This article discusses the distinguishing features of the event – its unusually efficient precipitation generation process and the unique runoff characteristics of the urbanized area. Although in recent years Fort Collins city officials had invested significant resources toward better flood-plain management, those efforts seemed insufficient in this case. An investigation followed to determine why flooding was particularly severe in certain areas; preliminary results are presented.
This event also provided an opportunity to learn important, although sometimes hard lessons about precipitation production, remote sensing of precipitation, runoff, and the impact of human activity. However, with rainfall amounts comparable to the Big Thompson flood (Figure 2), the Fort Collins event also enlightened emergency planners about those flood control measures that did work and future needs.
Figure 1. One of the rescue scenes during the Fort Collins flood on 28 July 1997. (Photo courtesy of John F. Weaver)
Figure 2. An isohyet analysis of precipitation (in inches) in the Fort Collins vicinity, 2330 UTC 28 July – 0500 UTC 29 July 1997 (5:30-11:00 p.m. MDT 28 July). Spring Creek is highlighted in blue, major roads in red, and the railroad in black. The two red "x's" show where the five fatalities occurred, four in a mobile home park (left "x") and one in a nearby residential area. Highway 287 is College Avenue in Fort Collins.
Figure 3. Water vapor imagery with 500-mb heights (solid green) and mean sea-level pressure analysis (dotted yellow) from the Eta 12-hour forecast valid at 0000 UTC 29 July 1997.
On 28 July very well-defined tropical moisture hovered over the area at all levels, as seen in the surface and upper-air observations and the water vapor imagery (Figure 3). Both Denver and Grand Junction atmospheric soundings contained 3.5 cm (1.4 in) of precipitable water, exceptional values for the Great Basin and Front Range locations. At the surface, moist easterly flow covered the eastern Colorado plains where a weak cool front had moved through and become quasi-stationary. Just above the surface layer, south-southeasterly flow advected moist tropical air onto the Front Range, with 700-mb dewpoint temperatures exceeding 10oC (50oF). Above that, in the middle and upper levels, winds were more south-southwesterly, suggesting that storm cell motion may be to the north-northeast. From the perspective of forecasters, the prestorm environment suggested the probability for cells to move off the higher terrain and away from a common generation point. Conversely, south-southeasterly steering flow during the Big Thompson event kept cells moving along or just west of the higher terrain, thus maintaining a common generation point where the rich source air was funneled into the higher terrain.
Heavy rains had already fallen in the Fort Collins area from thunderstorm echoes that were training south to north along the foothills during the night of 27-28 July. The heaviest accumulations up to this time had occurred mostly west of the city, but some heavy rainfall had extended to the west side of the city proper. The large-scale system changed very little during the day. In fact, during the afternoon general easterly flow across eastern Colorado and eastern Wyoming seemed to suggest that new convection would be concentrated over the foothills.
Thunderstorms erupted along the Front Range that afternoon and appeared to be reinforced by a weak shortwave moving northward through Colorado. This shortwave disturbance was noted in the wind pattern at the Aztec profiler, near the Four Corners area, and in model forecasts of quasi-geostrophic fields (discussed in a related article by Fernando Caracena). Drawing on unusually moist air through a deep layer, the thunderstorm cells formed a line and were making slow but steady progress toward the east-northeast away from the foothills in the Denver area. However, as that line of thun-derstorms progressed eastward, outflow generated and moved back toward the foothills. This upslope flow supplied the low-level support for storm initiation. Above the low levels, tropical moisture and atmospheric lift created by the northward-moving shortwave disturbance became superimposed over the low-level generation and evolved into a small area of highly efficient, regenerating convective cells on the southwest side of Fort Collins. Both satellite and radar showed the quasi-stationary cell complex near Fort Collins hanging back as an intense line of convection propagated to the northeast. An important part of understanding the evolution of this event is understanding the evolution of the thunderstorm outflows and the subsequent effects on the Fort Collins convection. Observations at the Denver Weather Forecast Office at 0100 UTC 29 July (7 PM local time, 28 July) showed a strong southeasterly outflow which enhanced the low-level upslope flow into the Fort Collins vicinity. However, without good low-level resolution of the velocity data, it is difficult to determine how this outflow affected the Fort Collins storm evolution either directly or indirectly.
Two other Doppler radars were sampling this storm: the Cheyenne, Wyoming, WSR-88D (roughly 75 km north) and the Colorado State University (CSU) CHILL radar (roughly 40 km east). Dual Doppler analyses using these radars provide much greater detail about the low-level flow supporting storm cell development, detail that was not available to the real-time forecasters. It appears that enhanced upslope flow toward Fort Collins developed after about 0215 UTC (8:15 PM local) from a particularly intense area of convection, including a bow echo signature near Greeley (east of Fort Collins). The enhanced flow from the east and southeast reached Fort Collins about 0235 UTC (8:35 PM local). Although thunderstorm cells in western Fort Collins had already been regenerating and moving very slowly to the north-northeast, an intensification of storm cells occurred over the southwestern side of the city after 0235 UTC. Individual cell movement became northeastward while the nearly stationary complex moved very slowly to the northeast following the Spring Creek drainage. By 0430 UTC the storm complex was dissipating and moving to the northeast.
