Water Resources Research

Extreme hydrometeorological events and the urban environment: Dissecting the 7 July 2004 thunderstorm over the Baltimore MD Metropolitan Region

Authors


Abstract

[1] Observational analyses and mesoscale modeling studies, using the Weather Research and Forecasting (WRF) model, are used to dissect the mechanisms associated with record lightning, rainfall, and flooding over the Baltimore metropolitan region on 7 July 2004. Storm evolution on 7 July 2004 exhibited characteristic features of warm season thunderstorms producing flash flooding over the Baltimore–Washington DC metropolitan region. The storm system was initiated along the Blue Ridge mountains, with model simulations suggesting that convergence-induced spin-up of a meso-low was responsible for initial thunderstorm development. Observations and model analyses show that thermal effects associated with Chesapeake Bay had a pronounced impact on storm evolution and rainfall distribution. Analyses of radar reflectivity and lightning observations suggest that the urban environment played a significant role in storm evolution and heavy rainfall distribution. Model analyses show that urban canopy effects from both the Baltimore and Washington DC urban regions play an important role in determining the storm environment associated with heavy rainfall. Urban Heat Island effects did not play a significant role in the storm evolution. Observations of aerosols and drop-size distributions from a vertically pointing LIDAR and a disdrometer and model analyses suggest that the aerosols may have played an important role in stimulating efficient precipitation mechanisms and extreme rainfall rates for the 7 July 2004 storm.

1. Introduction

[2] The United Nations projects a global population of 8.3 billion in 2030. An estimated 50% of the global population currently resides in urban areas. By 2030, this number is projected to increase to 61%, leading to approximately 5.1 billion urban residents [United Nations, 2006]. The Intergovernmental Panel on Climate Change (IPCC) in its 4th assessment report in 2007 concludes that an increase in the frequency of heavy precipitation events (and thus flooding), due to global warming, is “very likely” to be observed during the 21st century [IPCC, 2007a]. Commenting on the degree of the vulnerability of society, the IPCC ranks the areas where rapid urbanization is occurring among the most vulnerable [IPCC, 2007b]. Studies of extreme meteorological events, focusing on urban environments, are critical in improving our understanding of these phenomena and in mitigating their effects in the uncertain years to come. The characterization of the spatial and temporal properties of rainfall for such extreme storms have important implications for assessing the evolving flood hazards in urban environments.

[3] The vulnerability of the urban environment with respect to flooding and especially flash flooding arises from both hydrologic and hydrometeorological sources. From the hydrologic perspective, increasing impervious cover and elaboration of the drainage network through the storm drain network result in increased flood peaks and decreased response times of urban watersheds [Leopold, 1968; Graf, 1977; Smith et al., 2002]. An important consequence is that warm season thunderstorms become the principal agents of flash flooding in many urban environments [Ntelekos et al., 2007; Doswell et al., 1996; Baeck and Smith, 1998; Chang, 1976; Smith et al., 2005]. From the hydrometeorological perspective, the urban environment itself may have an impact on the evolution and rainfall distribution of warm season thunderstorm systems, as elaborated below.

[4] Rainfall modification by urban environments has been an active area of research for the last 40 years, with the METROpolitan Meteorological EXperiment (METROMEX) in St. Louis, Missouri [Changnon et al., 1971; Semonin and Changnon, 1974] stimulating extensive research in the area. The main findings of the METROMEX experiment included (1) an enhancement of precipitation in the downwind part of the city in a “fan-shaped” area that extended approximately 30 km from the city center; (2) higher rainfall anomalies during the warm season; and (3) more pronounced rainfall modification for high intensity storms. Interpretations of these and similar results have focused on three main mechanisms: (1) the temperature difference between the urban and nonurban areas with the urban region being warmer and thus creating centers of convergence over and downwind of it (Urban Heat Island, UHI); (2) the increased roughness of the urban environment due to urban infrastructure; and (3) the altered chemical properties and size distribution of urban aerosols due to emissions from the urban environment.

[5] Research results have enhanced our understanding but have also led to the realization that the processes involved in rainfall modification by urban environments are complicated and that our current understanding is far from complete, as summarized in the reviews by Lowry [1998], Shepherd [2005], and Collier [2006]. Reviews of prior research highlight the need for detailed case studies to isolate the effect of the urban environment on rainfall distribution [e.g. Bornstein and Lin, 2000; Hold and Pullen, 2007]. Reviews also point to the lack of understanding of the role of aerosols on rainfall modification. Prior studies have highlighted the role of aerosols in suppressing precipitation [Ramanathan et al., 2001; Givati and Rosenfeld, 2004]. Other studies have suggested theoretical mechanisms by which anomalous aerosol populations might lead to the formation of deeper convection, more efficient precipitation mechanisms and heavier surface rainfall [Williams et al., 2002; Rosenfeld, 2006; Khain et al., 2005].

[6] The St. Louis study region of METROMEX was ideal for examining urban impacts of rainfall in part because of the relatively simple geographic setting. Yet major urban cores in the US and around the world are often built in close proximity to complex terrain, especially in coastal and mountainous regions. A review of the literature on METROMEX and subsequent studies shows that “downwind amplification” of rainfall is a common form of rainfall modification (see, for example, the review of Lowry [1998]). In the case of urban environments situated over complex terrain, it is evident that such results might not be applicable. In this paper, we focus on the analysis of an extreme rainfall and flooding event over the Baltimore, MD metropolitan region on 7 July 2004. Baltimore is located in the eastern United States and is surrounded by complex terrain typical of the urban corridor of the northeastern US.

