TMCnet News
The Evolution of Convective Storms Initiated by an Isolated Mountain Ridge [Monthly Weather Review](Monthly Weather Review Via Acquire Media NewsEdge) ABSTRACT The evolution of convective storms over the Black Hills, an isolated mountain ridge in South Dakota and Wyoming and a regional convection hotspot, is investigated using a 10-yr observational climatology and quasi-idealized numerical simulations. Radar-observed diurnally forced mountain-convection events are classified according to their maximum cell-track length and duration, which are quantified using an automated cell-tracking algorithm. Environmental conditions during these events are obtained from operational radiosonde and model-analysis data. These data suggest that mountain-forced convective cells generally struggle to survive in the convectively inhibited flow downwind of the Black Hills. Those cells that do survive downwind prefer environments with strong bulk vertical shear over the 0-6-km layer, which favors organized multicellular or supercellular convection. Under slightly weaker shear, the cells tend to dissipate rapidly as they propagate downwind. Relatively weak winds aloft, when coupled with low-level winds aligned with the long terrain axis, support longer-lived, quasi-stationary cells with flash-flooding potential. The weak winds favor slow cell propagation while the along-ridge flow limits the negative feedbacks of storm outflow on the elevated convergence over the ridge, allowing convection to repeatedly initiate in the same location. The storm evolution is relatively insensitive to the background thermodynamic profile, provided that sufficient moist instability exists to support deep convection. Convection-permitting numerical simulations reinforce that changes in the background wind profile alone can explain the observed variations in cell evolution. They also suggest that the longevity of convective cells downwind of the ridge is sensitive to terrain-induced modifications to the vertical wind shear. (ProQuest: ... denotes formulae omitted.) 1. Introduction Mountains are highly effective at initiating cumulus convection, both by forcing mechanical uplift and by acting as elevated heat sources that drive upslope flow (e.g., Banta 1990). The prevalence of mountain con- vection has motivated recent field campaigns such as the Mesoscale Alpine Programme (Bougeault et al. 2001), the Cumulus Photogrammetric, In Situ, and Doppler Ob- servations (CuPIDO) experiment in southeastern Arizona (Damiani et al. 2008), the Convective and Orographically Induced Precipitation Study (COPS) in central Europe (Wulfmeyer et al. 2008), and the Dominica Experiment (DOMEX) over the Caribbean island of Dominica (Smith et al. 2012). Although these and other projects have intensively investigated convection initiation over complex terrain, they have generally paid less attention to the subsequent storm evolution. As a consequence, the factors controlling the latter remain less understood. Houze (1993) suggested that once mountain-forced convective storms initiate, they may behave similarly to convection over flat terrain. Fortunately, a large body of knowledge already exists on this topic. Of particular importance to the issue of storm maintenance are the strengths of moist instability and vertical wind shear. Weisman and Klemp (1982) used the bulk Richardson number (BRN): ... (1) where CAPE is the convective available potential en- ergy of a low-level air parcel and DU is the magnitude of the bulk wind difference between low- and mid- to upper levels (over the 0-6-km layer above ground level), as a nondimensional control parameter to distinguish dif- ferent cell structures. Provided sufficient CAPE exists to support deep convection, weak vertical shear (BRN . ;500) favored single-cell storms, moderate shear (BRN ; 50-500) favored multicell storm complexes, and strong shear (BRN ; 10-50) favored rotating supercells. Be- cause the latter two modes are able to maintain them- selves for prolonged time periods, stronger shear thus favored longer-lived convective storms. While the findings of Weisman and Klemp (1982) have largely stood the test of time, recent studies suggest that BRN is not the only factor influencing storm lon- gevity. Bunkers et al. (2006b) found that in addition to low BRN, stronger 0-8-km bulk vertical wind shear, lower lifting condensation levels (LCLs), and certain mesoscale to synoptic-scale environments favored longer- lived supercells. Moreover, some have since debated the precise role of moist instability on storm type. Observa- tional evidence suggests that the product (rather than the quotient) of moist instability and vertical shear may better predict storm type than BRN (e.g., Rasmussen and Blanchard 1998). Given the large variations in environmental conditions over complex terrain, the applicability of traditional metrics like BRN (which implicitly assume horizontal homogeneity) to mountain convection is not straight- forward. The convective boundary layer extends higher into the atmosphere over mountains than over the plains, which may locally erode convective inhibition. The land surface may also abruptly change over mountains (e.g., from farmland/prairie to forest), modifying the surface energy balance and the local stability. Furthermore, mountain convection is promoted by strong boundary layer ascent over the terrain feature (e.g., Banta 1984; Kirshbaum 2013). Clouds that rely on this forcing for their initiation may decay downwind if they cannot generate comparable low-level circulations (e.g., from cold pools or internal pressure perturbations). Finally, vertical wind shear may be locally enhanced by obstacle effects and/or thermal circulations, which may enhance storm organization exclusively near the orography (e.g., Houze et al. 1993). A unique characteristic of mountains is their tendency to lock convection in place, which raises the likelihood of flash flooding (e.g., Akaeda et al. 1995). Various studies have investigated quasi-stationary convection driven by mechanical forcing, where impinging airflow is forcibly lifted by a terrain obstacle. Maddox et al. (1978) found that two flash floods over the western United States (the Rapid City flood of 1972 and the Big Thompson flood of 1976) were associated with strong, conditionally un- stable cross-barrier flow at low levels, combined with unusually weak winds at upper levels. The heavy pre- cipitation was generated by sustained windward ascent at preferred locations followed by slow cell translation. Ducrocq et al. (2008) found that evaporatively cooled storm outflow sustained quasi-stationary convection upwind of the Massif Central during the French Gard flood of 2002 by acting as a barrier