Simulation of tropical cyclone impacts to the U.S. powersystem under climate change scenarios

Andrea Staid & Seth D. Guikema & Roshanak Nateghi &Steven M. Quiring & Michael Z. Gao

Received: 8 May 2014 /Accepted: 29 September 2014 /Published online: 15 October 2014# Springer Science+Business Media Dordrecht 2014

Abstract The links between climate change and tropical cyclone behavior are frequentlystudied but still uncertain. This uncertainty makes planning for climate change a difficult task.Here we focus on one area of climate-related risk: the impact of tropical cyclones on UnitedStates power systems, and we evaluate this risk through the simulation of impacts to the powersystem under 12 plausible scenarios in which climate change may affect tropical cycloneintensity, frequency, and location. We use a sensitivity analysis based approached grounded inthe literature rather than directly simulating from specific GCM output due to the high degreeof uncertainty in both the climate models and the climate-hurricane relationship. We show howchanges in tropical cyclone activity influence extreme wind speeds, probability of poweroutages, and the proportion of people without power. While climate change and its impacts areoften discussed globally, this work provides information at a much more local scale. Thesensitivity of an individual area can be assessed, and the information presented here can beincorporated into planning and mitigation strategies for power systems faced with an uncertainfuture in a changing climate.

1 Introduction

Tropical cyclones, and hurricanes in particular, have been the cause of extensive damage andfinancial loss in many regions of the United States. They rank among the most destructivenatural hazards for coastal areas (Huang et al. 2001; Pielke Jr et al. 2005; Vickery et al. 2009).Hurricane Sandy, for example, left more than 8 million customers without power, resulting inestimated costs of $65 billion (Zamuda et al. 2013). Power outages caused by tropical cyclones

Climatic Change (2014) 127:535546DOI 10.1007/s10584-014-1272-3

Electronic supplementary material The online version of this article (doi:10.1007/s10584-014-1272-3)contains supplementary material, which is available to authorized users.

A. Staid (*) : S. D. Guikema : R. Nateghi :M. Z. GaoDepartment of Geography and Environmental Engineering, Johns Hopkins University, Baltimore, MD, USAe-mail: staid@jhu.edu

S. M. QuiringDepartment of Geography, Texas A&M University, College Station, TX, USA

R. NateghiResources For the Future, Washington, DC, USA

are one of the biggest concerns for affected communities; a lack of power can result in businessinterruptions, healthcare stresses, and cascading effects for dependent infrastructure systems,such as telecommunication or water networks (Zimmerman and Restrepo 2009). While therehave been strong developments in power-outage prediction models that have proven useful forlocal utility companies (Liu et al. 2005, 2007; Nateghi et al. 2011; Winkler et al. 2010; Hanet al. 2009a, b), the focus has been on outage-forecasting in the days before a storm and not onlong-term changes in risk. There is a need for infrastructure providers and emergencymanagers to plan for hurricanes on much longer time scales, i.e., on the order of decades.This planning must consider how future climate conditions may influence storm behavior.

The relationship between tropical cyclone hazard and climate change has been studiedextensively, but there is still a great deal of uncertainty involved (Emanuel et al. 2008;Henderson-Sellers et al. 1998, Knutson et al. 2010, 2008a, Mann and Emanuel 2006,Mendelsohn et al. 2012, Pielke Jr et al. 2005, Yonetani and Gordon 2001, Mudd et al.2014). For example, the physics-based models developed by Knutson et al. suggest that thefrequency of Atlantic hurricanes and tropical storms will likely be reduced in the future(Knutson et al. 2008b). Results obtained by downscaling IPCC AR4 simulations also suggesta reduction in the global frequency of hurricanes in a warmer future climate scenario, with apotential increase in intensity in some locations (Emanuel et al. 2008). Statistical models arguethat the intensity and frequency of TCs will likely increase in a warmer future (Emanuel 2005;Saunders and Lea 2008). These examples from the literature highlight the deep uncertaintyremaining; the direction of change (e.g., more storms vs. fewer storms) and, to a greater extent,the magnitudes of change remain uncertain.

