Building back bigger in hurricane strike zones

Despite decades of regulatory efforts in the United States to decrease vulnerability in developed coastal zones, exposure of residential assets to hurricane damage is increasing — even in places where hurricanes have struck before. Comparing plan-view footprints of individual residential buildings before and long after major hurricane strikes, we find a systematic pattern of ‘building back bigger’ among renovated and new properties. US coastlines that are exposed to hurricanes are subject to a myriad of regulations regarding building and rebuilding of structures, yet satellite imagery shows that the footprint of residential buildings increases after hurricane events for both new and renovated structures. Such an effect poses a challenge for vulnerable coastlines to build resilience in the face of growing hazards and houses.


Despite decades of regulatory efforts in the US to decrease vulnerability in 23
developed coastal zones, exposure of residential assets to hurricane damage is 24 increasing -even in places where hurricanes have struck before. Comparing plan-25 view footprints of individual residential buildings prior to and long after major 26 hurricane strikes, we find a systematic pattern of "building back bigger" among 27 renovated and new properties. 28 Storm impacts on developed coastlines are expected to increase with climate change 1 . In 29 coastal counties around the United States, policies intended to mitigate coastal risk are 30 competing with population growth and development pressures 1-5 that render places more 31 vulnerable and less resilient to major storm events. 32 Research into the repercussions of hurricane impacts has examined regional-and local-33 scale socioeconomics and demographics 6-8 , housing stock and types 8,9 , planning and design 34 requirements (and variances from them) 10-13 , tax and insurance policy 3 , and real-estate 35 market recovery 14 . But one indicator of increasing vulnerability in hurricane zones is 36 especially enigmatic: residential footprints are growing even in places with legacies of past 37 impacts, including a systematic pattern of "building back bigger" among renovated and new 38 properties. 39 Here, we investigate broad development trends in hurricane alleys. We measure changes 40 over 5-14 years in residential building footprints at five locations on the US Atlantic and 41 Gulf Coasts that have been struck by one or more hurricanes since 2003 (Fig. 1). Each 42 location occupies a developed coastal barrier in a different state, is characterised primarily 43 2/7 by single-family residential buildings, and is demarcated in FEMA flood-risk maps a Special 44 Flood Hazard Area. Collectively, the locations have weathered six different hurricane 45 systems between 2003 and 2012, and sustained damage from multiple types of impacts 46 (e.g., wind, storm surge, waves). Each location has also had multiple years (5 or more) over 47 which residential recovery could occur. Using satellite imagery captured before the last 48 major hurricane event (or events) at each locale and again in 2017 (the most recent year of  49 coverage available at all five locations, and collected prior to the 2017 hurricane season), we 50 digitised the plan-view footprints of individual residential buildings in the pre-storm and 51 2017 imagery and compared their respective areas. 52 The resulting statistical distributions of footprint size yield the same pattern at all five 53 locations: since the last major hurricane strike, larger residential buildings have tended to 54 replace smaller ones ( Fig. 2a-e). Among buildings whose footprints change ( Fig. 2f- patterns, typically stemming from local variances that circumvent newer planning rules 11 . 97 Practices of assessment, appraisal, compliance, and enforcement hinge on local and 98 individual discretion and interpretation 10-12 . The development pattern we show across the 99 locations in Fig. 1 surely arose from a number of mechanisms 3,5,7-14 . However, the aggregate 100 effect of those mechanisms -including the tendency to "build back bigger" in hurricane 101 corridors and demarcated coastal flood-risk zones -appears insensitive to their particulars. 102 By demonstrating an emergent pattern of increased exposure in high-risk coastal 103 development, we intend for our analysis to complement local case studies of land-use 104 policy effects and hazard-mitigation strategies.   Building footprints were digitised manually and their areas calculated using GIS software. 208 We digitised the roofed footprint of every residential building in the first three rows back 209 from the "ocean-side" shorefront. At Mantoloking, north/south town boundaries set the 210 sampling space. At Santa Rosa Island, we sampled the reach of coastline between the 211 causeways at Pensacola Beach and Navarre Beach (west/east, respectively). At 212 Hatteras/Frisco, Dauphin Island, and the Bolivar Peninsula (immediately northeast of 213 Galveston), we sampled the full alongshore extents. These data (pre-storm and 2017 214 combined, ~4800 footprints) therefore represent a large sample or all of the single-family 215 residential buildings at each location. Footprints were matched between images using a 216 spatial join, then reviewed manually. Given inherent variability in pre-storm image quality 217 (resolution or image tilt), we use a compensatory envelope of ±15%, which assumes that a 218 building's 2017 footprint must change more than ±15% to be distinguishable from 219 potential error. This envelope is nearly four to five times greater than the ~3-4% error 220 variance attributable to our manual digitisation, and is therefore a conservative measure. 221 Summary magnitudes of change in footprint area do not correlate with elapsed time 222 between images, nor do they indicate a geographic control (i.e., Atlantic versus Gulf Coast). 223 Although we did not control for building characteristics (or demographics), we applied the 224 same method to five distinct locations (each with ~10-30 km of shoreline extent) and 225 found the same pattern everywhere, suggesting that contextual biases in any one sample are 226 not strong enough to skew the aggregate findings. 227 In the Supplementary Methods, we further discuss our locations, and compare a subset of 228 our measured footprints to total living area reported in tax records (Supplementary Fig. 1;  229 Supplementary Table 3). 230 Statistical analysis -To quantitatively distinguish between pre-storm and 2017 distributions 231 of building size (Fig. 2f-

