Performance of Hurricane-Resistant Housing during the 2022 Arabi, Louisiana, Tornado

Tornadoes are currently ignored in the design of typical buildings [i.e., considered Risk Category II buildings per ASCE 7 (ASCE 2022)], including residential structures, throughout the United States. One exception is Moore, OK, where local officials have enhanced the building code in direct response to previous tornado impacts (Ramseyer et al. 2016; Simmons et al. 2015). This leaves much of the US building stock highly vulnerable to tornadoes. For example, an EF-2 tornado has maximum wind speeds of up to $60\mathrm{m}/\mathrm{s}$ (135 mph) according to the enhanced Fujita scale (McDonald et al. 2012). A tornado of this magnitude produces wind loads as much as 38% greater (based on the square of the velocity ratio) than design wind loads for typical buildings in nonhurricane-prone regions, where the basic wind speed is around $51\mathrm{m}/\mathrm{s}$ (115 mph) or less (ASCE 2017).

Between 1993 and 2022, an average of 1,225 tornadoes have struck the United States each year (NOAA 2023), approximately 20% of which are rated EF-2 or higher (Fan and Pang 2019), and nearly all occur east of the continental divide. Due to their smaller footprints relative to hurricanes and other synoptic wind storms, the annual probability of a tornado striking a given point in the United States, even in the more tornado-prone regions, is relatively low (approximately 5E-4 annual probability, or 2,000-year mean reoccurrence interval for a relatively tornado-prone region such as Birmingham, AL (Stoner and Pang 2021). However, despite the low point probability, tornadoes cause 77 direct fatalities and billions in economic losses on average each year, with impacts likely to grow due to both changes in exposure and hazard (Ashley and Strader 2016). Continuing to ignore tornadoes, and the impacts they cause, in our building codes is a failure of policy (Prevatt et al. 2012).

Recognizing this policy failure, and the impacts on fatalities, injuries, as well as economic losses, there have been many studies highlighting the need for enhancing the resistant capacity of building structures (Honerkamp et al. 2022; Masoomi et al. 2018; Prevatt et al. 2012; Standohar-Alfano and van de Lindt 2016) and the benefits of enhanced construction (Ghosh et al. 2023; Liang 2001; Sandink et al. 2019; Simmons et al. 2015; Sutter et al. 2009) for mitigating tornado impacts. Many of these studies focus specifically on light wood-frame construction, as it comprises over 90% of the housing market and was valued at $33.3 trillion in 2018 (Adhikari et al. 2020). Wind damage mitigation studies frequently tie the enhanced construction to specific elements of the structural load path, notably the use of hurricane straps instead of toe-nails in the roof-to-wall connections, and the use of anchor bolts with nuts and washers instead of concrete cut nails or other weak anchorage options at the wall-to-foundation connection. It is implied by these studies that making some simple changes in the design of connections would significantly improve the resistance of wood-frame buildings to tornadoes. These studies are complemented by recent experimental and numerical studies on fundamental tornado-structure interaction. Laboratory tornado simulators have greatly advanced our knowledge of tornado structure and interaction with individual buildings (Ashrafi 2021; Bezabeh et al. 2018; Chen et al. 2023; Haan et al. 2010; Sabareesh et al. 2019; Wang et al. 2020), including the effects of building geometry (Case et al. 2014; Razavi and Sarkar 2021) and presence of openings (Chen et al. 2022; Jaffe and Kopp 2021; Wang et al. 2020). The data from these simulators are being used to investigate differences in wind loads induced by tornadoes and straightline winds (Kopp and Wu 2020; Roueche et al. 2020) and develop new codes and standard provisions for tornado-induced loading (Narancio et al. 2023; Wang and Cao 2021). Numerical simulations of tornadoes have also advanced tremendously in recent years, including the use of large eddy simulation to simulate the interaction of tornadoes with both individual buildings (Honerkamp et al. 2022; Liu et al. 2021; Verma et al. 2022) and clusters of buildings (Kawaguchi et al. 2020) to better understand the structural loads that must be resisted.

