The Red River Flood of 1997 in Grand Forks, North Dakota

Predicting Flooding Based on Snow Accumulation

Written by Jennifer Sorenson, December 6, 2002

   

An extremely harsh winter of 1996-1997 resulted in nearly 100 inches of snowfall in Grand Forks, ND.  Everyone wondered where all of the water would go when the snow finally melted.  No one expected quite so much of it to invade our city.  The 1997 Red River flood resulted in record breaking crest levels and flood devastation.  Most cities observed crests of two feet above previous records; Grand Forks observed a crest level over five feet above previous records.  The area is so flat that heavy precipitation is analogous to pouring water on a tabletop.  The Red River, usually 100 feet wide, became a shallow lake 15 miles across, in places.  Overall, the Red River Flood of 1997 was classified as a 100 year flood.  Some cities along the river saw peak discharges with over 500 year recurrence levels.

The devastation did not stop there.  A fire erupted downtown.  It was caused by an electrical problem triggered by the floodwater, and destroyed much of the historic down-town area.  Firefighters attempted to fend off the fire but conventional methods were out of the question.  The streets surrounding the fire were flooded and there was no water pressure in the fire hydrants.  Eventually, they were able to extinguish the fire by dropping fire retardant from an airplane, along with 120,000 gallons of water dropped from a helicopter in 60 massive dumps.

The estimated damages for the entire Red River flood totaled about $4 billion including all U.S. portions of the Red River.  The Grand Forks/East Grand Forks area claimed $3.6 billion of that total.  To put this in perspective, the annual flood disaster relief budget for the entire nation is on the order of $1 billion.  Throughout the flood and the rebuilding process, the people of the city clung to each other and to hope.  The city now stands on new feet as a growing community, a little more united and a little more aware of their blessings.

Immediately above:  This picture was taken in East Grand Forks on April 28, 1997, more than a week after the Red River crested.  Only 27 of 2500 single-family homes in East Grand Forks were untouched by the flood water.  [Photographer: J. Albert Diaz, taken from "Come Hell and High Water"]  Other flood pictures are taken from the Grand Forks Herald.  The far right picture at the top shows the main bridge between Grand Forks and East Grand Forks completely flooded.

Note:  In this paper, I use Grand Forks and East Grand Forks somewhat interchangeably.  Grand Forks is on the west side of the Red River in North Dakota, and East Grand Forks is on the east side of the Red River in Minnesota.   However, they both border the same section of the Red River.  Thus, streamflow and crest levels are identical for both cities.  Also, the geographic proximity of the cities lends itself to unity between the two communities.

 

What caused the flood?

Water, water, and more water!  The fall of 1996 was packed with precipitation.  Grand Forks (GF) and East Grand Forks (EGF) received substantial rainfall through October and into early November.  Despite a dry summer, the ground was now saturated to a depth of five feet.  The first of eight blizzards struck in mid-November.  The blizzards are shown below, with the total amount of snowfall on GF/EGF equal to nearly 100 inches.  As evident from the chart below, not all of the blizzards dropped an enormous quantity of snow; some dropped less than one inch of snow.  A blizzard is characterized by wind speeds of 35 mph or more, considerable falling and/or drifting snow, and visibility near zero.

                               

The pictures below capture a few moments from this harsh northern winter.

 

Above left:  Blizzard Betsy rages on December 17, 1996.  The blizzard had 40-mph winds and 50-below wind chill.  [Photographer:  Dan Diedrich, taken from "Come Hell and High Water"]

Above right:  Blizzard Elmo leaves behind 20-foot-high snow drifts on January 15, 1997, in Lyon County, MN.  This blizzard brought very little snow, but harvested wind chills of 70-below.  [Photographer:  Bill Alkofer, taken from "Come Hell and High Water"]

 

How is snow accumulation used to predict flooding?

The National Operational Hydrologic Remote Sensing Center (NOHRSC) is part of the National Weather Service (NWS).  NOHRSC gathers satellite and airborne remote sensing information, such as snow cover, snow water equivalent, and surface air temperature, and displays this data using geographic information systems (GIS).  The calculated SWE value is entered into a snow model to estimate the volume of snowmelt runoff.  A rating curve is then used to relate the the total flowrate of the river to an estimated crest level.  The NWS then uses this information to issue river and flood forecasts for the nation.  I created the following schematic to illustrate the process used to predict crest levels for a given river.

This process was utilized to predict the crest level of the Red River during the flood of 1997.  Snow water equivalent (SWE) is equal to the amount of water that would exist should all the snow melt instantaneously.  In general, 10 inches of snow would yield 1 inch of water.  This rule of thumb is meant to provide some idea of the relationship between snow and SWE and should not be used without consideration of temporal and other factors.