[7] The objectives of this study are to understand the mechanisms and processes associated with the genesis and evolution of such an extreme rainfall event and to isolate and assess the importance of mountainous terrain, the land–water interface and the urban environment of Baltimore and Washington DC on the formation and evolution of the 7 July 2004 storm. It is also the intent of this study to stimulate further discussion and research on the area of urban rainfall modification over complex terrain. The paper is structured as follows: in Section 2, data sources, the domains of analysis and the details of the mesoscale model simulations are presented. Results are presented and discussed in Section 3 grouped in subsections on the synoptic environment and storm evolution, multi-instrument observations and mesoscale model simulations. Section 4 summarizes the key findings and provides some concluding remarks.

2. Study Region, Data, and Model Implementation

[8] Analyses of the 7 July 2004 storm focus on study domains around the Baltimore metropolitan region (Figure 1). The largest domain, denoted d01 in Figure 1, includes much of the Appalachian Mountain region and is used for modeling on a coarse scale (6 km resolution). The nested domain, denoted d02 in Figure 1, is used for detailed model analyses (2 km resolution) of the region close to the urban environment of Baltimore and Washington DC and includes the Blue Ridge Mountains to the west and the Chesapeake Bay to the east. The domain focused over the urban cores of Baltimore and Washington DC (hereafter “urban focus” domain, smaller rectangle in Figure 1; also see Figure 2) contains observing systems used in this study. The Dead Run watershed (14.3 km2 in drainage area at the U.S. Geological Survey (USGS) stream gaging station at Franklintown) received the largest rainfall accumulations from the 7 July 2004 storm and is highlighted in Figure 2. Dead Run is a tributary of Gwynns Falls, the principal study watershed of the Baltimore Ecosystem Study (BES), a Long Term Ecological Research (LTER) project focusing on the ecology and hydrology of the urban environment [see Groffman et al., 2003; Smith et al., 2005]. The topography of the study area is complex. The Appalachian Mountains are a prominent feature of the region with their southwest to northeast orientation and maximum elevations that exceed 1100 m. To the east of the Blue Ridge Mountains, the urban corridor of Baltimore and Washington DC is located along the edge of the land–water boundary created by the Chesapeake Bay.

Figure 1.

WRF-ARW simulation domains 1 (denoted d01, 6 km) and 2 (denoted d02, 2 km) and “urban focus” domain (yellow rectangle, see also Figure 2). The urban environments of Baltimore and Washington DC are also shown (yellow shaded areas).

Figure 2.

“Urban focus” domain (yellow rectangle in Figure 1) and instrumentation site locations. Light green shaded areas depict the urban perimeters of Baltimore and Washington DC. Dark green shaded areas depict Baltimore County and Washington DC.

[9] Radar reflectivity data from the Sterling, VA Weather Surveillance Radar-1988 Doppler (WSR-88D; see Fulton et al. [1998] and Krajewski and Smith [2002] for a discussion of rainfall retrieval algorithms from Doppler radars and Figure 2 for the radar location) were used for obtaining rainfall estimates over the Baltimore–Washington DC region. Rainfall accumulation products at 15-min time interval were obtained from the Hydro-NEXRAD system [Krajewski et al., 2007]. Bias correction for radar rainfall estimates utilized a network of 17 rain gages in the Dead Run watershed [Smith et al., 2005, 2007].

[10] Radar reflectivity data from the Terminal Doppler Weather Radar (TDWR) of the Baltimore–Washington International Airport (BWI) are used for analyses of storm evolution over the Baltimore study region. The TDWR is a 5 cm wavelength Doppler radar located 10 km south of BWI (see Figure 2). An important element of the TDWR scanning strategy is that a base scan is taken every minute. After a base scan is finished, the radar returns to vertical scans of the atmosphere. The scanning strategy is ideal for tracking storm development and evolution at low levels, but not for examining vertical profiles of storm structure. The TDWR has an azimuthal resolution of 1° with 250 m radial resolution. Due to the potential for attenuation by raindrops, quantitative rainfall estimates from the TDWR are not presented in this study. Intercomparisons of TDWR and Hydro-NEXRAD rainfall estimates were carried out and they are in good qualitative agreement, despite the attenuation problems. TDWR observations are used for examining the initiation, evolution and dissipation of storm cells close to the urban environment of Baltimore and Washington DC due to its very high spatial and temporal resolution. Data were available for the time period of rainfall around Baltimore from 1801 UTC to 2115 UTC of 7 July 2004.

[11] Lightning data from the National Lightning Detection Network (NLDN; Cummins et al. [1998]) were used for examining the evolution of the 7 July 2004 thunderstorm system. The network records, among other variables, the date, time and location of Cloud-to-Ground (CG) lightning strikes over the conterminous United States [see Orville and Huffins, 2001]. The observing period for the NLDN data used in this study extends from 1991 through 2007.