to impinging south- erly flow. To identify the general parameters controlling such mechanically forced orographic convection, recent stud- ies have investigated conditionally unstable flow over an idealized ridge using numerical simulation (Chu and Lin 2000; Miglietta and Rotunno 2009, 2012). These studies have found that, at least for flows with moderate to large CAPE, the system evolution depends on the interaction between deep convection and its evaporatively driven cold pool. For a given mountain shape and moist sta- bility, this evolution is largely dictated by the back- ground cross-barrier wind profile. For weak low-level winds, the cold pool propagates upstream, as does the locus of convection initiation. Under strong low-level winds and weak winds aloft (and hence strong reverse vertical wind shear), the upstream cold-pool propaga- tion is balanced by background-wind advection, pro- ducing a quasi-stationary cold pool that repeatedly initiates convective cells upwind of the ridge [as in Ducrocq et al. (2008)]. For stronger winds over a deeper layer, the cold pool is swept downwind, allowing quasi- stationary convection to form directly over the high terrain. As a complement to the previous studies highlighted above, the objective of this study is to identify the en- vironmental controls on the evolution of diurnally forced mountain convection. Such convection, which is driven primarily by thermally forced circulations and/or ele- vated destabilization over heated terrain, dominates the summer climatology of many continental, midlatitude mountain ranges (e.g., Banta 1990). In particular, we aim to identify the environmental characteristics supporting two types of high-impact events: (i) quasi-stationary, terrain-locked storms with flash-flooding potential and (ii) long-track cells that may affect populated areas downwind. To this end, we perform a radar-based ob- servational climatology and numerical investigation of convective storms over the Black Hills mountains of South Dakota and Wyoming, located about 200 km east of the front ranges of the Rocky Mountains (Fig. 1). The Black Hills represents an excellent laboratory for the study of convective-storm evolution due to its rea- sonably simple, quasi-two-dimensional shape and the lack of significant terrain features nearby. Given its position in the northern Great Plains, it provides ex- cellent sampling of diverse convection events. Never- theless, with flat terrain to the east and the Rocky Mountains to the west, the ridge is embedded on a large- scale terrain gradient that influences the regional airflow. As a result of elevated daytime heating and nocturnal cooling over the Rockies, the low-level winds exhibit diurnal variations in strength and direction. In addition, the ridge may straddle a west-east moisture gradient where dry mountain air meets humid air to the east. On some occasions, a pronounced dryline separating these two air masses coincides with a local solenoidal circulation over the Black Hills, producing enhanced vertical motion over the crest. Kuo and Orville (1973) produced a 4-yr (1967-70) observational climatology of storm occurrence over the Black Hills using a combination of surface wind data and hourly radar observations. Consistent with Banta and Schaaf (1987), who studied storm genesis over the Rocky Mountain Front Range, they found that storms prefer- entially formed in the southeast or northeast quadrants of the ridge under northwesterly or southwesterly winds, respectively. In contrast, we focus on cell duration and track length, and the relationship of these characteristics to the atmospheric vertical profile. This requires fine- resolution radar data as well as upper-air profiling mea- surements that were not available to Kuo and Orville (1973). Section 2 describes the observational data sources and analysis methods, including a simple automated rou- tine used to determine cell-track duration and length. Section 3 presents the cell-tracking results, classifies the events based on their meteorological characteristics, and compares the background conditions for the dif- ferent event classes. In section 4, hypotheses derived from the observational analysis are critically evaluated using high-resolution numerical simulation. Conclusions are presented in section 5. 2. Observations Our observational climatology uses two main data sources: (i) operational radar data to identify convection events and analyze their cell tracks and (ii) radiosonde data to characterize the background environmental conditions during the events. We also considered oper- ational model analyses [from the National Centers for Environmental Prediction's (NCEP's) North American Mesoscale Model (NAM)] and reanalyses [NCEP's North American Regional Reanalysis (NARR)] to provide some larger-scale context for the events. How- ever, direct comparisons of wind and thermodynamic profiles between these data and the Rapid City, South Dakota, soundings revealed large errors, particularly for the NARR data and the NAM thermodynamic profiles (Soderholm 2013). Because the NAM wind profiles were more consistent with observations, we do incorporate some of these data into our investigation. a. Radar data We obtained operational Next Generation Weather Radar (NEXRAD) data from the KUDX station at Rapid City (Fig. 1) from the National Oceanic and At- mospheric Administration's (NOAA's) National Cli- matic Data Center (NCDC) data server over the 10-yr period 2003-12. This enabled the tracking of convective- cell motion for a large number of events and estima- tion of precipitation accumulations using the default NEXRAD relationship Z 5 300R1.4, where Z is the re- flectivity and R is the precipitation rate (in mm h21). Because this Z-R relationship often underestimates precipitation rates in the western United States (e.g., Petersen et al. 1999), the precipitation is likely under- estimated on average. On the other hand, because we did not account for reflectivity enhancements associated with hail, the precipitation may be overestimated within hailstorms. We use composite reflectivity (the maximum reflectivity in the vertical column) for the cell tracking and base reflectivity (at 0.58 elevation angle) for the precipitation calculations. We also use the NCDC's echo-top product to estimate each cell's maximum echo- top height over its lifetime. 