Spectrum of storm event hydrologic response in urban watersheds



[1] We seek to improve scientific understanding of urban storm event hydrologic response through analyses of rainfall and discharge data for the Baltimore metropolitan region. High-resolution data, 1 km2 and 15 min radar rainfall and 1 to 5 min discharge, provide the detail necessary to accurately characterize storm event hydrologic response in small urban basins. We examine flood-producing rainfall properties and storm event hydrologic response for nine small watersheds in the Baltimore region including seven urbanized basins, a forested basin, and an agricultural basin. We find expected contrasts in flood peak distributions and storm event runoff production between the urban and nonurban watersheds, but we also find a spectrum of storm event hydrologic response among the urban watersheds. Moores Run and Dead Run are end-members of this urban spectrum, with Moores Run producing anomalously large flood peak magnitudes and Dead Run producing anomalously large storm event runoff ratios. Analyses show that runoff production and timing of hydrologic response are linked to stormwater management infrastructure and play a central role in the spectrum of storm event response. Detention basins in these watersheds appear to operate as intended by stormwater legislation to lower peak discharges but not runoff volumes. Antecedent moisture does not appear to significantly impact storm event hydrologic response in the urban or nonurban basins. The rainfall climatology of flood-producing storms varies from urban to nonurban watersheds with urban watershed flood frequency displaying a pronounced warm season maximum, highlighting the central role of warm season thunderstorm systems for urban flooding in Baltimore.

1. Introduction

[2] We examine the spectrum of storm event hydrologic response in urban watersheds through analyses of rainfall and discharge data for nine drainage basins in the Baltimore, Maryland metropolitan region. We will show that there are significant gaps in our understanding of storm event response in urban watersheds, especially in light of the heterogeneous mix of stormwater management infrastructure associated with evolving stormwater control approaches from storm drains to detention basins. An improved scientific underpinning of urban hydrologic response to storm events is essential for hydrologic design, flood hazard assessments, and flash flood forecasting in urban areas.

[3] We focus on nine watersheds in the Baltimore metropolitan area with drainage areas ranging from 3.8 to 14.3 km2, and base our analyses on a sample of flood events that include the 50 largest flood peaks for each basin during the 2000–2009 time period. We utilize instantaneous discharge data from U.S. Geological Survey (USGS) stream gauging stations. The approximately 10 km 2 drainage area of the nine watersheds reflects the scale for which we can accurately resolve the temporal and spatial variability of storm event rainfall [Smith et al., 2012]. Rainfall analyses are based on a 10 year (2000–2009) high-resolution (1 km 2 horizontal resolution and 15 min time resolution) radar rainfall data set [Smith et al., 2012]. The availability of high-resolution rainfall fields allow for more accurate characterization of storm event hydrologic response in small urban watersheds than was possible in previous studies.

[4] In this study, we focus on the aggregate hydrologic response associated with the distribution of impervious surfaces, stormwater management infrastructure, and alteration of the drainage network [see e.g., Smith et al., 2002; Beighley and Moglen, 2003; Smith et al., 2005a, 2005b; Sharif et al., 2006; Sheng and Wilson, 2009; Meierdiericks et al., 2010b]. Land use in the nine drainage basins ranges in development from Baisman Run, which is a forested reference watershed, to the intensely urbanized Moores Run, which has the highest frequency of occurrence of flood peaks greater than 1 m 3 s −1 km −2 in the contiguous United States [Smith et al., 2005b].

[5] The basins in this study also represent varying density of stormwater management infrastructure. Maryland passed its first stormwater regulations to deal with increased storm runoff from urbanizing areas in 1982. The Stormwater Management Act called for local ordinances on stormwater management to be implemented by 1984 [State of Maryland, 1982; Schueler and Claytor, 2000] and focused primarily on “peak management,” or reduction of peak stream discharges from flood events, with the intent that this would also improve stream quality. Later revisions to the act, passed in 2000, increased focus on water quality and more frequent flood events, resulting in a new Stormwater Design Manual [Schueler and Claytor, 2000]. Within the Baltimore area, the increase in urban land use since 1984 is roughly proportional to the amount of stormwater management infrastructure within a basin [Meierdiercks et al., 2010a]. In order to continue to improve such regulations, it is important to improve scientific understanding of storm event hydrologic response.

[6] Many of the study watersheds are part of the Baltimore Ecosystem Study (BES) [Pickett and Cadenasso, 2006], which is the first urban National Science Foundation Long-Term Ecological Research Site. Previous research on hydrologic response in the Baltimore metropolitan area has pointed to the importance of impervious land cover in increasing runoff ratios in Dead Run, the importance of stormwater management infrastructure in decreasing runoff ratios in Upper Gwynns Falls, and the relationship between these runoff ratios and peak discharge in the basins [Smith et al., 2005a; Meierdiericks et al., 2010b]. Runoff ratios are defined as the ratio of discharge that left the basin to rainfall that fell on the basin. Moores Run, which has extraordinary flood peak response due to the efficient concentration of water in storm drains, has a low runoff ratio for high peak discharge producing storm events [Smith et al., 2005b]. Studies of the Dead Run basin suggest that the presence of modern stormwater detention may affect peak discharges on the same scale that impervious surfaces do [Meierdiercks et al., 2010a] and that antecedent moisture may affect the hydrologic storm response even with high impervious percentages in the basin [Smith et al., 2005a]. Previous research on flood-producing rainfall in the Baltimore metropolitan area has shown that flash flooding in the region is primarily caused by thunderstorm systems that initiate along the Blue Ridge Mountains, propagate eastward across the Piedmont, and interact with urban areas, in Baltimore and Washington D. C., and land-sea boundaries of the Chesapeake Bay and Atlantic Oceans [Ntelekos et al., 2007, 2008]. These convective thunderstorms in the Baltimore area have a strong diurnal cycle with maximum lighting strikes between 2100 and 2200 UTC, or 4 P.M. and 5 P.M. EST [Ntelekos et al., 2007].

