chongming and Hengsha islands. Though strategy 1 can accommodate all the elderly at city scale, it does not effectively mobilize evacuees due to the spatial mismatch between shelter capacity and evacuation demand. Almost all shelters will be heavily or fully occupied and a great number of cross-island and long-distance travels can be observed in Fig. 4a. Conversely, the O-D pattern of evacuation with strategy 2 is similar to that of NYC. Evacuees can be transferred locally to a nearby facility in an orderly manner, by deploying shelters in lower-capacity zones such as the downtown area of Shanghai (Fig. 4b). As a result, tens of thousands of evacuees in Chongming island do not have to be transported to shelters in mainland Shanghai via bridges, tunnels and ferries. Moreover, the total capacity of the shelters reaches up to approximately 686K people, providing a certain degree of redundancy at over 30% to improve system reliability in the case that some facilities are unavailable during extreme weather conditions or if there are meet more demands from other vulnerable groups such as people with very bad health.
Figure 5 depicts the spatio-temporal distributions of evacuee transfers and travel times between O-D pairs via the city’s transport system for large-scale coastal flood evacuation plans with the first and second strategies, respectively. An immediate finding is that the estimates with the optimization strategy, as expected, greatly outperform that of strategy 1, in terms of both evacuation efficiency and equity. In the first strategy, the general pattern of evacuation routes and times is similar to what has been found in Fig. 3a. The critical evacuation routes are still in place and will be further overloaded to transfer more evacuees from Chongming island to Mainland Shanghai and from the city center to the periphery (Fig. 5a). Despite the low efficiency in the remote islands, the estimates of evacuation time for 90% of the elderly is approximately 80 minutes under the baseline traffic condition (Fig. 5c), a slight reduction when compared to the initial evacuation plan in Shanghai (Fig. 3c). This can be explained by the presence of additional shelters which provide alternative options to transfer evacuees more promptly in less crowded routes, particularly in Huangpu River floodplain. If more shelters are deployed using this methodology, the performance of the emergency evacuation could be gradually improved (Supplementary Fig. 4). In contrast, major emergency routes will be offloaded even more with the second strategy for optimal coastal flood evacuation (Fig. 5b). The long-distance evacuation arteries will be almost totally replaced by capillary-like local roads in the city. Consequently, the travel times for 90% of total evacuees will be further reduced to only 23 minutes under baseline traffic condition and up to around 1 hour when significant congestion occurs (Fig. 5d).