Because many traditional risk and decision analysis methods struggle under such deepuncertainty, more robust planning tools are needed in this area (Ranger et al. 2013). Onepromising approach is to focus on actions that perform well under a range of scenarios that arewithin the realm of physical possibility as supported by the research literature. These scenarioscan be modified over time as conditions change and new information comes to light (Walkeret al. 2013). Long-term planning for major infrastructure projects, such as updates to theelectric grid, can leverage scenarios to assess the robustness of possible actions rather than as abasis for determining an optimal solution by assigning fixed probabilities to the scenarios. Akey component of this is the ability to estimate the performance of infrastructure systems underfuture climate scenarios in a way that builds upon accurate models of system behavior andoffers both broad-area insight and locally detailed performance estimates.

This paper offers insight into one aspect of this long-term planning problem for areas alongthe Gulf and Atlantic coasts of the United States. It focuses on what tropical cyclone impactsmight look like if climate change causes changes in storm behavior. We assess impacts for awide range of future storm scenarios, including the status quo if the climate remains stable andtropical cyclone seasons remain within the observed historic variability. Here, the termscenario is used to represent potential realizations of climate-induced changes to tropicalcyclone behavior in the North Atlantic basin. Using a simulation informed by the historicalhurricane record, we compare baseline impacts to the outcomes under various scenarios for 23states lying along the U.S. Gulf and Atlantic coasts. We create plausible scenarios of futuretropical cyclone behavior based on the literature on the relationship between climate changeand hurricanes. This literature offers a wide range of possible realizations. Regarding changesin intensity, for example, the general consensus is that storms will strengthen, although thedegree of change varies (Knutson et al. 2010; Pielke 2007). There is less consensus on changesin storm frequency, with the literature showing both increases and decreases (Knutson et al.2008a; Webster et al. 2005). We vary storm intensity, storm frequency, and the distribution oflandfall locations. For each scenario, the simulation results represent tropical cyclone impacts

536 Climatic Change (2014) 127:535546

to the United States in terms of wind-induced power outages. The analysis is done at thecensus-tract level, resulting in localized projections of extreme wind speeds, the fraction ofcustomers without power, and probabilities of power outages. In a field where much of thefocus is global in scale, we provide more localized information for decision-makers that canaid in long-term planning for their specific area of concern.

The sensitivity to changes in tropical cyclone hazards can vary greatly among regions. Thispaper provides insight into how power systems along the Gulf and Atlantic coasts of theUnited States may be affected by climate changes, which areas should be most concerned, andwhich areas are unlikely to see substantial changes under any tested scenario. The range ofpotential impact is a key component of informed planning models, since potential actions canbe tested across this range to ensure robust and sustainable solutions.

2 Climate change and tropical cyclone activity

Depending on the model used, the projected climate change impact on North Atlantic tropicalcyclones can vary significantly. Several studies suggest that the frequency of storms maydecrease in the North Atlantic Basin, but the intensity may increase, perhaps substantially(Mendelsohn et al. 2012). There is also concern that storm genesis location and trackmovement will be influenced by climate change, but there is not enough information to assessthe nature of this impact with a reasonable degree of spatial precision (Dailey et al. 2009).There is potential for substantial changes in tropical cyclone hazard, but there also remainssubstantial unresolved uncertainty. Under such conditions, knowing the range of reasonablepossible outcomes at a local level can result in more robust planning decisions. Scenario-driven planning can help local communities understand if they are particularly sensitive toclimate-induced changes in tropical cyclone hazards or if they are in a relatively insensitivearea. Plans can then be designed to perform adequately across this range of impacts, instead ofoptimizing for a single future or a small set of futures driven by highly uncertain insights intothe climate-tropical cyclone relationship (Walker et al. 2013).

We develop a range of plausible tropical cyclone scenarios and assess the impacts of thesescenarios in terms of electric power outages and extreme wind speeds in the U.S. We simulatea large number of replicated tropical cyclone seasons in the United States for 12 plausibletropical cyclone scenarios. A replicated tropical cyclone season represents one virtual year oftropical cyclone activity, in which the number of storms can vary from replication toreplication. Virtual storms are independently generated and the resulting impacts aremodeled through a power outage estimation model that was developed using the data ofHan et al. (2009a, b) and Nateghi et al. (2013).

