Introduction

The rainstorm on 17–18 July 1996 in northern Illinois produced three rainfall records and was the second most costly weather disaster in Illinois behind the 1993 flood (Changnon 1998). The 43.0-cm total storm rainfall at Aurora, Illinois, was the greatest point rainfall recorded for storm durations of 24 hours or less in this century in Illinois and other Midwestern states. Only a 46.2-cm storm in Missouri exceeded this amount, according to records at the National Climatic Data Center (NCDC 1997). The 27.9-cm storm rainfall recorded in southwestern Chicago was the heaviest 24-h amount ever recorded in that city based on data from recording rain gauge networks that date back to the late 1940s. The July 1996 storm also produced the heaviest 24-h mean rainfall recorded in Illinois over areas of 5200 and 13 000 km² immediately surrounding the storm center. Changnon and Kunkel (1999) define the areal extent of the total storm.

The 17–18 July 1996 storm resulted primarily from two distinct mesoscale precipitation systems centered in approximately the same region. The first system occurred on 17 July in mid- to late afternoon at the time of maximum daytime heating. The second system occurred in the early morning hours of 18 July. Changnon and Kunkel (1999) present a more complete description of the synoptic features of this event. Changnon (1999) describes the storm impacts and responses.

The Illinois State Water Survey in Champaign, Illinois, received many inquiries regarding the magnitude and timing of this extraordinary storm. Immediately after the storm, the concern was to define the magnitude and areal extent of the storm and to try to put it into historical context. Record rainfall from this storm fell over a wide area and exceeded the designed capabilities of many water-handling structures, such as retention ponds, which led to widespread flooding. Once it became obvious that this event was beyond current design criteria, the question was raised as to whether the rainfall frequency estimates needed to be changed to take this storm into account. Finally, engineers requested information on the spatial and temporal characteristics of this storm for the evaluation and calibration of hydrological models. This paper reviews the significance of the 1996 storm in relation to previous storms in Illinois, the storm’s general spatial and temporal characteristics, and its impact on design criteria.

Spatial distribution of the storm

How does the spatial distribution of the 1996 storm compare with that of other severe storm events in Illinois? One tool for summarizing the spatial distribution of storms is the area–depth curve, which provides a measure of average rainfall over the entire storm area, the maximum point rainfall, and the average rainfall at any selected subarea about the storm center. Area–depth curves have been used to assess heavy rainstorms that produce floods (Huff et al. 1958). The relation was closely approximated by a regression curve in which rainfall depth was related to the cube root of the area (Fig. 1). This transformation has most frequently provided the best fit for such large, severe rainfall events in Illinois. Figure 2 shows the assessment of the July 1996 storm’s maximum 24-h area–depth curve and those for four other rainstorms that are among the most severe rain events ever observed in Illinois. Each storm has been subjected to detailed analysis as part of the Water Survey’s research program in hydrometeorology (Huff et al. 1958; Huff and Changnon 1961). Curves were constructed using rainfall values over the 5200 km² area immediately surrounding the storm centers.

Based on Fig. 1, the July 1996 rainstorm had an estimated maximum 20 km² average rainfall of approximately 44.2 cm at the storm center. The maximum point rainfall, measured at a rain gauge, was 43.0 cm at Aurora. The derived maximum value almost always exceeds the maximum recorded point rainfall because the measured point rainfall rarely occurs at the exact center of the storm. Therefore, the curve-derived maximum may be considered a better estimate of the actual point maximum. The mean rainfall around the storm center gradually decreased to 42.9 cm at 130 km² and to 13.2 cm at 26 000 km².

Storms providing the two heaviest 24-h point amounts ever recorded in Illinois were those on 17–18 July 1996 and on 14 June 1957. Steepest rainfall gradients for these storms were around the storm centers, as illustrated by their slopes (Fig. 1). Although the other three storms are among the most severe rain events in this century, their rainfall gradients are much smaller and more typical of other large-area severe rain events.