Despite these advancements in our understanding of tornado-induced wind loads and structural response, there is still a lack of empirical evidence to demonstrate that simple changes to building construction will enhance the structural performance of buildings under tornado hazards. The International Code Council International Residential Code (IRC) has effectively required roof-to-wall connections beyond toe-nails since 2015 by including rafter/truss uplift connection force tables, which provide required uplift forces dependent on roof slope, roof span, wind exposure, and truss/rafter spacing. Reconnaissance reports have noted hurricane straps (or equivalent) and anchor bolts in recent construction in noncoastal regions after tornadoes (Pilkington et al. 2021; Wood et al. 2020), confirming the code impact at the local level. However, discerning whether these enhanced construction indicators actually improve performance is challenging because wind speeds in tornadoes are estimated based on the damage they cause (Mehta 2013) and not based on direct near-surface wind field measurement. Thus, there is a circular reference involved in estimating tornado wind speed intensity based on damage to enhanced construction unless sufficient independent wind intensity indicators are used (e.g., Lombardo et al. 2015). Furthermore, empirical evidence from field studies (Marshall 2002; Wood et al. 2020) suggests that enhanced construction indicators (e.g., hurricane straps and anchor bolts) are not necessarily indicative of an enhanced, continuous load path, as other weak links along the load path can compromise the benefits of strengthened links elsewhere.

With such questions, there remains a need to document and evaluate the performance of enhanced structural load paths in wood-frame residential buildings under tornado-induced wind loads to elucidate the true benefits of the enhanced construction and subsequently calibrate and inform future analytical and numerical models.

The objective of this study is to contextualize the performance of hurricane-resistant housing subjected to a strong tornado with respect to key causal factors, including the regulatory environment, hazard conditions, and observed structural load path. This study will take Arabi, LA, a community in a hurricane-prone region that was struck by a strong tornado on March 22, 2022, as an illustrative demonstration/hindcast for the proposed method. The rest of the paper is organized as follows: First, a brief meteorological summary of the 2022 Arabi, LA, tornado is provided. Next, the historical and regulatory context related to construction in Arabi is summarized. This is followed by a description of the reconnaissance strategy and technologies utilized and a summary of the resulting data set. The next section summarizes the near-surface wind environment based on analysis of wind indicators. The typical structural load path observed in the homes that were investigated is next provided. Failure observations are next summarized, followed by a fragility analysis of a benchmark building representative of tornado-impacted residential structures in Arabi and illustrating the impact of various construction alternatives on system performance. Finally, the paper concludes with a summary discussion and conclusions.

The Arabi, LA, tornado was part of a larger spring tornado outbreak across the Southeast United States that occurred over March 21–22, 2022 (Roueche et al. 2023). A total of 43 tornadoes were reported, including seven rated EF-2 and two EF-3 by the National Weather Service (NWS) using the enhanced Fujita scale (McDonald et al. 2012). The Arabi tornado touched down at 7:21 PM CDT, traveling a distance of 18.5 km (11.5 mi) across Jefferson, St. Bernard, and Orleans Parishes before lifting at 7:38 PM. The NWS survey teams found that most of the damage was concentrated within a relatively narrow path (300 m wide) through the community of Arabi (NOAA 2022) and estimated peak wind speeds to be $257\mathrm{km}/\mathrm{h}$ (160 mph). Approximately 200 structures were damaged by the Arabi tornado, causing an estimated $30 million in economic losses, and tragically there were two reported fatalities, one direct and one indirect (NOAA 2022). The tornado primarily affected single-family homes.

The New Orleans area, including Arabi, has witnessed several hurricanes and tornadoes in the past two decades, summarized in Table 1. Most notably, Hurricane Katrina impacted the region in 2005 leading to widespread flooding and costing over $15 billion in direct economic losses to residential buildings in the New Orleans region (Link 2010). The entire area affected by the 2022 Arabi tornado was flooded with inundation depths up to 2.14 m (7 ft) (NOAA 2020), although wind speeds were well below design levels. Since Katrina in 2005, the New Orleans area, including Arabi, has been impacted by Hurricanes Isaac (2012), Zeta (2020), and Ida (2021), although in each case wind speeds were well below design, and flooding was minimal. As a result, damage from these storms was primarily confined to building envelope failures and power infrastructure failures.

