In April 1997 the Red River of the North reached a record crest at Grand Forks, North Dakota, more than five feet above the 49-foot figure the National Weather Service had issued as a spring outlook in February. The dikes were overtopped, and most of Grand Forks and nearly all of East Grand Forks, Minnesota, were evacuated. A fire burned part of downtown while the streets were under water. This guide covers how the outlook was produced, why officials and residents read it as a fixed ceiling when it was an estimate with a wide band around it, and what that misreading means for any agency that builds protection to a forecast number.

A flat valley, a northward river and the 1979 benchmark

The Red River of the North runs north along the border between North Dakota and Minnesota and crosses into Manitoba on its way to Lake Winnipeg. Because it flows north, the southern part of the basin usually thaws first while the river downstream is still frozen, and meltwater coming from the south can meet ice and backwater farther north. The valley floor is the bed of glacial Lake Agassiz and is about as flat as inhabited land gets. Water that leaves the channel spreads out over a very wide area, and a rise of a foot or two in the river can put a great deal of additional ground under water.

Grand Forks sits on the west bank, and East Grand Forks faces it on the Minnesota side, where the Red Lake River joins the Red. River levels at the Grand Forks gauge are reported as stage, meaning feet above a fixed reference point, which is different from the depth of the water. The NWS flood stage for that gauge is 28 feet. A stage in the high 40s or 50s therefore describes a river standing more than twenty feet above the level at which it begins to cause flooding at that location.

The historical benchmarks shaped how people heard the 1997 outlook. The NWS historical crest listing for Grand Forks puts the 1979 crest just under 49 feet and the 1897 crest a little above 50 feet. Readers who need exact values should take them from that listing rather than from me. The 1979 flood mattered more than the 1897 one because it was within living memory, which meant that many residents and officials in 1997 had already fought a flood near 49 feet and come through it.

The winter of 1996 to 1997 brought heavy snow across the basin through a series of storms, and Fargo, upstream, is commonly reported to have had one of its snowiest winters on record. Accounts of the season also describe wet soils going into the freeze the previous fall. Wet soil that freezes absorbs little meltwater in spring, so more of the snowpack runs off into the river. By February, forecasters and local officials agreed that a major spring flood was coming. What they did not have, and could not have had, was a precise figure for how high it would go.

The February outlook: 49 feet, conditions attached

River forecasts for the Red are produced by the NWS North Central River Forecast Center, and the local NWS forecast office passes them to the public and to emergency managers. In late winter the forecast center issues spring flood outlooks, which are a different product from the short-range crest forecasts issued once water is moving down the channel. An outlook issued in February has to rest on what is known then, such as the water content of the snowpack and soil conditions, together with assumptions about weather that has not yet happened, including how fast the snow will melt and how much more rain or snow will fall before it does.

The outlook for Grand Forks that entered public memory was 49 feet. Roger Pielke Jr., then at the National Center for Atmospheric Research, examined how it was produced and used in a 1999 paper titled “Who Decides? Forecasts and Responsibilities in the 1997 Red River Flood.” As his account and later summaries describe it, the outlook was conditional on future weather, and some accounts describe a lower companion figure that assumed more ordinary conditions between the outlook and the melt. I am not reproducing the complete set of outlook figures and dates here, because popular retellings reduce them to a single number and I have seen them reported inconsistently. The NWS service assessment of the 1997 Red River floods is the place to read exactly what was issued and when.

The detail that matters most for planners is that the figure held at 49 feet from February until shortly before the crest. A number repeated in briefings, news coverage and planning meetings for about two months comes to feel settled, even when the product that carried it was never intended to be read that way. By April, 49 feet was the figure the city’s dike building was measured against and the figure residents used to decide whether to move belongings out of basements. It had also become the yardstick for judging whether the effort was on track.

Beyond the record: rating curves, confined flow and the band around the number

A stage forecast is usually produced in two steps. Hydrologic models estimate how much water will pass the gauge, measured as flow in cubic feet per second, and the forecasters then convert that flow to a stage with a rating curve. The rating curve is built from paired field measurements of flow and stage taken over years at the gauge. Within the range of measured floods the curve is well supported, but once a forecast calls for more water than has ever been measured at that point, the curve has to be extended beyond the data. The extended part is an engineering estimate, and its error compounds whatever error is already in the flow forecast.

