AbstractThis study presents a model-based framework to assess the time-varying accessibility of a roadway network on a system-wide level during extreme flood events. A regional MIKE 21 hydrodynamic model consisting of 1,912,576 computational points with mesh cell resolutions ranging from 70 to 15 m is utilized to compute regional inundation during an extreme 500-year flood scenario. This approach allows for an explicit accounting of the impact of pluvial flooding on roadway network accessibility. Accessibility conditioned on flood depth is then computed using a raster approximation of the roadway network model employing the flood-fill search method. The approach is demonstrated in the flood-prone low-gradient region of Lafayette, Louisiana, which experienced a devastating flood event in August 2016. The findings suggest that the main evacuation points enjoy a greater degree of accessibility compared to medical facilities within the urban core of the city during the flood peak. Significant improvements in network accessibility can be made by targeted mitigation of specifically flood-prone roadway segments. However, the analysis demonstrates the adverse impacts of such mitigation activities in low-gradient urban floodplain systems when implementing routine drainage compensation steps. The approach provides key insights into the role played by pluvial flooding and flood duration on network accessibility and can help inform emergency response and transportation systems planning and design efforts.IntroductionExtreme flood events are expected to increase in wet regions as a result of climate change (Tabari 2020). Relative sea level rise, shifts in population patterns, and changes in land use create additional uncertainty in the management of future risks (Wing et al. 2018). These factors create serious challenges for flood risk managers, engineers, and environmental planners in their efforts to sustainably work with water while protecting public safety and infrastructure in flood-prone communities. Roadway networks are a critical aspect of urban infrastructure systems and express unique vulnerabilities to flood hazards (Abdulla et al. 2020). Spatial constraints on roadway coverage in a given region (e.g., property ownership, cost, and vertical design considerations) can lead to roads being built in flood-prone areas. Beyond the immediate hazard to motorists associated with inundation hotspots, the indirect impacts of a loss of accessibility to critical resources may affect entire communities (Miller 2019).The direct impact of flood hazards on motorists, roads, and transportation disruption has been analyzed. The widespread availability of regional and continental flood hazard data sets (Wing et al. 2018) facilitates the identification of flood-prone roadway segments. These flooded segments may present a life-threatening situation for motorists. Ashley and Ashley (2008) report that vehicles were involved in 63% of the flood-related fatalities when the setting of occurrence or activity was known. Pavement lifespan and long-term maintenance costs are also impacted by road flooding. Lu et al. (2020) studied the effects of flood hazards on road pavement and showed that damage ratios increase as the number of flood cycles increase. Structural failure of drainage infrastructure (e.g., bridges and culverts) has also been linked to flood submergence. In a study in the Pacific Northwest, the USDA found that plugged culverts accounted for 28% of storm-induced road damage based on their analysis of 352 sites (Copstead and Johansen 1998). As suggested by Coles et al. (2017), there is a need to better understand the broader consequences of flooded roads beyond the direct impacts associated with flood hotspots. The proliferation of two-dimensional (2D) hydrodynamic modeling simulations coupled with recent advancements in remotely sensed digital topographic surface products make it possible to analyze the effects of flood hotspots as well as time-varying flood phenomena on local, regional, and continental scales. These methods can be coupled with high-resolution roadway network data to detect latent vulnerabilities related to street flooding. One such vulnerability is the inaccessibility of critical locations to motorists during an extreme flood event.The accessibility of roadway networks during flood events is a well-studied topic. The problem of flood-induced disruptions on roadway networks can be initially conceptualized as a specific type of impairment in a broader context of transportation network robustness (Schneider et al. 2011). Albano et al. (2014) used a GIS-based approach to estimate the degree of accessibility to emergency response structures in support of strategic planning and emergency response efforts for the floodplain of the Bradano River located in Italy. Their study evaluated the impacts to ambulance and fire services during the peaks of various design storm flooding scenarios based on MIKE FLOOD simulations. Coles et al. (2017) also integrated hydrodynamic modeling to evaluate flood-induced vulnerabilities including impacts on emergency service areas and travel times during the flood peak. Gori et al. (2020) examined time-varying accessibility impacts in Houston, Texas, due to flooding from Hurricane