Impact of Compound Flood Event on Coastal Critical 1 Infrastructures Considering Current and Future Climate

. The changing climate and anthropogenic activities raise the likelihood of damages due to compound flood 9 hazards, triggered by the combined occurrence of extreme precipitation and storm surge during high tides, and 10 exacerbated by sea -level rise (SLR). Risk estimates associated with these extreme event scenarios are expected to be 11 significantly higher than estimates derived from a standard evaluation of individual hazards. In this study, we present 12 case studies of compound flood hazards affecting critical infrastructure (CI) in coastal Connecticut (USA) . We 13 based the analysis on actual and synthetic (considering future climate conditions for the atmospheric forcing, sea 14 -level rise, and syntheticforecasted hurricane tracks) hurricane events, represented by heavy precipitation 15 and surge combined with tides and SLR conditions. We used the Hydrologic Engineering Center’s River Analysis 16 System (HEC-RAS), a two-dimensional hydrodynamic model to simulate the combined coastal and riverine flooding 17 on selected CI sites. We forced a distributed hydrological model (CREST-SVAS) with weather analysis data from the 18 Weather Research and Forecasting (WRF) model for the synthetic events and from the National Land Data 19 Assimilation System (NLDAS) for the actual events, to derive the upstream boundary condition (flood wave) of HEC- 20 RAS. We extracted coastal tide and surge time series for each event from the National Oceanic and Atmospheric 21 Administration (NOAA) to use as the downstream boundary condition of HEC-RAS. The significant outcome of this 22 study represents the evaluation of changes in flood risk for the CI sites for the various compound scenarios (under 23 current and future climate conditions). This approach offers an estimate of the potential impact of compound hazards 24 relative to the 100-year flood maps produced by the Federal Emergency Management Agency (FEMA), which is vital 25 to developing mitigation strategies. In a broader sense, this study provides a framework for assessing the risk factors 26 of our modern infrastructure located in vulnerable coastal areas throughout the world. includes shifted tide and SLR. 330 Five of the CIs were outside the FEMA 100-year flood zone, but they present flooding for FL1 and SD3. For FL2 all 331 of the study sites were more vulnerable (positive % change), as compared to the FEMA map. Similar findings are 332 presented for SD5, with the exception of except for CI8.

and. potential changes due to climate change. 71 The scenario-based analysis of this study formed the basis on which to address two questions: 72 (2) Will future climate (including SLR and intensification of storms due to warmer sea surface temperatures) bring a 75 significant increase in flood impact for the power-grid coastal infrastructures? 76 The proposed framework offers a multi-dimensional strategy to quantify the potential impacts of tropical storms, thus 77 enabling for a more resilient grid for climate change and the increasing incidence of severe weather. 78 We investigated these questions based on eight case studies of CI in Connecticut (USA), distributed on the banks of 79 coastal rivers discharging along the Long Island Sound. 80 2 Materials and methods 81

Study sites 82
This study focused on seven coastal river reaches (Fig. 1, Table 1), where eight power grid substations lie in proximity 83 to riverbanks and are prone to flooding caused by both coastal storms (such as hurricanes) that combine heavy 84 precipitation and high surge. These power grid substations are codedlabeled on the map CI1 through CI8. 85 For each river reach adjacent to a CI, we developed a hydrodynamic model domain, and we applied a distributed 86 hydrological model for predicting river flows from the upstream river basin. Table 1 shows the specification of each 87 river reach, associated drainage basin, the correspondent domain extent for the hydrodynamic simulations, and the 88 hydrological distance [distance along the flow paths] of each power grid substation from the coastline. This distance 89 was derived using the 30m National Elevation Dataset (NED) for the continental United States (USGS 2017). 90 Among the case study sites, two CIs are relatively inland [CI3 and CI4] (table 1: see hydrologic distance. belong to watersheds ranging from 10 to 300 km 2 basin area, which are sub-basins of the Connecticut River basin. 94 The hydrodynamic model simulation domains ranged from 3.7 to 8.3 km in river length and 2.2 and 20.7 km 2 in area. 95