1) CELL TRACKING We began by qualitatively inspecting composite radar reflectivity for the summer months (June-August) of each year to identify days where isolated convection de- veloped directly over the Black Hills between the morning and early evening hours [1500-0300 UTC, or 0900-2100 mountain daylight time (MDT)]. This time period places the focus on thermally forced convection, although it does not rule out mechanical forcing. Days with no precipitation, large-scale stratiform precipita- tion, widespread convection initiation, and preexisting disturbances crossing the Black Hills region were re- moved from consideration. Because frontal precip- itation and organized convection events were largely ruled out, our climatology omits some of the heavier precipitation events over the region. Of the 92 days per year that were considered, around 30 developed signif- icant precipitation, of which approximately half fea- tured localized cell initiation over the Black Hills. Each of these days (numbering 137 in total) is considered a unique convection event. For each event, we analyzed composite reflectivity fields using an automated storm-tracking algorithm, deployed over 1500-0300 UTC. To ensure maximum control over its operation, we developed our own code rather than using existing software like the Thunder- storm Identification, Tracking, Analysis and Nowcasting (TITAN; Dixon and Weiner 1993). Nonetheless, our procedure shares many of the same principles. It first converts the reflectivity data from radar coordinates onto a Cartesian grid with 1-km resolution. For each radar scan, it surveys the entire radar area (a circle centered on Rapid City with a 230-km radius) for con- vective cells. It identifies individual cells as regions of interconnected points with reflectivities exceeding 40 dBZ and areas exceeding 10 km2. This reflectivity threshold, which corresponds to a rain rate of approxi- mately 12 mm h21, is commonly used to distinguish deep convection from stratiform precipitation in con- tinental, midlatitude regions (e.g., Johnson et al. 1998). Comparison of 35- and 40-dBZ thresholds for a sam- pling of 2012 events revealed virtually no differences in cell tracks, suggesting that the results are not highly sensitive to the threshold. The area threshold is used to avoid ground clutter and focus on larger deep-convective cells. The routine analyzes the radar scans chronologically, identifying all cells at each new scan time and then checking for overlap with cells from the previous scan. When two cells at successive times exhibit overlap, the identification (ID) of the latter is reassigned to that of the former. When a cell does not overlap with any prior cells, it is assigned a new ID and considered newly initiated. Repeated over the entire event, this pro- cedure allows for the construction of cell tracks, for which the precise cell location at any time is given by its reflectivity-weighted mean (applied just to the points exceeding 40 dBZ). Because the algorithm does not distinguish between isolated cells and multiple inter- connected cells, an individual cell track may corre- spond to a single cell and/or a multicell cluster. To limit consideration to mountain-forced convection, only cells that initiate inside the rotated rectangular region encompassing the ridge in Fig. 2 (or the ''analysis box''), are considered. These cells are tracked until dissipation or propagation outside the radar scan area. To restrict analysis to deep convection, cells must at- tain a maximum echo-top height of at least 9 km. Be- cause most cells surpassing 40 dBZ easily exceed this height, only about 5% of the cell tracks were omitted on this basis. The assumption underpinning the tracking algorithm is that the radar scan frequency is sufficient for the cells of interest to overlap in successive frames. For a 5-min scan interval and a cell diameter of 10 km, this requires a translation speed of less than 35 m s 2 1. While the possibility exists that some of the smaller, faster-moving cells were not continuously tracked throughout their lifetimes, inspection of all 137 events ensured that the algorithm generally functioned properly: only in 5 events were cells identified by eye not tracked continu- ously from initiation to dissipation. These cells were reclassified manually based on radar animations. Simple routines are used to detect cell splits and mergers, which allow tracks to be maintained despite transformations in cell identity. Splitting is detected whenever two distinct cells at one time overlap with a common cell at the previous time; merging corre- sponds to the opposite. The track length and duration of any cell that undergoes splitting and/or merging is traced back to the very first cell(s) that formed in that sequence of events. Because the evolution of convective cells is often chaotic, many splits were detected that do not qualitatively conform to the classical picture of splitting supercells (e.g., Weisman and Klemp 1984). 2) CELL CLASSIFICATION We partitioned the events into three categories ac- cording to their potential hydrometeorological impacts (Table 1). The short-track long-duration (STLD) cate- gory corresponds to quasi-stationary cells that travel short distances over long time periods and thus possess high flash-flooding potential. The criterion for STLD is that at least one cell must travel less than 25 km over a 90-min period, which corresponds to a translation speed of 4.6 m s21. The 25-km distance corresponds to the cross-ridge length scale of the Black Hills (Fig. 2) and 90 min is sufficiently long for heavy precipitation to accumulate. For a stationary cell with a characteristic rain rate of 50 mm h21, this duration yields a 7.5-cm accumulation. The long-track long-duration (LTLD) category corresponds to storms that propagate far downwind, with the potential to cause severe weather over populated regions. For this category, at least one cell must survive for at least 120 min and travel at least 100 km. This time threshold is 2-4 times the characteristic duration of a single-cell thunderstorm (e.g., Doswell 2001) and the distance threshold is four mountain half- widths, indicating propagation far downwind. Finally, events in which no cells satisfy the STLD or LTLD cri- teria are classified as short-track short-duration (STSD). This type of event, which is by far the most common, is considered the least likely to produce severe weather. Although the STLD and LTLD classifications are not mutually exclusive, we did not encounter any ''hybrid'' events that satisfied the criteria of both. While justified by basic meteorological consider- ations, these classifications are admittedly arbitrary. The observed convection events exhibited a continuum of track lengths and durations, suggesting that real events do not naturally conform to our trimodal distribution. This is not surprising given that our criteria are based on macroscale event outcomes rather than fundamental as- pects of cell structure. Because the results obtained herein are sensitive to the thresholds chosen, we do not base our