Harvey. Their multidisciplinary study focused primarily on fluvial flooding sources using Hydrologic Engineering Center’s River Analysis System (HEC-RAS) models associated with the Brays and Greens Bayou subregions. Key differences were revealed in the time-varying accessibility impacts owing to social vulnerability factors (Gori et al. 2020). Abdulla et al. (2020) also selected a neighborhood in the City of Houston to analyze flood-induced road network vulnerabilities under different impairment scenarios. Their methodology included the usage of an SIS network diffusion model that used the giant connected component (GCC) framework to measure the flood impacts on the network. However, Loreti et al. (2022) studied the effects of local impacts on roadway networks during extreme floods and concluded that the GCC may not be as relevant as the accessibility of specific localities where critical services can be obtained. The previous studies suggest that the local accessibility to specific refuge points is a key aspect in terms of accurately evaluating the total impact of flood-induced disruptions on transportation networks. Such considerations should also factor into the development of equitable mitigation strategies that account for existing disparities in access to critical resources (Gori et al. 2020). The emphasis of the preceding studies has been mainly on impacts induced by fluvial (i.e., flooding from river/channel overflow) flood hazards.In comparison with fluvial flooding, pluvial floods (i.e., localized surface/flash-type flooding) are generally more challenging to predict (Houston et al. 2011). The time-varying impacts of pluvial flooding on road network accessibility have been relatively understudied in comparison with the more well-defined riverine and coastal flood hazards. Also, fluvial-centric studies tend to be limited to specific subwatersheds. This may lead to an underestimation of the system-wide flood-induced accessibility impairment on municipal jurisdictional scales (e.g., city and county boundaries) that constrain planners and managers but do not align with hydrologic boundaries.The main objective in this paper is to develop a raster-based methodology capturing the time-varying accessibility of urban roadway networks on a countywide scale during an extreme flood scenario. The flood scenario considered here takes into account pluvial flooding on a jurisdictional scale (i.e., county/parish-wide basis), focusing on the accessibility of typical motorists to emergency centers and key hurricane evacuation points. The method is demonstrated in Lafayette Parish, Louisiana—one of the fastest growing municipalities in the state, which also experienced catastrophic damage during the historic flood event of August 2016. The time-varying accessibility of the entire network is quantified, including smaller neighborhood streets that are often neglected in large-scale studies. The impact of strategic flood mitigation of critical roadway corridors frequently impaired by pluvial flooding, including potential adverse effects, is also analyzed. The paper is organized in the following manner. Section “Methods” describes the study area, hydrodynamic modeling effort, and raster-based search procedure used to quantify regional accessibility. Section “Results” focuses on the main results of the work, and section “Discussion” discusses the key findings situated in the broader context of related studies and more general settings. Section “Conclusions” summarizes the main takeaways of the study and its limitations, as well as potential opportunities for future research.MethodsStudy Area DescriptionLafayette Parish (696 km2) is a south Louisiana municipality located at approximately 30.21N, 92.06W (Fig. 1). As of 2019, Lafayette Parish possessed a total population of 244,390 (USCB 2022), making it the fourth largest municipality in the state of Louisiana. Despite population declines at the state level, Lafayette has experienced a 10% increase in population since 2010 (Potter 2021).Hydrology and ClimateLafayette has a subtropical humid climate and averages approximately 152.4 cm of precipitation per year (NWS 2022). Soils are generally poorly drained (Hydrologic groups C and D) with slopes generally on the order of 0.1%. The low-gradient topography and soils coupled with frequent intense rainfall events makes flash flooding a common occurrence in Lafayette Parish. Lafayette was affected by the great flood of 1927 as well as a historic event occurring in August 1940. Recent notable events include major flooding in January 1993 and June 2001, moderate floods in May 2008 and May–June 2014, and a historic event during August 2016.Analysis FrameworkThe analysis framework evaluates the entire county roadway network including residential streets. The rationale is that pluvial inundation of key residential streets creates unsafe driving conditions, restricting access to critical resources for residents whose properties may be otherwise protected from flooding. The roadway data set consisted of a one-dimensional (1D) polyline vector roadway network spanning 2,603.9 km available from the effective Flood Insurance Study (FIS) published by FEMA. The 1D polylines were transformed in a GIS into a 2D rasterized representation. Although this