 

Snow Cover Data Collection

There are three methods presently used to estimate snow water storage:  manual surveys, automated recording devices, and aircraft and satellite remote sensing.  Manual surveys are often used in mountainous areas of western United States, and are performed with a standard federal snow sampler.  Automated recording devices, such as SNOTEL, are used by the U. S. Soil Conservation Service.  They attain precipitation and temperature measurements, as well as SWE measurements through use of snow pillows.  Snow pillows have a stainless-steel plate surface (1.22m x 1.52 m) and are filled with antifreeze solution.  The pressure in the pillow is indicative of the water-equivalent depth of snow.  Aircraft and satellite remote sensing perform best for less than 30 cm in SWE in non-forested areas.  This is the method used by NOHRSC.

NOHRSC develops daily digital maps depicting the aerial extent of snow cover from two image sources.  The first of these sources is Advanced Very High Resolution Radiometer (AVHRR) image data from National Oceanic and Atmospheric Administration (NOAA) polar orbiting satellites.  As the NOAA satellites fly at about 800 km above the earth's surface, data is instantaneously sent to NOHRSC's AVHRR ground receiving center in Chanhassen, Minnesota.  The second source is image data from Geostationary Operational Environmental Satellites (GOES).  GOES data is downloaded hourly from two satellites stationed over the east and west coasts, about 35,800 km above the earth's surface.  When the land is blocked by cloud cover, NOHRSC reverts to data collected by the Defense Mapping Satellite Program (DMSP) Special Sensor Microwave/Imager (SSM/I), which is able to penetrate non-precipitating cloud cover. 

Images from the various sources are radiometrically calibrated to generate one inclusive coverage for the U.S.  The image below was taken from the AVHRR sensor aboard the NOAA-14 and NOAA-12 satellites on April 21, 1997.   [AVHRR flood image provided by NOHRSC]  Snow cover images are shown below, along with the snow water equivalent data.

                                 

 

Snow Water Equivalent Data Collection

Radiation detection systems on low-flying aircraft measure the gamma radiation in potassium, uranium, and thorium radioisotopes in the top 20 cm of soil.  A gamma ray behaves somewhat like an x-ray: it is a penetrating radiation, it does not carry an electric charge, and it travels at the speed of light.  However, it differs from an x-ray in that it is sourced by nuclear processes rather than atomic processes.  Water mass in the snow blocks the terrestrial radiation signal reaching the soil.  Therefore, comparisons between bare ground and snow covered ground radiation measurements will yield snow water equivalent data, to within 1 cm accuracy. 

This technology can also be used to make airborne soil moisture measurements in the upper 20 cm of the ground within an accuracy of 4%.  Historical data is used to establish the connection between the measured gamma radiation count and the actual soil moisture.  Ground samples are collected for each flight line at the time of the flight. 

gamma  

Both figures above are taken from the 1997 North American Airborne and Satellite Snow Data CD.  The figure on the left illustrates the process by which uncollided terrestrial gamma count rates are measured.  The rates are used in the following equation to calculate SWE.  The figure on the right is an example of the satellite image NOHRSC collected.  It shows gamma counts ranging from zero to 2 x 109.  This image was particularly challenging to access.  NOHRSC stores and distributes their Arc/Info raster data files in the form of .arz files.  The .arz extension is indicative of a file compressed with the UNIX "tar" and "gzip" utilities.  In a UNIX system, it is relatively straightforward to uncompress the file and open it in ArcEdit.  On a DOS system, it is necessary to first rename the file with a .tgz extension, followed by execution of the "gzip" and "untar" utilities.  Then, using Arc Toolbox, the image can be converted to a grid by importing the file to a raster.  Once in raster format, black contour lines were added using the Spatial Analyst.

SWE Equation

        CK = Radiation attenuation coefficient in water, cm2/g
        Ko =  Uncollided terrestrial gamma count rates over bare ground
        K   =  Uncollided terrestrial gamma count rates over snow-covered ground
        Mo =  Percent soil moisture over bare ground
        M   =  Percent soil moisture over snow-covered ground

The GIS images below are provided by NOHRSC and display the snow water equivalent data  and satellite snow cover in early April and mid-April, prior to the flooding.  A tremendous volume of snow melted very rapidly, creating major problems for the cities along the river.

  

  

            

Displaying Information in a Geographic Information System

I created the images shown below in ArcMap 8.2.  First, I created my basemap by importing state, county, and hydrologic unit coverages (HUCs) from USGS, as well as river reaches from the EPA.  In order to retain only the portion of the rivers and waterbodies within the North Dakota and Minnesota state boundaries, I used the Clip tool available through the Geoprocessing Wizard.  Next, I created a new point feature class at the monitoring point for each city along the Red River.  I did this by first creating a .dbf file in Excel.  In this spreadsheet, I included the city basin code, name, latitude, longitude, and SWE data collected in the spring of 1997.  I found the latitude and longitude of each city from the location of the NWS basin of the associated NOHRSC flight lines.  I imported the table and converted it to an event, confirming that all features in my map were presented in the same coordinate system (GCS North American 1983).  Grand Forks county is highlighted in the map below.