Levees and dikes change the relationship between flow and stage. When a river at high water can spread across its floodplain, each additional unit of flow raises the stage only a little because it has room to go sideways. When levees and emergency dikes hold the water in a narrower corridor, the same flow has to stand higher. In 1997, communities up and down the Red raised their protection as the season went on, so the river at Grand Forks was more confined than in earlier floods. That raised the stage a given flow would produce, and a rating relationship built partly from those earlier floods described a river with more room than it actually had.

I want to be careful about the next point. My understanding from the service assessment and from later reviews is that the Grand Forks stage forecast depended on assumptions of this kind, both about how far the rating curve could be extended and about how much water would stay in the channel. The specific assumptions the forecasters made, and the specific figures they used, are set out in the assessment. Anyone who needs them should read that document rather than rely on a paraphrase. The general mechanism is standard river hydraulics and applies to any stage forecast made for a confined river at a level above its record.

Together those factors put a wide band of uncertainty around the outlook. Part of the band came from weather that had not yet occurred, and part came from the step of converting a flow nobody had measured into a stage. Pielke’s argument, as I read it, was that forecasters understood the outlook could be wrong by a substantial margin, but that the uncertainty never reached users in a form they could plan around. The public product carried a single stage, and that single stage was what got passed along.

How 49 feet became a ceiling: dike heights, freeboard and memory

The flood fight at Grand Forks was organized around the outlook. The city, the U.S. Army Corps of Engineers and very large numbers of volunteers raised emergency dikes on top of the permanent levees and filled sandbags through the spring. Accounts generally put the target protection at roughly 51 to 52 feet, which was the outlook figure plus several feet of freeboard. I have seen the exact height reported differently, and it varied from one section of dike to another, so I give it as a range.

The planning problem lay in how the margin was used. Freeboard is normally there to absorb wave action, settling of the fill, low spots and the ordinary imprecision of building a dike under time pressure. In Grand Forks the same few feet also had to absorb any error in a two-month-old outlook for a stage above the record, which means the margin was being asked to cover two different kinds of risk at once. When the river came in more than five feet above 49, the freeboard could not cover both, and in practice it could not cover even the forecast error by itself.

The 1979 benchmark reinforced the reading. A 49-foot outlook sounded like a repeat of a flood many residents had already fought successfully, and that comparison shaped how much effort households put into moving property, how seriously businesses took the threat and how ready people were to leave. Pielke’s paper describes local officials and residents treating the number as a prediction of the actual crest, and it describes the NWS position afterward that the outlook had been issued with caveats that users did not take on board. His account presents those as two sides of a disagreement over whose job it was to understand the forecast, and I would encourage anyone who uses forecasts to read it for that reason.

Nobody in this sequence was careless. The forecasters issued what their methods produced, and the city built protection with a margin above the forecast as flood fighters normally do. Residents trusted a figure that matched their own experience. The failure lay in the gap between what the number meant to the people who made it and what it meant to the people who used it, and no one owned that gap until the water came over the dikes.

The April blizzard and the forecast changes in the last week

In early April, while the flood fight was under way, a severe blizzard crossed the region, beginning with freezing rain and ice and followed by heavy snow and high winds. Many accounts call it Hannelore. It brought down power lines across the valley and left a large number of customers without electricity. It also halted dike work, and it added more water to a basin that already held more than it could drain. The storm came after the 49-foot outlook had been issued, so any condition in that outlook based on normal weather after February was overtaken by events.

As the meltwater moved north and the river at Grand Forks rose, the NWS raised its forecast crest more than once during the final week or so before the crest, and the later figures approached what the river actually did. Because popular accounts disagree on the sequence of forecast figures and their times, and the service assessment holds the record, I am not giving that sequence here, and the operational point does not depend on it: each increase came when little time was left to act, so the crews raising dikes were working against a target that kept moving upward while the river kept rising.