representation differs from the typical polyline-based graph theoretic framework, the rasterized approach (described later) was taken as a first step to account for the rerouting problem whereby motorists will typically seek any available pathway to reach their destination during a flood. Such considerations should factor into realistic evaluations of roadway network accessibility problems (Loreti et al. 2022). Light detection and ranging (LiDAR) digital topographic surface data were sampled onto the roadway network raster to develop the roadway digital surface elevation model.A multiresolution MIKE 21 hydrodynamic model was developed for the entire parish, thereby producing time-varying inundation surfaces for various precipitation events (Fig. 1). MIKE 21 is a proprietary numerical modeling program that simulates 2D flow problems by solving the St. Venant equations of shallow water dynamics using the finite-difference method (DHI 2017). The 500-year return period was utilized given its significance in the protection of critical facilities in the US and its local relevance as a close approximation to the 2016 flood. The resulting composite inundation surfaces are then intersected with the roadway surface model, thereby estimating depths of flooded roads throughout the network. A maximum safe navigability depth is imposed, which defines the impassable partition of the network. The accessibility of all locations in the network relative to fixed resource points is then established using the flood-fill search algorithm. Geospatial maps and summary accessibility statistics are developed for further analysis.Raster-Based Accessibility Analysis and Data ProcessingThe geospatial coordinate location of each point within the roadway network was first transformed into a georeferenced raster (matrix) map. Topographic elevations were then assigned to each point within the roadway network via sampling from the available LiDAR data. The preceding steps yield a three-dimensional surface representation of the roadway network relative to a datum. Flood depths at each roadway point were then estimated by subtracting the simulated water surface elevation data from the roadway raster elevation. Given a specified navigability threshold (which should depend on vehicle ride height), a new raster depicting the drivability (or equivalently, navigability) of each network location was developed at each time increment throughout the flood event. The D8 flood-fill algorithm was then utilized to identify all locations in the network from which motorists would not be able to access (i.e., reach) a resource target point (Fig. 2). Given a target point, an arbitrary location on the roadway grid is theoretically accessible/reachable if and only if there exists a continuous path to the target. This method notably gives equal consideration to all possible maneuvers (within the roadway network) and all potential pathways for motorists under the assumption that motorists may ignore conventional traffic regulations during floods in response to an emergency/crisis situation. This framework also tacitly assumes that motorists would explore all possible within-network pathways if necessary to reach the target point. Although some types of ad hoc bypassing maneuvers (e.g., motorists crossing medians, parking lots, etc.) were not considered, the raster-based framework can be easily extended to examine such flood-induced effects.The roadway surface model raster consisted of square cells having a 20-m width. Special care was taken to enforce network continuity across channel features that may cause artificial sinks. As a first step, nonroadway transportation features were eliminated from the data set and roadways that were known to be absent from the data set were manually added. An estimation of over 900 drainage crossings was made by intersecting the roadway network and the available drainage channel feature vector data. The topographic data at each crossing location was inspected and the road raster elevations were manually corrected as needed to eliminate any spurious discontinuities and to factor in overpasses and bridges in the raster network. A similar processing step was performed to deal with discontinuities arising from the LiDAR resampling operation. The final processing step was an initial sweep of the flood-fill algorithm to confirm 100% accessibility of the raster network under unflooded conditions.Hydrologic and Hydrodynamic ModelingA 2D modeling approach was employed utilizing the MIKE 21 inland flooding simulation package. The model domain consisted of 16 subwatersheds encompassing the parish roadway network and proximal regions outside the parish jurisdictional boundary. Nearest-neighbor resampled bare-earth 5×5 LiDAR digital elevation model (DEM) was utilized to develop the digital topographic surface model for each subwatershed. Subwatershed boundaries were delineated to account for interbasin exchange (transfer of floodwaters across hydrologic divides) known to occur in Lafayette’s extremely flat terrain.Square grid cells were used to construct the computational mesh whose resolutions ranged from 70×70 m to 15×15 m, with the coarser resolutions used for areas with larger channel widths. A total of 1,912,576 computational points were