In addition to the recent hurricane impacts, 11 tornadoes have also affected the New Orleans area in the past two decades, including the 2022 Arabi tornado. Notably, these include an EF-3 in February 2017, two EF-2 tornadoes in February 2007, an EF-2 in February 2006, and the EF-3 tornado that struck Arabi in March 2022 and is the focus of this paper.

Recognizing the significant threat presented by windstorms, the state of Louisiana has been proactive in adopting and maintaining regulatory practices, including building code adoptions, targeted toward natural hazard mitigation (IBHS 2021). The International Building Code (2000 version) was first adopted in 2004, but there were no statewide requirements for residential buildings. In the aftermath of Hurricane Katrina, in 2006, the statewide Louisiana State Uniform Construction Code (LSUCC) was adopted, and it required enforcement of the latest International Residential Code (IRC) for new construction and voluntary adoption of Appendix J for existing buildings. At the time of the tornado in 2022, LSUCC enforced the 2015 version of the IRC.

As per current and previous versions of IRC enforced by the LSUCC, Arabi has a 700-year mean reoccurrence interval basic wind speed of $233\mathrm{km}/\mathrm{h}$ (145 mph), triggering the requirement for wind load design for residential structures based on the Wood Frame Construction Manual (AWC 2018), ICC 600 (ICC 2014, p. 600), or similar, and precludes the use of the prescriptive provisions of the IRC. In essence, residential structures built according to the IRC in Arabi could be considered engineered structures. In addition to these design requirements, St. Bernard Parish, where Arabi is located, has also enforced additional requirements to prevent flooding of residential buildings, such as requiring that the lowest floor of residential buildings be above the base flood elevation. In the most recent version of the building code, this requirement has changed to a minimum of 0.46 m (18 in.) above the base flood elevation. This requirement results in elevated buildings supported on foundations akin to pier and beam systems. However, elevating building foundations may alter wind loads on buildings, which is not well understood at present. Furthermore, since the elevation requirement is primarily focused on flood hazard mitigation, the elevated foundations are not required to resist uplift caused by wind loads (except in special flood zones). The presence of wind and flood hazards results in competing design requirements (Hughes et al. 2022; Kameshwar and Padgett 2014; Padgett and Kameshwar 2016). Elevating buildings is beneficial for preventing flood damage; however, it may have detrimental effects on the wind performance of buildings, especially without sufficient anchoring of the foundation to resist wind-induced uplift and drag forces.

Posttornado reconnaissance activities were performed on March 25–29, 2022, by researchers from six universities. The response was coordinated by the structural extreme events reconnaissance (StEER) network (Kijewski-Correa et al. 2021). The response strategy focused on deploying rapid imaging technologies to capture the entirety of the tornado path from aerial and terrestrial perspectives, conducting rapid on-site assessments of structural performance, conducting forensic load path assessments of potential case study buildings, and characterizing the near-surface wind field through documentation of wind indicators. Table 2 summarizes the teams and members involved in data collection.

Direct measurement of the near-surface wind environment of tornadoes remains a considerable challenge due to the infrequent occurrence of a tornado for any single location and the short duration and extreme intensity of tornadoes coupled with the sparsity of near-surface measurement instrumentation. Existing networks of near-surface wind measurements, such as the various mesonets (e.g., Brock et al. 1995; Schroeder et al. 2005), are not nearly dense enough to regularly sample tornadoes, and even when they are impacted, they are often damaged by the tornado (Roueche et al. 2019), although a few reliable measurements exist (Lombardo 2018). Fixed radar, including the WSR-88D network (Heiss et al. 1990), has broader coverage, but even at the lowest beam elevations (e.g., 0.5°), the beam heights are typically sampling tornadoes a kilometer or higher above the surface, and there is insufficient knowledge at present to reliably bring these wind speeds down to surface levels through modeling. Mobile radar units have been effective at sampling lower elevations but are less effective in treed areas and areas with significant topography, and even in the more ideal settings of the US plains, relatively few high-quality data sets exist. In light of all these challenges, indirect methods such as the enhanced Fujita (EF) Scale (Edwards et al. 2013; McDonald et al. 2012), which is damage-based, and directional patterns of wind indicators such as trees and signs (Chen and Lombardo 2019; Lombardo et al. 2015; Rhee and Lombardo 2018) are typically used for estimating tornado intensity, while direct measurements are rare. For the Arabi tornado, researchers conditioned a tornado wind field using directional patterns of felled trees to keep the estimates independent of the building damage, as no direct measurements were available. The nearest fixed radar (KLIX) was 40 km from the impacted areas, with the lowest beam height of 0.5 km (1,600 ft), and no automated surface observation station (ASOS) or other near-surface measurement stations were close to the tornado.