[12] Microphysical analyses of the 7 July 2004 storm were supplemented by raindrop-spectra observations from a Joss-Waldvogel disdrometer [see Steiner et al., 2004] and a vertically pointing LIDAR (LIght Detection And Ranging) system, both of which were located on the University of Maryland Baltimore County (UMBC) campus (see Figure 2 for instrument location). Disdrometer data were available from 1830 UTC to 2117 UTC. Lidar observations of the total attenuated backscatter coefficient (at 532 nm wavelength) were available for the prestorm environment from 1305 UTC to 1619 UTC.

[13] The Weather Research and Forecasting (WRF) model, developed by the National Centers for Atmospheric Research (NCAR), is a fully compressible, nonhydrostatic, next-generation mesoscale model. In this study, the Advanced Research version of WRF, ARW V2.2, was used for all model analyses [see Skamarock et al., 2007]. Initial and boundary conditions for the WRF-ARW model analyses carried out in this study are obtained from Eta model output analysis fields [Rogers et al., 1996] from the archive of the National Centers for Environmental Protection (NCEP). The data are archived on the #212 NCEP-Eta postprocess grid on a 40 km resolution. Both three-dimensional atmospheric variables such as wind, temperature, humidity and cloud water content and two-dimensional near surface, surface and subsurface variables such as soil temperature and soil water content in four depth levels, sea surface temperature and pressure are included in the model fields. The temporal resolution of the model's boundary conditions is three hours.

[14] WRF model simulations were performed on two grids, domain 1 (d01) and domain 2 (d02), shown in Figure 1. Domain d01 (parent domain) had a resolution of 6 km while nested domain d02 had a resolution of 2 km (parent/grid ratio 1:3). The center-point latitude and longitude of the parent domain was 39.00° and −79.00° respectively. The south-to-north extend of d01 and d02 was 120 and 169 grid cells, respectively. The west-to-east extend for the two domains was 171 and 232 grid cells, respectively. A one-way concurrent ARW simulation with two domains [see Skamarock et al., 2007] was implemented. The fine grid initialization was performed from an external file that contained high resolution meteorological and terrestrial data. The lateral boundary conditions of the fine grid were given at each coarse grid time step forecast. The time step of the parent grid was set to 36 s; for the nested domain the time step was 12 s. For both domains, the highest available resolution of 30 arc-sec was used as input for most of the static fields in the WRF Preprocessing System (WPS).

[15] For all the simulation results presented in this study, the model was initiated at 03 UTC of 7 July 2004 and was integrated for 24 hours. The physics and other options of the model were kept the same for both d01 and d02 domains to avoid inconsistencies at the boundaries. Table 1 presents the WRF-AWR model physics options used for all simulations performed in this study (for references on the physics options, see Skamarock et al. [2007]).

Table 1. WRF-AWR V2.2 Physics Options
SchemeOption
MicrophysicsWSM6
Longwave RadiationRRTM
Shortwave RadiationDudhia
Surface LayerMonin-Obukhov (Janjic)
Land SurfaceNoah
Boundary LayerMellor-Yamada-Janjic TKE
Cumulus ParameterizationNone

[16] The Noah land-surface model used herein includes a treatment of the urban environment through increased roughness height, reduced albedo, increased storage capacity, increased soil thermal conductivity and reduced evaporation for pixels designated as “urban or built-up land”. In assessing the role of the urban environment (Baltimore and Washington DC) for the 7 July 2004 storm, a “No Urban” simulation was implemented by replacing the “urban” pixels of the input files of the initial conditions for both d01 and d02 domains with the “cropland/woodland mosaic” characterization (the dominant background vegetation type in the vicinity of the Baltimore–Washington DC area).

[17] CG lightning analyses presented in Section 2 point to the importance of the Chesapeake Bay in the evolution of the 7 July 2004 thunderstorm. WRF was used to assess the impact of the land–water boundary and to obtain a better understanding of its role during the storm. A “No Chesapeake Bay” model simulation was implemented, identical in domain configuration and physics options with the “Control” simulation but this time by “removing” the effect of the Chesapeake Bay from the initial conditions. This involved changing variables in the input files of both the coarse and fine domains of the one-way concurrent nested run. The list of these variables includes most of the variables used by the Noah land-surface scheme such as, land-mask, soil type, vegetation type, vegetation fraction, skin temperature, sea-surface temperature, soil temperature at the lower boundary and finally soil moisture, temperature and liquid water in four different soil-depth levels. An implicit assumption was made that the effect of Chesapeake Bay on the boundary conditions was negligible because of the large distance of domain d01 from this area (see Figure 1). As described previously, the boundary conditions of the finer domain, d02, were provided at each time step from the model fields of the coarser domain.

3. The 7 July 2004 Storm Over Baltimore

[18] The 7 July 2004 storm produced more CG lightning strikes over the Baltimore metropolitan region than any other thunderstorm system during the period of NLDN observations (1991–2007). As shown later in this section, CG lightning production along the western margin of Baltimore City exceeded the annual average of 3.5 strikes km−2 yr−1 in less than two hours. Peak rainfall accumulations exceeded 150 mm, with the largest rainfall accumulations over the Dead Run watershed (see Figure 2 for watershed location). Extreme flooding was observed in several urban basins, with the most severe flooding in Dead Run. The flood peak in Dead Run, at a drainage area of 14.3 km2, was the largest for the USGS gaging station from a record that extends back to 1960. Basin-averaged rainfall accumulations at 60 and 120 min rainfall rates exceeded respectively the 100 and 300 year point rainfall accumulation for the Baltimore region. Basin average rainfall rates exceeding 40 mm h−1 were maintained for approximately 90 minutes, with the peak basin average rainfall rate exceeding 150 mm h−1.