Simulation framework 96
To evaluate the effect of compound events, we selected four tropical storms: two actual hurricanes (Sandy and Irene) 97 that hit Connecticut, and two synthetic scenarios based on actual hurricanes Sandy  To simulate the synthetic hurricane Florence with WRF, we used the GFS forecasts at 0.25° x 0.25° spatial resolution 120 as initial and boundary conditions. We used a three-grid setup with a coarse external domain of 18 km spatial resolution 121 and two nested domains with 6 km and 2 km horizontal grid spacing, respectively. Two-way nesting was activated for 122 both inner domains. Vertically, the domains stretched up to 50 mb with 28 layers. We parameterized convective 123 activity on the outer (resolution of 18 km) and the first nested (resolution of 6 km) domain using the Grell 3D ensemble 124 scheme (Grell and Devenyi 2002). Further details on the model setup are presented in Table 2. 125 For the future hurricane Sandy scenario, we used the hurricane Sandy simulations under future climate conditions 126

Hydrological modeling 133
To account for the river inflow (upstream boundary condition), we applied a physically-based distributed hydrological 134 model [CREST-SVAS (Coupled Routing and Excess Storage-Soil-Vegetation-Atmosphere-Snow)] described in 135 Shen and Anagnostou (2017). 136 To simulate river discharges for the synthetic hurricanes (Florence and future Sandy), we used the WRF simulations 137 at 6-km/hourly spatiotemporal resolution, as described above. To force the hydrological model for the actual events 138 (Sandy and Irene), we used data from Phase 2 of the North American Land Data Assimilation System (NLDAS-2) 139 (Xia et al., 2012) dataset. NLDAS-2 is a gridded dataset derived from bias-corrected reanalysis and in situ observation 140 data, with a one-eighth-degree grid resolution and an hourly temporal resolution, available from January 1, 1979, to 141 the present day. We derived the precipitation from daily rain gauge data over the continental United States, and all 142 other forcing data came from the North American Regional Reanalysis (NARR) by NCEP (Higgins 2000), to which 143 we applied bias and vertical corrections. To reduce the computational effort, we performed the hydrological simulation 144 using a hydrologically conditioned 30 m spatial resolution DEM (USGS 2017). 145 The hydrologic simulation includes the use of land use and land cover information retrieved from the Moderate