[7] Previous research on urban storm event hydrologic response has focused on quantifying the urban extent of basins and creating predictive equations for the n-year discharge of urban basins. Early efforts focused on the relationship between impervious cover and increased peak discharges with decreased time to peak [e.g., Carter, 1961; Leopold, 1968; Martens, 1968; Anderson, 1970; Liscum and Massey, 1980; Robbins and Pope, 1996]. The relationship between impervious surface and peak discharge, however, is not simple. The connectedness of impervious surface is important [Shuster et al., 2005], and impervious area may be difficult to quantify [Moglen and Kim, 2007]. Other important urban land surface modifications found to impact hydrologic response include changes in channel and floodplain geometry [Leopold, 1973], increases in drainage density through the storm drain network [Graf, 1977], modifications of hydraulic properties of urban soils [Gregory et al., 2006; Pouyat et al., 2010], and implementation of stormwater management control structures.

[8] There have been attempts to create predictive equations for discharge with measures such as an urbanization factor for channel improvement and vegetation [Espey and Winslow, 1974] or an urbanization index for extent of street curbs, sewers, and channel modifications [Land et al., 1982]. Additional factors in urban basin response include basin shape [Veenhuis and Gannett, 1986] and long-term basin storage [Sauer et al., 1983]. A ranking of the importance of different measures of urban development found the increase in high-density urban development to be most important followed by the spatial development pattern, the total increase in urbanization, and finally the decrease in forested land use [Beighley and Moglen, 2003]. Development far away from the outlet was found to result in larger peak discharges. Despite these efforts to quantitatively define urbanization and its effects, it is difficult to attribute specific changes in time to peak and peak discharge to specific changes in land use [Crippen, 1965; Wright et al., 2012], and the spatial variation of these land use changes can be important to changes in hydrologic response [Riggs, 1965]. Additionally, historical changes in peak discharge values may be due, in part, to changes in rainfall patterns [Olivera and DeFedd, 2007]. Our study focuses specifically on the spectrum of hydrologic response in this collection of urban basins, with the understanding that no single value can quantify the urbanization of a basin or predict its response.

[9] We build upon the previous studies by addressing the following questions:

[10] (1) How does urbanization alter storm event hydrologic response? Specifically, how do response time scales and runoff generation vary between urban and nonurban basins and how do they vary within the urban basins?

[11] (2) How does stormwater management infrastructure affect storm event hydrologic response?

[12] (3) How does antecedent moisture affect urban flood response?

[13] (4) Do urban flood occurrences cluster in time?

[14] (5) How does the rainfall climatology of flood-producing storms vary between urban and nonurban basins?

[15] Contents of the sections are as follows. We describe the nine watersheds in section 2 and introduce the principal data sets used in the study. Results focusing on analyses from the 10 year discharge and rainfall data for the nine basins are presented in section 3. We present a summary and conclusions in section 4.

2. Methods

2.1. Study Area

[16] We focus on nine similarly sized watersheds in the Baltimore metropolitan area (Figure 1 and Table 1): Baisman Run (3.8 km 2), Minebank Run (5.3 km 2), Whitemarsh Run (7.1 km 2), Herring Run (5.5 km 2), Moores Run (9.1 km 2), Cranberry Branch (10.7 km 2), Herbert Run (6.4 km 2), Upper Gwynns Falls (11.0 km 2), and Dead Run (14.3 km 2). Each basin has a USGS stream gauging station at its outlet (see Table 1 for USGS ID numbers). The similar size of the basins allows for comparison of hydrologic response between basins without a considerable influence from basin scale. There is little variation in the range of average basin slopes (Table 1), with the steepest slopes in the two nonurban basins, Baisman Run and Cranberry Branch. The basins are located in the Piedmont physiographic province and elevations within the basins range from 18 to 308 m above mean sea level (Figure 1; elevations from USGS Digital Elevation Models [Gesch et al., 2002]).

Figure 1.

Overview image of the Baltimore study region. (a) Elevation in meters above average sea level and (b–d) 2006 land use land cover for individual study basins are shown. In Figure 1a, Baltimore city limits are outlined in yellow and the Chesapeake Bay is colored with blue. The location of Figure 1a is highlighted with the red box in Figure 1e.

Table 1. Summary of Watersheds Characteristics
 Percent Land Use (2006)  
Drainage BasinArea (km2)Average Slope (%)UrbanOpen SpaceAgricultureForestIncrease in Urban AreaPercent Impervious Area
No.NameUSGS IDHighMediumLow1984–20062001
1Baisman Run015835803.814.
2Minebank Run01583979675.311.724.813.77.419.92.330.10.920.4
3Whitemarsh Run015850907.
4Herring Run015852005.57.734.116.58.622.80.016.5−0.124.6
5Moores Run015852309.15.442.634.−0.233.4
6Cranberry Branch015855008.510.
7Herbert Run015891006.47.030.535.210.913.
8Upper Gwynns Falls0158919711.06.533.010.51.930.80.522.312.714.6
9Dead Run0158933014.35.545.620.39.818.

[17] Basin land use, for 1984 and 2006, is based on the USGS Chesapeake Bay Watershed Land Cover Data Series [U.S. Geological Survey (USGS), 2006]. Urban land use is defined as low intensity (20%–49% impervious), medium intensity (50%–79% impervious), or high intensity (80%–100% impervious). Impervious percentages are from the 2001 USGS National Land Cover Dataset [Homer et al., 2004].