We chose the scenario-based approach largely due to the high degree of uncertainty infuture climate projections and the resulting impact on hurricanes. This approach gives greatinsight into the range of possible consequences and what different regions of the country mayneed to prepare for. The sensitivity analysis conducted here allows us to study which factorsdrive changes in impacts, how these impacts may vary over location, and which areas might bemore or less sensitive to potential changes.

Catastrophe modeling is prevalent in industry, and it often focuses on financial losses forinsurance companies. Companies such as AIR, RMS, and EQECAT all have their ownversions of hurricane risk models for the U.S. Model details are not available publicly, butsome models do attempt to characterize future risks as the climate changes. AIR, for example,has a version of their model that is conditioned on warm sea-surface temperatures (AIRWorldwide 2012). RMS uses their model to assess short-term climate threats in the Risky

Climatic Change (2014) 127:535546 537

Business report, but publicly available details are limited and the results focus only on changesin financial loss (Risky Business Project June 2014). Validation is done on losses from paststorms, so their ability to validate under climate change conditions is limited (Risk Manage-ment Solutions 2012). Our scenario-based approach focuses more on long-term future out-comes that deviate from the historical record, so there is no longer an applicable validationdataset to be used.

3 Simulation methodology

The simulation uses historical hurricane and tropical storm data. The baseline runs take thehistorical data as input, and the alternative scenarios alter parameters from the historical data tosimulate plausible climate-induced changes to storm behavior. For each replication in thebaseline case, we first sample from a Poisson distribution, with a mean equal to the historicalmean, to determine the number of storms making landfall in that replicated year. For eachstorm, we randomly sample a landfall location from a smoothed distribution that assigns aprobability to each 50 km stretch of coastline from Texas through Maine on the basis of thehistoric landfall counts in each of the 50 km coastal segments. We randomly sample amaximum wind speed at landfall from the historical record. Based on which section of thecoastline the storm hits, we subset the historical tracks, keeping only those that made landfallin the same region. These tracks are then used to train a random forest model, which is astatistical model used to predict the storms movement in each six-hour time step. For eachtime step, the wind speed decays according to the hurricane decay models of Kaplan andDemaria (Kaplan and Demaria 1995) until the wind speeds fall below the tropical cycloneclassification level. This relatively simple model was chosen because it uses wind speed asinput (instead of pressure) and does not require a priori knowledge of storm size or movement.With the storm track and intensity determined, these parameters are fed into a wind field modelbased on the methods of Willoughby et al. and Holland to calculate the storm radius and winddistribution (Willoughby et al. 2006; Holland 2008). This model generates estimates for themaximum 3-second gust wind speed and the duration of wind speeds above 20 m/s for thecentroid of each census tract (Huang et al. 2001; Han et al. 2009a, b). This wind data is thenpassed to a statistical outage prediction model, which uses a random forest model that has beentrained on past hurricanes in different areas of the U.S. (Nateghi et al. 2013; Guikema et al.2014). The outage prediction model is a simplified version of the work of Nateghi et al.(2013)in that it uses only publicly available data and a reduced set of variables to estimate the numberof customers without power as a result of a hurricane.

Using the baseline case as a point of comparison, we also simulate different climate-induced storm scenarios to examine the influence of climate-induced changes in tropicalcyclone behavior in the North Atlantic Basin. The scenarios represent changes in intensity,frequency, and location. We vary intensity by taking the randomly sampled maximum windspeed for each storm and multiplying it by a factor. We simulate scenarios for intensity factorsof 0.8, 1.2, and 1.4, meaning a decrease in strength of 20 %, an increase of 20 %, and anincrease of 40 %. These intensity changes are based on bounding the estimates of intensitychanges in the climate literature. For scenarios of frequency, we adjust the mean of the Poissondistribution that is used to sample the number of storms in each replicated year. The baselinecase has a mean of 2, and we simulate scenarios for means of 0.5, 1, 3, and 4, again based onbounding the frequency change estimates that we found in the climate literature.