employed in the model. Steps were taken to ensure drainage continuity in each subwatershed by implementing a D4 flow path burning procedure prior to simulation execution. The D4 method was selected for consistency with the numerical method employed to integrate the 2D hydrodynamic model where fluxes are only calculated along the principle x- and y-axes.The 24-h 500-year return period frequency event was selected as the rainfall-runoff framework to provide scalable standardized reference conditions consistent with FEMA regulatory products and standard engineering design practice. The runoff inputs were developed using a rain-on-grid (RoG) approach whereby the precipitation scenario is modeled by applying the precipitation rate as a source directly to the grid cells. The 24-h soil conservation service (SCS) Type 3 dimensionless precipitation hyetograph was utilized with a 500-year total rainfall depth of 39.6 cm. The total storm depth was obtained via extrapolation based on a logarithmic regression relationship fitted to the 24-h design storm rainfall depths for various design storm frequencies provided for Region 1 in the current Louisiana DOT and Development (LADOTD) Hydraulics Design Manual (LADOTD 2011). Infiltration losses were accounted for by reduction of the precipitation inputs using an equivalent runoff curve number (CN) of 85 reflective of the dominant land cover and soil types in the region. This value is consistent with the predominant land cover and types of the Atchafalaya/Teche-Vermilion Basin hydrologic region as described in greater detail in a previous study (Miller 2022). Depression abstractions are accounted for directly within the computational grid.Hourly discharge hydrographs were extracted to provide headwater inputs for downstream subwatersheds. Varying computational time steps were employed based on numerical stability considerations (i.e., Courant number), but in general the time steps did not exceed 5 s. The simulation period was 3 days (72 h).Uncertain Parameters and Field Verification DataThe main uncertain parameters for the RoG approach were the Manning resistance and the numerical wet/dry depth tolerance. In this model, a static Manning resistance coefficient was applied from which depth-dependent Chezy numbers were calculated at runtime using C=h16/n. A value of n=0.067 was employed for all subregions and was found to provide satisfactory performance based on local prior studies in the region. Both the accuracy and computational efficiency of RoG simulations are sensitive to the wet/dry tolerance. A drying depth that is too large will increase computational efficiency but result in artificially low runoff hydrographs. For all simulations a wetting depth of 2 mm and drying depth of 3 mm was found to provide reasonable results based on initial test runs.The flood of 2016 was taken as a representative surrogate for the 500-year regional event based on the work by Brown et al. (2020), which provided a comprehensive assessment of intensity and duration of the event. The simulations suggest that much of the flooding in Lafayette Parish is pluvial, with peak 500-year depths ranging from 0.15 to 0.6 m with an approximate mean value of 0.3 m. Except for the Vermilion River, which is not the main source of residential flood risk, the entire parish was ungauged during the 2016 flood. At the time of this study, there were no real-time water level measurements available to support a full rigorous calibration of the shallow flood dynamics. As a result, a field campaign was undertaken during the flood event to collect flood pictures at strategic locations within the network. The photo database was used to document the effects of pluvial flooding on the road network and to aid in initial evaluation of the model performance. A total of 55 high water marks (HWMs) were extracted from the photos coupled with the LiDAR DEM data using methods similar to Wang et al. (2018). The photos also served to document impassable roadway conditions at various locations in the network.ResultsModel Performance EvaluationThe simulated mean peak 500-year water surface elevation was 8.75 m, while the mean HWMs were 8.61 m. The model standard deviation was 1.89 m, and the observed deviation among the HWMs was 1.93. The percent bias of the model was 14%, while the Nash-Sutcliffe efficiency (NSE), root mean square error (RMSE), and coefficient of determination (R2) relative to the HWM data were 0.96, 0.37, and 0.968, respectively for all comparison points overall. The preceding statistics suggest that the 500-year 24-h hydrodynamic model simulation results reasonably approximate the data observed in the August 2016 event (Fig. 3). The average peak deviation of 14 cm suggests that the model performs well in comparison with US regulatory base flood elevations, which have a nominal vertical accuracy of 30 cm (Maidment et al. 2009).Impact of Pluvial Flooding on AccessibilityAn initial analysis was conducted to evaluate the impacts of pluvial flooding on regional road flooding and impassibility. Employing a 20.3-cm safe driving depth for the typically most vulnerable and common vehicle type (cars), the impassible zones were delineated for both the 500-year hydrodynamic simulation event and the FEMA regulatory 500-year