[18] Two basins with little urban development, Baisman Run and Cranberry Branch, are included in the study for comparison to the urban basins. Baisman Run is the forested “control watershed” for the BES [Pickett and Cadenasso, 2006] with 75.8% forested land use and only 1.5% urban land use resulting in a 0.2% impervious area coverage. Cranberry Branch is a primarily agricultural watershed with 55.1% agricultural land use and 2.9% urban land use resulting in a 0.6% impervious area coverage.

[19] The remaining seven basins have significant urban development. Minebank Run is the second least developed of the urban basins. Open space and forested space are primarily concentrated around the downstream portion of the channel in Minebank Run. Whitemarsh Run and Herbert Run have predominantly medium and high intensity urban land use (over 30% of each) with some low intensity urban land use, while Herring Run and Upper Gwynns Falls have less medium intensity urban land use (under 17%). Herbert Run has the highest percentage of impervious surface by area at 35.6%.

[20] Development in these urban basins occurred at different times. Whitemarsh Run had a 7.4% increase in urban area between 1984 and 2006 and Upper Gwynns Falls had a 12.7% increase. Herring Run and Moores Run had essentially no urban development between 1984 and 2006 (Table 1). This distinction is important due to the Maryland Stormwater Management Act, which required municipalities to adopt storm water management programs to reduce peak flows by 1984 [Comstock and Wallis, 2003]. Because a significant portion of Upper Gwynns Falls was developed after 1984, it has a high density of stormwater detention basins [Meierdiercks et al., 2010a]. Upper Gwynns Falls has a lower percentage of impervious cover than the other urban basins (14.6% versus more than 20% for other urban basins), but the urban land use in Upper Gwynns Falls is nearly identical to the percentage of urban land use in Minebank Run. This similarity between Upper Gwynns Falls and Minebank Run land use will be important when comparing differences between the storm event hydrologic response of Upper Gwynns Falls and that of the other urban basins.

[21] All of the basins included in this study represent the spectrum of hydrologic response, but three urban basins stand out in their land use and hydrologic response. Upper Gwynns Falls, mentioned above, is the basin with extensive stormwater detention infrastructure. Moores Run and Dead Run represent end-members of the urban hydrologic response spectrum, as will be detailed in section 3. Moores Run is the most highly urbanized basin in the study region at 82% urban and has the highest frequency of flood peaks that exceed a discharge of 1 m 3 s −1 km −2 in the contiguous United States [Smith et al., 2005b]. Moores Run contains mostly dense residential land use and was developed more than 100 years ago, long before stormwater management regulations were implemented in Maryland. The basin has no significant detention structures and stormwater is quickly transported through the watershed via storm drains and surface channels [Smith et al., 2005b].

[22] Dead Run is the third most urbanized basin in the study. Intensive development in Dead Run began in the late 1950s [Nelson et al., 2006] and includes residential and commercial areas with modern stormwater management infrastructure (i.e. detention basins) to the north and west and residential areas without modern stormwater management infrastructure to the south and east [Meierdiercks et al., 2010a]. The Baltimore Beltway, built in the early 1960s, runs through Dead Run [Nelson et al., 2006 and creates a large, connected impervious area.

2.2. Data

[23] Discharge data were obtained from the USGS Instantaneous Data Archive (IDA, http://ida.water.usgs.gov/ida) for the 10 year period from January 2000 to December 2009. Discharge data were available in 1 or 5 min time increments for each of the nine watersheds. All discharge data were interpolated to 1 min time resolution.

[24] The IDA data required some minor corrections. In the case of Minebank Run, the gauge was moved upstream in October of 2001, reducing the drainage area from 7.5 km 2 to 5.3 km 2. The data from January 2000 to October 2001 were adjusted by decreasing the discharge proportionally to the decreased drainage area. The rating curve for Moores Run has been corrected multiple times by the USGS and the adjusted rating curves are not reflected in the IDA discharge data. In order to update the IDA discharge data, stage was reconstructed from the original discharge data using the original rating curves. Then the updated discharge was calculated from the stage using the updated rating curve. The impact of this update was generally small, but did decrease discharge values for the largest flood events.

[25] High-resolution (1 km 2 and 15 min) rainfall fields were developed from Sterling, Virginia Weather Surveillance Radar 1988 Doppler (WSR-88D) radar reflectivity fields using the Hydro-NEXRAD system [Krajewski et al., 2011]. Rainfall fields were bias corrected on a daily basis using a network of rain gauges in the Baltimore metropolitan area [Smith et al., 2012]. Basin-averaged rainfall time series were derived at 15 min time intervals for the nine drainage basins from the 1 km 2 rainfall fields. Partial coverage of the basin by a 1 km 2 grid is reflected by using a fraction of the grid cell rainfall in the basin-averaged rainfall rate computations.

[26] The 50 largest peak discharge events in the IDA data over the 10 year period were identified for each basin. Any peak over a specified threshold was identified as a flood event. All peaks were required to be separated by at least 6 h. The threshold was chosen so that each basin would have 50 flood events, for an average of five events per year. After the peak-over-threshold flood events were chosen, they were checked with the USGS annual peak data. In the rare cases where an annual peak greater than the threshold was not included in the peak-over-threshold events, the peak was added without timing information, and was excluded from any analyses that involve timing, while the lowest peak-over-threshold was removed. This occurred for a small number of large flood events for which the stream gauging station was damaged by the flood, but the USGS determined the peak discharge by other measures.