The location scenarios are more subjective, and there is even more uncertainty about trackchanges than about frequency and intensity changes. We adjust the probability distributions

538 Climatic Change (2014) 127:535546

while still retaining the general shape of the spatial probability distribution of landfall locationsbecause it is based on actual geographical characteristics. For example, some land areas aremore prone to hurricanes because they jut out into the path of oncoming storms. We createdfour modified distributions to assess the changing impacts as storm location changes. The firstscenario shifts storms further up along the mid-Atlantic coast, the second shifts them furtherdown into the Gulf of Mexico, the third spreads the distribution out more evenly to reduce thenatural peak around Florida, and the fourth concentrates the peak around Florida, therebyreducing the probabilities in the Gulf and in the Northeast. All scenarios maintain the originalshape of the smoothed distribution.

For each scenario, we ran the simulation for 1600 replications in order to reach conver-gence. The aggregated results from 1600 simulated years of tropical cyclone activity allow usto calculate expected return periods for the output values. We calculate the 100-year, 50-year,25-year, and 4-year, and 2-year return periods for maximum wind speed, duration of windsabove 20 m/s, and the fraction (and number) of customers without power for each census tract.We also calculate the probability of each census tract having at least 10 % of customerswithout power in a given year. The aggregated results allow for many such calculations, but wechose these parameters to portray the potential climate change impacts on both a large andsmall spatial scale.

4 Results

The simulation results show the impact under both the status quo of the baseline case and theclimate-change induced scenarios that were evaluated. The scenarios demonstrate the sensi-tivity of various areas of the country to potential changes in tropical cyclone behavior, and theresults can be evaluated on a local level.

4.1 Baseline impact

For an initial baseline, we simulate tropical cyclone impact assuming that the frequency,intensities, and locations of tropical cyclones follow the observed historical distributions asdiscussed above. For each storm, we forecast wind-induced power outages in each census tractwithin the range of the storm. We repeat this process for 1600 simulated years, yielding aprobabilistic estimate of the impact of tropical cyclones on power systems in the United Statesat the census tract level using the power outage model of Guikema et al. (2014). The keyphysical hazard input to the power outage model is the estimated spatial wind field of a givenhurricane. The impacts of surge, rainfall, and inland flooding are incorporated only indirectly;outages due to these causes were included in the training data, but the differences in surge,rainfall, and inland flooding between storms is not explicitly modeled. We have shown in otherwork that excluding explicit modeling of surge impacts does not substantially impact theaccuracy of the power outage predictions (Guikema et al. 2012).

Figure 1 shows the impact for 1600 replicated years with impact measures of (1) the 100-year wind speed (the wind speed with an annual probability of exceedance of 0.01), (2) theprobability of at least 10 % of customers losing power in a given year, and (3) the 100-yearfraction of customers without power from a given storm, as assessed for each census tractindividually.

These results show the estimated conditions under the historic state of tropical cycloneactivity. We assume both a static population and a static power system (i.e. no upgrades or newtechnologies), so this level of impact represents the status quo. Some regions are expected to

Climatic Change (2014) 127:535546 539

be more heavily impacted than others. The annual probability of at least 10 % of the populationbeing without power is 0.21 when averaged across the census tracts in the state of Florida. Incontrast, when averaged over the entire region evaluated, this probability is 0.06. As expected,our simulation shows that the tropical-cyclone-induced 100-year wind speed drops off sharplyas you move inland. The fraction of customers without power tells a slightly different story.This calculation depends not only on wind speed and duration of high winds, but also on thepopulation of each census tract. Tracts with very low populations can appear as discrepanciesamong neighboring tracts. Information regarding the number of customers without power,instead of the fraction without power, is easily obtained as well and may be most useful forthose planning storm responses, but it is more difficult to visually see impact trends duepopulation variability among tracts.

As one point of comparison, our estimated baseline 100-year wind speeds can be comparedto standard design criteria wind speeds given by the American Society of Civil Engineers(ASCE) design manual, ASCE 710, that are based on historical data (ASCE 710). Forhurricane-prone coastal areas, the simulation output matches well with the ASCE 710 100-year wind speeds. The 100-year wind speed for Houston, TX, for example, is only 1 % lowerin our simulation output than in the ASCE design standard. Similarly, our estimated 100-yearwind speed for Orlando, FL is about 2 % higher than in ASCE 710. Most areas of interest arewithin 10 % of the ASCE wind speeds, although some inland areas have larger deviations. Thesimulation has some inherent variability. This, along with the choice of the smoothed landfallprobability distribution and the wind decay model used may account for some of the largerdifferences. However, this generally close match to independent estimates of 100-year windspeeds gives confidence that we are estimating the long-term wind environment well, at leastin the baseline situation.