peak scenarios. It is noted that the relative navigability is sensitive to the safe driving depth threshold, and the selected 20.3-cm depth is within the 14–60-cm safe driving depth for cars considered by Abdulla et al. (2020) and is also below the maximum of 30 cm as determined by Pregnolato et al. (2017). The FEMA regulatory product was selected as the basis of comparison for inundation impacts given that the FEMA models have generally underestimated pluvial flooding in the Lafayette region. The results suggest that 24.8% and 23.8% of the roads in Lafayette are impassable during the 500-year scenario for the pluvial model and FEMA model, respectively (Fig. 4). The modest 4.1% underestimation of the FEMA model, however, fails to reflect the substantial difference in the spatial pattern that emerges when considering pluvial flooding. The flooded roads are more evenly distributed throughout the parish when considering pluvial flooding. The pluvial approach also detects potentially large swaths (209 ha bounding region) of impassable roads within the inner transportation core of the city that are not captured by the FEMA model.Accessibility to Evacuation PointsThe model was used to evaluate the time-varying accessibility for four target evacuation points. Accessibility from each raster grid cell in the roadway network was calculated relative to each target location based on the simulated flood depths at various points in time. In this manner, the motorist-based accessibility calculation can also serve as a proxy for accessibility relative to building locations (e.g., homes and business) located within the network. The accessibility target locations correspond with key evacuation points, located at interstate corridors crossing the north, east, south, and western boundaries of the parish. These corridors have supported hurricane-induced contraflow evacuations in the past.The results show that the evacuation points are nearly all equally accessible from a motorists’ standpoint during the 500-year event (Fig. 5). A sharp 75% reduction in network accessibility occurs over a 4-h period leading up to the storm peak for all evacuation points. The southerly site experiences the most extreme loss of accessibility, with the model suggesting that this site would be completely inaccessible to motorists during the 500-year event for up to 6 h. In general, more than 50% of the network would not be able to access the evacuation points for up to 12 h and 75% of the network would not have access for at least 6 h.Accessibility to Emergency FacilitiesAn evaluation of the accessibility to the four major medical facilities in Lafayette was also performed. Despite the relative proximity to one another in comparison to the distance between the evacuation points, the medical facilities exhibit significantly more variation in motorist accessibility (Fig. 5). All medical sites except LGMC-5 would likely be inaccessible to motorists for 4–6 h. Up to 80% of the network would not be able to access any of the medical facilities during the critical 6-h period. Site SWST-7 is particularly vulnerable to loss of access, with floodwaters potentially restricting all motorist access for 18–20 h. All the preceding accessibility results are based on the 20.3-cm safe driving depth criteria being imposed. Although this target takes the perspective of an average motorist, the 20.3-cm threshold is slightly below the 25-cm threshold for emergency vehicles used by Green et al. (2017) in their analysis of flood rescue operations in Leicester, UK.Sensitivity of Impacts to Storm IntensityThe accessibility evaluation was also conducted for the 100-year 24-h event (32 cm) (LADOTD 2011) to evaluate the sensitivity of the impact to design storm intensity. For brevity’s sake, the 100-year analysis was restricted to Target points 4 (WEST) and 7 (SWST). Fig. 6 provides a comparison of the 500-year versus 100-year simulation results. The results show that reducing the total storm precipitation by 19% lowers the peak flood-induced nonnavigability of the roadway network from 25% to 21% (Fig. 6). However, this reduction yields comparably marginal reductions in the inaccessibility to the selected target points when evaluating 500-year versus 100-year flood conditions. The evacuation target point (WEST-4) exhibited a larger reduction in inaccessibility resulting from a reduction in storm intensity compared to the emergency facility (SWST-7). The differences at both sites were more pronounced during the rising limb (i.e., the time when floodwaters are rising) compared to the falling limb. A comparison of the inaccessibility versus navigability at both locations [Figs. 6(a and b)] suggests that higher levels of inaccessibility occurred during the rising limb compared to the falling limb for similar (and lower) levels of overall navigability.Strategic Mitigation EffectsEmergency facility SWST-7 is located on Ambassador Caffery Parkway, which is the most heavily traveled state highway within the city limits of Lafayette (Fig. 7). The cause of the high level of flood-induced inaccessibility is the inundation of the nearby roadway intersection and adjacent areas during heavy rain events. This causes an impassable condition to occur directly at the entrance of the facility. Here