For the tree-fall method, the location and fall direction of 36 trees were recorded either on-site or using the orthomosaics generated from the photographs taken by the UAS. Most of the downed trees were in the southern half of the tornado path through Arabi. The Chen and Lombardo (2019) method was utilized to estimate wind speeds from tree-fall. A steady vortex model is assumed to move along the tornado path with a translation speed $V_T$, meaning the model-generated tree-fall pattern on all transection lines is identical. In this way, the tree-fall patterns are condensed to a profile line defined by $L$, which is the minimum distance from the location of the fallen tree to the tornado path, and $\beta$, which is the angle of the fallen tree. For exploring the relationship between $L$ and $\beta$, the modified Rankine vortex model is applied with various parameters ($\eta$, $G_{\max }$; $\alpha$; $R_{\max }$; $\phi$) for estimating the tree-fall transection profiles as shown (Chen and Lombardo 2019). The input parameters are defined as follows: $\eta$ is the ratio between the tree-fall critical speed to the vortex translational speed $V_T$; $G_{\max }$ is the ratio between the maximum vortex rotational speed to the $V_T$; $\alpha$ is the angle of the vortex rotational speed; $R_{\max }$ is the maximum radius where the full rotational speed occurs, and $\phi$ is the modification parameter for Rankine Vortex profiles.

For each trial, one parameter combination is picked, and an estimated $L$ is generated for each input $\beta$ from the transection line. Then, the simulation error for this trial is the difference between the estimated $L$ and the practical distance obtained from the damage survey. The best-fit parameter combination with the minimum error is picked ($\eta =1.2$ $G_{\max }=1.3$; $\alpha =82$; $R_{\max }=50\mathrm{m}$; $\phi =0.8$) for the entire region. The peak wind speed was estimated through Eq. (1), based on the tree-fall model analysis, and for the whole area was approximately $45\mathrm{m}/\mathrm{s}$

Fig. 4 shows a cross section cut through the peak gust wind field. The maximum wind gust profile is skewed to the right, with the peak gust close to the radius to maximum wind speeds, because the best fit for $\alpha$ was close to 90°, suggesting predominately tangential flow. With predominately tangential flow, the translational and maximum horizontal wind speeds align near the right radius to maximum winds.

Most structures impacted by the tornado were wood-frame, single-family residential buildings, and thus these structures also made up a majority of the performance assessments (192 out of 210). The structural load paths observed amongst these were consistent by era, broadly defined as pre-Katrina and post-Katrina, representing homes built prior to and after Hurricane Katrina (2005), respectively. The focus of this paper is on the post-Katrina homes due to their enhanced construction and relevance to tornado-resistant construction.