Additional information on the magnitude of the July 1996 storm, with respect to other extreme rain events in Illinois, was obtained from a review of area–depth relations for more than 400 storms of various durations in which the 100-yr frequency of point rainfall was equaled or exceeded (Huff 1993). All events that met the above criteria for storms of 24-h or less duration were selected, and their area–depth curves were compared with that for the July 1996 storm. Because of a major interest in small basin hydrology, this previous study of 400 storms involved spatial distributions (area–depth relations) over areas up to 5200 km² surrounding the centers of the maximum storm rainfall. Table 1 shows the 10 storms with the heaviest 24-h rainfalls over areas of 5200 km². The July 1996 storm had the greatest mean rainfall among the 10 storms, followed closely by storms in 1910 and 1946. Six of the top 10 storms occurred since 1950, when such extreme events became the subject of field surveys conducted to improve the data samples and to provide detailed analysis of the rainfall distribution. The top three storms also ranked as the top three storms based on the mean rainfall for the 13 000 km² surrounding the storm centers.

Table 2 provides additional information about the magnitude of the July 1996 storm using the estimated 26 km² rainfall obtained from area–depth analyses for the top 10 storms. Thus, Table 2 provides a measure of the rain intensity at or near the storm center. Nine of the 10 storms are the same as those in Table 1, but the order is different. The July 1996 storm ranked first again in Table 2. Although the top four storms are the same in both tables, their order is different as well.

These statistics show that the three major, 24-h storms of 1996, 1946, and 1910 were the most severe storms experienced in Illinois since the start of this century with respect to both intensity and areal extent. Of the three storms, the July 1996 storm ranked first in intensity for both the 5200 and the 13 000 km² area immediately surrounding the storm center.

Based on the most current analysis of rainfall frequency information for northeastern Illinois (Huff and Angel 1989), and on the spatial distribution of rainfall in the 1996 storm, an area of approximately 12 000 km² experienced 24-h point rainfall amounts that exceeded those expected to occur, on the average, once in 10 years. Similarly, the 25-, 50-, and 100-yr frequency values of 14.0, 16.4, and 19.3 cm, respectively, were exceeded over areas of 6730, 4920, and 3500 km², respectively. These statistics further illustrate the magnitude of this storm.

Temporal distribution of the storm

Temporal characteristics of the July 1996 storm were evaluated using hourly rainfall amounts from recording rain gauges. Data from six recording rain gauges along a northwest–southeast line, extending from Sycamore to southeastern Cook County were analyzed. This line was only a short distance north of the main axis of the total rainfall pattern (Fig. 3).

Figure 4a shows the time distribution of the rainfall for the Sycamore gauge, which is located 75 km northwest of the storm center at Aurora. Five easily discernible rain bursts occurred. The heaviest convective burst (0200–0300 LST on 18 July) reached a maximum intensity of 5.0 cm h⁻¹. Total storm rainfall at Sycamore was 24.2 cm, with an hourly maximum of 5.0 cm, representing 21% of the storm total. During the 3-h period (0000–0300 LST on July 18) 53% of the 24-h total rainfall was recorded. Nearly 78% of the total storm rainfall occurred in the major nocturnal rain period. The rain gauge at St. Charles in Kane County had a time distribution similar to that at Sycamore. The maximum hourly amount of 4.0 cm occurred from 0100 to 0200 LST on 18 July; the storm consisted of four bursts; and the nocturnal storm period produced 62% of the total rainfall of 17.4 cm.

Rain at gauge 31 showed a distinctly different distribution than that at the Sycamore, De Kalb, and St. Charles gauges. Time distribution curves at gauge 31 (Fig. 4b) indicate two nearly equivalent major rain periods. However, the nocturnal period continued to dominate in the production of rainfall and accounted for 64% of the total storm rainfall. Gauge 31 is located 13 km east of Aurora and is the closest recording gauge to the maximum point rainfall at Aurora. Assuming that the rainfall distribution from gauge 31 would most closely correspond to that at the nonrecording rain gauge at Aurora, the extreme value at Aurora may have been caused by the larger rainfall contribution by the earlier daytime storm period, and is compared with amounts in most of the immediately surrounding storm area.