4.2 Potential climate impacts

If climate were to impact tropical cyclone intensity, strength, or track, the resulting impactwould not be felt equally across the country. The regional effects also vary depending on themeasure of interest. To model the effects of varying intensity, we simulated an additional 1600hurricane years but multiplied the intensity of each generated storm by an intensity factor, k.We repeated this for k=0.8, 1.2, and 1.4, generating 1600 hurricane years for each k, torepresent a reasonable range of changes in intensity as suggested by the literature. Whenlooking at wind speeds, the effects of varying intensity are felt primarily in coastal areas. Thiscan be seen in Fig. 2, where the changes are seen primarily along the coasts. The biggestchanges are in those areas that receive the most frequent hurricanes indicating that they areparticularly sensitive to changes in hurricane intensity. The fraction of customers without

Fig 1 Baseline impacts of (a) 100-year wind speed, (b) annual probability of at least 10 % of customers losingpower, and (c) 100-year fraction of utility customers without power plotted for each census tract

540 Climatic Change (2014) 127:535546

power, however, depends both on wind speed and storm size. Stronger storms are generallylarger, and our simulations show the effects of storm size as the reach of the stronger (orweaker) storms creates bands of increased (or decreased) impact. These are the areas on themargins of the impacted area and are areas of the country particularly sensitive to changes inhurricane intensity. Figure 3 shows this result, as, on average, the stronger storms impact areasfurther inland than in the baseline case. The areas that fall on the edge of the impacted area seethe largest changes when the average storm intensity varies. This is of particular interestbecause may of these areas are farther inland and do not have a strong history of experiencewith hurricanes. Additional consideration of hurricane preparation planning may be appropri-ate in such areas.

To examine the effects of varying hurricane frequency, we simulated an additional 1600hurricane years but substituted in a different value for , the mean number of tropical cyclonesmaking landfall per year. We repeated this for =0.5, 1, 3, and 4, simulating 1600 hurricanesyears for each, to represent a reasonable range of changes in frequency as suggested by theliterature. A change in tropical cyclone frequency, as opposed to intensity, brings aboutchanges of a different nature. The 100-year wind speeds change slightly, but not substantially.Extreme winds are driven by the strongest storms, not the frequency of more moderate storms.However, there is a more substantial impact on the annual probability of power outages, sincemore (or fewer) storms directly results in a larger (or smaller) probability of any given tractbeing impacted by a storm, and subsequently losing power. Figure 4 shows a comparison ofthese changes under different storm frequency conditions. The historical baseline is approx-imately 2 tropical cyclones making landfall in the U.S. per year (=2). An average change ofjust one storm per year (more or less) can change the annual probability of at least 10 % of thepopulation without power by over 15 % in some census tracts.

Although there is less evidence to support shifts in tropical cyclone location as a result ofclimate change, there is speculation that the locations of tropical cyclogenesis may shift in awarmer climate (Dailey et al. 2009). It is worth assessing how such changes may impact the

Fig 2 Changes in 100-year wind speeds for varying storm intensity away from baseline

-1.0

-0.8

-0.6

-0.4

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

(a) Intensity Factor 0.8 (b) Intensity Factor 1.2 (c) Intensity Factor 1.4

Change

in10 0-Y ear

Fr act ion

With ou t P

owe

Fig 3 Changes in 100-year fraction of customers without power for varying storm intensity away from baseline

Climatic Change (2014) 127:535546 541

U.S., but the scientific understanding is too weak to confidently support direct simulation ofspecific scenarios. Instead, we examined the sensitivity to changes in landfall location byadjusting the smoothed historical spatial probability distribution of landfall locations to shiftstorms further north and south, and also to spread out or concentrate the distribution (seesupplemental material for details). As expected, the areas hardest hit by tropical cyclones shiftalong with the shifting distributions, but our modified distributions are all based on the originaldistribution. This ensures that we still account for the geography of the coastline in determin-ing landfall probabilities. Even with the modified distributions, some areas are still stronglyimpacted even when the probabilities of landfall are sharply reduced. For instance, Fig. 5shows the change in the probability of at least 10 % of customers without power as the averagelandfall location distribution varies. We see that areas along both the Gulf Coast and southernAtlantic coast area particularly sensitive to changes in landfall locations, but in all scenarios weexamined, the Florida peninsula continues to have a relatively high annual probability of atleast 10 % of the population losing power. The changes in maximum wind speeds and fractionof customers without power follow similar patterns, but there is more local variability.