[27] In order to analyze rainfall and discharge for the 50 flood events in each of the nine drainage basins, water balance calculations were carried out by computing the maximum x-hour discharge (with x between 0.25 and 24) and the maximum x-hour rainfall for each event. For durations less than 6 h, the x-hour rainfall was constrained to be within 6 h of the peak discharge time. For durations longer than 6 h, the x-hour rainfall was constrained to be within 24 h of the peak discharge time. Maximum x-hour discharge values were constrained to contain the peak discharge. For water balance calculations, events with missing rainfall data were excluded. These events were identified as having any maximum x-hour rainfall less than 0.1 mm h −1; it was assumed that average rain rates less than 0.1 mm h −1 could not produce floods of the magnitudes included in the study. Approximately two to three events were excluded for any given basin.

3. Results

3.1. Peak Discharge

[28] The distributions of flood magnitudes for the peak-over-threshold flood events exhibit large variation across the nine basins (Figure 2) with a clear difference between urban and nonurban basins. Baisman Run, the forested watershed, and Cranberry Branch, the agricultural watershed, have smaller flood peaks and less variability in peaks than the other basins. The 0.75 quantiles in Baisman Run and Cranberry Branch are smaller than the 0.10 flood peak quantiles in the seven urbanized basins. Baisman Run has the least variability around its median value of 0.22 m3 s−1 km−2.

Figure 2.

Box plot of flood magnitudes. The limits of the box represent the 25th and 75th percentiles, while the line inside represents the median (50th percentile). The whiskers indicate the 10th and 90th percentiles.

[29] Flood peaks in Upper Gwynns Falls, which has extensive stormwater management infrastructure, are relatively small for an urban watershed. The median peak discharge in Upper Gwynns Falls, 0.93 m 3 s −1 km −2, is smaller than the 0.10 quantiles in the other urban basins, and Gwynns Falls' 0.75 quantile is smaller than the 0.25 quantiles in the other urban basins. A possible interpretation of these results is that modern stormwater management infrastructure, specifically detention basins, lowers peak discharges and reduces variability of flood peaks in Upper Gwynns Falls.

[30] At the other end of the urban spectrum, Moores Run has exceptionally large flood peaks. The median flood peak in Moores Run, 4.00 m3 s−1 km−2, is larger than the 0.75 flood quantiles in all other watersheds. The remaining five urban watersheds have median flood peaks ranging from 1.52 m3 s−1 km−2 in Minebank Run to 2.27 m3 s−1 km−2 in Dead Run. Dead Run has a relatively narrow distribution of flood peaks, although its maximum flood peak is the largest in the region (see discussion below). The extreme flood peaks in Moores Run are not caused by efficient runoff production (compared with Dead Run; see discussion below) but by runoff production in hydraulically connected portions of the watershed in close proximity to the storm drain and surface channel network. Early 20th century stormwater infrastructure in Moores Run rapidly transports stormwater through the storm drains and surface channels in the basin [Smith et al., 2005b].

3.2. Storm Event Hydrologic Response

[31] The spectrum of storm event hydrologic response is illustrated in Figure 3, with Baisman Run and Moores Run as end-members. Relationships between maximum three-hour rainfall rate and peak discharge for the nine basins are not strictly linear, but the robust regression lines, computed by iteratively reweighted least squares [Holland and Welsch, 1977], highlight the varying responses across basins. The 3 h time frame was used to capture an x-hour rainfall rate at which most basins showed high correlation between peak discharge and rainfall rate (see Figure 3). The lines extend over all observed values for each basin, so the line for Baisman Run ends at a much lower peak discharge value than does the line for Moores Run. The regression line for Baisman Run has a steep slope, and Baisman Run has few peak discharges greater than 0.7 m3 s−1 km−2, even for rain rates greater than 25 mm h −1. Moores Run, on the other hand, has no peak discharges below 2 m3 s−1 km−2, despite having no three-hour rain rates greater than 22 mm h−1. The peak discharge values in Moores Run increase much more rapidly with an increase in rainfall rate than do the peak discharge values in Baisman Run.

Figure 3.

Maximum 3 h rainfall rate versus peak discharge with regression lines.

[32] Other basins fall within the envelope created by Baisman Run and Moores Run. Slopes for the trend lines in the other urban basins decrease in steepness from Herbert Run, to Upper Gwynns Falls, to Minebank Run, to Whitemarsh Run, to Dead Run, to Herring Run. The regression line for Cranberry Branch is somewhat misleading as the bulk of events fall along a nearly vertical line, but a handful of events without very high rainfall rates have high peak discharges (this anomalous behavior is also reflected in the low correlation coefficients presented below). Dead Run has the largest peak discharge, corresponding to an exceptional 3 h rainfall rate over the watershed. This large discharge determines the slope of the regression line for Dead Run; without this large discharge the slope drops from 5.62 to 3.29. The variation in regression line slopes is not directly related to urban land use. Herbert Run has the highest impervious percentage, but its regression slope is in the central portion of the response spectrum. These analyses highlight the observed hydrologic response in a basin for a given rainfall and allow for comparison between the basins' hydrologic responses independently of rainfall depth. The results strongly suggest that there are basin characteristics other than impervious fraction, which significantly impacts the storm event hydrologic response.

[33] Analyses of the correlation between basin-averaged rainfall rate and peak discharge for different averaging times illustrate the variability of characteristic s for the nine basins. The Spearman correlation coefficient between maximum 3 h rainfall and peak discharge for the Baltimore area basins (Figure 4) ranges from 0.12 for Cranberry Branch to 0.66 for Dead Run. The urban watersheds generally show decreasing correlation with increasing rainfall averaging time. The decrease is monotonic for Herring Run, Whitemarsh Run, and Minebank Run, while values for Moores Run and Herbert Run increase before decreasing. The correlation coefficient for Upper Gwynns Falls is flat with values fluctuating between 0.2 and 0.3 over the time scales. Cranberry Branch is monotonically increasing, with lower correlations in the 0.5 to 3 h time scales than even Baisman Run.