At gauge 31, 54% of the 24-h total storm rainfall occurred in the 3-h period from 0100 to 0400 LST on 18 July. Adding the 17% from 1500 to 1600 LST in the earlier daytime storm, these four hours (17% of the total rain period) accounted for 73% of the total storm rainfall.

Nearly equivalent daytime and nocturnal storm intensities were found at gauge 15 in southwestern Cook County, where a rainfall total of 27.9 cm fell (Fig. 4c). The daytime maximum intensity of 5.5 cm h⁻¹ exceeded the nocturnal maximum, but the total nocturnal storm contributed 52% of the total storm rainfall.

Farther east at gauge 24, the distribution was very similar to that at gauge 15. However, gauge 25 in extreme southeastern Cook County reverted to the Sycamore type of distribution. The nocturnal maximum intensity well exceeded the daytime maximum, and the nocturnal period contributed 68% of the storm total.

Overall, the nocturnal storm was the greatest producer of rainfall in the heaviest part of the total storm rainfall pattern. For the most part, the nocturnal storms also contributed the heaviest hourly rainfall intensities in the storm system. Typical of extreme convective rainfall events, a major portion of the rainfall occurred during a relatively small portion of the total storm duration and at night in Illinois (Huff 1979).

Maximum 3- and 6-h rainfall distributions

Combining the data of 91 available recording rain gauges, and the 165 total rainfall measurements throughout the storm area, isohyetal maps and area–depth relations were derived for the maximum 3- and 6-h rainfall amounts. This information is valuable in analyzing and evaluating the hydrologic consequences of this unusually severe flood-producing event.

The rainstorm was divided into two distinct rain periods. The first period was in the mid- to late afternoon of 17 July, and the second period peaked in the early morning hours of 18 July. The maximum 3- and 6-h amounts for the two rain periods were based on the maximum 3- and 6-h amounts at each station. Therefore, the 3- and 6-h time periods when the maximum rainfall occurred varied only slightly between stations. Differences in the timing of the maximum 3- and 6-h rainfall were most noticeable during the first period, with starting times ranging from 1100 to 1600 LST on 17 July. However, for a majority of the rain gauges on 17 July, the time of the 3-h maximum for the first rain period was 1300–1600 LST, and the time of the 6-h maximum was 1200–1800 LST. For the second rain period on the morning of 18 July, the starting times varied by only one or two hours between gauges. For most stations, the time of the 3-h maximum was 0100–0400 LST, and the time of the 6-h maximum was 0000–0600 LST.

Figures 5 and 6 show the isohyetal patterns within the heavier portions of the storm for the 6-h periods of both the afternoon and evening storms. Greater 3- and 6-h rainfall totals occurred during the nocturnal storm period. The nocturnal rainfall maximum exhibited a general northwest–southeast orientation, while the daytime storm period had a general west–east orientation.

Comparison of the July 1996 storm with the expected frequency of extreme rain events

Design engineers use expected rainfall amounts at select return periods to determine the design of water-handling structures such as dams, storm drains, and retention ponds. Most of the design considerations are based on return periods of 100 years or less. A concern raised by a major rainfall event, such as the July 1996 storm, is the potential impact on the estimates of the expected rainfall amount.

As shown in Fig. 6, the 43.0-cm amount is well above any previously recorded amount in the 97-yr precipitation record at Aurora. If the region’s rainfall frequencies (Huff and Angel 1989) are extended to 200 years and beyond for both Aurora and northeastern Illinois, and the 43.0-cm storm is fit to the line, the storm appears to exceed the 1000-yr, 24-h storm event. However, this approach assumes that the statistical distribution fit to the frequencies out to 100 years would also be appropriate out to 1000 years. This assumption cannot be validated because of our limited historical records; however, the result shows that the storm far exceeds what is normally expected during any 100-yr period. Therefore, the 43.0-cm amount would be considered a true statistical outlier.