Compound scenarios 209
We modeled four types of synthetic compound event scenarios, as well as actual events by (1) simulating the synthetic 210 hurricanes; (2) introducing a climate change factor, in the form of SLR (~0.6 m), as projected for 2050, as a prediction 211 Two scenarios were created for hurricane Irene. IR1 was the actual hurricane Irene, that made landfall in Connecticut 216 during high tide, and IR2 was the IR1 scenario with future SLR added to the tidal water level as a downstream 217 boundary condition in HEC-RAS. 218 For hurricane Sandy, we generated five scenarios. SD1 was the actual Sandy. For SD2, we shifted the peak high tide 219 to coincide with the maximum storm surge recorded, as derived from the local NOAA stations (hereafter referred to 220 as 'shifted tide water levels'). We further added SLR to the shifted tide water levels from SD2 to create the third 221 scenario (SD3). The remaining two scenarios for hurricane Sandy represented future climate conditions. Specifically, 222 SD4 was the future hurricane scenario simulated with the GFS (Chapt. 2.2.1) and shifted tidal water level. SD5 was 223 the future Sandy with shifted tide water levels and SLR. 224 For the synthetic hurricane Florence event, we simulated two scenarios. FL1 was the synthetic Florence event, based 225 on the GFS track that gave landfall in Connecticut and Long Island (Chapt. 2.2.1). FL2 was the same synthetic event, 226 with SLR added to the coastal water levels. 227 Table 3 shows, for each scenario, the basin-averaged event accumulated precipitation (mm) and the simulated peak 228 flow (m3/s) used as an upstream boundary condition in HEC-RAS, along with the recurrence interval of the peak 229 flows derived using a Log-Pearson probability distribution fitted using yearly maxima from the long-term simulated 230 flows (1979-2019) from CREST. This shows how significant the precipitation forcing was for each considered 231 scenario. For CI1, for example, the future Sandy (SD4/5) scenario, with a peak flow of 242.4 m3/s, was the most 232 extreme event with a recurrence interval of 316 years, followed by Irene (158.5 m3/s) and Florence (51.3m3/s) with 233 a recurrence interval of 56 and 2 years respectively, whereas, for CI8, Florence and future Sandy had similar 234 magnitudes with peak flows of 93.1m3/s (6) and 94.7m3/s (6), respectively. In table 3 The study further looked at whether the depth of water at a station would change for various scenarios. Figure 6 shows 261 an example of the flood depth over simulated time at CI3 for the scenarios of Sandy. PreWe investigated pre-defined 262 criticalhazardous water levels were investigated for each station, as hypothetical values representing the height 263 between the floor and the critical electric system in the station. Specifically, we considered 0.5 m, 1.5 m, and 2.5 m 264 for threshold levels. As a measure of the potential threat to the electric infrastructure, we determined the percentage 265 of time that the flood level was over each specific threshold (Figure 7 9). This data was then used to assess potential 266 flooding problems associated with on-site inundation: we associated the changes in risk posed to the CI from the 267 different examined scenarios based on the changes in those percentages. 268 3 Results and Discussion 269

