Hurricane Hugo, Puerto Rico, the Virgin Islands, and Charleston, South Carolina, September 17-22, 1989 (1994)

Earl J. Baker, Florida State University, Tallahassee

Although the worst of Hugo affected one of the more sparsely populated reaches of the South Carolina coast, it was necessary for residents in a much larger area to assess and respond to the threat. Forecast information provided by the NHC was good during most of the crucial response period, making decision making much easier than might otherwise have been possible. Inundation maps and evacuation clearance-time calculations produced in pre-storm studies proved useful and generally accurate. Computerized and graphical decision aids were utilized extensively and contributed an impression of high-tech performance and credibility to elected officials, but some users had dangerous misconceptions about the functions and capabilities of these forecast tools. There was very little use of forecast uncertainties in the decision process. Evacuations went well, evidenced in part by the low loss of life from flooding. However, in many areas prone to surges, evacuation was not as complete as is widely believed. Had Hugo's eyewall strayed and struck any of the major population centers within the predicted range, many homes would have been flooded with occupants still in them.

After Hugo left the Caribbean, forecasts indicated that the storm would follow a northwesterly course. From September 18 to 19, long-range forecasts should the storm approaching various locations along Florida 's east coast.

At 0600 Wednesday, September 20, the forecast track was moved farther north, anticipating landfall along the South Carolina coast between Beaufort and Charleston in approximately 60 hours. Charleston's probability of being affected by Hugo was put at 12 percent.

During the day on Wednesday, the forecast track was altered slightly, placing Hugo more to the northwest, then more to the north just before landfall, which in

the 1800 advisory was anticipated at Beaufort in less than 36 hours. A hurricane watch was issued for the area from St. Augustine, Florida, to Cape Hatteras, North Carolina, at that time, indicating that landfall was expected within 36 hours.

Throughout Wednesday, communities on Florida's east coast and in Georgia and South Carolina monitored the storm, watching for a more westerly track or an increase in forward speed. No significant response actions were implemented, although in some locations such as Beaufort, officials suggested as a precautionary measure that residents go to friends and relatives farther inland if they would feel more comfortable doing so.

By Wednesday evening, officials in Charleston County were recommending that residents evacuate. The forecast that prompted the evacuation call was considered speculative, but officials felt the 11 p.m. news programs would be their last opportunity to reach residents via the mass media until the following morning.

The next morning at 0600 the center of Hugo was forecast to reach Beaufort in 24 hours, and the NHC issued a warning from Fernandina Beach, Florida, to Cape Lookout, North Carolina The watch remained in effect south to St. Augustine and north to Cape Hatteras. Charleston 's landfall probability was 30 percent, and Hugo's sustained winds were reported to be 95 knots (109 mph), just shy of being a category 3 storm.

The Governor of South Carolina ordered evacuation of barrier islands, beaches, and peninsulas, except for the city of Charleston. Officials in Charleston County changed their recommendation to an order. Local governments disseminated the order, assisted by the National Guard. Most, if not all, locations in South Carolina evacuated for a category 3 hurricane. Coastal Georgia also began to evacuate in response to the warning, as did parts of North Carolina.

At 1200 on Thursday, Hugo was upgraded to a category 3 storm, with winds reported and forecast to remain at 100 knots (115 mph). The track moved slightly north, taking the center over Charleston.

At 1500 Thursday, a special advisory was issued to report that Hugo 's winds had unexpectedly increased to 110 knots (126 mph) and that the forward speed had increased (from 15 to 20 knots [17 to 23 mph]). The hurricane warning area was extended to Oregon Inlet, North Carolina, and the watch was extended to Cape Henlopen, Del. The track was shifted a bit farther north (with landfall predicted near Georgetown, South Carolina), and the areal extent of the windfield was expanded. The intensification prompted evacuees in at least one Myrtle Beach shelter to be relocated farther inland.

At 1800, Hugo was reported having intensified to 120 knots (138 mph) sustained winds, making it a category 4 hurricane (116 knots [133 mph] the threshold). Forward speed decreased slightly to 17 knots (20 mph). The evacuation was nearly complete by that time, however, and few preparations were altered.

The public was predictably responsive to the information disseminated by the media and local officials. A telephone sample survey was conducted in January 1990 to document how households responded in the surge-prone areas of three South Carolina locations: Beaufort, Charleston, Mt. Pleasant/Sullivans Island/Isle of Palms, and Myrtle Beach (Baker, 1990). The following account is based primarily upon that study.

Overall evacuation rates varied from 62 percent in Charleston to 81 percent in Mt. Pleasant/Sullivans Island/Isle of Palms, but there were variations within these areas. From high-risk Sullivans Island and Isle of Palms 96 percent left, and there were probably comparable successes in other high-risk barrier islands. If more residents had heard official orders to evacuate and realized that they were orders —rather than notices—more would have left. Overall, 89 percent of the residents in the areas ordered to evacuate said they evacuated. This compares with an evacuation rate of 70 percent among those who said they heard only a recommendation, and 61 percent among those saying they heard neither. Those living within a block of bays and beaches were most likely to evacuate (84 percent), compared with those in other locations.