Fig 4 Changes in the probability of at least 10 % of customers without power for varying storm frequencies

Fig 5 Changes in the probability of at least 10 % of customers without power for varying landfall distributions

542 Climatic Change (2014) 127:535546

We conducted additional analysis to compare the relative impacts of changes in frequency,intensity, and landfall location based on graphs of the empirical distribution of outages at thecensus tract level. The details of this process and the results are available in the supplementalmaterial. The overall conclusion is that changes in storm intensity have a greater potentialimpact on the U.S. power system than changes in frequency or landfall location. The impacts,both in terms of customers without power and maximum wind speeds, grow substantiallyworse as storms get stronger, on average. Changes in frequency or landfall, on the other hand,show relatively tightly grouped overall impacts, although the changes do depend strongly onlocation. When evaluating the entire United States as a whole, intensity largely dictates theseverity of the outcome. Thus, characterizing the nature of future climate impacts on stormintensity should be an important focus of future work.

4.3 Metropolitan area impacts

Changes at the national level provide insight into the overall impact of potential future changesin tropical cyclone activity and are useful for federal agencies. Local decision-makers,however, are more concerned with smaller areas of potential impact, and the simulation resultscan also provide information at a much higher resolution. We demonstrate this by examiningthe impacts at the scale of several metropolitan areas: Houston (TX), New Orleans (LA),Miami (FL), Washington (DC), and New York (NY).

As expected, the results depend strongly on location because some areas of the country aremore prone to tropical storms and others are more sensitive to climate changes affecting tropicalcyclones. Figure 6 shows the return periods for the fraction of customers without power for theselected metropolitan areas under different hurricane intensities, together with the averagereturn periods for the entire coastal area. The Houston, New Orleans, and Miami metropolitanareas are heavily impacted even for scenarios of lower intensity storms. These cities sit inalready hurricane-prone areas, and strong storms only increase the already significant impacts.On the other hand,Washington and NewYork behave very differently. Both have relatively lowimpacts for low intensity storms, but NewYork sees sharp increases in the fraction of customersout as storm intensity increases, whereas Washington has a small range across all scenarios.

Fig 6 Mean return periods for the fraction of customers without power as intensity varies, plotting the averagefor five metropolitan areas and for all census tracts (bottom right) evaluated

Climatic Change (2014) 127:535546 543

NewYork is highly sensitive to changes in storm intensity, muchmore so thanWashington. TheWashington area is relatively protected from changes in intensity by virtue of its more inlandlocation. The severe effects of higher wind speeds from stronger storms will be felt primarilyalong the coasts, as shown previously in Fig. 2. The resulting power outages have the potentialto be widespread in many regions of the country. In addition, the difference between the 100-year and 50-year storm impacts is minimal, and this has strong implications for design andmitigation to withstand tropical cyclones and the potential effects of climate change. NewOrleans, Miami, and New York all see greater than 90 % of customers without power for boththe 50- and 100-year return periods.

We conducted additional analysis of the sensitivity of different regions to changes in frequencyand landfall location, and we direct the reader to the supplementary material for the details. Wefind that shifts in landfall location will have strong regional effects. Increases in storm frequencywill worsen the impacts in already hurricane-prone areas. However, in general, the strongestchanges in tropical cyclone hazard are seen in the scenarios of high storm intensity. For mostregions of the country, increases in storm intensity will result in the most severe damages, butfrequent, milder damages could occur if the frequency of tropical storms increases.

5 Conclusion

Planning for future climate change is a difficult task due to the high degree of uncertainty aboutpotential changes in tropical cyclone frequency, intensity and track. This paper examines arange of potential changes in tropical cyclone activity and quantifies how these changes couldinfluence power outage risk. Tropical cyclones can cause substantial damage to power systeminfrastructure and can leave local utility companies and government agencies with high repaircosts in the aftermath of a storm. Anticipating the nature of potential impacts allows forproactive mitigation against damage, reduced costs, and a more resilient power grid.