Raising an emergency dike by a foot or two over several miles in a matter of hours is a very large job, and it gets harder as the river climbs, because crews and equipment are working on saturated ground beside high water. A forecast increase that would have been easy to absorb in March became, in the last days, a race the flood fighters had little chance of winning. The NWS historical crest listing for the Grand Forks gauge records the 1997 crest as 54.35 feet on April 22, 1997, which was more than five feet above the outlook and roughly four feet above the 1897 record. If you are citing the exact figure, verify it against the NWS listing.

Overtopped dikes and the evacuation of two cities

Beginning on April 18, dikes in Grand Forks and East Grand Forks were overtopped or gave way, and water moved into neighborhoods that had been protected until then. In a valley as flat as the Red’s, water that gets behind a dike spreads far and keeps spreading, so the area at risk grew over hours rather than staying near the breach. Grand Forks Mayor Pat Owens and city officials ordered evacuations of neighborhood after neighborhood as conditions worsened. East Grand Forks, which is smaller and lies lower relative to the river, was emptied almost completely over the same period.

The scale is commonly reported as most of Grand Forks’ roughly 50,000 residents and nearly all of East Grand Forks’ roughly 9,000. Those are widely repeated figures for the evacuated population, and they are not counts I can tie to a single authoritative tally, so I give them as approximate. Grand Forks Air Force Base, west of the city, is commonly reported to have sheltered thousands of evacuees, many of them in an aircraft hangar. Many other residents went to relatives or to nearby towns and stayed there for weeks.

The evacuation came late because it was triggered by dikes failing rather than by the forecast. For weeks the plan had assumed that protection to roughly 51 or 52 feet would hold against a 49-foot crest, so the need to move tens of thousands of people became clear only when the margin had already been used up. That left evacuations running in deteriorating conditions, some of them overnight, with water already in streets. It also left people with little chance to move belongings, medications and vehicles out of the way of the water.

Critical infrastructure went with the neighborhoods. The Grand Forks water treatment plant was lost to the flood, which left the city without safe water, and the sanitary sewer system failed with it. Those losses meant that residents could not return as soon as the water dropped, and they bear directly on the fire described next. I am not giving a total damage figure. Published estimates vary widely depending on whether they cover the two cities, the wider basin or the Canadian side as well, and I do not have one I would stand behind.

The downtown fire of April 19, fought from floodwater

On April 19, with downtown Grand Forks under several feet of water, a fire broke out in a commercial block. Accounts generally identify the Security Building as where it started. The fire spread through adjoining buildings over parts of several downtown blocks, and one of the buildings destroyed was the home of the Grand Forks Herald. Accounts commonly put the loss at about eleven buildings, though I have seen different counts depending on how adjoining structures are counted. Some accounts attribute the fire to an electrical cause. I have not seen a final determination I can cite, so I am not stating a cause here.

Fighting the fire required departing from normal methods because of the flood. Engines could not drive into the affected blocks in the usual way, and the municipal water system that supplies hydrants was failing along with the treatment plant, which made hydrant pressure unreliable or absent. Firefighters worked from apparatus positioned in or at the edge of the water. Helicopters dropped water from buckets onto the burning buildings, and accounts describe those aircraft as military and National Guard helicopters. Floodwater was all around the fire, but that did not make it usable for suppression without pumps and equipment able to draft it, and the fire burned for a long time before crews could bring it under control.

The Herald kept publishing. Its staff worked from temporary quarters outside the city, the paper was printed at the St. Paul Pioneer Press in Minnesota, and copies were distributed free to evacuees. The Herald received the 1998 Pulitzer Prize for Public Service for its coverage during the flood. For emergency managers, that matters because the local paper remained a working channel of public information even after its own building had burned and flooded.

A fire inside a flood combines hazards that planning documents usually treat separately. The flood annex generally assumes fire protection continues as normal, and the fire department’s plans generally assume streets that apparatus can drive on and hydrants that work, but on April 19 in Grand Forks neither assumption held. Electrical systems in flooded buildings, gas service and occupants who had left stored materials behind all raise the ignition risk at the moment suppression capacity is lowest. The fire department ends up depending on drafting equipment, high-clearance vehicles and aircraft that nobody had included in its ordinary response plans.