scientific conclusions exclusively on the observational analysis. However, the trends that emerge from this anal- ysis facilitate physical hypotheses that can be tested by numerical simulation. b. Soundings To characterize the environmental conditions during the convection events, we used radiosonde data from Rapid City (or RAP) downloaded from the University of Wyoming's online archive. These soundings are nor- mally available twice daily at 1200 and 0000 UTC (0600 and 1800 MDT) and occasionally at other times (e.g., 1800 or 2100 UTC). Because these data are only avail- able at a point location in the lee of the Black Hills (Fig. 2) at widely spaced time intervals, they may not always accurately represent the conditions encountered by developing clouds. Nevertheless, such uncertainties are tolerable given that these soundings represent the best available vertical profiling observations in the Black Hills vicinity. For each event, we first found the available sounding closest in time to the first-cell initiation. To ensure that the sounding was supportive of deep convection (and to remove soundings contaminated by convective outflow), we required mean-layer CAPE . 100 J kg21 (based on a parcel initialized with the mean conditions over the lowest 500 m) or for a conditionally unstable layer deeper than 2 km to exist above the boundary layer. If a sounding failed to satisfy these criteria, we discarded it and re- peated the analysis on the next closest sounding to the initiation time. If no soundings between 1200 and 0300 UTC passed this test, the event was discarded. Seven events were discarded on this basis, leaving a total of 130 with a corresponding radiosonde profile. Because of its availability and closeness to the initiation time, the 0000 UTC sounding was selected for 89% of the cases, withthe 1200 UTC(6%) and 1800UTC (4%)composing most of the remainder. 3. Observational results a. Examples from each category For a qualitative picture of the differences between the three event types, we provide radar snapshots from each category. An archetypal STSD event occurred on 2-3 July 2012, where convective cells formed just east of the Black Hills crest starting at around 2100 UTC. At 2151 UTC three cells were apparent, which are numbered based on their time of origin (Fig. 3a): C3 had just ini- tiated while C2 and C1 had already existed for 25 and 60 min, respectively. From 2151 to 2223 UTC C1 dissi- pated while C2 and C3 moved approximately 30 km to the east (Fig. 3b). Both C2 and C3 were weakening by 2251 UTC (Fig. 3c) and dissipated completely by 2318 UTC. The cell tracks for this event, which mostly originate just to the east of the main ridgeline, do not extend outside the analysis box (Fig. 3d). Using the default NEXRAD Z-R relationship cited earlier we obtain a modest maximum precipitation accumulation of 1.5 cm within the analysis box for this event. An archetypal STLD event occurred on 30 June-1 July 2012, where an intense isolated cell (C1) formed over the southern edge of the ridge at 1930 UTC. This cell enlarged and intensified to attain a maximum re- flectivity exceeding 70 dBZ by 1943 UTC (Fig. 4a). Over the next 50 min, C1 traveled only 8 km south-southeast from its initiation point while sustaining its high re- flectivity (Fig. 4b). Forty minutes later, C1 finally began to detach from the ridge and dissipate while a new cell (soon to be C2) began to develop in its original place (Fig. 4c). From 1930 to 2116 UTC, C1 traveled only 22.3 km, satisfying the criteria for STLD. The cell tracks reveal extremely tight clustering at the southern edge of the Black Hills, with the longest-lived cell C1 shown in blue (Fig. 4d). This event produced the largest pre- cipitation accumulation within the analysis box (8.8 cm) of the entire database. An archetypal LTLD event occurred on 27-28 August 2011, where cell C1 initiated along the southern ridge flank at around 2106 UTC. Approximately 20 min later (2134 UTC), C2 initiated roughly 75 km to the north along the ridgeline (Fig. 5a). Over the next 30 min, both cells traveled southeastward with maximum reflectivities exceeding 60 dBZ. At around 2201 UTC C2 split into a left mover moving due eastward (C3) and a larger right mover (retaining the C2 label) moving southeastward (Fig. 5b). By 2224 UTC, C1 also split into a left mover (soon to be C4) and a right mover retaining the C1 label. The left movers had mostly dissipated by 2306 UTC while the right movers remained coherent and intense (Fig. 5c). C1 (C2) dissipated at 0100 (0038) UTC the following day, lasting 234 (184) min and traveling 134 (126) km. The cell tracks in Fig. 5d show the two isolated, long- track supercells initiating over the high terrain and propagating southeastward over the plains. The maxi- mum estimated precipitation accumulation within the analysis box for this event was 5.4 cm. Some insight into the environmental conditions re- sponsible for the different cell evolutions in these three cases is provided by radiosonde hodographs in Fig. 6, each at 0000 UTC on the evening of the convection event. While the STSD event of 2-3 July 2012 exhibits moderate shear and weak clockwise curvature, the STLD event of 30 June-1 July 2012 exhibits relatively weak winds at all levels, along with northerly low-level flow. Storm motion in both of these cases is slower than the mean 0-6-km winds, suggesting that the terrain may serve to slow the downwind cell propagation. The LTLD event of 27-28 August 2011 exhibits a straight hodo- graph with much stronger vertical shear than the other two cases, along with deviant rightward cell motion. b. Summary of cell-tracking results Basic statistics from the 10-yr database are compared for the three event categories in Table 1, showing that 77 cases were classified as STSD, 28 as STLD, and 25 as LTLD. Thus, the majority of events do not produce quasi- stationary cells or long-track cells. While the mean ini- tiation times of the first cell (tinit) are similar for all three event types (around 1900 UTC), the mean dissipation times of the final cell (tdiss) differ significantly: convec- tion dissipates earlier for the STSD events (2322 UTC) than for the STLD events (0054 UTC) or the LTLD events (0106 UTC). As expected from the event classi- fication scheme, the maximum track duration (Dt)max and track length (Dd)max are largest for the LTLD events, while (Dt)max is also relatively large for the STLD events. The average maximum radar-derived rainfall accumulation (Pmax) within the analysis box is the largest in the STLD and LTLD events (3.2 cm) and much less in the STSD events (1.6 cm). Table 1 also provides the fraction of events where cells that formed within the analysis box evolved into super- cells. The