 

 

 

 

 

 

 

        

I was then able to graphically display the graduated color circles shown below.  [City names are shown to the upper right of the circles.]  As indicated in the legend, the snow water equivalent increases with increasing intensity in color.  I wanted to graphically present the SWE variance in a raster format, and found that I was able to do so using kriging, an option available through the Spatial Analyst toolbar.  Kriging is an interpolation method used to producing a surface of predicted values.  The multi-step mathematical process weights the surrounding measured values to derive a prediction for an unmeasured value.  It is assuring that the contours of the kriges generally align with the SWE point values.

 

Snow Water Equivalent Along the Red River during the Spring of 1997

      

 

 

To more closely examine the trend in snow water equivalent data, I plotted the SWE data from the above dates at the National Weather Service (NWS) East Grand Forks monitoring location.  The SWE data decreases as snow cover decreases (i.e. the snow melts).     

Looking at these five images, it is evident that SWE data decreases first along the lower portion of the Red River.  Intuitively, this makes sense because more snow stays on the ground longer along the northern portion of the Red River.  The Red River is unique in that it flows north.  For this reason, snowmelt from regions south of Grand Forks contribute to the streamflow of the Red River at the Grand Forks location.

                          

Forecasting Methodology

Using the SWE number calculated above, River Forecast Centers are able to estimate the actual volume and temporal distribution of the snow melt runoff.  They arrive at their estimates by using a snow model, which simulates snow accumulation and melting based on future predictions of precipitation and surface air temperature.  Future precipitation is usually based on the normal precipitation for the area, but can be based on zero future precipitation as well (i.e. for river transportation issues).  Future temperature is reflective of normal spring warm weather, resulting in a single snowmelt peak in the month of April.  Based on historic forecasted and observed crest stages, future precipitation estimates are equaled or exceeded 50% of the time.

The Snow Estimation and Updating System (SEUS) is the means by which ground-based, airborne, and satellite snow cover observations are entered into the snow model.  SEUS is also used to improve accuracy and understand uncertainty associated with the snow model.  The snow estimation portion of SEUS consolidates snow cover data into a gridded field of snow water equivalent (SWE), and "normalizes" the data.  Each cell in the grid represents one square kilometer.  Since a 1 sq. km grid cell may not be representative of the actual SWE behavior of that area, the estimation technique standardizes the SWE observations and includes the deviation from the long-term normal SWE data.  The updating portion of SEUS tweaks the SWE estimate before it is used in the snow model.  Tweaking is based on historical analysis and streamflow simulations, and is meant to account for the bias of both the SEUS and the snow model.

Flow volumes are characteristic of hydrologic modeling, but river crest levels are generally used in forecasting flooding and public service announcements.  Rating curves relate flow volumes to river crest (or stage) levels, as shown in the rating curve below [Taken from "The Red River of the North 1997 Floods - Service Assessment and Hydraulic Analysis"].   Rating curves are empirical and are based on observations from previous floods.  The NWS uses stream flow values and rating curves published by USGS.  However, USGS bases their rating curves only on observed flows, so it is up to NWS to do the forecasting.  When predicted crest levels are higher than historic records, NWS must extend the rating curve by extrapolation.  Of course, errors may be introduced in this process, especially since changing hydraulic conditions may result in unexpected changes to the rating curve.

 Plot of DAILY MEAN STREAMFLOW, IN CUBIC FT PER SEC at 000001

The Daily Mean Streamflow graph for the month of April is taken from the USGS water resources website.  It is interesting to relate the date from the discharge graph to the stage level on the rating curve based on the Red River discharge.  For example, consider April 15, 1997.  From the USGS graph, this date corresponds to a mean streamflow of approximately 48,000 cfs.  From the rating curve 45,000 cfs corresponds to a stage level of about 44-45 feet.  According to the NWS, the crest level on April 15, 1997 was 45.3 feet. 

 

How did the city of Grand Forks prepare for the flood?

The National Weather Service originally predicted the Red River would crest at 49 feet (28 feet is flood stage).  Grand Forks and East Grand Forks cautiously prepared for the river to crest at 52 feet.  Sandbagging and dike-building began on April 3, 1997, even before the final blizzard struck (on April 5th).  On April 7th, President Clinton declared North Dakota a disaster area.  The National Guard was activated to help with the sandbagging efforts, and many volunteers emerged from the community.  As the Red River continued to rise, the city began sandbagging 24 hours a day.  Nearly every able body was working to save the city. 