Figure 4.

Spearman correlation coefficients between maximum 0.5, 1, 3, 6, 12, and 24 h rainfall and peak discharge for the nine watersheds.

[34] Correlation coefficients between maximum 1, 3, 6, and 12 h rainfall (mm) and runoff depth (mm) for the nine watersheds (Figure 5) are generally increasing with duration. Correlation coefficients at the 12 h time scale range from a minimum of 0.40 for both Cranberry Branch and Baisman Run to 0.91 for Herring Run. Cranberry Branch again has lower correlation values at short time intervals than Upper Gwynns Falls or Baisman Run. For the other six urban watersheds, the 12 h correlation coefficients have a relatively narrow range from 0.72 to 0.91. Scatterplots of maximum 12 h rainfall rate and runoff rate for Dead Run and Baisman Run (Figure 6) illustrate the striking contrasts in storm event water balance between urban and forested watersheds in the Baltimore region.

Figure 5.

Spearman correlation coefficients between maximum 1, 3, 6, and 12 h rainfall and runoff depth for the nine watersheds.

Figure 6.

Maximum 12 h rainfall rate and maximum 12 h runoff rate for Dead Run and Baisman Run.

[35] Basin response times (Table 2) were computed as the time between the rainfall centroid (centroid of the 48 h of rainfall data around the time of peak discharge) and the peak discharge and provide a more quantitative measure of the characteristic response times for the basins. In these analyses, discharge has a 1 min resolution while rainfall has a 15 min resolution; this means that for an individual storm event, response times may be constrained by the rainfall time resolution and differences of up to 15 min may be due to the resolution discrepancy. With this in mind, the response time for each storm was rounded to the nearest 15 min and then mean and median values were computed. In cases where the median value was zero, it was labeled as “less than 0.25” h. Baisman Run, Upper Gwynns Falls, and Cranberry Branch clearly have the longest median response times with values greater than 1 h. Baisman Run and Cranberry Branch have the largest mean basin slopes (Table 1), suggesting that runoff production processes have a greater impact on response times than channel slope does. Upper Gwynns Falls, which contains a dense network of stormwater detention basins, has longer median response times than the forested watershed, Baisman Run. With the exception of Upper Gwynns Falls, the urban basins are split between the extremely short response times of Minebank Run, Herring Run, and Moores Run (median values less than 15 min as explained above) and the longer response times of Dead Run, Herbert Run, and Whitemarsh Run (median values greater than 0.5 h). Response times are not purely a reflection of the small differences in basin area among these six watersheds as Moores Run is larger than Herbert Run and Whitemarsh Run, but has a shorter response time.

Table 2. Response Time Measures and Managed Drainage Areas for the Nine Basins
BasinResponse Time (h)Volume-to Peak (h)Managed Drainage Area (%)
Baisman Run2.221.258.487.200.00
Minebank Run0.47less than
Whitemarsh Run0.600.503.152.2819.52
Herring Run0.47less than 0.251.741.320.96
Moores Run0.31less than 
Cranberry Branch2.622.256.676.14 
Herbert Run0.980.503.052.4210.85
Upper Gwynns Falls2.181.755.685.3841.35
Dead Run1.700.753.572.6528.59

[36] Similar patterns are shown by the volume-to-peak ratio (Table 2), a measure of response time that depends only on discharge observations [see also Bradley and Potter, 1992]. The volume-to-peak ratio is computed as the volume under the hydrograph divided by the peak discharge, with the volume under the hydrograph defined by any discharge between the peak and the 12 h minimum discharges. While volume-to-peak ratios are larger than response times, the three distinct groups are the same as for response time. Baisman Run, Cranberry Branch, and Upper Gwynns Falls have median ratio values greater than 5.3 h, Minebank Run, Herring Run, and Moores Run have median ratio values less than 1.6 h, and Whitemarsh Run, Herbert Run, and Dead Run have ratios greater than 2.25 h. While land use is certainly important for response times in urban watersheds, the differences in response times between these urban watersheds are not determined by land use. For example, we would expect the watershed with the least urban land use to have the slowest response time, but Minebank Run has the lowest urban land usage of all six basins and the shortest response time. The urban basins with longer response times do, however, correspond to the basins that have experienced urban development since 1984 (Table 1), suggesting that this longer response time is a function of modern stormwater detention infrastructure.

[37] To understand the efficiency of runoff production in the basins, runoff ratios were computed for each of the 50 flood events as the ratio of maximum 12 h rainfall (mm) to maximum 12 h discharge (mm). The distribution of runoff ratios (Figure 7) shows Dead Run with much larger storm event runoff ratios than other watersheds. The median runoff ratio in Dead Run, 0.58, is larger than the 0.75 quantile runoff ratio in all other watersheds. The 0.25 quantile runoff ratio in Dead Run, 0.47, is larger than the median runoff ratio in all watersheds and is larger than the 0.75 quantile runoff ratio in all but Herbert Run and Whitemarsh. Baisman Run has the smallest runoff ratios of any watershed, with a median value of 0.09, and the narrowest distribution of runoff ratios. Cranberry Branch has the second smallest median runoff ratio, 0.18, but a large variability in runoff ratios. Upper Gwynns Falls does not fit with the nonurban Cranberry Branch and Baisman Run in distribution of runoff ratios. Its median runoff ratio, 0.36, is the fourth largest among the nine watersheds, and it has a large variability in runoff ratios. The presence of stormwater detention ponds generally causes lower peak discharges but not lower discharge volumes in the watershed; these results demonstrate the desired impact of the “peak management” requirements in the original Maryland Stormwater Management Act [State of Maryland, 1982; Schueler and Claytor, 2000].

