On the evening of Saturday 31 July 1976, thunderstorms stalled over the upper Big Thompson drainage in the Colorado Front Range and put something close to a foot of rain in about four hours into a narrow canyon that had US Highway 34 and a river at the bottom of it. The Weather Service recognised that a flash flood was underway and said so on the wire that night. Most of the people who died never received that information in any form, because in 1976 nothing existed that could put a message in front of a tourist standing in a canyon campground. The commonly published death toll is 144.

The storm that would not move off the drainage

The Big Thompson River runs east out of Rocky Mountain National Park through Estes Park and then drops through about twenty five miles of canyon to the plains at Loveland, in Larimer County, Colorado. US Highway 34 runs the length of that canyon beside the river, crossing it repeatedly, and in the narrowest reaches the walls close in to a few hundred feet with the road and the water occupying most of the available floor. That geometry is the whole story, because a canyon with those proportions converts a rainfall event into a moving wall of water rather than a spreading sheet.

On the afternoon and evening of 31 July, moist low level air moved west against the foothills under weak steering winds aloft, so the thunderstorms that built over the drainage regenerated in roughly the same place instead of translating downwind. Published rainfall analyses put the heaviest accumulations in the upper canyon and the Glen Haven area at somewhere between about ten and fourteen inches in roughly four hours, with twelve inches the figure most often cited, and the United States Geological Survey professional paper on the storm and flood of 31 July to 1 August 1976 in the Big Thompson and Cache la Poudre basins is the place to get the isohyetal map and the gauge record rather than a secondary summary like this one.

Flow from the main stem and from the North Fork coming down through Drake combined into a flood wave that swept the canyon during the late evening and into the early hours of 1 August. The peak discharge near the mouth of the canyon is commonly published at around 31,000 cubic feet per second against an ordinary late July flow in the low hundreds, and published accounts describe water depths in the narrow reaches on the order of twenty feet. Highway 34 was destroyed in long sections, the bridges went, and the canyon’s telephone and power lines went with them early enough in the event that the ordinary means of talking to people inside the canyon were failing while the water was still rising.

What the Weather Service could see in 1976

It is worth being precise about which part of the warning chain worked, because the case is often taught as a detection failure and that is not what the record supports. Flash flood watch language covering the Front Range foothills was in effect that evening, and the Denver forecast office issued flash flood warning products during the night as reports of flooding came in. Published timelines differ by some minutes on exactly when each product cleared, and if you need the issuance times for a training deck, work from the NOAA natural disaster survey report produced after the flood rather than from retellings, because that report is where the sequence was reconstructed contemporaneously.

The tools available for seeing the storm were poor by any modern standard. The Doppler radar network that now gives forecasters rainfall rate estimates and storm relative motion did not exist, and the radar of that period was designed to find echoes rather than to quantify accumulation, with terrain between the radar and a storm sitting against the foothills. There was no telemetered rain and stream gauge network in the drainage feeding a forecast office in real time, which meant the office had no way to convert what it suspected into a number of inches on a specific basin. Automated Local Evaluation in Real Time gauge networks, the technology that now makes a Front Range canyon observable minute by minute, spread in the years after this flood partly because of it.

What the office largely had instead was telephone reports from people in and near the canyon, which is a detection method with an obvious defect: it depends on the people at greatest risk staying alive, staying connected and thinking to call. The other trap in this event is hydrological rather than technological. The heavy rain fell in the upper and middle canyon, and at the canyon mouth and out at Loveland the evening was comparatively quiet, so the people best placed to act on a warning and the people who could see a reason to act on one were not the same people.

The dissemination tools that existed, and what each one could not do

NOAA Weather Radio was operating in 1976, and the 1050 Hz tone alert that wakes up a receiver was part of the system, but two things limited it badly. Receivers were uncommon in private hands and almost unheard of in a tent or a rented cabin, and the VHF transmitters that carry the service have the same line of sight behaviour as any other VHF signal, which means a deep east-west canyon in the Front Range is exactly the kind of terrain that produces coverage holes. Specific Area Message Encoding, which lets a receiver stay silent unless the alert is for its own county, did not arrive until the 1990s.