3.1. Synoptic Environment, Storm Structure, and Evolution

[19] The synoptic environment of the 7 July 2004 storm was dominated by an eastward propagating extratropical cyclone with its low-pressure center located roughly over the Great Lakes during the storm. Evolution of the extratropical cyclone was examined from analyses of conventional weather maps and WRF-ARW model analyses on a 40 km national grid. The maximum geopotential height depression at the 500 hPa level from the model simulation was 120 m and occurred at 1200 UTC on 7 July. The standard deviation of 500 hPa geopotential height values for the area is 80 m [Blackmon, 1976], placing the 7 July 2004 cyclone at the high end of the spectrum of summer extratropical cyclones. Deepening of the low-pressure system resulted in intensification and veering of the winds in a direction perpendicular to the Appalachian mountain range. This aspect of the low-level flow field played an important role in orographic mechanisms associated with storm evolution, as detailed below.

[20] The structure and evolution of the 7 July 2004 storm shares elements that are typical of thunderstorm systems that produce flash floods over the Baltimore region [Ntelekos et al., 2007]. Storm initiation (Figure 3; lightning activity is shown in strikes km2 hr−1) occurred shortly after 1600 UTC over the Blue Ridge mountains. The initiation location of the storm reveals the central role of the Blue Ridge mountains for the development of the 7 July storm. During the next two hours (1700–1900 UTC) the storm intensified and moved to the east, reaching the Baltimore region at about 1900 UTC. During these three hours (1600–1900 UTC) the storm covered a distance of approximately 70 km, traveling with an average speed of more than 20 km h−1. From 1900–2200 UTC, the storm interacted with the urban environments of Baltimore and Washington DC and with the land–water boundary of the Chesapeake Bay region. The most intense period of the storm for both lightning and rainfall production was between 2000 and 2100 UTC while it was situated over the two cities. During the three hours, from 1900 until 2200 UTC, net storm motion slowed appreciably to an average speed of less than 10 km h−1. This observation points to the important role of the urban environment and of the Chesapeake Bay on storm evolution.

Figure 3.

Hourly aggregated contours of cloud-to-ground lightning strikes during the 7 July 2004 storm (in strikes km−2 hr−1).

3.2. Multi-Instrument Observations

[21] Storm evolution over the Baltimore region as reflected in radar and CG lightning observations from 1845–2115 UTC (Figure 4) highlights the continuous development of convective elements along the western margin of Baltimore City, especially over the Dead Run watershed. Three central points arise from analyses of radar and lightning observations (Figure 4): (1) the western margin of Baltimore City is a center of persistent redevelopment of convective storm elements during the period from 1845 to 2115 UTC; (2) storm cells develop west and southwest of the city and either remain stationary along the southern margin of the city producing large rainfall accumulations, or decay as they try to propagate over the city; (3) the storm system develops into a squall line that sweeps over the region from 2015 to 2045 UTC (Figures 4d4f), producing the final period of extreme rainfall rates over Dead Run and adjacent regions.

Figure 4.

30-min interval snapshots of the evolution of the 7 July 2004 storm based on TDWR reflectivity fields (in dBZ) and cloud-to-ground (CG) lightning (each bolt shows the location of a CG strike within the minute of the snapshot) over the “urban focus” domain.

[22] Extreme CG lightning activity occurred on the west side of both Washington DC and Baltimore (Figure 5; contours of storm total CG lightning strikes are in strikes km−1), with the lightning contours following the orientation of the urban environment. Storm total lightning exceeded the annual total for a large area of western Baltimore, located northwest of the Dead Run watershed. The line of elevated CG flash densities extending from the area south of Dead Run to the eastern boundary of Washington DC (Figure 5) was produced during a short period of time (2015–2045 UTC) and was associated with the convective line that developed over the region (Figure 4). As model results suggest (see Section 3), the squall line was associated with a mesoscale boundary that formed with the Blue Ridge meso-low. The squall line did not propagate over the central part of Baltimore and decayed rapidly as it approached Chesapeake Bay. As shown later (Section 4), southerly winds were continuously diverging when passing over the city, making the western part of the city a center of convergence and the environment over the city itself very “unfriendly” for westerly propagating storm cells.

Figure 5.

Storm total cloud-to-ground lightning contours over the “urban focus” domain (in strikes km−2).

[23] The spatial pattern of storm total rainfall (Figure 6) is similar to storm total lightning (Figure 5), with rainfall contours wrapping around the western sections of the urban environments of Baltimore and Washington DC. Bias-corrected rainfall estimates were obtained from 15-minute Hydro-NEXRAD rainfall fields with bias correction utilizing 17 rain gages from the Dead Run rain gage network. The bias correction was applied to obtain accurate rainfall estimates close to the core of the heaviest rainfall over the urban environments of Baltimore and Washington DC. The most extreme rainfall is located over the Dead Run watershed and is offset from the lightning maximum by approximately 10 km. Although there is a close relationship between CG lightning and rainfall for mesoscale convective systems, the locations of spatial maxima of rainfall and CG lightning often show offsets for extreme flood producing storms [Tapia et al., 1998; Petersen et al., 1999; Smith et al., 2000].