supercell identification is performed manually based on reflectivity data following Bunkers et al. (2006a), who used multiple criteria to distinguish these cells (hook echoes, bounded weak echo regions, deviant motion, etc.). In cases where supercellular features were not obvious or only briefly evident, the events were clas- sified conservatively as nonsupercellular. The largest su- percell fraction (60%) was observed in the LTLD cases, followed by the STSD (30%) and STLD (21%) events. c. Environmental conditions 1) THERMODYNAMIC PROFILES To evaluate whether the thermodynamic conditions systematically differed between the three event categories, we compare their composite temperature and dewpoint profiles in Fig. 7. Remarkable similarities are apparent: all three are conditionally unstable below 600 hPa and transition to moist neutral up to the tropopause. Based on a mean-layer parcel initialized with the mean ther- modynamic conditions over the lowest 500 m, we obtain slightly larger CAPE for the LTLD (613J kg21) and STLD (654 J kg21) soundings than for the STSD sound- ing (502 J kg21). The CIN is similar for all three cate- gories (;140 J kg21), suggesting that convective cells encounter a hostile environment as they move off the ridge. The LCLs of the STLD (2120m AGL) and LTLD (2019 m AGL) soundings are slightly lower than that of the STSD (2444 m AGL) sounding, which is consis- tent with previous findings that lower LCLs are more supportive of longer-lived convective cells (e.g., McCaul and Cohen 2002; Bunkers et al. 2006b). Similar results are obtained by computing convective parameters individually for each sounding and then av- eraging the results over each event class, conditioned on soundings with CAPE . 100 J kg21 (Table 2). We obtain average mean-layer CAPE (CIN) of 808 (121), 1095 (154), and 907 (111) J kg21 for the STSD, STLD, and LTLD events, respectively. As with the composite soundings, the LCLs are again the highest for the STSD events and slightly lower for STLD and LTLD events. In addition, the mean precipitable water vapor (PWV) varies by less than 1% between the three event classes. Of these subtle differences, the only statistically signif- icant one at the 95% confidence level is the higher LCL in the STSD soundings, compared to the STLD and LTLD soundings. The above moist-instability calculations at RAP likely represent lower bounds for the Black Hills region. The sounding location in the lee of the Black Hills typically places it in the descending branch of mechanical and/or thermal orographic circulations, both of which produce adiabatic warming aloft. Also, because the most com- mon sounding time is 0000 UTC (typically 4-5 h after convection initiation), thunderstorm outflow may some- times contaminate the sounding. The filtering described in section 2b likely removes some, but not all, of these contaminated soundings. This combination of adiabatic warming aloft and/or low-level cooling in some sound- ings likely imparts a stable bias to the mean climatology. 2) WIND PROFILES To compare the background wind profiles for the three different event types, we present wind roses av- eraged over three different layers defined by their height above ground level (AGL): 0-2 km (low levels), 2-5 km (midlevels), and 5-11 km (upper levels) in Fig. 8, along with layer-averaged wind speeds jvji in Table 2, where i denotes the layer of interest. Although for obvious reasons we place the higher-level winds in the upper panels of the figure, we begin our discussion with the lower-level winds and progress upward. Over the 0-2-km layer (Figs. 8g-i) the winds have similar mean speeds (around 5 m s21) and alignments (north-south) for all event types. This preferred wind orientation, which is perpendicular to the local east-west terrain gradient at Rapid City (Fig. 2), suggests that channeling around the Black Hills is the dominant low-level wind signature at Rapid City. This effect likely renders the low-level winds from the RAP soundings unrepresentative of the true background flow conditions. To obtain more representative background low-level wind estimates, we use NAM model analyses, taken at the same times as the corresponding radiosonde data. Although model-analysis data are inherently uncertain, they characterize the mesoscale flow pattern surround- ing the Black Hills more reliably than a single downwind point sounding. For this analysis, we first define a rect- angular region encompassing the Black Hills with a width of 2.48 and height of 1.6 8 , centered at 44 8 N, 103.88W (not shown). Along each edge we create a five- point evenly spaced stencil and compute the mean 0-2-km AGL NAM winds at each point. We then average these values around the rectangle's perimeter to obtain a rough estimate of the upstream wind direction ( b). Based on b we assign the upstream wind to one of the eight principal directions (northerly, northeasterly, easterly, etc.) separated by 458 increments. We then recompute the mean wind vectors over just the five points along the upstream edge of the perimeter, which filters out the downwind flow perturbations induced by the terrain. For example, if b 5 228 (measured clockwise from due north), the winds are considered to be north- erly and we average along the northern edge. However, if b 5 238 the winds are considered to be northeasterly and we average over the northeastern corner, including the three northernmost (easternmost) points of the eastern (northern) edge, including a shared corner point. Following the above procedure we obtain NAM- based 0-2-km wind vectors for each case, from which we create wind roses in Figs. 9a-c analogous to those for the sounding data in Figs. 8g-i. As in the sounding data, no significant differences in low-level wind speeds are apparent: the mean STSD, STLD, and LTLD wind speeds are 4.5, 4.4, and 4.6 m s21, respectively. However, in contrast to the sounding data, the winds are no longer oriented exclusively north-south. The winds for the STSD events exhibit no dominant low-level direction (Fig. 9a) while the winds for the LTLD events favor a northwest-southeast orientation (Fig. 9c) that is offset about 308 from the long axis of the ridge (approximately 158 counterclockwise of due north in Fig. 2). By contrast, the winds in the STLD events are predominantly north- northwesterly or south-southeasterly (Fig. 9b), aligning very closely with the ridge axis. This suggests that low-level winds aligned with the long terrain axis may favor terrain-locked convection, an important result that will receive further attention. At midlevels, where the horizontal winds are typically less affected by the orography, the flow becomes pre- dominantly westerly, with significant differences