What changed afterward: probabilistic outlooks and permanent protection

The NWS has since moved toward long-range river outlooks that state uncertainty in numbers. Its Advanced Hydrologic Prediction Service, which grew through the late 1990s and 2000s, produces outlooks giving the probability that a river will exceed a given stage over a coming period, built from many runs of the hydrologic models under a range of past weather conditions. Spring flood outlooks for the Red River now present stages with probabilities attached, not a single figure. A planner can read the stage with a given chance of being exceeded and set protection and triggers against it. The exact form of the products has changed over the years, so for how they look today check with your NWS forecast office.

The communities rebuilt differently. Grand Forks and East Grand Forks carried out a permanent flood protection project with the Corps of Engineers that combined levees and floodwalls with buyouts of properties in the most exposed areas, and much of the land along the river became open space instead of houses. The project was completed in the 2000s. The Corps district office and the two cities are the right sources for its design level and current status, which I am not stating here.

The forecast lesson goes beyond the Red River, and it holds for any agency in the country. A crest forecast, a hurricane surge estimate, a dam inundation map and a wildfire spread projection all carry uncertainty that often drops out between the forecaster and the decision maker. Probabilistic products help only when someone at the receiving end reads the probability and plans against it. A local emergency manager can still take a probabilistic outlook, pick out its most likely value and brief that alone, which repeats the 1997 mistake with better tools behind it.

The 1997 record also shows where responsibility for interpretation sits. After the flood, forecasters and local officials disagreed in public over whether the uncertainty had been communicated, as Pielke’s paper describes. For a working agency, the practical answer is that the emergency manager has to ask for the range and its assumptions every time, because the most a forecast office can do is offer them, and whether they get used depends on someone at the receiving end asking.

What to do at your agency

• The emergency management coordinator calls the warning coordination meteorologist at the local NWS forecast office and asks, for each forecast river gauge in the jurisdiction, what the record stage is and whether the rating curve is considered reliable above it, then writes the answers into the flood annex.
• The emergency management coordinator checks the flood annex for every protection height or evacuation trigger tied to a single forecast crest, and rewrites each one so that it names the probabilistic outlook stage it is set against and the freeboard added for construction tolerance.
• The fire chief or a designated officer adds one paragraph to the department’s flood response plan naming the drafting-capable apparatus, high-clearance vehicles and helicopter bucket resources available through mutual aid if hydrant pressure fails during a flood, with the phone numbers for requesting them.
• The utilities director, or the emergency management coordinator on the utilities director’s behalf, confirms at what river stage the water treatment plant goes out of service and sends that figure to the fire chief, so that the fire department knows the stage at which it loses hydrants.
• The public information officer writes one paragraph into the public information annex requiring that every public statement of a forecast crest include the range or probability the forecast office provided and the date and time the forecast was issued.
• The emergency management coordinator adds a forecast uncertainty briefing to the agenda of the next flood season planning meeting already on the calendar, with the NWS office asked to walk through one current probabilistic outlook product for a local gauge.

Takeaways

• The NWS spring outlook for Grand Forks that entered public memory in 1997 was 49 feet, and accounts describe it as conditional on weather after it was issued in February.
• The NWS historical crest listing records the 1997 Grand Forks crest as 54.35 feet on April 22, 1997, which was more than five feet above the outlook and above the 1897 record.
• A stage forecast above the record depends on an extended rating curve and on assumptions about how much water levees keep in the channel, and the specific assumptions used in 1997 are in the NWS service assessment.
• Emergency dikes were built to roughly 51 to 52 feet, so a few feet of freeboard had to cover both construction tolerance and the error in a two-month-old outlook.
• Roger Pielke Jr.’s 1999 paper describes users treating the outlook as a prediction of the actual crest and the NWS position that the caveats had been there for users to read.
• Most of Grand Forks and nearly all of East Grand Forks were evacuated after the dikes failed starting April 18, which put the evacuation on the worst possible timeline.
• A fire on April 19 destroyed downtown buildings including the Grand Forks Herald’s, while floodwater kept apparatus out and the water system could not supply hydrants, and I have seen no final determination of its cause that I can cite.
• Probabilistic river outlooks exist now, but they protect a community only if the emergency manager plans against the stated probabilities rather than briefing the most likely value alone.