Certain characteristics of the Fort Collins storm complex were likely quite deceptive to those observing and forecasting its development and evolution. First, the strong southeasterly surface flow that fed the storm was in rain-cooled air behind the initial line of thunderstorms. However, surface dewpoint temperatures were >15oC (~60oF) demonstrating the continued potential in the low-level air. Furthermore, above the low levels, tropical moisture continued feeding up from the south-southwest, unaffected by the initial activity. The net result was exceptional precipitation efficiency in a small area just at the beginning of the sharp elevation rise, although the storm cell intensities were not especially noteworthy compared to other cells to the east.
Radar Guidance – Radar reflectivity values were less intense than the hail-producing convection to the east. Indeed, compared to the thunderstorm cells to the east, the Fort Collins storm was associated with warmer cloud-top temperatures and less lightning strike activity (Figure 4), suggesting relatively shallow convection. Many observations from Fort Collins confirmed the seemingly strange absence of lightning, given the intensity of the rainfall rates. Observers also noted a lack of hail and raindrops that were smaller and warmer than those typically observed in local thunderstorms. These observations and the somewhat atypical satellite and lightning characteristics were indicative of low-centroid storm cells and suggested highly efficient precipitation production.
Radar observations confirmed the low-centroid character of the storm complex in Fort Collins. The Front Range WSR-88D indicated that the storm intensity decreased rapidly between the 2.4o and 3.4o tilt angles (Figure 5). This suggested that most of the precipitation growth was occurring in a 4-km deep layer between cloud base (around 500 m above ground level (AGL)) and 4.5 km (15,000 ft) AGL. This pattern of convection is characteristic of tropical convection in the southern and eastern United States and in landfalling tropical storms, but is uncommon in Colorado. The local radar products are derived using thresholds that are intended to represent the more typical thunderstorms of a semi-arid continental location. Thus, forecasters were particularly challenged because of storm-scale complexity and unrepresentative radar guidance for thunderstorms with such unusual tropical characteristics. This is especially true given the concurrent development of more typical thunderstorm activity in the area associated with stronger signals in the satellite, radar, and lightning guidance.
Figure 4. Infrared satellite imagery and lightning strikes at 0330 UTC 29 July 1997. Less lightning strike activity around Fort Collins (upper left) suggests shallow convection there.
Forecasting Challenges Using Radar-Derived Radar Guidance – Radar-derived rainfall accumulation guidance indicated heavy amounts of accumulation, but was still significantly less than the ground reports (Figures 2 and 6). Although low-centroid convection often results in underestimated intensity because of below-beam effects, that is not what occurred on 28 July. The 0.5o and 1.5o tilt angles were sampling well in the most intense part of the convection in Fort Collins, and these two tilt angles were used in the WSR-88D precipitation algorithms. The problem is apparently from the physical processes occurring within the cloud as the precipitation grows, in particular, the warm rain process. Simply stated, energy backscattered to the radar and used to derive rainfall rates is related to both the mean diameter and the density of the drops in a given volume. Increasing the mean diameter will have a more notable effect on the backscattered energy than increasing the drop density. Thus, given two storm cells with equivalent rainfall rate potential, a mixture of large hydrometeors (water and hail) can be seen as a more intense storm cell than one with a higher concentration of smaller raindrops. Derived rainfall rates for the high concentrations of small-diameter drops in tropical convection is typically underestimated by the default setting of the WSR-88D. Consequently, a different reflectivity to rainfall (Z-R) is permitted in areas where tropical rainfall is common.
Figure 5. Four-panel reflectivity from the Front Range radar with rivers/reservoirs/ highways background. Tilt angles clockwise from top left are 0.5o, 1.5o, 2.4o and 3.4o.
In the Front Range area, convection typically contains large drops and hail which are more accurately represented by the standard WSR-88D parameters. A rainfall rate cap is designed to prevent anomalous rainfall rates associated with hail. On 28 July, the rainfall rate cap and perhaps the standard Z-R relationship contributed to the underestimated rainfall guidance in Fort Collins. Although observed conditions that day strongly suggested these problems, the solution is unfortunately not as simple as changing a few parameters. The other storms of the day did not have tropical characteristics quite as impressive as the Fort Collins storm. Adaptation of the parameters to tropical convection may have resulted in significantly overestimated rainfall guidance in some of these other areas. Another factor that users must recognize is that if the Fort Collins storms had occurred 50 km farther away from the radar, then range degradation would have likely overwhelmed any attempted improvements to the rainfall guidance because the radar would have been sampling above the heavy precipitation region of this low-centroid storm. In real time the forecasters had the difficult task of recognizing the unique characteristics of the Fort Collins storm evolution that resulted in exceptionally efficient rainfall production. Accurate evaluation of cases requiring this type of real-time assessment and reaction will be a major challenge for AWIPS-era forecasters. Parameters will not always be representative of the event at hand, and the variety of AWIPS products must be used to diagnose and forecast atypical events, as was the case on 28 July 1997.