in mean wind speed between STSD (9.1 m s21), STLD (6.5 m s21), and LTLD (10.8 m s21) events (Figs. 8d-f and Table 2). These differences suggest that stronger vertical shear favors more organized, self-sustained convective cells and weaker winds aloft favor slow-moving or quasi- stationary convective cells. The trends in the midlevel winds are largely reinforced at upper levels, where the LTLD events possess the strongest winds and STLD events the weakest (Figs. 8a-c and Table 2). These layer- averaged wind differences between the three event classes are all statistically significant at the 95% confi- dence level. The sensitivity of event type to vertical shear is borne out by direct calculations of wind-velocity differences DU over the 0-6-km layer AGL, often considered to be the most relevant layer for cell organization (e.g., Markowski and Richardson 2010). First we subtract the mean winds over the lowest 500 m AGL from the winds at 6 km AGL to give the bulk shear jDUj1. Averaging over all soundings for each event class, we obtain jDUj1 5 16.3, 13.8, and 21.7 m s21 for the STSD, STLD, and LTLD events, respectively. Another measure of vertical shear (jDUj2) is obtained by subtracting the mean wind over the lowest 500 m AGL from the density-weighted mean wind from 0-6 km AGL (Weisman and Klemp 1982), which gives magnitudes of 7.9, 6.3, and 10.1m s21 for the STSD, STLD, and LTLD events, respectively. These differences in vertical-shear magnitude are all significant at the 95% confidence level. To examine the relationship between BRN and cell- track length/duration, we compute BRN [using Eq. (1) with jDUj2 in the denominator] individually for each event and then average over each class, which gives values of 45 (STSD), 100 (STLD), and 20 (LTLD) (Table 2). The relatively large BRNs in the STLD events are associated with relatively short track lengths and durations. By contrast, the smaller BRNs in the LTLD events favor longer-lived multicells and supercells, which is consistent with their frequent supercell occur- rence (Table 1). Although based on Weisman and Klemp (1982, 1984), the STSD events also possess sufficiently small BRN to support supercells in the majority of events, very few of these events actually develop them. This may result from the horizontal inhomogeneity of the regional environment, namely, the large CIN (.100 J kg21) that cells encounter downwind. For cells to survive in such environments, stronger internal dy- namical forcing, and thus stronger background vertical wind shear, may be required. Table 2 also compares the median Froude number (Fr 5 U/Nh) of the three event classes, where U and N are the mean wind speed and static stability of the 200- 1200 m AGL layer and h 5 1200 m is the approximate Black Hills relief. An elevated layer base of 200 m for these layer averages helps to avoid surface-based tem- perature inversions and superadiabatic layers. Because of the lack of representativity of the low-level winds at RAP, we again use NAM data (following the averaging procedure described earlier) for the background condi- tions. Although the median Fr is subunity for all three event classes, a substantial number of events (36% overall) are characterized by Fr . 1. Thus, diurnally forced Black Hills convection events may occur within mechanically blocked or unblocked flows. d. Physical hypotheses Because of limitations such as poor representativity of the soundings, arbitrary thresholds in the radar-based event classification, and the neglect of synoptic-scale forcing in the climatology, the observational results above are not definitive. However, the signals that emerged provide a basis for physical hypotheses that can be tested by numerical simulation. A first hypothesis is drawn from the finding that, apart from small variations in the mean LCL, no significant thermodynamic differences were found between STSD, STLD, and LTLD sound- ings (Fig. 7 and Table 2). We thus hypothesize that ther- modynamic differences are not the primary cause for the differing cell evolutions in the three event classes. However, significant differences in the vertical wind profiles and BRN were identified: (i) LTLD events were favored by stronger winds aloft, stronger vertical shear, and low BRN; and (ii) STLD events were favored by weaker winds aloft, weaker vertical shears, and larger BRN. Given that the LTLD events were also associated with a high frequency of supercells (Table 1), it is rea- sonable to suspect that the stronger shears and low BRNs in these cases favored more organized convective structures (supercells and multicells) that were better equipped to survive downwind. By contrast, the long- lived cells in the STLD events mostly developed in weaker-shear environments and were mostly not su- percellular. Because the weaker vertical shears were less favorable for long-lived cells, we suspect that the Black Hills terrain played a role in sustaining the cells and locking them in place. Specifically, the terrain may have favored repeated storm initiation in the same location. Although each cellular pulse may have been short-lived, the merging of radar echoes from successive training cells may have given the appearance of a single, quasi- stationary cell. A hypothesis for the STLD events is drawn from Fig. 9b, which shows that such events are favored by low- level background winds aligned with the ridge axis. We hypothesize that this particular wind orientation di- minishes the negative feedbacks of storm outflow on the thermally driven upslope inflow feeding the convection, allowing for persistent cell generation over the ridge. This physical mechanism is depicted schematically in Fig. 10, which compares cell development for cross- barrier versus along-barrier low-level winds (upper- level flow is typically westerly, so that is held fixed). For the sake of illustration, we assume that the impinging flow rises freely over the ridge and is thus unblocked (Fr . 