Influenced by the region being in a floodplain, and being in a hurricane-prone region with basic wind speeds of $233\mathrm{km}/\mathrm{h}$ (145 mph), the post-Katrina homes had similar load paths (conceptually illustrated in Fig. 5), being constructed on elevated foundations, and utilizing steel straps to connect the roof to the walls and walls to the foundation. The homes were constructed atop concrete masonry piers anchored to continuous concrete footings [0.3 m (1 ft) wide by a depth that varied] by steel rebar extending from the footing into the cells of the masonry piers. The observed spacing of the piers varied between 1.2 m (4 ft) and 2.3 m (7.5 ft). A pressure-treated $6\times 6$ timber beam (14 cm by 14 cm) typically spanned across each masonry pier and was anchored to each pier using one 12.7-mm (0.5-in.) diameter A36 steel bolt embedded into a grouted cell of the masonry block pier, or in some cases by steel straps embedded into grouted cells of the masonry piers that wrapped over the beams and fastened with nails. $2\times 10$ (3.8 cm wide by 23.5 cm deep) or $2\times 12$ (3.8 cm wide by 28.6 cm deep) wood floor joists spanned across the timber beams, spaced 40.5 cm (16-in.) on center and connected to the timber beams by a galvanized steel hurricane strap (Simpson StrongTie H2.5A or equivalent). The plywood subfloor was nailed to the floor joists (exact nail size and spacing were not recorded). Light wood-frame stud walls were fastened to the subfloor with nails through the bottom plate into the subfloor and sometimes the floor joists underneath. Studs (typically $2\times 4$) were typically fastened to the bottom plate and the lower top plate by two nails (4 mm by 89 mm; 0.165 in. by 3.5 in.) into the end grain of the studs. The upper top plate was fastened to the lower top plate by nails at approximately 30 cm on the center, but the exact spacing varied. Engineered wood sheathing, usually 11.1 mm (0.44-in.) thick oriented strand board (OSB), was fastened to the exterior edge of the wall studs. In some cases, the OSB sheathing was continuous from the bottom of the rim joist to the upper top plate, but it was common for the OSB to only span from the lower of the top plates to the bottom edge of the bottom plate, with a separate sheathing panel covering the rim joist. The roof structure was either wood roof joists and rafters, or engineered wood trusses. When trusses were used, the individual truss members were observed to be $2\times 4$. When roof joists and rafters were used, the size of the members varied according to the span, but they were typically $2\times 6$ or $2\times 8$. The exact connection of the roof structure to the wall varied but typically consisted of a steel hurricane strap (either H2.5A or H8) in combination with three toe-nails. Both a hurricane strap and toe-nails are typically present in a connection because during construction, the toe-nails are used to initially set the roof framing member, and then the hurricane straps are installed at a later date. In rafter construction, the ridge was typically framed with a ridge board, with the rafters on one side end-nailed, and on the other side toe-nailed, to the ridge board. A collar tie was typically observed on every other rafter. Roof sheathing was either plywood or OSB panels [1.22 m by 2.43 m by 11.1 mm (4 ft by 8 ft by 0.44-in.) thick] and was nailed to the roof structure with a variety of fastener sizes and spacings. Ignoring some minor variations, the presented encompasses the majority of the load paths encountered in post-Katrina homes impacted by the tornado.

Despite the use of anchor bolts and steel hurricane straps as described in the previous section (common indicators of wind-resistant construction) in both pre-Katrina and post-Katrina homes, failures were widespread throughout the tornado path regardless of the construction era.

To illustrate the impact of various weak links in the structural load path on the overall system performance, a fragility analysis is performed for an illustrative benchmark building. The chosen benchmark building is a 9.1 m by 18.3 m single-story home with a hip roof and $7:12$ roof slope (30°), representing a typical home in the Arabi reconnaissance data set. The structural system is taken to be wood-frame, while other details of the structural load path are varied to represent the field observations. The fragility analysis considers two independent limit states, the first being the failure of a single roof sheathing panel, and the second being the failure of any connection within an idealized vertical load path other than roof sheathing. The failure limit state distribution is defined in Eq. (2) as follows:where $\mathit{W}$ = random variable representing the uplift wind load; $\mathit{R}$ = random variable representing the resistance of various load path elements; and $\mathit{G}$ = random variable representing the self-weight of the elements. The random variable for uplift wind load, $\mathit{W}$, is expressed as a pressure for the roof sheathing analysis, or a force per unit length along the house perimeter for the remainder of the vertical structural load path. While lateral wind loads would contribute to uplift forces and member failures, particularly at the foundation level, their contribution is ignored in this study for simplicity and since the purpose is to provide relative comparisons between various load path options. Subsequently, the resulting fragility functions likely overestimate the true capacity.