Flood extent 270
The inundation extents shown in figure 6 represent an aggregation of the overall runs rather than a specific simulation 271 time, and it represents the extent reached when all pixels had the maximum inundation depth. Total flood extent ranged 272 between less than 1 km 2 to more than 7 km 2 , with a minimum extent of 0.4 km 2 for the actual Sandy (SD1) at C8, and 273 a maximum extent of 7.1 km 2 for the future Sandy (SD5) at C3. The results showed consistent agreement that the 274 flood extent increased with increasing intensity of the event and an increase in the recurrence intervals of the flows 275 (Table 3). 276 Changes across the study sites relative to the FEMA 100-year flood extend (Table 4, Figure 7a-c) ranged from -87.8% 277 (for CI8 for SD1) to 192.2% (for CI2 for IR2). Overall, the sites with a return period of fewer than 100 years, showed 278 consistently less flooding than that of the FEMA map, a finding best represented by the comparison of actual events, 279 such as IR1. 280 where tens of meter-scale absolute differences were found between the FEMA estimated flood extent for hurricane 284 Sandy. The strength of correlation (dCorr) between changes in the upstream (flow peak) or downstream (surge peak) 285 components, and the absolute differences with FEMA extent, gives an idea of the importance of each every single 286 driver of change. For the cases investigated in this study, the percentage difference mostly depends on the surge: 287 surge height explains more than 80% of the variation in the differences to FEMA extent (dcorr=0.8 in median). CI6 288 appears to be the sites where the surge has the strongest correlation with the absolute difference in flood extent, as 289 compared to FEMA maps. The differences with FEMA maps are less related to the peak flows (median correlation 290 0.5, with max correlation recorded for CI3). As expected, the correlation with surge increases at the decreasing of the 291 hydrologic distance to the coast, while the correlation with the flow increases the further a site is from the coast, even 292 though this relationship is not linear. 293 As we proceeded with the synthetic scenarios, adding compound and future climate, the results indicated the additional 294 impacts of the joint flood drivers (shifted tide, surge, SLR). 295 For the same event, peak storm-tide levels occurring near local high tide ( i.e.,. SD2) resulted in more flooding than 296 that of events happening at low-tide (like actual Sandy, SD1). Climate change related SLR exacerbates extreme 297 event inundation relative to a fixed extent (FEMA) with variability that ranged from 8.3% (CI4/5) to as high as 425% 298 (CI8). CI8 is the site hydrologically closer to the coast (see the hydrologic distance in Table 1), making it the most 299 susceptible to the altered scenario. Nonetheless, the shifted tide increased the inundation relative to the FEMA 100-300 year flood map also for CI2 and CI4/5. 301 The effects of compound events emerged drastically with the combination of both shifted tide and SLR. With the 302 exception of Except for CI3 and CI8, all other CIs showed an increase in the percentage change from FEMA (Table  303 4). In comparison to SD1, SD3 exhibited increased inundation for all the CIs. The inundated area was about 146% 304 more (1.9 km 2 ) for SD3 than SD1 (0.9 km 2 ) for CI1, for example. The river flood peak for hurricane Sandy had a 305 recurrence interval of about two years, but the flood hazard associated with this event became more devastating if 306 simulated in a compound way, including SLR and shifted tide. This result suggests that events of lower river flood 307 severity (from less fewer rain accumulations) can produce an aggravating impact, as the intensity of major storm 308 surges increases due to shifted timing and SLR. 309 For the synthetic hurricane Florence and hurricane Irene, we saw an increased flooded area in comparison to FEMA 310 (Table 4); for CI2, for example, the increase was almost 200% from IR1 to IR2. Again, this result confirms that 311 accounting for river peak flow frequency alone does not effectively capture the severity of a flood hazard in the case 312 of coastal locations. 313 For all the study sites for future Sandy, we saw consistent increases in flood extent (Table 4)  The CDFs of water depth for the whole domain (Figure 8), confirm that the water depths derived for coupled events ( 339 i.e.,. high tide coinciding with surge peak, or SLR and future climate) are generally higher than those derived 340 from events with independent drivers Note that for some cases ( i.e.,. IR1 and IR2, for CI2 in Fig. 8) water 341 depths increase very consistently as SLR increase. Larges changes in the CDFs appearsappear for lower water 342 depths. Thus, regions with generally lower hazard (depth), will likely experiencesexperience larger impacts under 343 SLR. Results also confirm that scenarios with simultaneous high values for all these parameters implicated a higher 344 vulnerability of the CIs. Comparing these changes in pairs [ i.e.,. IR1 vs IR2, or SD1 vs SD3] also highlights that 345 compound scenarios changeschange in the frequency of extreme values that go far beyond the average are much 346 more pronounced than the related changes of the median depths (cumulative probability=0.50). In particular, it may 347 be asserted that more expressed changes in extremes could lead to corresponding "hazard shift" for all CIs, as 348 represented in Figure 8. 349

350
These results suggest that fluvial flow is not the only driver determining flood risk. Actual Irene (IR1) and synthetic 351 Florence (FL) had higher river flood return periods than did actual Sandy (SD1) ( Table 2). Nonetheless, the CDFs of 352 the flood depth showed different behavior in terms of severity. For CI1, for example, IR1 had higher probabilities for 353 for example, which is more coastal than the other CIs, presented increasing flood depth due to tidal timing. 357 As expected, and as previously highlighted when considering the flood extent (Table 4)