SUPPLEMENTARY METHODS
Locations -We examined five locations on the US Atlantic and Gulf Coasts that have been struck by one or more hurricanes since 2003. Our selection of locations was determined in part by date and image suitability: satellite imagery collected prior to 2002 tended to lack resolution crisp enough for reliable digitisation. Collectively, these five locations have weathered six different hurricane systems between 2003 and 2012, and sustained damage from multiple types of impacts (e.g., wind, storm surge, waves). Each location occupies a developed coastal barrier in a different state, and, although FEMA flood-risk maps are known to vary in their quality and accuracy (see Supplementary Ref. 1), each location is demarcated in FEMA flood-risk maps a Special Flood Hazard Area -either Zone A (hundred-year flood zones) or Zone V (hundred-year coastal flood zones likely to experience "velocity" from storm surge or wave action). These traits thus lend the locations similar physical environmental settings and federal designations, but potentially different state and local land-use planning contexts. Furthermore, by spanning the longest period possible (5 years or longer) since the last major hurricane event at a given location, our analysis allows for both rapid and slow paces of residential recovery. (That is, aerial images from the 2017 hurricane season, for example, might show damage but not reconstruction.) Each location is characterised primarily by single-family residential buildings: where possible, we confirmed this building-type classification with tax records ( Supplementary  Fig. 1). To sample conservatively, we did not digitise buildings with visible adjacent parking lots, assuming they served either multi-unit condominiums or commercial buildings. Although a given building may have changed from a single-family residence to a commercial space (or vice versa), we expect the impact of any such buildings on the statistical analysis is negligible, given the large number of individual buildings we sampled. This analysis is preliminary: it is limited to five US sites, and is not an exhaustive list of all sites on developed coastal barriers that have sustained hurricane damage (even in the US). Nevertheless, these preliminary results are instructive and motivate further work. Depending on imagery and data availability, the same comparative-footprint approach could be extended to other locations prone to cyclones (or other hazard types), and even applied in the absence of any recent cyclone (or other hazard) activity. Integrating a deliberately simplified analysis like ours with detailed collation of tax records, permits, construction types, and code variances (see Refs. 8,11 & 12 in the main text) would reflect the influence of political, legal, planning and other policy mechanisms in the coastal zone. But even in the absence of such detailed homeowner data (particularly outside the US), our methodology still quantifies broad development trends in ways that may help the wider sustainability-science research community to identify, understand, address the economic and policy forces that shape decision-making and risk evolution in hurricane alleys (and other hazard zones).
Comparison of total living area to measured footprints -Property taxes (and national Census statistics) report the total living area of a house, not its plan-view footprint. The roofed footprint that we digitise might approximately match the total living area for a single-storey house, but will almost certainly under-predict the total living area of a multi-storey building. For a building with deep covered porches, which do not count toward living area, our measurement of the roofed footprint will tend to over-predict the size of the actual (taxed) total living area.
To estimate how our footprint data scale relative to total living area, we compared the 2017 footprints of front-row properties from Hatteras/Frisco and Santa Rosa Island to total living area reported in property tax records as of 2016 ( Supplementary Fig. 1). We use these two locations because their tax records are publicly available online: Hatteras/Frisco via Dare County (https://tax.darecountync.gov/parcelcard.php?parcel=); Santa Rosa Island via the Florida Geographic Data Library (https://www.fgdl.org/metadataexplorer/explorer.jsp).
We find that total living area is, on average, ~94% larger than footprint area at Hatteras/Frisco, and ~39% higher at Santa Rosa Island ( Supplementary Fig. 1, insets). Applying these scaling factors, respectively, to our mean footprints allows us to compare our measurements to national statistics (via the US Census Bureau) for mean total floor area in new single-family houses (Supplementary Table 3). By direct comparison, according to the sizes reported in tax records, the mean size of (front-row) single-family residential buildings in our 2017 sample from Hatteras/Frisco are 28% larger than the 2016 national average for new single-family houses; our sample from Santa Rosa Island are 47% larger than the 2016 national average. Note that our measured samples include all existing buildings, not just those built most recently. Hypothetically, a location where development exactly matches the national trend in new houses each year will, over time, end up with an overall mean house size that is smaller than the mean size for the most recent year.  a In a two-sample Kolmogorov-Smirnov test, the null hypothesis is that data in the two samples come from the same continuous distribution. The alternative hypothesis is that the two samples are from different continuous distributions. The result of 1 indicates that the test rejects the null hypothesis at the α = 5% significance level. b p is the probability of observing a test statistic as extreme as, or more extreme than, the observed value under the null hypothesis.
c In a two-sample t test, the null hypothesis is that data in the two samples come from independent random samples from normal distributions with equal means and equal but unknown variances. The alternative hypothesis is that the data comes from populations with unequal means. The result of 1 indicates that the test rejects the null hypothesis at the α = 5% significance level. d Hypothetically, total footprint area could change with preferential destruction or removal of buildings of a given size, without otherwise altering footprints of existing buildings. We test for this effect by comparing the mean pre-storm footprint of "surviving" buildingsthose present in both images -with the mean pre-storm footprint overall. The only significant difference we find is at Bolivar, where smaller buildings were disproportionately affected. However, the preferential loss of smaller footprints only accounts for a 9% increase in mean footprint size.