In Eq. (2), W is obtained using the ASCE 7 provisions in two ways, once using the requirements of ASCE 7-16 (straight-line wind provisions that would have been applicable to the most recently constructed homes in Arabi, LA) (ASCE 2017) and a second time using the ASCE 7-22 tornado provisions (ASCE 2022). While the tornado provisions have not been adopted in Arabi and therefore do not apply to the homes affected by the tornado, they are used here as a best estimate of tornado-induced wind loads on a typical building. W can be defined then as follows:where $K_z$ = terrain and height adjustment factors; $K_d$ = directionality adjustment factor; $K_{zt}$ = topographic factor; $K_e$ = elevation factor; $V$ = wind speed in $\mathrm{m}/\mathrm{s}$ (an independent variable in the fragility analysis); $K_{vT}$ =tornado pressure coefficient adjustment factor; and $G{C}_p$ and $G{C}_{pi}$ = external and internal pressure coefficients for components and cladding. The $T$ or Tor subscript found in many of these variables for $W_{7–22}$ indicate that these parameters have been modified in ASCE 7-22 from their original values based on straightline winds to account for tornado effects, including the atmospheric pressure deficit, nonmonotonic vertical velocity profile, and vertical wind component. For the sheathing failure limit state, the sheathing panel area was used to evaluate the ${GC}_p$ taken from Chapter 30 Components and Cladding, while for the vertical load path analysis, the gust effect factor is a separate variable and $C_p$ is taken from Chapter 27 Main Wind Force Resisting System provisions of ASCE 7.

The details of G and R in Eq. (2) differ between the sheathing and vertical load path analysis frameworks. For the sheathing failure limit state, G is the gravity load contributed primarily by the weight of the roof covering (assumed to be asphalt shingles). The resistance R of the panel is taken from Datin et al. (2011), specifically utilizing the capacities for 8d common and 8d ring-shank nails at $6:6$, $6:8$, and $6:12$ spacing in a combined lognormal distribution to account for the variability encountered in the field reconnaissance. An alternative configuration is also considered in which the resistances from Datin et al. (2011) are adjusted to account for the overdriving of the fasteners as observed in the field using Eq. (5)where $R_{\mathrm{adj}}$ = adjusted sheathing resistance; and $R_{\mathrm{typ}}$ = unadjusted sheathing resistance from Datin et al. (2011); $F_{od}$ = overdrive factor, defined as $t_{od}/t_n$, where $t_{od}$ is the thickness of the sheathing panel considering the overdriven fastener, and $t_n$ is the nominal thickness of the panel (taken to be 11.1 mm). $F_{od}$ is defined as a normally distributed random variable with a mean of 0.85 and a COV of 0.2, truncated within the range of [0.7 1.0]. The sheathing fastener configurations considered are summarized in Table 4. Roof sheathing fragilities are developed via Monte Carlo simulation at different levels of the wind gust velocity (V) as an intensity measure. Failure of any panel constitutes the failure of the roof.

Roof sheathing is considered in a separate fragility analysis because the sheathing resistance model is formulated differently from the other elements in the vertical load path and because the wind loads are derived from ASCE 7 for Component and Cladding with wind load statistics as given in Table 5. The remainder of the typical load path illustrated in Fig. 5 is modeled as a load chain with connection resistances expressed in terms of force per unit length of wall and wind loads derived from ASCE 7 Main Wind Force Resisting System approaches. A more detailed description of this resistance modeling approach is given in Rittelmeyer and Roueche (2022). The main steps in the approach are as follows. (1) Uplift resistance of individual connections is taken as the sum of connection capacity and cumulative dead load, where capacities are computed either from LRFD-based strength equations provided in applicable design specifications or from a comparable mechanics-based model. (2) Capacities of parallel connections, such as the stud-to-plate and overlapping wall sheathing connections, are summed to form a load chain of connections in series. (3) After consolidating parallel connections, the system-level uplift resistance of the load path is taken as the resistance of the weakest connection in series. Rather than developing fragility functions separately for each load path connection, the use of a combined limit state applied to the whole load path (excluding roof sheathing) as a series system offers the advantage of evaluating the effective impact of various configurations on building performance. As the failure observations surveyed in the preceding section indicate, system-level performance is often limited by characteristic weaknesses, typically in the roof-to-wall or wall-to-foundation connection. Wherever the characteristic weakness is present, the system-level fragility closely resembles what the component-level fragility for that connection would be in isolation.