the accessibility impact for SWST-7 is demonstrated by strategically mitigating the intersection and the surrounding roadways. This is done by simulating a grade adjustment by elevating the hydrodynamic model grid cells above the 500-year peak flood elevation. Cuts were also placed near defined stream locations in the elevated corridor in the model to simulate the conventional design provision to ensure positive drainage via equalizers and other similar drainage structures. The roadway surface model was adjusted by imposing a continuously elevated path along the mitigated corridors. Although the continuous elevation of a roadway corridor for several miles may be costly and difficult to achieve in practice, various practical emergency flood-fight countermeasures (e.g., aqua-dams and sandbag dykes) can generate similar effects. Table 1 shows that the corridor mitigation alternative reduced the percent of impassible roads from an average of 8.2% to 8.1%, which is a very modest 1.2% reduction. However, this caused the mean accessibility to increase from 57% to 70%, which is a 22.8% increase. Notably, during the critical period of nearly total inaccessibility occurring between 12 and 32 h of the storm, the grade adjustment allows an average of 37% accessibility with a minimum level of accessibility of 17%. These results indicate that substantial gains can occur by strategically mitigating flood-prone roadway corridors.Table 1. Summary of impacts associated with strategic mitigation of the critical transportation corridor shown in Fig. 7Table 1. Summary of impacts associated with strategic mitigation of the critical transportation corridor shown in Fig. 7Hours from start of flood event% impassable% accessible% impassable% accessible00%100%0%100%41%97%1%97%84%78%3%79%1214%1%14%18%1625%0%25%17%2019%0%19%44%2416%0%15%49%2812%1%12%58%3210%1%9%68%368%73%8%74%407%75%7%75%446%76%6%77%486%78%6%78%525%79%5%79%565%82%5%83%605%83%5%84%645%84%5%84%685%84%5%84%725%84%5%84%The impact to surrounding peak flood elevations was also analyzed. The impact Δpeak was taken as the difference between peak water surface elevation postalternative and peak existing water surface elevation. Fig. 7 suggests that ensuring positive drainage may still result in an objectionable increase (>1 cm) in peak water surface elevations as a result of strategic mitigation of roadway segments. This point is particularly salient for residential areas located outside of the regulatory flood zone because residents may not be required to carry flood insurance in this case.DiscussionBy incorporating pluvial flooding, this study complements the study by Gori et al. (2020) that focused exclusively on fluvial flooding from Hurricane Harvey in 2017. The results show that fluvial models (e.g., FEMA Flood Insurance Studies) may estimate a similar percentage of peak flooded roads but may not accurately represent the spatial distribution in comparison with models that account for pluvial flooding. This study also demonstrates a specific example of ancillary human costs associated with pluvial flood hazards, which are expected to increase with climate change (Houston et al. 2011).The results show that the 500-year 24-h duration event applied uniformly over Lafayette Parish approximated the August 2016 flood event in terms of the available HWM data (NSE=0.96). This result corroborates the regional frequency classification arrived at by Brown et al. (2020). The 500-year RMSE of 0.37 m obtained in comparison with the HWM data from the 2016 event is consistent with the 0.28–0.32-m performance range presented by Neal et al. (2009) for whole city urban-flood simulations. Although the analysis accounts for depth-dependent roughness, the base n resistance factor (0.067) falls well within the minimum RMSE contours as shown in Neal et al. (2009) for the Carlisle, UK, region. Spatial areal reduction factors (ARFs) were not applied, however the work by Kao et al. (2020) suggests that an ARF in the vicinity of 90% would be reasonable for this region given the size of the watershed. The usage of a design storm–based approach facilitates representative comparisons with other more frequent precipitation scenarios (e.g., 10- and 25-year). Actual storm events can also be easily incorporated using spatially heterogenous radar-based precipitation products and coupling with national hydrologic forecasting models as demonstrated in a recent study in this region by Saad and Habib (2021).The results showed that the main evacuation routes exhibit nearly identical accessibility profiles despite their relatively large physical separation. The model showed that the evacuation points could be equally inaccessible to 80% of the network during a critical 6-h period near the flood peak. A comparison of the effects of storm intensity showed that both the 500- and 100-year design storm conditions led to very similar consequences from a network accessibility standpoint in this study region. The evacuation target exhibited a larger response to overall storm intensity compared to the emergency facilities. However, the results showed that the degree of inaccessibility to an evacuation target exceeded 75% during both the 500- and 100-year peak conditions. This means that at least 75% of the roadway network