Figure 6.

Storm total bias-corrected radar-rainfall contours from the Sterling, VA WSR-88D station over the “urban focus” domain (in mm).

[24] The western boundary of Baltimore's urban core had an important impact on the creation and propagation of storm cells. Convective cells were continuously created around this area but did not propagate over the city (Figure 7). At 1843 UTC, the locations of the center of mass of three storm elements, based on reflectivity values exceeding 40 dBZ, are marked in Figure 7 (circles 1, 2, and 3). The figure also includes the storm tracks (mass centroid locations; see Dixon and Wiener [1993] and Smith et al. [1996]) of these elements back in time. Elements 1 and 2 were part of the same multicell storm that initially formed southwest of the city. The storm element propagated toward the city and reached the southwest boundary at 1819 UTC where it split into two storms elements (1 and 2). Storm element 2 remained stationary at the southwestern boundary of the city for more than half an hour before it dissipated. After reaching the southwest corner of Baltimore City, storm element 1 veered to the northwest and propagated along the western margin of Baltimore City. Storm element 3 illustrates the continuous initiation of storm cells at the southwestern section of the city and northeastern propagation of the storm elements.

Figure 7.

TDWR base reflectivity field (in dBZ) at 18:43 UTC over southwestern Baltimore (see Figure 2 for spatial reference). Three storm elements (marked with circles 1, 2, 3) and their back trajectories in time (white lines) are also shown.

[25] Maximum rainfall accumulations in the Dead Run watershed exceeded 150 mm based on raingauge observations over the watershed. Time series of basin-averaged radar-rainfall estimates and unit discharge (discharge normalized by drainage area) observations from the USGS station at the outlet of the Dead Run basin (Figure 8) highlight the timing of extreme rainfall and flooding over the western sections of the Baltimore region. Basin-averaged rainfall rates exceeding 40 mm h−1 were sustained for more than 90 minutes over the 14.3 km2 basin. The maximum unit discharge associated with the event reached 17.2 m3 s−1 km−2 at 2055 UTC. The second largest flood peak in Dead Run occurred in June 1972 (Hurricane Agnes) and had a peak unit discharge of 14.6 m3 s−1 km−2.

Figure 8.

Dead Run watershed, basin-average radar-rainfall rate time series (green shaded area, in mm h−1) from the Sterling, VA WSR-88D station and unit discharge time series (blue, black-dotted line, in m3 s−1 km−2) at the basin outlet.

[26] Raindrop size distribution measurements for the 7 July storm were taken with a Joss-Waldvogel disdrometer located on the UMBC campus 6 km southeast of the maximum precipitation in Dead Run (see Figures 2 and 6). Peak rainfall rates exceeded 100 mm h−1 at the disdrometer location and there were systematic changes in raindrop size distributions (Figure 9) over the period of peak rainfall rates (2035–2110 UTC). There was a general increase in drop arrival rate, reaching a maximum value of almost 5000 drops s−1 m−2 and a general decrease in mean diameter. Peak rainfall rates for the storm are thus tied to a striking increase in the raindrop number concentration (compare with analyses of evolution of raindrop size distributions for a “leading line–trailing stratiform” system by Uijlenhoet et al. [2003]).

Figure 9.

Joss-Waldvogel distrometer observations (see Figure 2 for disdrometer location): rainfall rate (top, in mm h−1), drop mean diameter (middle, in mm), and drop arrival rate (bottom, in drops s−1 m−2).

[27] Observations from a vertically pointing lidar system on the UMBC campus (Figure 10) suggest that anomalous aerosol populations may have played a role in efficient precipitation mechanisms associated with extreme rainfall rates. Lidar observations of the prestorm environment are converted to total attenuated backscatter coefficient (in km sr−1) in the visible channel (532 nm) and cover the period from 1305 UTC to 1600 UTC. A striking element of this figure is the elevated aerosol layer at about 2 km above ground level which is present after 1425 UTC and shows steady growth in aerosol backscatter until 1600 UTC. Time series of the model-simulated vertical profile of relative humidity close to the location of the UMBC lidar facility (Figure 11) show that relative humidity increased rapidly during the period of increasing aerosol backscatter. These results draw links between hygroscopic growth of aerosols and efficient precipitation mechanisms that produce extreme rainfall rates under moist, highly convective conditions. Although anomalous aerosol populations may play a role in promoting heavy rainfall, there is no evidence that urban aerosols from local sources play an important role. Back-trajectory wind field model analyses (not shown) suggest that the layers of elevated aerosol concentration have sources remote from the Baltimore–Washington DC metropolitan region.

Figure 10.

Vertically pointing lidar observations of the prestorm environment (see Figure 2 for lidar location): total attenuated backscatter coefficient (532 nm, in km−1 sr−1).

Figure 11.