The input parameter set used to compute connection resistances is summarized in Table 6, which contains probability distributions for the nondeterministic parameters in the roof, wall, and floor systems of the characteristic load path in Fig. 5. Within this resistance framework, distributions may be continuous or discrete and numerical or categorical. Certain parameters are coupled, such as wood species and specific gravity. Some categorical parameters control connection configurations by selecting from a defined list of possible fastener options. Discrete distributions may be weighted according to the relative frequency of each outcome. Distributions are assigned to be representative of actual structural observations in Arabi in terms of the set of possible values and relative probabilities of occurrence.

Load path fragility is evaluated using the wind load provisions of ASCE 7-16 and the tornado provisions of ASCE 7-22 and using the Main Wind Force Resisting System (specifically Chapter 27) approaches for both. Load statistics for both models are presented in Table 7. The tornado provisions of ASCE 7-22 include several amplification factors to account for the higher wind pressures that can be experienced in tornadoes compared to straightline winds for a given wind speed, such as the atmospheric pressure deficit, nonmonotonic velocity profile, and vertical winds.

The load path configurations considered in the vertical load path fragility analysis focus on the roof-to-wall connection type (toe-nail, H2.5 with toe-nail, and H8 with toe-nail) and whether the wall sheathing overlaps both stud wall top plates or only the lower top plate or whether wall sheathing is continuous across the wall-to-floor connection. The various configurations are summarized in Table 8.

Fragility curves resulting from the roof sheathing and load path analyses are presented in Figs. 10 and 11, in which Monte Carlo simulation results based on 50,000 model runs for each configuration have been fitted to a lognormal CDF with a logarithmic mean $\mu$ and standard deviation $\sigma$ as listed in the Supplemental Materials. The roof sheathing fragility results in Fig. 10(a), based on the ASCE 7-16 load model, indicate upper and lower bound median failure wind speeds of 89 and $72\mathrm{m}/\mathrm{s}$ for the limit state of at least one panel failure. Including a factor to account for overdriven fasteners reduces median failure wind speeds by about 9%. Results based on the ASCE 7-22 load model, shown in Fig. 10(b), represent an average reduction in median failure wind speed of 16% relative to the ASCE 7-16 cases, with upper and lower bounds of 74 and $60\mathrm{m}/\mathrm{s}$. Model results are also compared to empirical fragilities in Fig. 10: an upper-bound curve based on observed roof substrate damage and a lower-bound curve based on observed roof substrate damage but excluding roof structure damage.

Results of the load path fragility analysis in Fig. 11 expectedly demonstrate higher variance than those of the roof sheathing analysis. The ASCE 7-16 load model produces upper and lower bound median failure wind speeds of 93 and $60\mathrm{m}/\mathrm{s}$, while the ASCE 7-22 loading reduces median failure wind speeds by 14% on average and leads to upper and lower bounds of 81 and $39\mathrm{m}/\mathrm{s}$. Across all configurations considered, the toe-nailed roof-to-wall connection is consistently a limiting link in the load path; continuous sheathing across the wall-to-foundation connection (Case T-0-1) does not meaningfully improve system performance, while overlapping wall sheathing at the upper top plate (Cases T-1-0 and T-1-1) improves performance slightly. The addition of an H2.5 hurricane tie strengthens the roof-to-wall connection but only marginally increases load path resistance by 7% (as measured by the expected failure wind speed) if the double-top plate connection remains unreinforced by overlapping wall sheathing. Because the H8 hurricane tie also strengthens the double top plate connection, configurations that include an H8 substantially improve performance relative to H2.5 cases in the absence of a top plate sheathing overlap (e.g., Case H8-0-0 improves the expected failure wind speed by 30% relative to H25-0-0) and marginally improve performance otherwise (e.g., Cases H8-1-0 and H25-1-0). Cases H8-0-0, H25-1-0, and H8-1-0 are indistinguishable for the same reason since the limiting link for these configurations is in the wall-to-foundation connection, which in these cases is not reinforced by continuous sheathing. The fragility fit parameters for all cases are available in the Supplementary Materials.

For all combinations and limit states considered, the analytical fragility modeling revealed that median failure wind speeds when considering ASCE 7-16 straightline wind loads were around 15% higher than when considering ASCE 7-22 tornado load provisions. In other words, the tornado provisions tend to produce higher wind loads and thus lower failure wind speeds when compared to straightline wind loads.