Local risk for CI 374
Much of the flood damage in CI is incurred by components being submerged for a long period. Investigating the 375 duration of the flood depth at the CI location (Figure 9) should be considered in planning for any protective measures, 376 such as elevating or waterproofing equipment. If a critical infrastructure shows 0%, it means that for that 377 scenario/event the water didn't did not reach the substation at all, at least during the simulated timeframe. This 378 could be due to the water flooding other upstream locations, and therefore draining away from the station, or because 379 the topography of the landscape actually prevented water from reaching the area for some specific events. 380 According to our analysis, none of the scenarios has an actual impact on CI1. For the other CIs, comparing individual 381 events we could see an increase in risk due to the compound hazard scenarios-that is, shifted tide and SLR. Important 382 to note is that, for most of the sites, the compound risk due to SLR and tide timing was always higher for the lower 383 water-level thresholds (0.5 m). This implies a higher risk for CI components currently positioned closer to the ground. 384 Damage to the CI components is dictated by both the flood depth and the duration of submergence. The suggested 385 high values of risk [increase percentage in inundation duration] (Figure 9) further imply differences in the timing of 386 repairs. In the cases of CI7 and CI8 (Figure 9), the CIs remained submerged with 0.5 m of water for about 20% of the 387 event period for actual Sandy, but for. For the worst-case future Sandy scenario, the location was flooded for more 388 Another importantcritical insight was provided by the hurricane Florence scenarios. As mentioned earlier, Florence 391 did not affect the study area, although an early GFS storm forecast track predicted landfall in Long Island and 392 Connecticut. For this event, the estimated measure of risk was about 20%, and it was shown to increase to up to 40% 393 for the lower water depth (0.5 m) threshold in some locations. The result of the simulated scenario allows for an 394 assessment of potential damage and for an identification of equipment that might be affected by future events under 395 current climatic conditions. In this regard, comparing the results for the different CIs during the Sandy scenarios 396 revealed an interesting pattern. While we might have expected a greatermore significant impact over the whole domain 397 when shifting the tide (Figure 9, Table. 3), we found different impacts in the CI locations. Notably, the risk appeared 398 lower when the tides were shifted (Fig. 9) for some of the CIs (for example, CI5 and CI7). This can be explained by 399 the fact that higher water levels in the domain were changing the water flows, allowing the flood to follow different 400 drainable ways. This can be a very useful piece of information for deciding whether to and where to take measures in 401 terms of flood occurrence and potentially relocating CIs to avoid catastrophic compound flood events. 402 From table 1 we can see that CI8 is the closest to the coastline followed by CI7, CI6, and CI5. From figure 9 we can 403 see that all the CIs that are closer to the coastline are susceptible to changes in the downstream water level condition 404 (Shifted tide/ SLR) (Table 3). CI4 is the farthest from the coast followed by CI3. Both the CIs show minimal response 405 to changes in the coastal water level compared to CI5/ CI6/ CI7. This analysis gives us conclusive evidence of risk 406 associated with the location of the CI from the coastline. 407

Concluding Remarks 408
Preparing for the challenges posed by climate change requires an understanding of current actual, & possible, and 409 future scenario of tropical storm impacts, and a correct understanding interpretation of the hazard imposed by 410 compound flooding. In this work, we have developed and implemented a modeling framework that allows to address 411 addressing this task, focusing on coastal electric grid infrastructure (substations). To date, the design of these facilities 412 typically has assumed the current climatic conditions. However, a changing climate, as well as the co-occurrence of 413 compound drivers, and the resulting more extreme weather events mean those climate bands are becoming outdated, 414 leaving infrastructure operating outside of its tolerance levels. 415 We explored a range of actual and synthetic hurricane scenarios, offering a system that could inform short-and long-416 term decisions. For the short-term decision, the framework allowed to investigate the characteristics of the hurricane-417 related inundation, considering the compound effect of riverine and coastal flooding coinciding, or not, with peak high 418 tides. Generally, hurricanes affect large areas, and the specific locations at which damage will occur are often difficult 419 to anticipate. Simulation of different scenarios can provide system operators with the ability to prepare for damage 420 and respond quickly once it has occurred-for example, by pre-positioning repair crews. Furthermore, by simulating 421 the impact using possible storm paths, the framework allows us to understand the potential impacts on the CI. The 422 framework proposed in this study evaluates the extent of flood nearby a critical coastal infrastructure caused by 423 possible extreme compound events. Each type of infrastructure system has specific elements vulnerable to specific 424