WRF-ARW simulation results over the point of model maximum rainfall accumulation to the southwest of Baltimore (see Figure 13): relative humidity profiles at the beginning of each hour from 14:00 to 21:00 UTC (in %).

3.3. WRF-ARW Modeling

[28] WRF-ARW simulations of the 7 July 2004 storm over Baltimore provide insights on the mechanisms of storm initiation and evolution. The simulated 900 hPa wind field for domain d01 (Figure 12) illustrates key elements of the wind field at critical times for storm development. At 1200 UTC, rapid intensification of the synoptic scale low-pressure system over the Great Lakes is critical in turning the low-level wind field almost perpendicular to the Appalachian Mountain range (see Figure 12, top). Model simulations show a strong gradient in horizontal velocities between the two sides of the mountain range at 1200 UTC (color-shaded areas on the upper left corner of Figure 12, top). On the windward side of the mountain, wind velocities are about 14 m s−1 (∼50 km h−1). On the lee side, wind velocities are as low as 2 m s−1 (∼7 km h−1). Although the turning of the low-level wind is related to the intensification of the low-pressure system over the Great Lakes and the velocity gradient is related to the existence of the mountain range, there is some synergy between these two elements: the turning of the winds to a direction almost perpendicular to the mountain range accentuates the effect of the mountains to the wind field, making the wind field differences between the lee and the wind side mountain more pronounced.

Figure 12.

WRF-ARW simulation results for domain d01: 900 hPa wind field streamlines and wind magnitude (small shaded subplots, in m s−1) at 1200 UTC (top) and 1600 UTC (bottom).

[29] Low-level convergence over the Blue Ridge peaks around 1200 UTC (Figure 12, top) and leads to the creation of a meso-low on the lee-side of the mountain (O'Handely and Bosart [1996] have noted the creation of Appalachian lee-side cyclogenesis in extreme cases). Shortly thereafter, a meso-high forms to the east, over the Atlantic Ocean (Figure 12, bottom). Although the formation of the meso-low is masked by the presence of the mountains to the west, the cyclonic rotation of the wind field is evident in the bottom panel of Figure 12. By 1600 UTC, the flow field has changed markedly from the 1200 UTC configuration, with strong southerly winds that bring warm, moist air over the storm area. The simulated wind field is consistent with Velocity Azimuth Display (VAD) observations from the Sterling, VA WSR-88D radar (not shown).

[30] Model analyses suggest that the storm initiated over the Blue Ridge Mountains as the result of spin-up of a meso-low formation along the convergence line over this area at about 1600 UTC (Figure 12). This feature is in good agreement with the time and location of initial lightning observations (Figure 3). The observed formation of a meso-low to the west and meso-high to the east is the optimum configuration for strong southerly flow and moisture advection. The creation of the meso-low is accompanied by a mesoscale boundary that propagated to the east and played an important role in organizing the squall line that passed over the Baltimore and Washington DC metropolitan areas (see also Figure 4).

[31] Figure 13 shows the storm total rainfall accumulation for the 7 July 2004 storm from both the Sterling, VA WSR-88D (top) and the model's “Control” simulation (bottom) over domain d02. Storm total rainfall from the model simulation (Figure 13, bottom) captures the rainfall maximum over the western margin of Baltimore City. The model predicts an area of heavy rainfall to the southwest and northwest of Baltimore, close to the location of the observed rainfall maximum over Dead Run. Peak rainfall accumulations over this area from model simulations reach about 80 mm. The model also predicts heavy rainfall accumulations on the south and northwest regions of Chesapeake Bay. The results are in reasonable agreement with Hydro-NEXRAD radar-rainfall estimates from the Sterling, VA WSR-88D (Figure 13, top). The model also captures important elements of the timing of the storm over Baltimore, especially when considering the convective nature of the 7 July storm. Model simulations indicate that rainfall over Baltimore began around 1830 UTC on 7 July 2004 and persisted for almost two hours (Figure 14). WRF predicts rainfall rates that exceeded 100 mm h−1 at 1930 UTC and remained over 20 mm h−1 for almost one hour (1850 to 1940 UTC). Relative humidity progressively increased as winds become southerly at about 1500 UTC and peaked at about 1900 UTC when the model suggests an almost complete saturation of the atmospheric column above Baltimore (Figure 11g).

Figure 13.

Comparison of radar-rainfall observations and WRF-ARW simulation results for domain d02: storm–total rainfall accumulation contours (in mm) from the Sterling, VA WSR-88D station (top) and the model's “Control” simulation (bottom).

Figure 14.

WRF-ARW simulation results for the point of maximum rainfall accumulation to the southwest of Baltimore (see Figure 13): rainfall rates (in mm h−1).

[32] The difference in rainfall accumulation between the “Control” and the “No Urban” model simulations is shown in Figure 15 over the “urban focus” domain (see Figures 1 and 2 for domain location). The results of Figure 15 show a large increase in the model simulated rainfall for the “Control” case at the southwest (more than 40 mm) and northwest (more than 30 mm) regions of Baltimore but also on the northwest portions of Washington DC (more than 20 mm), following the orientation of the urban environment. Notice the striking similarities in the distribution of positive rainfall contours between Figures 6 and 15.

Figure 15.