Figure 7.

Box plot of flood event runoff ratios in the nine watersheds. The limits of the box represent the 25th and 75th percentiles, while the line inside represents the median (50th percentile). The whiskers indicate the 10th and 90th percentiles.

[38] The unusually high runoff ratios in Dead Run may be attributed to land cover and stormwater infrastructure. Dead Run has a large connected impervious area associated with transportation corridors including the Baltimore Beltway. Additionally, the basins with some urban development since implementation of stormwater regulations (Dead Run, Upper Gwynns Falls, Whitemarsh Run, and Herbert Run) have higher runoff ratios than the other basins. The area of the watersheds controlled by stormwater management infrastructure was measured from the 2010 Baltimore County Stormwater management Geographic Information Systems (GIS) layers, which give upstream areas to all stormwater management infrastructure in Baltimore County. Moores Run and Cranberry Branch were excluded from these analyses because they do not lie entirely within Baltimore County. The managed drainage area in Table 2 was calculated as the percentage of each watershed upstream of any detention basin or underground detention unit. The Spearman correlation coefficient between the managed drainage area and the median runoff ratio for the six urban watersheds in Baltimore County is 0.58. The Spearman correlation coefficient between the percent impervious area and the runoff ratio of these six watersheds is 0.64. Given that the percent impervious area of a watershed is a commonly used measure of urban extent and certainly has a large effect on runoff ratio, this suggests that the presence of detention basins signifies higher runoff ratios. It appears likely that by collecting stormwater to route to detention basins, the stormwater management infrastructure is actually increasing the connectivity of these basins and ensuring a higher percentage of the runoff reaches the outlet.

[39] The role of antecedent soil moisture for storm event hydrologic response in urban watersheds is poorly understood [Cronshey et al., 1975; Pitt and Lantrip, 2000; Shi et al., 2007]. We attempt to examine how impervious surface affects the role that antecedent moisture plays in runoff ratios through analyses of antecedent rainfall, computed as rainfall from 144 to 24 h before peak discharge, as a surrogate for antecedent soil moisture. Spring events (February–April) were removed from analyses under the assumption that the ground remains wet in the spring. Events with runoff ratios greater than 1.0 were also removed. There does not appear to be a clear relationship between 5 day antecedent rainfall and runoff ratio (Figure 8). Many of the watersheds, with the exception of Upper Gwynns Falls, have increasing trend lines, but these lines appear to be determined mainly by the presence of several outlying points. These points with high runoff ratios and very high antecedent rainfall are the result of clustering of flood events in sequential days as discussed below. Oddly, Moores Run shows a relatively large slope, despite the fact that it is one of the most heavily urbanized watersheds. Baisman Run, the forested watershed, showed no increase in runoff ratio with antecedent rainfall despite its lack of impervious surface. These results may be affected as much by the clustering of storm events as by the effects of antecedent soil moisture on runoff production, and 5 day antecedent rainfall may not be a good proxy for soil moisture. Additional plots were created for varying time periods of antecedent rainfall, and no clear relationship was found for any length of time.

Figure 8.

Five-day antecedent rainfall versus runoff ratio for the peak-over-threshold events in nine watersheds.

[40] Peak discharge events were chosen to have an average of five floods per year, but the number of events can vary significantly from year to year (Figure 9). The year 2003 was exceptionally wet [Smith et al., 2005a], with a median of 11 flood events and a range from 8 (Herring Run) to 12 (Upper Gwynns Falls), and 2002 was exceptionally dry with a median of two events and a range from 0 (Upper Gwynns Falls) to 4 (Minebank Run and Herbert Run). The index of dispersion, calculated as the variance divided by the mean, of annual flood counts is greater than 1 for all basins and ranges from 1.3 in Cranberry Branch to 2.6 in Whitemarsh Run. The index of dispersion for Poisson distributed occurrences is 1, and values greater than 1 suggest that clustering is an important element of the flood-occurrence process. The clustering of flood events may be due to the clustering of rainfall events or to time-varying changes in basin state (including antecedent moisture). During the 10 year period of record, there are a number of “flood episodes” in which multiple flood events occur within sequential days; notable examples include 11–13 June 2003 and 1–3 June 2006. This may be caused by both rainfall event clustering and basin characteristics such as antecedent soil moisture and reduced basin storage.

Figure 9.

Box plot of the number of flood events in each of the 10 years. The limits of the box represent the 25th and 75th percentiles, while the line inside represents the median (50th percentile). The whiskers indicate the 10th and 90th percentiles.

3.3. Flood-Producing Rainfall

[41] Storm-averaged rainfall maps for the 20 largest flood peaks with available rainfall data in Dead Run, Moores Run, Upper Gwynns Falls, and Baisman Run (Figure 10) illustrate differences in rainfall properties of flood-producing storm events between urban and nonurban basins. The rainfall maps are generated for storms which produce flooding in individual basins, so the variations in these average rainfall maps reflect the differing properties of the basins as well as basin-independent spatial heterogeneities of the rainfall.

Figure 10.

Storm-averaged rainfall maps for the top 20 flood peaks in (a) Dead Run, (b) Moores Run, (c) Upper Gwynns Falls, and (d) Baisman Run. Storm-averaged rainfall is defined as total rainfall within 12 h of the peak discharge for all storms divided by 20 storms.