The Emergency Broadcast System existed and could be activated at state and local level, though local activation for weather was rare in practice and the system had the structural weakness that it only reaches a person who is already tuned to a participating station. Ordinary broadcast reception in the canyon was its own problem, since canyon walls block FM and attenuate AM groundwave, and a family eating dinner in a motel in the Narrows had no particular reason to have a radio on at all. There were no sirens in the canyon, and an outdoor siren would have had limited value there anyway, because sirens are an outdoor alerting device that tells people to seek information without telling them what to do.

Everything a modern emergency manager would reach for first was still years or decades away. Commercial cellular service in the United States did not begin until 1983, reverse notification calling systems came into county use in the 1990s, the Integrated Public Alert and Warning System was established in the 2000s, and Wireless Emergency Alerts went live in 2012. That left the telephone, which reached only residences and businesses with a line and which failed as the flood took out the plant, and it left officers in vehicles with public address systems and hands to knock on doors.

Centennial weekend and a population nobody had a list of

Colorado entered the Union on 1 August 1876, so 31 July 1976 was the eve of the state centennial, and the canyon was full. Estimates of how many people were in the canyon that evening vary widely across published accounts, generally in the range of two thousand to three thousand five hundred, and I would treat any single figure with suspicion because nobody was counting. What is not in dispute is the composition of that population: campers, motel and cabin guests, fishermen, people at a dance hall, families driving between Loveland and Estes Park on a Saturday night, and a comparatively small number of permanent residents.

List based notification is useless against that population, and this is the durable lesson of the case for a working agency. A county can enrol residents in a notification system, publish a weather radio programme, build a phone tree for the canyon homeowners association and still have done nothing whatever for the four people asleep in a tent who arrived from Nebraska that afternoon. The people at highest risk in a recreational corridor are almost always the people with the weakest connection to local information sources, and the density of that population peaks on exactly the holiday weekends when staffing is thinnest.

Visitors also carry a knowledge deficit that changes what a warning has to say. A canyon resident knows which side of the road climbs, knows where the river has come up before and knows what the sound of rock moving in the channel means. A tourist knows the road, because the road is the only thing he has seen, and when told to get out he will get in his car and drive on the road, which in the Big Thompson meant driving along and repeatedly across the river he was trying to escape. Many of the dead were found in or near vehicles, and survivors were disproportionately those who abandoned the road and climbed the canyon wall.

The warning that reached people was carried by people

The warning that actually reached anyone in the canyon that night was delivered by Colorado State Patrol troopers and Larimer County deputies driving the highway, using public address, stopping cars and knocking on doors. Sergeant W. H. Purdy of the Colorado State Patrol was doing that work when he was caught by the water and killed. Accounts of his final radio transmission vary in their exact wording, and the substance common to them is that he reported he was in the middle of the water and could not get out, which is the reason I will not put quotation marks around a specific sentence here; if you want the wording, go to the Colorado State Patrol’s own memorial material and the contemporary reporting rather than to a version that has been through several retellings.

Door to door notification is the most effective method available for reaching people who are not connected to any system, and it is also the most dangerous and the slowest, which is a combination that should make any incident commander uncomfortable. It puts a responder into the hazard area at the moment the hazard is arriving, it covers a small fraction of the population per unit time, and it degrades to nothing the instant the access route fails. In this canyon the access route and the hazard were the same strip of ground.