WRF-ARW simulation results for the “urban focus” focus: storm–total rainfall accumulation difference contours between the “Control” and the “No Urban” simulations (in mm). Positive (negative) contour values indicate more rainfall over this area for the “Control” (“No Urban”) simulation.

[33] Model analysis of the low-level wind (Figure 16, top row) and resulting divergence (Figure 16, bottom row) fields over the “urban focus” domain show that the urban environment has a significant effect in altering the locations of convergence and divergence zones over the Baltimore region, mainly through the effect of the urban canopy layer. Convergence is stronger and more widespread along the western boundary of Baltimore during the “Control” run (Figure 16, bottom row, left) when comparing with the “No Urban” run (Figure 16, bottom row, right). The role of the urban environment is especially pronounced as the mesoscale boundary passed over the region. The Washington DC metropolitan region also played a role in the convective development over the region west of Baltimore, due to the strong low-level southerly flow that is retarded when passing over Washington DC. These features are likely linked to anomalous storm motion around the city (Figure 7).

Figure 16.

WRF-ARW simulation results for the “urban focus” domain: snapshot of the 950 hPa wind (top row, in m s−1) and resulting divergence (bottom, in s−1) fields at the time of the passage of the squall line over the area (1920 UTC) for the “Control” (left column) and “No Urban” (right column) simulations.

[34] Simulated temperature fields at low atmospheric levels did not show anomalies over the area of the urban environments in the “Control” simulation suggesting that the Urban Heat Island (UHI) had a negligible impact on storm properties. The high wind speeds associated with the 7 July 2004 storm (see Figures 12 and 16) created a homogeneous temperature field that suppressed any impact that the UHI might have on the approaching storm.

[35] Observations of hourly CG lightning strikes (Figure 3) and storm motion from TDWR reflectivity fields (Figure 4) suggest a pronounced link between the evolution of the 7 July 2004 storm and the land–water boundary of the Chesapeake Bay. A “No Chesapeake” WRF-ARW model run was designed, as described in Section 2, to study the effect of the Chesapeake Bay on the storm. Model results of the 1000 hPa temperature and wind field, for the “Control” (Figure 17, top row) and the “No Chesapeake” (Figure 17, bottom row) model runs, show significant alteration of the storm environment by the Chesapeake Bay (notice that Figure 17 focuses over the Chesapeake Bay only). The results suggest that the Chesapeake Bay acted as a thermal boundary that altered the low-level flow by enhancing divergent motion around it. The effect of the Bay was more pronounced in the afternoon and late afternoon hours of the storm (Figures 17c and 17d, first row). When the effect of the Bay is removed (Figure 17, bottom row), the low-level wind field over the Chesapeake area is smoother with no divergent flow to the east and west of the Bay. Model simulations of the 900 hPa water vapor mixing ratio and temperature (not shown) show the Chesapeake Bay to be a cool pool with a “tongue” of low values of mixing ratio for the late afternoon.

Figure 17.

WRF-ARW simulation results over Chesapeake Bay: snapshots of the 1000 hPa wind (in m s−1) and temperature (in °C) fields at four different times in the simulation. Top row plots show the results of the “Control” simulation at the beginning of the following hours: (a) 0900 UTC, (b) 1200 UTC, (c)1500 UTC, and (d)1800 UTC. Bottom row plots show the results of the “No Chesapeake Bay” simulation for the same times.

[36] The difference in rainfall accumulation between the “Control” and the “No Chesapeake” WRF simulations (Figure 18) illustrates the pronounced role of Chesapeake Bay in controlling the distribution of heavy rainfall from the 7 July storm. For clarity of presentation, only differences exceeding 20 mm are shown. Figure 18 reinforces the observations made before. The “Control” run shows more precipitation on the west side of Chesapeake Bay. When the effects of the Bay were removed, storm total rainfall was shifted further to the east over the Chesapeake Bay area.

Figure 18.

WRF-ARW simulation results for domain d02: storm–total rainfall accumulation difference contours (in mm) between the “Control” and the “No Chesapeake Bay” simulations. Positive (negative) contour values indicate more rainfall over this area for the “Control” (“No Chesapeake Bay”) simulation.

4. Summary and Conclusions

[37] In summary, the major findings of this paper are:

[38] • 7 July 2004 marks one of the most intense thunderstorm systems over the urban corridor of the eastern US. The storm produced record CG lightning flash densities over the Baltimore region (for the 1991–2007 observing period of the NLDN). Peak storm total CG flash densities, which were concentrated in a period of two hours, exceeded the annual average. The storm resulted in record flooding of urban watersheds with rainfall accumulations exceeding 150 mm in about two hours. The return intervals of peak 60 min and 120 min rainfall rates exceeded 100 and 300 years respectively.

[39] • Evolution of the 7 July 2004 storm can be characterized by initiation over the Blue Ridge (16–17 UTC), intensification and westward propagation toward the urban environment of Baltimore and Washington DC (17–19 UTC), interaction with the urban environment and the land–water boundary of the Chesapeake Bay (19–21 UTC) and dissipation along the western margin of Chesapeake Bay (21–22 UTC). Storm evolution for the 7 July storm exhibits similarities with many warm season thunderstorm systems producing flash floods over the urban environment of Baltimore. Storm motion slowed markedly as the system approached the Baltimore metropolitan region and Chesapeake Bay. Similar patterns of initiation and evolution for flash-flood producing storms presumably hold for much of the urban corridor of the eastern US, which is characterized by a coastal region of dense urban development bordered by the Appalachian Mountains to the west.