[42] The rainfall maps for the urban end-member basins, Moores Run and Dead Run (Figures 10a and 10b), exhibit a southwest to northeast pattern of rainfall. Both show a local maximum to the northeast of the city, and Dead Run has an additional, larger maximum over the basin itself. This maximum to the northeast of the city has been observed in other studies of rainfall in the Baltimore region [see Ntelekos et al., 2007, 2008; Smith et al., 2012], and the “downwind” (northeastern) maximum in rainfall has been examined for a number of urban settings in the United States [Shepherd, 2005]. Rainfall maxima for the urban end-member basins are more concentrated than maxima for Upper Gwynns Falls and Baisman Run (Figures 10c and 10d); maps for Dead Run and Moores Run show a smaller area of maximum rainfall and higher spatial gradients surrounding the maximum.

[43] Measurements of the Spearman correlation coefficient between the rainfall at the basin center and the rainfall at a certain distance upwind or downwind of the basin are shown in Figure 11 for the storm events depicted in the average rainfall maps. These plots show that the two urban end-member basins, Moores Run and Dead Run, have correlation coefficients that decrease more rapidly with distance from the basin, suggesting that the rainfall depths decrease more rapidly with distance for these urban end-member basins. Moores Run does decrease more slowly than Dead Run in the positive, downwind direction due to the rainfall maximum to the northeast of the city. The correlation plots quantify the concentration of rainfall in the urban end-member basins across the population of storms and show that the storm-averaged map characteristics are not the properties of a few dominating large storms but common properties across the population of storms. The mean rainfall maximum over Moores Run in Figure 10, of approximately 42 mm, is smaller than the mean rainfall maxima for Dead Run, Upper Gwynns, and Baisman Run, which all exceed 50 mm. Storm-averaged rainfall for the nonurban Baisman Run has the largest area with accumulations exceeding 50 mm and relatively smooth gradients over the region. The averaged rainfall for Upper Gwynns Falls also exceeds 50 mm and has spatial gradients that fall between the gradients in Dead Run and Baisman Run. The concentrated nature of rainfall in the urban end-member basins and the rainfall maximum to the northeast of the city suggest that the observed flood-producing rainfall is from warm season convective storms. Urban heat island, urban canopy, and urban aerosol impacts on warm season thunderstorms have been identified as sources of spatial heterogeneities in warm season rainfall climatology [Shepherd, 2005].

Figure 11.

Spearman correlation coefficients between rainfall at center of basin and rainfall at distance along the primary direction of rainfall movement for each storm included in Figure 10. Rainfall direction is indicated by the white arrow in Figure 10d.

[44] Maximum values of basin-averaged rainfall rate for the nine basins also show systematic spatial variation. In Figure 12, we present box plots of maximum 1 and 3 h rainfall rates for the 50 largest flood events in each of the nine basins. The basins located to the northeast of Baltimore (Moores Run, Herring Run, Whitemarsh Run, and Minebank Run, see Figure 1) have the highest short-term median rainfall rates (Figure 12). This is consistent with analyses of storm-averaged rainfall fields for Dead Run and Moores Run (and previous studies in Smith et al., 2012]. The rainfall rate distributions in Figure 12 represent both the spatial patterns of rainfall in metropolitan Baltimore and the sampling of storms that cause flood events in the particular basin. The intensely urbanized basins are more likely to flood from storm events with short-term, high-intensity rainfall. For example, Dead Run has larger rainfall rates than Baisman Run due to the sample of storms that make up the flood record in the two basins. Moores Run ranks fourth in median 1 h rainfall rate and sixth (behind Dead Run and Upper Gwynns Falls) in 3 h median rainfall rate. As will be discussed below, Moores Run responds to rainfall rates at time scales of less than 1 h, resulting in the lower position for 3 h rainfall rates and the lower storm-averaged rainfall for flood-producing storms (Figure 10). These differences in short-term rain rate partially account for differences in the flood peak magnitudes (Figure 2), particularly for flood peaks greater than the mean value. For example, Herbert Run and Whitemarsh Run have similar areas and extents of urban development, but the.75 and 0.9 flood quantiles in Whitemarsh Run are larger than in Herbert Run, reflecting the increased short-term rainfall rates in Whitemarsh Run. This points to the importance of rainfall spatial heterogeneities when considering flooding in an urban area.

Figure 12.

Box plots of maximum 1 and 3 h rainfall. The limits of the box represent the 25th and 75th percentiles, while the line inside represents the median (50th percentile). The whiskers indicate the 10th and 90th percentiles.

[45] Seasonality of flood occurrence (Figure 13) shows clear differences between flood-producing storm events in urban and nonurban basins. Urban watersheds, other than Upper Gwynns Falls, have a pronounced seasonal peak in flood occurrence during the middle of the warm season (late July). This reflects the prominent role that warm season thunderstorm systems play in flooding of urban watersheds [see also Ntelekos et al., 2007]. Upper Gwynns Falls, Cranberry Branch, and Baisman Run exhibit a more uniform distribution of intra-annual flood occurrences, suggesting that floods are produced by a different population of storms, with fewer warm season thunderstorms.

Figure 13.

Intra-annual distribution of flood peak occurrences.

[46] The link between urban flooding and warm season thunderstorms can also be seen in the diurnal distribution of peak discharges (Figure 14). Peak discharge time is determined by both the time of the rainfall and the lag time, or time between peak rainfall rate and peak discharge, of the basin. Lag times for 10 km2 urban basins are short. Moores Run, for example, has a characteristic lag time of approximately 15 min [Smith et al., 2005b]. For all basins, the maximum peak discharge time is between 3 P.M. and 9 P.M. EST. For the six urban watersheds, excluding Upper Gwynns Falls, the maximum is between 6 P.M. and 9 P.M. EST, and for Baisman Run, Cranberry Branch, and Upper Gwynns Falls the maximum occurs earlier in the day. Because nonurban basins have longer lag times, the peak rainfall rates that produce these discharge peaks must occur earlier in these three basins. The evening maxima in the urban basins suggest the dominant flood-producing rainfall is from organized convective systems [Ntelekos et al., 2007].