would not be able to safely navigate to this evacuation target during the peak of such extreme rain events. Anarde et al. (2018) noted that inaccessibility due to flooded bridges on evacuation routes can also jeopardize disaster response activities in hurricane regions. The combination of these factors underscores the latent hazards associated with intense rain events occurring during hurricane evacuation operations.The emergency facilities showed greater variance in the percent of accessibility despite their relatively close proximity compared to the evacuation points. In general, the evacuation routes were more robust to accessibility impairments compared to the emergency facilities. This may reflect the impact of disparate roadway design standards (e.g., federal/state highways versus local road designs) on the urban-flood accessibility profiles of communities.The preceding results provide quantifiable consequences that can be factored into resiliency planning and design efforts. Albano et al. (2014) provide a model of such urban flood–induced accessibility consequence analyses for the Puglia region of Italy. The model in the current study shows that mitigating the adverse consequences of street flooding on accessibility to emergency facilities can be achieved by targeting specific vulnerable transportation corridors. However, the analysis also demonstrates that such countermeasures have the potential to increase flood risks on adjacent communities in extremely flat topographic environments as a result of long and narrow floodplain obstructions. Such obstructions traditionally represent elevated roadways, but it is worthwhile to note that the more recent development and deployment of temporary hydraulic flood barriers can produce similar dam-like effects. More research is needed to evaluate the indirect hydraulic impact of these and other types of flood-fight devices, e.g., Huong et al. (2002), used in emergency flood scenarios.ConclusionsThis study presents a raster-based method to analyze the time-varying effects of flooded roads on system-wide network accessibility to discrete target points. Taking into account pluvial flooding, the method is demonstrated for the roadway network of Lafayette, Louisiana, during the 500-year design flood scenario. The results demonstrate that the peak percentage of inaccessibility of the network could be three to four times larger than the percent of impassibility when considering a safe driving depth for a typical motorist. The model is also used to identify and evaluate strategic mitigation alternatives while factoring in the hydraulic impact on surrounding areas. Hence, the presented framework yields insights that can inform the development of more flood-resilient transportation networks in the future. However, several improvements can be made. First, the raster-based accessibility calculation did not discriminate between paths, treating all potential pathways to a target resource as being equally viable routes. In reality, driver fatigue, traffic congestion, and other factors (e.g., one-way streets, medical facility preference, familiarity with potential route) may constrain the potential viable options for motorists. Flood impacts on driver speed and perceived path optimality were also not analyzed, but the present framework can be extended to incorporate such considerations. Smaller yet more frequent flood scenarios may also be beneficial to community planning efforts, and the design storm framework presented herein can be scaled to these higher storm frequencies straightforwardly. However, for these types of scenarios, the subsurface drainage system (not analyzed here given the size of the extreme event) will become more significant. The strictly 2D hydrodynamic model framework can be easily extended to a hybrid 1D/2D approach incorporating such local effects. The flood-fill search algorithm was selected for convenience, but other network search methods including diffusion models could have also been used (Abdulla et al. 2020).The incorporation of continuous gauge measurements for stage and flow would also help provide the foundation for a more rigorous calibration of the hydrodynamic model. Such calibration is especially important for accurate representation of smaller, more frequent flood events. Also not analyzed was the deluge of crowd-sourced information (e.g., photo streams from pedestrians, drivers, and residents; commentary) on social media platforms that accompanies significant flood events. When coupled with video streams from traffic cameras and surveillance footage, these diverse data streams may reveal new insights about human/flood interactions that cannot be gleaned from gauge measurements. Methods aimed at improving the capture and synthesis of these time-limited data sets, while facilitating rapid access across a diverse spectrum of regional stakeholders, would help address this gap and enhance the articulation of flood consequences in the future.Data Availability StatementSome or all data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request. Examples of such data may include flood photos, high-water-mark data, or time-varying accessibility rasters for the storm scenarios considered herein.AcknowledgmentsThe author would like to thank two anonymous referees for their helpful suggestions that led to an improved manuscript. This work was sponsored in part by the Louisiana DOT and Development Louisiana Transportation Research Center Transportation Innovation for Research Exploration (LTRC TIRE) program under Grant No. DOT-LT-1000269, and by an award from the National Academies of Sciences Gulf Research Fellowship Program.References Abdulla, B., A. Kiaghadi, H. Rifai, and B. Birgisson. 