Differential reflectivity (ZDR) and differential propagation (DP) are radar tools that sense the size and shape of hydrometeors. These tools can be very useful with atypical rain events and for distinguishing between hail and rain. These are not currently implemented within the WSR-88D system, but they are available for research purposes from the CSU radar here. The CSU radar data show promise for better rainfall estimation with radar polarization tools.
Figure 6. Radar-derived total accumulation (in.) as of 0500 UTC 29 July 1997 with rivers, reservoirs (Horsetooth Reservoir left-center), and roads background.
Despite the city's planning efforts, the brave rescues by the fire authority personnel, and the National Weather Service warnings, it is natural to question the "reason" for loss of lives. All five fatalities occurred in a very small area, and none in cars, showing the unusual circumstances surrounding this unfortunate occurrence. Like many Front Range communities, Fort Collins in recent years dedicated a great deal of resources to flood-plain management, particularly for Spring Creek. The effort included moving residences out of the floodway and designing retention pond areas along the flood plain that could hold excess water during a major rainfall event. The casualty count and residential damage along Spring Creek would very likely have been much greater if these flood control measures had not been implemented. Flattened trees and damaged roads along the floodway painted an inevitable prognosis of much greater tragedy. Creekside Park, a devastated greenway park along Spring Creek between the railroad and U.S. Route 287 (College Avenue, Figure 2), is testimony to the importance of flood-plain management because this area had once been a trailer park.
What Went Wrong? – Unfortunately, an adjacent trailer park area had not been moved because it was not located directly in the designated 100-year floodway. Four of the five deaths occurred here. (This area is depicted by the red "x" between the railroad and Highway 287 in Figure 2.) What happened that made this site such a disaster area? The railroad embankment along the west side of this trailer park served as the eastern barrier for one of the city's planned water retention areas. A small culvert in this embankment allowed only a small amount of water through. The rest of the flow, including any floodwaters from Spring Creek, would be directed south through a large opening under the railroad and through Creekside Park to another large opening under College Avenue. This planned redirection of Spring Creek can be seen in Figure 2 as a jog to the south just west (left) of the railroad. The magnitude of this rainfall event filled the water retention area to capacity and caused the water to over-flow the railroad embankment into the trailer park around 0430 UTC (10:30 PM local). Then the culvert blew out with tremendous force destroying at least one trailer and flooding the trailer park. As a result, the water level behind the embankment dropped briefly, thus ending the overflow and allowing a train to move along the tracks. Then debris began to clog the blown-out culvert just as excess floodwaters from an adjacent drainage surged into the water retention area. Shortly before 0500 UTC (11:00 PM) the water surged over the railroad embankment again, derailing the freight train and unleashing a massive flood surge through the trailer park, down through Creekside Park and Spring Creek, then under College Avenue. As trail-ers, vehicles, and debris got swept along, fires erupted from severed gas lines (Figure 1). Then the partially blocked opening underneath College Avenue caused a damming of the floodwaters, which later surged over the roadway. The fifth flood fatality occurred just downstream of the surge over College Avenue.
Although the low-centroid, warm rain processes presented an unusual challenge for the local WSR-88D adaptation parameters, adjusting the threshold to match the Fort Collins storm characteristics would invoke its own drawbacks. Other storms of the day had fewer tropical characteristics and may have been poorly represented by the more tropical settings. As mentioned earlier, these types of situations present one of the biggest challenges in the AWIPS era because they demonstrate the difficult task facing forecasters — the need to recognize and react quickly to atypical events. A positive aspect from the forecasters' point of view of this experience is that, even with extensive activity throughout eastern Colorado and the subtle storm evolution, timely watches and warnings were issued.
Some serious questions remain about the dissemination of these warnings to those who needed them. These issues are being addressed by the FSL Local Data Acquisition and Dissemination (LDAD) project (November 1997 issue of the FSL Forum). The LDAD systems automatically acquire local data, quality control datasets before integration into AWIPS, and disseminate critical weather information (using the latest visualization and integration techniques) to emergency response officials.
Finally, there is the issue of flood control. The flood control measures taken in Fort Collins certainly prevented greater disastrous consequences from this historic rainfall. City rescue teams should be commended, especially the Poudre Fire Authority personnel for reducing the casualty count by placing their own lives at risk in the rescue of scores of people. The fatal flood surge and extensive damage caused by the failed floodwater retention serve to remind us that flash flood danger cannot be completely eliminated. Effective warning and dissemination must complement well-planned flood control.
(Matthew Kelsch is a scientist in the Local Analysis and Prediction Branch, headed by Dr. John A. McGinley. Mr. Kelsch can be reached here.)
Web design and html by: Will von Dauster