1). However, the processes of interest are largely the same in blocked flows (Fr , 1) where the low-level air detours around the terrain. Figure 10a shows the impinging cross-barrier flow directly ascending the windward slope and then colliding with upslope flow on the lee. This upslope lee flow is only possible under weak low-level winds and strong solar heating, which allows the adverse lee pressure gradient (arising from diabatically generated low pressure over the crest) to overcome the fluid inertia and reverse the winds. As the airstreams converge over the high terrain, they separate from the surface and vent through the boundary layer to initiate moist convection (e.g., Kirshbaum 2011). The clouds mature and drift eastward, where their plunging cold pools replace the upslope flow over the lee slope (Fig. 10c). This acts to reduce the convergence over the crest and suppress further con- vection initiation until the upslope flow reestablishes (e.g., Demko and Geerts 2010). As a result of this neg- ative feedback, such cross-barrier flows are less favor- able for quasi-stationary mountain convection. For the case of along-ridge low-level flow (Fig. 10b), the air is drawn upslope as it parallels the ridge axis. This produces an organized line of convergence over the ridge crest, over which clouds initiate and deepen as they propagate lengthwise. As they mature, the clouds obtain a westerly component and begin to detach from the ridge. However, in contrast to the cross-barrier case, these clouds do not pass over the upslope flow that gave rise to them. Their outflow instead descends the down- wind side of the ridge and does not interact with the convergence line along the ridge axis (Fig. 10d). Because the circulation responsible for convection initiation is not disrupted by storm outflow, new cells can repeatedly ini- tiate in the same location. Under sufficiently weak upper- level winds, this can produce a continuous line of cells with a common anchor point or a quasi-steady terrain-locked convective plume. Thus, we expect the combination of along-barrier low-level winds and weak winds aloft to be more favorable for quasi-stationary, STLD events. For the handful of STLD cases that were super- cellular, it is possible that the long-lived convection was owing to the internal supercell organization (as opposed to the merging of multiple cellular pulses), and that the slow cell motion was owing purely to the background wind profile (as opposed to being anchored by the ter- rain). To examine this alternate hypothesis, we used the hodograph-based technique of Bunkers et al. (2000) to predict the deviant motion of the right and left movers for each such event using its corresponding RAP sound- ing. We found that in only one of the STLD supercell events was the predicted motion of either supercell slow enough to satisfy the STLD criteria (#4.6 m s21). The fact that the observed cell translation was typically much slower than these predictions supports the hypothesis that the mountain ridge is pivotal for locking the cells in place. 4. Numerical simulations a. Numerical setup The numerical simulations are quasi idealized in that they use a realistic terrain profile and background en- vironmental conditions but exclude many of the com- plexities of real flows. They are conducted with the Bryan Cloud Model version 16 (Bryan and Fritsch 2002), a compressible, fully nonlinear, and nonhydrostatic mesoscale model. For time integration a split time step is used, with a larger time step for the integration of mete- orological processes and a smaller time step to maintain stability of acoustic modes. We use sixth-order centered horizontal advection along with fifth-order vertical ad- vection with implicit diffusion. To diminish the spurious effects of poorly resolved waves, we apply sixth-order horizontal diffusion with a coefficient of 0.12. Unless otherwise specified, all simulations use the following subgrid parameterization schemes: a 1.5-order TKE sub- grid mixing scheme, the Morrison two-moment micro- physics scheme, and a bulk aerodynamic surface drag with a nondimensional drag coefficient representative of land surfaces (Cd 5 0.011). The Coriolis force is applied to perturbations from the geostrophically balanced initial state using the f-plane approximation with a Coriolis parameter of f 5 1024 s21. The domain has dimensions of Lx 5 480 km, Ly 5 480 km, and Lz 5 18 km in the x, y, and z directions, respectively, with open lateral boundaries and a 6-km- deep Rayleigh sponge layer at the model top. The grid resolution is Dx 5Dy 5 1 km in the horizontal, giving nominally 480 grid points in each direction. The spacing of the terrain-following vertical grid is Dz 5 100 m from the surface up to 4 km and 400 m from 8 km to the do- main top, with a linear stretching layer between 4 and 8 km, giving a total of 81 levels. The terrain at the lower boundary is based on the Black Hills but simplified to avoid undue complexity and minimize terrain forcing at the lateral boundaries, which can cause spurious waves to radiate into the do- main interior. As shown in Fig. 11a, the ridge is em- bedded on a gradually sloping terrain with a base height of around 1 km and a peak of around 2.2 km, giving a relief of about 1.2 km. We isolate the ridge by creating an ellipse encompassing it with axis lengths of ax 5 50 km and ay 5 100 km, rotated counterclockwise by a 5 358 (Fig. 11a), then conducting the following operation: ... (2) where horig is the original terrain height, r2 5 x02/a2x 1 y02/ay2 , and x0 and y0 are the axes of a Cartesian reference frame centered on, and rotated with, the ellipse. The hmod field contains a largely isolated ridge with some modest terrain to the west, most of which is below 1.2 km (Fig. 11b). To level the surrounding terrain, we raise it to a minimum height of 1.2 km. We then then rescale the ridge to a peak height of 2.4 km to recover its original relief (Fig. 11c) and filter out all scales below 6Dx to avoid forcing the model at short wavelengths (Fig. 11d). Finally, we lower the entire terrain field by 1200 m so that the plains surrounding the ridge are at zero elevation. The simulations are initialized at 0600 local time (1200 UTC) using the mean 1200 UTC RAP thermody- namic sounding from the 41 events over 2010-12 (Fig. 12a). We did not use the full 10-yr database because the sim- ulations were conducted prior to its completion. How- ever, this sounding was very similar to the full 10-yr mean (not shown), so it was deemed satisfactory. As with the composite soundings in Fig. 7, it contains a conditionally unstable layer from 800 to 600 hPa and a tropopause at around 200 hPa. However, because of the early morning timing, the boundary layer is stably stratified with a noc- turnal inversion. By neglecting larger-scale variations associated with the Rocky Mountains as well as spatial moisture gradients that often exist in the region, this horizontally homogeneous initial flow focuses the analy- sis on the mesoscale dynamics of the mountain ridge. The model vertical coordinate is defined relative to ground level rather than sea level, so that both the flat terrain surrounding the ridge and the base of the initial sounding are at zero height. A plausible diurnal cycle is obtained by applying horizontally uniform and sinusoi- dally time-varying surface sensible heat and moisture fluxes with an amplitude of 250 W m22, similar to that found in full-physics simulations over the Black Hills with the Weather Research