2020. “Characterization of vulnerability of road networks to fluvial flooding using SIS network diffusion model.” J. Infrastruct. Preserv. Resilience 1 (6): 1–13. https://doi.org/10.1186/s43065-020-00004-z. Albano, R., A. Sole, J. Adamowski, and L. Mancusi. 2014. “A GIS-based model to estimate flood consequences and the degree of accessibility and operability of strategic emergency response structures in urban areas.” Nat. Hazards Earth Syst. Sci. 14 (4): 2847–2865. https://doi.org/10.5194/nhess-14-2847-2014. Anarde, K., S. Kameshwar, J. Irzza, J. Nittrouer, J. Lorenzo-Trueba, J. Padgett, and P. Bedient. 2018. “Impacts of hurricane storm surge on infrastructure vulnerability for an evolving coastal landscape.” Nat. Hazard. Rev. 19 (1): 04017020. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000265. Brown, V., B. Keim, W. Kappel, D. Hultstrand, A. Peyrefitte, and A. Black. 2020. “How rare was the August 2016 South-Central Louisiana heavy rainfall event?” J. Hydrometeorol. 21 (23): 773–790. https://doi.org/10.1175/JHM-D-19-0225.1. Coles, D., D. W. Yu, and Z. Herring. 2017. “Beyond ‘flood hotspots’: Modelling emergency service accessibility during flooding in York, UK.” J. Hydrol. 546 (Dec): 419–436. https://doi.org/10.1016/j.jhydrol.2016.12.013. Copstead, R., and D. Johansen. 1998. Water/Road interaction: Examples from three flood assessment sites in Western Oregon. Portland, OR: United States Department of Agriculture. DHI (Danish Hydraulic Institute). 2017. MIKE 21 flow model & MIKE 21 flood screening tool. Lakewood, CO: DHI Water & Environment Inc. Gori, A., I. Gidaris, J. Elliot, J. Padgett, K. Loughran, P. Bedient, and A. Juan. 2020. “Accessibility and recovery assessment of Houston’s roadway network due to fluvial flooding during Hurricane Harvey.” Nat. Hazard. Rev. 21 (2): 04020005. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000355. Green, D., D. Yu, I. Pattison, R. Wilby, L. Bosher, R. Patel, and T. Ryley. 2017. “City-scale accessibility of emergency responders operating during flood events.” Nat. Hazards Earth Syst. Sci. 17 (4): 1–16. https://doi.org/10.5194/nhess-17-1-2017. Houston, D., A. Werritty, D. Bassett, A. Geddes, A. Hoolachan, and M. McMillan. 2011. Pluvial (rain-related) flooding in urban areas: The invisible hazard. New York: Joseph Rowntree Foundation. Kao, S.-C., S. DeNeale, E. Yegorova, J. Kanney, and M. Carr. 2020. “Variability of precipitation areal reduction factors in the conterminous United States.” J. Hydrol. 9 (22): 100064. https://doi.org/10.1016/j.hydroa.2020.100064. LADOTD. 2011. 2011 hydraulics manual. Baton Rouge, LA: State of Louisiana Department of Transportation and Development. Loreti, S., E. Ser-Giacomi, A. Zischg, M. Keiler, and M. Barthelemy. 2022. “Local impacts on road networks and access to critical locations during extreme floods.” Nat. Sci. Rep. 12 (1552): 1–15. https://doi.org/10.1038/s41598-022-04927-3. Maidment, D., D. Brookshire, J. Brown, J. Dorman, G. Galloway, B. Imam, and T. Fong Yee. 2009. Mapping the zone: Improving flood map accuracy—Report in brief. Washington, DC: National Academy of Sciences. Miller, R. 2019. A model-based approach to detect zones of inaccessibility during extreme flood events: A case study in the Teche-Vermilion watershed of South-Central Louisiana. Baton Rouge, LA: Louisiana Transportation Research Center. Neal, J., P. Bates, T. Fewtrell, N. Hunter, M. Wilson, and M. Horritt. 2009. “Distributed whole city water level measurements from the Carlisle 2005 urban flood event and comparison with hydraulic model simulations.” J. Hydrol. 368 (Jan): 42–55. https://doi.org/10.1016/j.jhydrol.2009.01.026. Pregnolato, M., A. Ford, S. Wilkinson, and R. Dawson. 2017. “The impact of flooding on road transport: A depth-disruption function.” Transp. Res. Part D Transp. Environ. 55 (Jun): 67–81. https://doi.org/10.1016/j.trd.2017.06.020. Saad, H., and E. Habib. 2021. “Assessment of Riverine dredging impact on flooding in low-gradient coastal rivers using a hybrid 1D/2D hydrodynamic model.” Front. Water 3 (Mar). https://doi.org/10.3389/frwa.2021.628829. Schneider, C. M., A. A. Moreira, J. S. Andrade Jr., S. Havlin, and J. Herrmann. 2011. “Mitigation of malicious attacks on networks.” Proc. Natl. Acad. Sci. 108 (10): 3838–3841. https://doi.org/10.1073/pnas.1009440108. Wang, Y., A. Chen, G. Fu, S. Djordjevic, C. Zhang, and D. Savic. 2018. “An integrated framework for high-resolution urban flood modelling considering multiple information sources and urban features.” Environ. Modell. Software 107 (Jul): 85–95. https://doi.org/10.1016/j.envsoft.2018.06.010. Wing, O., P. Bates, A. Smith, C. Sampson, K. Johnson, J. Fargione, and P. Morefield. 2018. “Estimates of present and future flood risk in the conterminous United States.” Environ. Res. Lett. 13 (3): 034023. https://doi.org/10.1088/1748-9326/aaac65.