The communications engineering point underneath this is one that anyone who has worked a mountain county already knows. A vehicle working a deep canyon is at the edge of coverage on most terrestrial systems, talk-in from a portable is worse than talk-out, and a unit that drops below the rim may be heard by nobody at all. Published accounts describe reporting out of the canyon degrading as the night went on, and by the time the water had taken the road, the bridges and the wire plant, the county had lost both the ability to push warnings into the canyon and the ability to learn what was happening inside it. If you plan warning delivery by patrol unit, you have to plan the radio path that unit will use and test it in the terrain before you need it.

Counting the dead, and why the number moves

The figure most commonly published for the Big Thompson Canyon flood is 144 dead. Reputable accounts, including state and federal retrospectives and academic treatments of the event, carry figures in the range of roughly 139 to 145, and I am deliberately not attaching a specific number to a specific agency from memory, because that is the kind of detail that gets an instructor caught. If you are putting a number in front of a class or in a written plan, open the source you intend to cite, use the number that source publishes, and say in the same breath that published totals differ by a few.

The reasons the count moves are ordinary and worth understanding, because they recur in every mass fatality event. Several bodies were never recovered, with five or six the number usually given, so whether a missing person is carried as a death is a decision rather than an observation. Deaths from injuries and from cardiac events during the event or shortly after may be counted or not. Flooding also occurred in the Cache la Poudre basin the same night, and whether those deaths are folded into a Big Thompson total depends on who is doing the folding. Roughly 150 injured is the figure usually published, damage estimates are commonly given at about 35 million dollars in 1976 dollars, and several hundred homes and dozens of businesses were destroyed, with the exact counts varying by source.

The honest framing for teaching purposes is that this was the deadliest flash flood in Colorado history and that it killed roughly a hundred and forty five people in a few hours on a Saturday night. Precision beyond that is false precision unless you are quoting a named source, and the reader who later finds a different number in a government document will decide the rest of your material is equally loose.

What got built after 1976

The most visible product of this flood is the simplest. Colorado put up the canyon signs telling people to abandon their vehicles and climb, which is a protective action message pre-positioned in the hazard area so that it does not depend on any electronic path reaching anyone. That is a genuinely good piece of warning design, and it costs almost nothing compared with the systems that came later.

The technical response took longer and was less visible. Telemetered rain and stream gauge networks using the ALERT protocol spread through Colorado and much of the western United States in the years that followed, and the Front Range now has dense gauge coverage feeding both the Weather Service and local flood warning programmes, with the Urban Drainage and Flood Control District in the Denver metropolitan area building a flash flood prediction programme at the end of the 1970s that became a model elsewhere. The Weather Service’s flash flood programme was reworked over the following decades, NOAA Weather Radio transmitter coverage expanded, SAME encoding arrived in the 1990s so that receivers could be made to alert only for their own area, and the Emergency Broadcast System was replaced by the Emergency Alert System in 1997.

Larimer County and the Colorado Department of Transportation also built the administrative machinery that did not exist in 1976, including canyon evacuation planning and the authority and the practice of closing the highway before a predicted event rather than after an observed one. When the September 2013 Front Range floods hit the same canyon, damage was extensive and the death toll statewide was in the single digits, with published counts differing depending on which deaths are attributed to the flooding. I would not present that contrast as proof that the warning improvements worked, because the two storms were not comparable: 2013 was a multi-day, widespread rainfall event with days of lead time and 1976 was a stalled convective storm that produced a lethal flood wave within a few hours of the first raindrop.

What still does not reach everyone

Wireless Emergency Alerts are the closest thing we have to the tool Big Thompson needed, because they reach a handset by geographic targeting without the owner having enrolled in anything, and a visitor from out of state gets the alert on the same terms as a resident. The limits are real and they bite hardest in exactly the terrain this article is about. The handset has to be on, powered, and attached to a cell site, which in a Front Range canyon or an Appalachian hollow is not a given, and the alert category has to be one the user has not switched off. Over-alerting produces exactly that switching off, which is a system-level cost paid by whoever sends the next alert.