   

Above left: Human chains were formed along the Red River and around people’s homes throughout Grand Forks and East Grand Forks to build sandbag dikes.  [Photagrapher: John Stennes, taken from "Come Hell and High Water"]

Above right:  This man is patrolling the dikes, looking for any sign of dike failure or water overflow.  [Taken from the Grand Forks Herald]  At one of the homes I volunteered at, we were required to form our human chain on top of the dike to continue to build the dike.  On one side, was the raging river, and on the other was a cement patio.   Neither option seemed too enticing.

Various parts of the city were evacuated as water began to ravage the streets and homes of the community.  When the river continued to rise to its final crest of 54 feet, the city's sandbagging efforts were devastated.    The map below is also taken from "Come Hell and High Water" and shows water overlaying most of the city of Grand Forks.

Areas flooded by midnight Friday, April 18 [these areas generally correspond to flood damage above the first level]

Areas flooded by midnight Saturday, April 19 [these areas generally correspond to flood damage in the basement and possibly the the first level]

  Areas flooded by midnight Sunday, April 20 [these areas generally correspond to flood damage in the basement]

 

 

 

Below is a plot of the actual stage level of the Red River at Grand Forks, ND during the month of April and into May.  As noted on the graph, the Red River is at flood stage in Grand Forks at 28 feet.  The river finally crested at 54 feet on April 21, 1997, nearly double the flood stage height.  The crest level exceeded the original 49 foot NWS outlook crest level by 5 feet.

The streamflow plots below are taken from the USGS water resources site.  The graph on the left illustrates the streamflow throughout the water year starting in 1996.  The peak streamflow of 137,000 cfs was reached on April 18, 1997, three days before the river crested.  In the graph on the right, the 137,000 cfs peak streamflow of 1997 far exceeded the previous peak streamflow records of 82,000 cfs in 1979 and 85,000 cfs in 1897.

 

 

What went wrong?

The original crest stage forecasted was 49.0 feet, and the river actually crested at 54.35 feet.  When assessing the potential flood, an error of this magnitude was devastating.  East Grand Forks city engineer Gary Sanders references the 1979 flood, which had a crest level of 48.8 feet and a peak streamflow of 82,000 cfs.  He rationed that predicting a crest level of 49.0 ft in 1997 was putting this flood in the same class, but the peak flow in 1997 reached 137,000 cfs.  Sandars said, "Missing by five feet may not sound like much, but when you talk about flow, the National Weather Service missed by almost 100 percent."

The NWS performed a hydraulic analysis to determine what caused the discrepancy between the forecast and actual crest level so that the same error would not be repeated in the future.  The crest stage forecasted on April 16, 1997 was 50 to 50.5 feet (adjusted based on updated streamflow records from USGS), which means the NWS underestimated the crest stage forecast by approximately 3.8 feet.  While unable to explain 0.6 feet of the forecast error, the hydraulic analysis attributes the remaining 3.2 feet to three factors:  unsteady backwater effect (2.0 feet), bridge effects (0.8 feet), and the effect of the first discharge peak (0.4 feet).

First, the Red River basin has a very mild slope, around 0.5 feet/mile in the Grand Forks area and around 0.2 feet/mile for most of the river downstream (north) of Grand Forks.  Even these mild gradients may have caused the river to experience backwater effects.  Also, the Red River was spread out in some places up to 15 miles, but in other places it was naturally constricted to a width of 600 feet.  This is another possible source of backwater effects.  Backwater effects produce a loop effect on the rating curve, referred to as hysteresis in mathematics.  This loop effect was not accounted for in the rating curves used by the NWS.  Secondly, there are four bridges within a two mile stretch along the Red River bordering Grand Forks and East Grand Forks.  The Red River covered these bridges, but the physical structure (volume) of these bridges was not included in the flow volume used to arrive at the crest prediction.  Finally, a "plug" on the Red River 20 miles upstream (south) of Grand Forks blew out.  This resulted in a premature rush of water flowing through a coulee, which ultimately dumped approximately 30,000 cfs of water into the Red River at the Grand Forks location. 

 

Acknowledgements

I would like to thank Dr. David Maidment for his guidance, assistance, and personal referral to Dr. Tom Carroll.  Dr. Carroll provided a wealth of information from the National Operational Hydrologic Remote Sensing Center regarding snow accumulation and the Grand Forks Flood of 1997 specifically.  Thank you!  I also received actual water level data throughout the city of Grand Forks from Peggy Feilen at the City of Grand Forks and DEM data along the Red River courtesy of Byron Williams at the US Corps of Engineers in St. Paul, MN.  I did not end up using the data they sent, but I appreciate their effort.

 

Resources