Few residents left before official evacuation notices had been issued. Most of those who evacuated in the face of Hugo did so on the morning of Thursday, September 21; by 1200, between 75 percent and 90 percent of the eventual evacuation total in the survey area had left, except the residents of Myrtle Beach, which responded that only 35 percent had gone. By 1600, almost everyone who left had already done so, except in Myrtle Beach, where departures continued until 1900. Almost two-thirds of the Myrtle Beach evacuees said they left between 1200 and 1900.

Very few evacuees went to public shelters (ranging from 2 percent in Mt. Pleasant/Sullivans Island to 13 percent in Myrtle Beach). More people went to motels than shelters, and, as in most evacuations, most of the evacuees (56 to 66 percent) went to the homes of friends or relatives. Officials in most communities

actively discouraged residents from using public shelters, out of concern over insufficient shelter space.

Shelter use is often associated with income, and that was the case in Hugo, although the relationship was not simple. Twenty-five percent of households reporting an annual income less than $10,000 used public shelters, and in no other income group did more than 8 percent go to shelters. Minorities were much more likely to use public shelters than whites (31 percent vs. 5 percent) regardless of income, although shelter use among minorities decreased with income more dramatically than among whites.

Officials encouraged evacuees to go inland and leave the coastal region entirely, and in the sample locations, between 64 and 78 percent of the evacuees went to out-of-county destinations. Of the evacuees staying in their own counties, 25 percent went to public shelters, compared with only 2 percent of those going out of county.

The evacuation proceeded as smoothly as could be expected, and the public evaluated the warning and evacuation performance by public officials very favorably (Baker, 1990). Traffic tie-ups on Interstate 26 leading west from Charleston prompted officials to devise a scheme to employ all lanes for westbound traffic. However, by the time the plan was completed traffic was moving more smoothly and the idea was not implemented.

A public school used as a shelter in McClellanville in Charleston County, South Carolina, flooded to a depth of approximately 6 feet with several hundred evacuees inside during the height of the storm, but there were no fatalities. Building drawings provided by the school board listed the elevation of the ground floor of the building as approximately 20 ft, whereas the actual elevation was closer to 10 ft. No ground survey was conducted as part of the hurricane evacuation studies to verify the actual ground elevation of the building. However, the greater planning failure was that no local officials questioned the 20-ft elevation during the review of the study, and despite the fact that the area was shown as flood-prone on flood insurance and surge maps generated by the study. With hindsight, residents of the McClellanville area felt it was “obvious” that the school site was not 20 ft high.

The single most widely used technical product in the local Hugo evacuation studies were surge-inundation maps. The surge heights predicted by SLOSH after

Hugo's landfall, using parameters determined afterward, matched very closely the actual surge heights. SLOSH maps generated earlier and used for evacuation planning were composites of different storm scenarios, and officials seemed satisfied with their validity in Hugo, although the panel has seen no specific data to verify this. Many local officials using the SLOSH model felt the surge zones should be mapped at a larger scale to show greater detail.

Another heavily employed planning product was clearance-time calculations, which were generated by a consultant for the Charleston District of the U.S. Corps of Engineers. In conjunction with storm forecasts, clearance-time analyses provide the basis response decisions such as when to call for evacuation. They are based upon assumptions about road and street networks and public response patterns. As Hugo threatened the Charleston area, the NWS suggested to the county emergency preparedness director that the clearance times calculated for the evacuation study were unrealistically pessimistic. However, the director was not able to find a countywide consensus among municipal officials regarding the accuracy of the clearance times and, hence, decided to enact the Corps of Engineers recommendations. Actual evacuation times observed in Hugo appeared to be very close to the calculated times, although the analysis was cursory.

The same study for the Charleston district of the corps made key assumptions regarding public response for use in clearance-time calculations and shelter planning. In general the assumptions derived for planning in South Carolina matched Hugo responses well, but not enough planning scenarios were addressed to fit all the conditions that prevailed in Hugo (Baker, 1990).

Public officials were aggressive in the Beaufort and Charleston areas in discouraging evacuees from using public shelters, suggesting instead that they evacuate to inland destinations. The behavioral analysis indicated that, in such a scenario, shelter use would be lower than normal, and more evacuees would go out of town, but no statistical modifier was provided. Evacuation rates, evacuation timing, and vehicle use were all predicted accurately throughout South Carolina. Shelter use was overpredicted, and evacuation out of county was underpredicted in Beaufort and Charleston, but predicted accurately in Myrtle Beach.

In developing behavioral assumptions, residents had been surveyed by telephone to document how the population intended to respond. Actual responses documented in post-storm sample surveys deviated greatly from these intended responses, demonstrating the general invalidity of using hypothetical response data to predict actual responses. Those surveyed overstated their likelihood of evacuating

before officials recommended or ordered evacuation by approximately 300 percent and overestimated public shelter use by a similar percentage. Fortunately, behavioral assumptions were derived for South Carolina by relying primarily upon an empirical modelling approach based upon response patterns observed in past hurricane threats elsewhere.