The scenarios assessed here were chosen to represent a range of plausible outcomes for a futureaffected by climate change and the resulting tropical cyclone impacts, but they are not tied tospecific climate projections (i.e., they are not associated with a specific emissions scenario orGCM results). The range of results demonstrates the sensitivity of the U.S. power system tochanges in storm behavior, and these impacts can be evaluated for small regional areas. Inlandareas are generally more protected from the strongest impacts, but some areas may still seeconsiderable changes in maximum wind speeds, power outage likelihood, and the number ofcustomers losing power during a hurricane. A shift in landfall location could result in impacts toareas of the country with little experience in dealing with hurricanes. Coastal areas are particularlysensitive to increases in storm intensity. 100-year wind speeds are projected to increase by morethan 50 % in some areas with a 20 % increase in storm intensity. The probability of customerpower outages in a given area increases slightly, but the actual number of customers losing powerwould change more drastically as a result of stronger, and often larger, storms. A reduction instorm frequency, on the other hand, is projected to have a corresponding reduction in the likelihoodof power outages. Future work remains, particularly in linking models such as this to specificclimate scenarios, but the results presented here provide a starting point for improved climateadaptation, and the framework can be extended to explicitly link to specific climate scenarios.

Acknowledgments This work is funded in part by the National Science Foundation (NSF) CMMI Grant1149460, NSF CBET SEES Grant 1215872, NSF SEES Grant 1331399, and NSF CMMI Grant 0968711.Thanks also to the National Oceanic and Atmospheric Administration from which much of the original data wasgathered.

544 Climatic Change (2014) 127:535546

References

AIRWorldwide (2012) Assessing U.S. Hurricane Risk: Do The Models Make Sense? AIR WorldwideASCE 7-10. ASCE 7-10 Wind Speeds. Available From: https://www.atcouncil.org/windspeed/Dailey PS, Ljung G, Zuba G, Guin J (2009) Probability of hurricane intensification and United States hurricane

landfall under conditions of elevated Atlantic sea surface temperatures. In: Hurricanes and climate change,Springer, 121138

Emanuel K (2005) Increasing destructiveness of tropical cyclones over the past 30 years. Nature 436(7051):686688Emanuel K, Sundararajan R, Williams J (2008) Hurricanes and global warming: Results from downscaling IPCC

AR4 simulations. Bull Am Meteorol Soc 89(3):347367Guikema SD, Udoh I, Irish J, Nateghi R (2012) The effects of hurricane surge in power system outage risk

models. Proceedings, Probabilistic Safety Assessment And Management 11/Esrel 12, Helsinki, FinlandGuikema SD, Nateghi R, Quiring S, Staid A, Reilly A, Gao M (2014) Predicting hurricane power outages to

support storm response planning. IEEE AccessHan S, Guikema SD, Quiring SM (2009a) Improving the predictive accuracy of hurricane power outage forecasts

using generalized additive models. Risk Anal 29(10):14431453Han S, Guikema SD, Quiring SM, Lee K, Rosowsky D, Davidson RA (2009b) Estimating the spatial distribution

of power outages during hurricanes in the gulf coast region. Reliab Eng Syst Saf 94(2):199210Henderson-Sellers A, Zhang H, Berz G, Emanuel K, Gray W, Landsea C, Holland G, Lighthill J, Shieh S,

Webster P (1998) Tropical cyclones and global climate change: A post-IPCC assessment. Bull AmMeteorolSoc 79(1):1938

Holland G (2008) A revised hurricane pressure-wind model. Mon Weather Rev 136(9):34323445Huang Z, Rosowsky D, Sparks P (2001) Hurricane simulation techniques for the evaluation of wind-speeds and

expected insurance losses. J Wind Eng Ind Aerodyn 89(7):605617Kaplan J, Demaria M (1995) A simple empirical model for predicting the decay of tropical cyclone winds after

landfall. J Appl Meteorol 34(11):24992512Knutson TR, Sirutis JJ, Garner ST, Vecchi GA, Held IM (2008) Simulated reduction in Atlantic hurricane

frequency under twenty-first-century warming conditions. Nat Geosci 1(6):359364Knutson TR, Mcbride JL, Chan J, Emanuel K, Holland G, Landsea C, Held I, Kossin JP, Srivastava A, Sugi M