Figure 14.

Diurnal distribution of flood peak occurrences.

4. Summary and Conclusions

[47] Storm event hydrologic response was analyzed for nine similarly sized (approximately 10 km2) watersheds in the Baltimore area. The basins range in land use from a forested reference watershed, Baisman Run, to an “old” (pre-stormwater management) urban watershed, Moores Run. The largest 50 flood events during the 2000–2009 period were examined for each basin using high-resolution radar rainfall fields and USGS discharge data. The following conclusions are drawn from the analyses.

[48] (1) There are large contrasts in flood peak distributions among urban watersheds in Baltimore in addition to the large differences between urban and nonurban watersheds. Moores Run has anomalously large flood peak magnitudes with median flood peaks larger than the 0.75 flood peak quantiles in other urban drainage basins. Large flood peaks in Moores Run are not tied to large runoff ratios (see item 2 below) or to anomalously large rainfall rates (see item 3 below) but are most closely linked to rapid removal of runoff through the storm drain network and to the absence of modern stormwater detention structures. Extensive stormwater management infrastructure in Upper Gwynns Falls creates a flood peak distribution, which more closely resembles that of the nonurban watersheds than the urban watersheds.

[49] (2) A spectrum of storm event hydrologic response, with Moores Run and Baisman Run as end-members, can be seen in relationships between maximum 3 h rainfall rates and peak discharge. Differences among hydrologic responses in urban basins are not fully accounted for by differences in urban land cover among basins, suggesting the importance roles of runoff production and timing of hydrologic response in the spectrum of urban storm event response.

[50] (3) Response times for the urban basins included in this study can be broadly categorized into two groups: extremely short, less than 0.1 h, and moderately short, more than 0.4 h. This categorization does not depend on land use, as Moores Run (82% urban) is in the extremely short group, with a response time of 0.02 h, while Dead Run (76% urban) is in the moderately short group, with a response time of 0.83 h. Response times, like runoff ratios, are linked to recent urban development involving modern stormwater management infrastructure. Interestingly, the response time for Upper Gwynns Falls of 1.75 h is longer than the response time of forested Baisman Run at 1.25 h. Stormwater infrastructure increases urban response times but may increase runoff ratios over intermediate time scales by increasing basin connectivity.

[51] (4) There are large contrasts in storm event runoff ratios among urban watersheds in addition to the large differences between urban and nonurban watersheds. Dead Run has anomalously large storm event runoff ratios with a median value (0.58) that is larger than the 0.75 quantiles in other drainage basins. Transportation corridors and stormwater management infrastructure provide hydraulic connectivity throughout the Dead Run watershed, which promotes large storm event runoff ratios. Although Upper Gwynns Falls resembles nonurban watersheds in terms of flood peak distribution, the distribution of storm event runoff ratios for Upper Gwynns Falls decreases within the urban cluster, suggesting that stormwater management infrastructure is working as was desired by the Stormwater Management Act—lowering peak discharges but not decreasing runoff volumes. Modern stormwater management infrastructure does effectively lower flood peaks across the basins, but the implementation of stormwater management infrastructure may actually increase connectivity thereby increasing stormwater runoff volumes.

[52] (5) Antecedent moisture, represented by 5 day antecedent rainfall, does not appear to significantly impact storm event hydrologic response. Results comparing runoff ratio to antecedent rainfall do not show clear trends. This may be due to the relative unimportance of antecedent moisture in urban hydrologic response to large storms, or it may be due to the use of an indirect measurement of antecedent moisture.

[53] (6) Flood event clustering is an important element of urban flood hydrology. Many of the peak-over-threshold events occur within short times of each other in wet years or during long rainfall events. The annual peak-over-threshold flood counts per basin range from zero in dry years to 12 in wet years. Annual flood counts in all basins have index of dispersion values greater than 1, suggesting that there is clustering of flood events. Future studies will examine the impacts of basin storage and of clustering of storms on flood event clustering.

[54] (7) Spatial and temporal properties of flood-producing rainfall vary markedly over the nine watersheds. The storm-averaged rainfall fields for the 20 largest flood peaks in Dead Run and Moores Run exhibit large spatial gradients and rainfall maxima northeast of Baltimore City. The rainfall fields for the 20 largest flood peaks in Baisman Run and Upper Gwynns Falls exhibit smaller spatial gradients and larger areas of mean rainfall exceeding 50 mm. Spatial heterogeneities of rainfall associated with urbanization may play an important role in spatial variation of flood hazards over the Baltimore region.

[55] (8) The rainfall climatology of flood-producing storms varies between urban and nonurban watersheds (with Upper Gwynns Falls more closely related to nonurban watersheds). The six urban watersheds have a pronounced warm season maximum in flood frequency while the nonurban watersheds and Upper Gwynns Falls exhibit a more uniform distribution of intra-annual flood occurrences. There is also a pronounced diurnal cycle to flood occurrences in the six urban watersheds, with floods occurring more often in the evening between 6 P.M. and 9 P.M. EST. This highlights the central role warm season thunderstorm systems play for urban flooding in Baltimore.

[56] (9) High-resolution radar rainfall fields provide an important tool for studying storm event hydrologic response in small urban basins. High-resolution in space allows for the representation of spatial variation of rainfall between basins, and high-resolution in time allows for the analysis of basin water balance at the short time scales of urban basins.


[57] The research was supported by the National Science Foundation (grants EEC-0540832, CBET-1058027), the Willis Research Network, and the NOAA Cooperative Institute for Climate Science. This work utilized the field infrastructure supported by the NSF Long-term Ecological Research (LTER) Program (under grants 0423476 and 1027188.).