Usually, for a station such as Aurora with 97 years of records, an outlier such as the 1996 value will not have a large influence on the resulting distribution of expected extreme rainfall because of the way that the rainfall frequency values are. The typical procedure is to find the highest rainfall for each year of the record. This annual maximum time series is then fit to a statistical distribution using various fitting techniques (e.g., method of moments, maximum likelihood, or L-moments methods). The resulting fitted statistical distribution is used to calculate the expected rainfall amounts at the return periods of interest. One result of the fitting process is to reduce the effect of outliers. Huff and Angel (1989) provide a more detailed explanation of the factors involved in determining the rainfall frequency values.

According to Huff and Angel (1989), the expected 100-yr, 24-h storm value for Aurora is 21.3 cm. Using the generalized extreme value distribution and both the maximum likelihood and L-moments methods of fitting the distribution (Huff and Angel 1992), inclusion of the 43.0 cm increased the estimated 100-yr, 24-h storm value to 25.2 cm (maximum likelihood method) and to 28.4 cm (L-moments method), as shown in Fig. 7. This sizeable increase would have a significant impact on the design of water-handling structures. However, the use of regional averages to reduce the variability associated with extreme values at any particular station is recommended (Huff and Angel 1989). The regional average 100-yr, 24-h storm value for northeastern Illinois is 19.3 cm (Huff and Angel 1989). Eight stations in northeastern Illinois were used to develop this regional average. Except for Aurora, the expected rainfall for the 100-yr, 24-h storm at the other seven stations has not changed appreciably with the addition of data through 1996, including the July 1996 storm. Inclusion of the July 1996 storm for Aurora increases the regional average of the estimated 100-yr, 24-h storm to approximately 19.7 cm (maximum likelihood method) and 20.1 cm (L-moments method). These changes correspond to increases of only 2%–4% at the 100-yr return period. Thus, the use of fitted statistical distribution and regional averages diminishes the influence of this large outlier storm on the rainfall frequency estimates. Therefore, the rainfall value of 43.0 cm for Aurora would not significantly change the expected regional average rainfall amounts found in Huff and Angel (1989).

Summary

The rainstorm on 17–18 July 1996 in northern Illinois produced three rainfall records. The 43.0-cm total storm rainfall at Aurora was the maximum point rainfall recorded for storm durations of 24 hours or less in this century in Illinois. The 27.9-cm storm rainfall recorded in southwestern Chicago was the heaviest 24-h amount ever recorded in that city by recording rain gauge networks dating back to the late 1940s. The July 1996 storm also produced the heaviest 24-h mean rainfall recorded in Illinois over areas of 5200 and 13 000 km² immediately surrounding a storm center since the start of the century. Statistics that further illustrate the magnitude of this storm are that the 100-, 50-, and 25-yr recurrence interval of 24-h point rainfall were respectively exceeded at 3500 km², 4900 km², and 6700 km².

The July 1996 rainstorm resulted from two distinct mesoscale precipitation systems located over approximately the same region. The first system occurred on 17 July in mid- to late afternoon, and the second system occurred on 18 July in the early morning hours, a time when any previous rainstorms have been most intense. The nocturnal storm was the major rain producer over most of the storm area. However, the afternoon storm was heavier in the vicinity of the point rainfall record at Aurora than over the rest of the storm area.

As in most heavy rainstorms, a large portion of the July 1996 rainfall occurred during a very small percentage of the total storm duration. The closest recording gauge to the storm center at Aurora had 73% of the total storm rainfall over a 17% time frame of the storm’s duration.

One concern resulting from a major rainfall event such as this storm is the impact on the estimate of the expected rainfall amounts at selected return periods used to design water-handling structures. Based on data from Huff and Angel (1989), the Aurora rainfall appears to exceed even the 1000-yr return period. However, the use of a statistical distribution and regional averages minimized the effect of this one storm on the frequency distribution. In the final analysis, the 100-yr, 24-h amount increased by only 2%–4%.

While the focus of this study is on Illinois, the results may be applicable to surrounding states and other regions with similar heavy rainfall climatology. Besides the record rainfall at Aurora, Illinois, the timing and spatial distribution of the large rainfall event over 24 hours in a well-developed area of Illinois can be compared with the impacts (Changnon 1999) to show the vulnerability of urban areas to such events.