A full-scale evacuation was implemented in coastal Georgia, and officials estimated that there were 175,000 evacuees, 6,000 of whom went to public shelters. If both figures are even close to actual response, the public shelter demand figures projected in the Georgia behavioral analysis were grossly exaggerated. The Georgia appears to have relied far too heavily on hypothetical response data. Although official estimates of other responses in Georgia are not available, the Georgia behavioral analysis projections for early evacuation (during a watch, prior to recommendations, or orders) also appear flawed.

The response decision-making process was discussed earlier, and it was noted that NHC forecasts were used in conjunction with planning-study clearance times. Graphical tools and computer software were employed in some locations to facilitate the computations. However, none of the tools appeared to have been used very effectively, particularly when forecast uncertainties had to be taken into account, and some users had gross misconceptions about the tasks performed by the aids. At least two local preparedness officials credited the aids with making accurate predictions of when the storm would arrive, when in fact they simply facilitated computations based upon input assumptions. The tools appeared to be accurate simply because the NHC forecasts and study-generated clearance times were accurate. Users were generally pleased with the computer software available to them, but in many locations the sysem was largely a means to impress elected officials and the media, rather than an actual decision-making aid.

Public officials face a tremendously difficult responsibility in deciding whether and when to recommend or compel evacuation during a hurricane threat. Unnecessary evacuations are expensive, disruptive, and unpopular, but waiting too late to leave can be disastrous. In Hugo, decisions were made much easier by the consistency and validity of the forecasts provided by the NHC.

Federal agencies began several studies earlier in the decade that provided the foundation of hurricane evacuation plans in South Carolina. The NHC simulated numerous hurricane scenarios to indicate the areas that would be inundated by storm surges. The U.S. Army Corps of Engineers and its contractors calculated the number of people who would need to evacuate, the length of time necessary to evacuate, and

the public shelter space needed in different storm categories. South Carolina mapped the surge-prone areas of each county, and FEMA paid for the creation of a computer database of this information.

In making response decisions, officials consider the strength of the storm that might affect their location and the amount of time requiredto evacuate for that storm category. They then compare this time to that remaining before the storm impacts the coast to determine when evacuation must begin. The coastal counties in Georgia, South Carolina, and North Carolina had graphical devices provided by the corps and in some cases, computer software provided by FEMA or private firms. Local officials appear to have employed these aids with varying degrees of proficiency.

However, interpreting storm effect computations is not the difficult part of decision making. Complications arise in determining specific inputs such as the strength and size of the storm, where and when it will make landfall, and how far inland it will penetrate. Each of these parameters is forecast by the NHC, but each is subject to error. Accounting for the forecast uncertainty and incorporating it into the decision process is the difficult part of decision making.

Since 1983 the NHC has included probabilities with its advisories, to indicate the likelihood of a storm's passing within 65 mi of certain locations during various time periods. These probabilities account for uncertainty in the direction and forward speed elements of forecasts. Commercial software in use by some of the counties in the threatened area provided a graphical depiction of the uncertainties of position forecast and calculated intensity.

Post-storm analyses of local decision making suggest that few officials systematically employed uncertainty information, particularly NHC probabilities, in responding to Hugo. Discussions of the decision-making process always centered upon the forecast itself and on clearance times, with only a general concern that the forecast or the clearance time calculations could be in error.

In South Carolina, coastal officials relied very heavily upon the Charleston office of the NWS for advice and judgment, and that interaction was more influential than any other input. The Charleston office interpreted NHC forecasts for local officials and in some instances offered second opinions. Facsimile connections to local governments and conference call capabilities would have facilitated the NWS office 's ability to interact with local officials. Computer software available to county officials in South Carolina included a module originally developed by the Charleston NWS staff that indicates appropriate response actions based upon staff judgments about acceptable risk and other factors. When Hugo increased to a category 4 storm at 1800 on Thursday, the Charleston NWS was influential in the decision not to attempt to evacuate a larger area. By contract, the governor' s office in South Carolina worked more closely with the Columbia NWS office.

In retrospect, decisions appeared “correct” largely because the assumed information (forecasts and clearance times) proved accurate. But, had Hugo increased forward speed earlier, for example, the retrospective might have been different. It should also be noted that the worst of Hugo did not hit the populated locations of Beaufort, Charleston, or Myrtle Beach. If this had occurred, 20 to 30

percent of the homes in the inundated areas of those locations would have been flooded while occupied. Although evacuation notices were timely, they were disseminated successfully only in the most hazardous beachfront and island locations.

The claim following Hugo that the relatively low loss of life in such a severe storm was attributable to the improved planning conducted since 1983 is not supported by fact, except perhaps that the studies provide a better indication of the areas needing evacuation. Most locations did not evacuate in Hugo until the NHC issued a warning, which has been the norm for at least two decades. The fact that the right side of the eyewall crossed the coast in one of the least populated reaches of South Carolina's coast was probably the greatest factor resulting in so few deaths. On average, however, the improved studies and plans will result in fewer deaths over time.

Baker, E. J. 1990. Evacuation in Hurricane Hugo in South Carolina. Quick Response Report, Natural Hazards Research and Applications Information Center. Boulder, Colorado: University of Colorado.