To facilitate a comparison between the analytical fragility functions and the field observations, empirical fragility functions were also generated, using the damage observed to individual homes in the reconnaissance study and the tree-fall-conditioned wind field described earlier. The limit states were chosen to correspond to the analytical fragility functions and consisted of (1) uplift of at least one roof sheathing panel, or (2) any roof structure or wall structure damage, where the structure is defined as broken or missing roof or wall framing elements (e.g., roof trusses, rafters, wall studs). During the field study, roof sheathing was assumed to be damaged if it was removed from the roof structure, and the damage ratio was defined as the area of missing roof sheathing divided by the total roof sheathing area. Since it is possible that the roof sheathing remained attached to the roof structure, but large sections of the roof structure itself failed, the upper and lower bounds of the roof sheathing damage were used for the limit state. The lower bound assumed that the true roof sheathing damage area is the difference between the total roof sheathing damage area and the roof structure damage area (i.e., only the roof sheathing missing from portions of the roof structure that remained was counted). The upper bound assumed that all missing roof sheathing was truly damaged. Given these limit states, the empirical fragility functions were fitted using a generalized linear model with a probit link function, as described in Roueche et al. (2017).

The empirical and analytical fragility functions overall lack agreement, both for sheathing failure (Fig. 10) and system failure (Fig. 11) limit states. In all comparisons, the empirical fragility functions are shifted to the left of the analytical functions, suggesting that the structures failed at lower wind speeds than those predicted by the analytical model. The empirical fragility functions also have higher variability than the analytically derived functions. The discrepancy between the empirical and analytical fragility functions could be attributed to one or more of the following causes that relate to both loads and resistance:

This study investigated the impacts of a strong tornado, rated EF-3 by the National Weather Service, that struck Arabi, LA, on March 22, 2022. The study integrated rapid imaging techniques, forensic damage assessments, and tree-fall conditioned wind field models to contextualize the performance of mostly modern (post-2005) light wood-frame structures constructed to hurricane-resistant building codes that were impacted by the tornado. The observed levels of damage, caused by a tornado, are unfortunately not rare in the United States, but the juxtaposition of the observed damage levels in a hurricane-prone region with enhanced construction techniques has important implications for understanding and mitigating tornado damage risks elsewhere. Further, the study pilots a framework for developing both empirical and analytical fragility functions for a tornado tuned to the specific load path observations from the forensic investigations that can be a model for future tornado assessments. The major findings from these efforts are summarized as follows:

The findings of this study are limited by several factors. The relatively few felled trees heighten the uncertainty and potential biases in the tree-fall-conditioned wind speed estimates. Further, the analytical fragility modeling was limited to a single aerodynamic archetype and only considered the vertical load path while neglecting the contributions of lateral loads toward failure. The wind load analysis was also based on static wind load design provisions, rather than wind tunnel data or computational fluid dynamics simulations tailored to the event. Nonetheless, the study has broader applicability beyond this specific tornado event for future tornado assessments and tornado risk mitigation efforts.

All data and code generated or used during the study are available in the Natural Hazards Engineering Research Infrastructure DesignSafe platform in accordance with funder data retention policies. (PRJ-3443, doi: https://doi.org/10.17603/ds2-pdfb-d686).

This work was sponsored in part by the National Science Foundation under grant CAREER-1944149 and research grants CMMI-1841667 and 2103550. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. The work was also partially funded by the National Oceanic and Atmospheric Administration through Grant Nos. NA19OAR4590213, NA19OAR4590212, and NA19OAR4590214. The authors acknowledge and appreciate the contributions of the following undergraduate students who served as Data Librarians in curating the reconnaissance data: Hannah Bartels, Morgan Aldridge, Andrew Golson, Fox Harris, JD Holt, and Cason Stevens. The authors also gratefully acknowledge Daphne LaDue (University of Oklahoma) for assisting with the field reconnaissance, and specifically engaging homeowners to facilitate access to the interior of homes for more detailed load path assessments. And finally, we gratefully acknowledge the homeowners and their willingness to support the study despite the recent impacts.