Opt-in mass notification systems remain useful and remain oversold. The number that determines whether yours will perform is the enrollment percentage of your at-risk population, and in most counties I have seen that number is far lower than the people who bought the system believe, and nobody checks it between contract renewals. It also reaches nobody who is visiting, which in a recreation corridor may be the majority of the people in harm’s way on the night it matters. NOAA Weather Radio still requires a receiver and still has terrain holes, and outdoor sirens still cannot be heard indoors, in a car with the windows up, or over the sound of a river in a canyon.

Door to door notification is still the backstop, and it is still slow and still dangerous, which means the planning decision worth making in advance is at what point you stop sending people into the hazard area to warn other people. That threshold belongs in writing, decided by an operations chief on a calm afternoon, because it is not a decision anyone makes well from a command post at nine o’clock at night with water rising and a trooper not answering.

What to do at your agency

• Have your emergency manager produce a one page list of every transient or seasonal concentration of people in your jurisdiction that appears on no notification roster, naming each campground, RV park, river outfitter, trailhead, fairground and event venue along with the manager’s name and number, and attach that list to the flood or severe weather annex that already exists.
• Ask your PSAP manager or alerting authority to show you the current IPAWS alerting permissions for your jurisdiction, who personally holds the credentials, when each of those people last sent a test or training message, and whether anyone has drawn a practice polygon over your highest risk valley in the last twelve months.
• Have your communications officer take a NOAA Weather Radio receiver and drive the low ground of your jurisdiction, logging where the signal fails, and give the map to your emergency manager so that the coverage holes are documented rather than assumed.
• Get the enrollment number for your opt-in notification system from the vendor administrator, expressed as a percentage of households, and put that single figure on the agenda of the next county emergency management or public safety committee meeting.
• Have a shift supervisor run a radio coverage test in the canyon, valley or bottomland where you would actually send units to warn residents, with a portable on the hip rather than a mobile in the cab, and write down where talk-in fails.
• Ask your operations chief to add one paragraph to the flood annex stating the condition under which door to door notification stops and units withdraw to high ground, and get it signed rather than discussed.
• If your training material uses Big Thompson, Rapid City, Johnstown or any other historical flood, have whoever owns the slides check the casualty figures against a named published source and add a line saying which totals are disputed.

Takeaways

• Thunderstorms stalled over the upper Big Thompson drainage in Larimer County, Colorado on the evening of 31 July 1976, producing published rainfall estimates of roughly ten to fourteen inches in about four hours, with twelve inches the figure most often cited.
• The commonly published death toll is 144, reputable accounts carry figures between roughly 139 and 145, and the count moves because several bodies were never recovered and because attribution of later deaths and of deaths in adjacent basins differs by source.
• The Weather Service issued flash flood watch and warning products that evening, so this case is a failure of dissemination and public response rather than a failure to recognise the hazard, and published timelines differ on the exact issuance times.
• Nothing available in 1976 could reach a transient population inside a canyon, since weather radio receivers were rare and terrain-blocked, the Emergency Broadcast System required a listener already tuned in, there were no sirens, and cellular service, reverse notification and Wireless Emergency Alerts were years or decades away.
• The canyon held an estimated two thousand to three thousand five hundred people on Colorado centennial weekend, mostly visitors who appeared on no list any agency held and who did not know the terrain well enough to choose the right escape route.
• Warning delivery fell to Colorado State Patrol troopers and Larimer County deputies driving the highway, and Sergeant W. H. Purdy of the Colorado State Patrol was killed doing that work.
• Survivors were disproportionately people who left their vehicles and climbed the canyon wall, which is why Colorado later installed signs in the canyons directing people to climb to safety rather than drive out.
• Telemetered ALERT gauge networks, expanded weather radio coverage with SAME encoding, the Emergency Alert System, IPAWS and Wireless Emergency Alerts have all narrowed the gap since 1976, and none of them removes the need for a plan that assumes some fraction of your at-risk population will be reached only by a person knocking on a door.