[40] • The synoptic environment of the 7 July 2004 storm was dominated by a warm season extratropical cyclone with surface low over the Great Lakes. WRF-ARW model analyses show that the low-pressure system deepened rapidly around 1200 UTC of 7 July, intensifying and turning the low-level winds almost perpendicular to the Appalachian Mountains. Model analyses show spin-up of a meso-low along the Blue Ridge soon after 1200 UTC. The time and location correspond with lightning and radar observations of storm initiation along the Blue Ridge. Lee-side cyclogenesis along the Blue Ridge was soon followed by development of a meso-high off the Virginia–Maryland coast in the Atlantic. The paired low to the west high to the east created ideal conditions for transport of warm, moist air from the south over the Baltimore region. The synoptic and mesoscale evolution of the 7 July 2004 storm provide a general depiction of the heavy rainfall environment for flash flood events in urban environments.

[41] • WRF-ARW analyses reveal the important role of the Chesapeake Bay for evolution of the 7 July 2004 storm. Chesapeake Bay acted as a thermal boundary that impeded the eastward propagation of the storm. It also acted as a divergence zone for the southerly winds enhancing convergence over Baltimore. “Removal” of the water boundary in the model simulations results in an eastward shift of the rainfall distribution. The “Bay” effect is more pronounced in the late afternoon hours acting as a “tongue” of lower temperatures and lower water vapor mixing ratios.

[42] • The rainfall and lightning distribution around the Baltimore metropolitan region (and Washington DC) and also the evolution of thunderstorm cells around Baltimore suggest that the urban environment played an important role in the local evolution of the 7 July storm over Baltimore. Storm total rainfall and lightning wrap around the western boundaries of Baltimore and Washington DC. Storm tracking analyses using 1-min, high-resolution reflectivity observations from the TWDR radar show that thunderstorms elements split around the Baltimore metropolitan region with one branch moving rapidly along the western boundary of the city and the second branch remaining stationary along the southern boundary of the city. Model analyses depict alterations of the low-level flow field in the urban area through frictional effects associated with urban canopy. Model analyses also show that the Washington DC metropolitan had an impact on the distribution of zones and convergence and divergence west of Baltimore as the storm systems developed. Alterations of the low-level wind field in the urban core of Baltimore interacted with thunderstorm outflows to determine the anomalous storm motion around the west side of the Baltimore region. The Urban Heat Island effect was not found to have an impact on the evolution of the storm due to the strong low-level winds that smoothed out any temperature anomalies over the urban area.

[43] • Extreme rainfall rates for the 7 July storm are associated with efficient precipitation mechanisms. Disdrometer observations show that peak rainfall rates are associated with anomalously high drop arrival rates. Lidar observations reveal layers of elevated aerosols in the prestorm environment. Lidar observations and model time series of the vertical profile of relative humidity that show an almost completely saturated water column, draw direct links between efficient precipitation mechanisms and the role of aerosols under moist convective conditions.

[44] • The influence of the urban environment on the evolution of the 7 July 2004 storm was not characterized by a “downwind” (northeastern) amplification of rainfall, as is typical in studies like the METROMEX experiment. The largest impact on rainfall distribution was on the western margin of the city. This result is not inconsistent with those from METROMEX, because complexity of the terrain of the northeastern corridor significantly alters storm evolution and dynamics. The Baltimore results provide insights into the question of how topographically complex urban regions affect the evolution of warm season thunderstorm systems and the associated distribution of rainfall and flash flooding.

[45] • The combination of mesoscale modeling and analysis of high-resolution remote sensing data sets provides a powerful tool for examining rainfall variability in complex settings like the Baltimore metropolitan region. As computational resources improve, it would be useful to examine climatological analyses of rainfall at high resolution derived from both mesoscale modeling and remote sensing sources, like the Hydro-NEXRAD system.

Acknowledgments

[46] The authors are pleased to acknowledge that the mesoscale model simulations reported in this paper were performed at the TIGRESS high performance computer center at Princeton University which is jointly supported by the Princeton Institute for Computational Science and Engineering and the Princeton University Office of Information Technology. The authors would also like to acknowledge Professor Ngar-Cheung (Gabriel) Lau of the GFDL and Dr. Jimy Dudhia of the NCAR for their valuable recommendations toward the completion of this paper and Ray Rogers for collecting the lidar data used in this study. Also, the authors would like to thank the WRF-help team for providing valuable feedback on various issues relating to WRF-ARW. NLDN data provided by the NASA Lightning Imaging Sensor (LIS) instrument team and the LIS data center via the Global Hydrology Resource Center (GHRC) located at the Global Hydrology and Climate Center (GHCC), Huntsville, Alabama, through a license agreement with Global Atmospherics, Inc. (GAI). The data available from the GHRC are restricted to LIS science team collaborators and to NASA EOS and TRMM investigators.

[47] This research was supported by the National Science Foundation (NSF grants EEC-0540832, EAR-0208269, EAR-0409501, and ITR-0427325).

Ancillary