(2010) Tropical cyclones and climate change. Nat Geosci 3(3):157163Liu H, Davidson RA, Rosowsky DV, Stedinger JR (2005) Negative binomial regression of electric power

outages in hurricanes. J Infrastruct Syst 11(4):258267Liu H, Davidson RA, Apanasovich T (2007) Statistical forecasting of electric power restoration times in

hurricanes and ice storms. Power Syst, IEEE Trans 22(4):22702279Mann ME, Emanuel KA (2006) Atlantic hurricane trends linked to climate change. EOS Trans Am Geophys

Union 87(24):233Mendelsohn R, Emanuel K, Chonabayashi S, Bakkensen L (2012) The impact of climate change on global

tropical cyclone damage. Nat Clim Chang 2(3):205209Mudd L, Wang Y, Letchford C, Rosowsky D (2014) Assessing climate change impact on the US East Coast

hurricane hazard: temperature, frequency, and track. Natural Hazards ReviewNateghi R, Guikema SD, Quiring SM (2011) Comparison and validation of statistical methods for predicting

power outage durations in the event of hurricanes. Risk Anal 31(12):18971906Nateghi R, Guikema S, Quiring SM (2013) Power outage estimation for tropical cyclones: improved accuracy

with simpler models. Risk Anal 34(6):10691078. doi:10.1111/risa.12131Pielke RA Jr (2007) Future economic damage from tropical cyclones: sensitivities to societal and climate

changes. Philos Trans A Math Phys Eng Sci 365(1860):27172729Pielke RA Jr, Landsea C, Mayfield M, Laver J, Pasch R (2005) Hurricanes and global warming. Bull Am

Meteorol Soc 86(11):15711575Ranger N, Reeder T, Lowe J (2013) Addressing deep uncertainty over long-term climate in major

infrastructure projects: four innovations of the Thames estuary 2100 project. EURO J DecisProcess 1(34):233262

Risk Management Solutions, Inc. (2012) Principles Of Model Validation: United States Hurricane ModelRisky Business Project (2014) Risky Business: The Economic Risks Of Climate Change In The United States.

Risky BusinessSaunders MA, Lea AS (2008) Large contribution of sea surface warming to recent increase in Atlantic hurricane

activity. Nature 451(7178):557560Vickery PJ, Masters FJ, Powell MD, Wadhera D (2009) Hurricane hazard modeling: the past, present, and future.

J Wind Eng Ind Aerodyn 97(7):392405Walker WE, Haasnoot M, Kwakkel JH (2013) Adapt or perish: a review of planning approaches for adaptation

under deep uncertainty. Sustainability 5(3):955979

Climatic Change (2014) 127:535546 545

Webster PJ, Holland GJ, Curry JA, Chang H (2005) Changes in tropical cyclone number, duration, and intensityin a warming environment. Science 309(5742):18441846

Willoughby H, Darling R, Rahn M (2006) Parametric representation of the primary hurricane vortex. Part II: anew family of sectionally continuous profiles. Mon Weather Rev 134(4):11021120

Winkler J, Duenas-Osorio L, Stein R, Subramanian D (2010) Performance assessment of topologically diversepower systems subjected to hurricane events. Reliab Eng Syst Saf 95(4):323336

Yonetani T, Gordon HB (2001) Simulated changes in the frequency of extremes and regional features ofseasonal/annual temperature and precipitation when atmospheric CO2 is doubled. J Clim 14(8):17651779

Zamuda C, Mignone B, Bilello D, Hallett K, Lee C, Macknick J, Newmark R, Steinberg D (2013) US EnergySector Vulnerabilities to Climate Change And Extreme Weather. U.S. Department Of Energy

Zimmerman R, Restrepo CE (2009) Analyzing Cascading Effects Within Infrastructure Sectors For ConsequenceReduction. IEEE Conference On Technologies For Homeland Security

546 Climatic Change (2014) 127:535546

Simulation of tropical cyclone impacts to the U.S. power system under climate change scenariosAbstractIntroductionClimate change and tropical cyclone activitySimulation methodologyResultsBaseline impactPotential climate impactsMetropolitan area impacts

ConclusionReferences