and Forecasting Model (WRF) (not shown). Broad boundary layer destabilization over the domain, combined with elevated heating over the mountain, enables convection to develop in the after- noon hours. The simulations are integrated for 12 h to 0000 UTC (a half diurnal cycle) with data written to file every 5 min (similar in frequency to the radar scans). b. Experimental design We conduct seven simulations with varying back- ground wind profiles to evaluate the hypotheses from section 3d. These wind profiles are shown in Fig. 12b, of which the first (case 1, or control) corresponds to the average 1200 UTC winds from the 2010-12 events. Be- cause STSD events are by far the most common, we expect this generic profile to give rise to such an event. With southwesterly winds over 0-2 km, case 1 is char- acterized by weak cross-barrier low-level flow. Case 2 uses the same cross-barrier winds as case 1 over 0-2 km but halves the vertical speed shear over 2-5 km to evaluate whether weak winds aloft favor STLD events. Similarly, case 3 doubles the vertical speed shear over 2- 5 km to evaluate whether stronger vertical shear favors LTLD events. Cases 4-7 are designed to examine the impacts of both low-level wind direction and upper-level wind speed on the storm evolution. Cases 4-6 are identical to cases 1-3 except that the 0-2-km winds are rotated clockwise by 1208 to a north-northwesterly direction roughly aligned with long axis of the Black Hills. Over 2-5 km the wind rotation is linearly reduced to 08, leaving the winds un- changed above 5 km. Finally, case 7 is identical to case 5 except that, rather than being rotated clockwise by 1208, the low-level winds are rotated counterclockwise by 608 to assume a south-southeasterly orientation. The purpose for conducting both cases 5 and 7 is to critically evaluate whether the combination of barrier-aligned low-level winds and weak winds aloft robustly favors STLD events. Table 3 presents CAPE, CIN, jDUj1, jDUj2, and BRN for all seven simulations at the time just before con- vection initiation (1800 UTC) and the approximate po- sition of Rapid City (x 5 300 km, y 5 240 km). The range of values of CAPE, CIN, jDUj1, and BRN are generally consistent with the observational climatology (Table 2), suggesting that the simulated flows are realistic and reasonably capture the key differences in vertical wind profiles between the three event classes. The simulated range of jDUj2 (4-10 m s21) is slightly smaller than that observed (6-11 m s21), which likely arises as a conse- quence of the averaging used to create the wind profiles. The larger jDUj2 in cases 1-3, compared to cases 4-7, arises from stronger mountain-induced flow perturbations under cross-barrier low-level winds, which are described shortly. c. Results The cell-tracking algorithm described in section 2a is applied to composite reflectivity data from the numeri- cal simulations. With the aid of this analysis, we present a summary of the simulation results in Table 4, which provides tinit, tdiss,(Dt)max,(Dd)max, Pmax, and the clas- sification of each simulation (using the same cell length and duration thresholds as before). These results con- form well to the hypotheses proposed in section 3d: all cases are STSD except when the vertical shear is strengthened in case 3 (LTLD) and when low-level winds paralleling the ridge and weak upper-level flow are combined in cases 5 and 7 (STLD). Perhaps counterintuitively, case 6 does not produce any LTLD cells despite pos- sessing stronger initial vertical wind shear than case 3 (this result will be addressed shortly). Similar to the radar analysis, we describe the simu- lated cell evolution for three exemplary events using composite reflectivity (and surface wind vectors) in Figs. 13-15. In the STSD example of case 1, an intense cell (C1) emerges over the northeast side of the ridge by 1940 UTC, just downwind of a convergent boundary between westerly cross-barrier flow and reversed lee flow (Fig. 13a). Although turbulent and irregular, the flow pattern is broadly consistent with the schematic in Fig. 10a, with cross-barrier flow over the windward slope and regions of reversed flow over the lee, causing ele- vated convergence over the lee slope. Between 1940 and 2120 UTC, C1 dissipates to the east of the ridge while a new cell (C2) forms in its place, only to also dissipate downwind (Fig. 13b) as a third cell (C3) forms over the ridge. The weakly divergent wind signature beneath C3 reveals that the upslope flow is being increasingly re- placed by convective outflow. Intermittent cell forma- tion over the ridge continues until 2230 UTC (Fig. 13c), by which time the convective outflow has overwhelmed the upslope inflow and the mountain convection ceases. This event is classified as STSD because, although deep convection persisted for 3-4 h over and downwind of the ridge, no quasi-stationary or long-track cells developed (Fig. 13d). In the STLD example of case 7, the first cell (C1) forms over the northern ridge edge at 1840 UTC (Fig. 14a), significantly earlier than that in case 1. Convection ini- tiation is facilitated by the weak vertical shear as well as the along-ridge motion of developing cumuli, which are supported by strong and sustained subcloud ascent as they translate along the ridge axis. C1 is remarkably stationary and anchored to the same initiation point until 1940 UTC (Fig. 14b). By 2010 UTC C1 starts to detach from the ridge (Fig. 4c) but remains within 20 km of its initial location (and thus registering the event as STLD) as a new cell (C2) takes its place. Over the next 2 h, cells repeatedly initiate in the same location as C1, persist there for 1-2 h, then detach and dissipate down- wind. Despite the intensifying resolved turbulence be- tween 1840 and 2010 UTC, the mesoscale flow pattern remains broadly reminiscent of Figs. 10b and 10d: con- vergent upslope flow paralleling the ridge axis culmi- nates in cell initiation on the downwind side of the ridge, producing outflow with little feedback upon the upslope inflow. This contrasts with case 1, where the con- vective outflow effectively cut off the thermally driven inflow. In the LTLD example of case 3, the first cell (C1) develops at 2000 UTC, slightly after that in case 1 (Fig. 15a). This delay is caused by stronger vertical shear suppressing the initial development of deep convection. A similar pattern of cell initiation and downwind dissi- pation as in case 1 takes place until about 2100 UTC, where a recently initiated cell (C2) develops over the northeast edge of the mountain (Fig. 15b). Rather than dissipating, however, this cell continues to intensify as it travels downwind, eventually splitting into two cells (C2 and C3) at around 2210 UTC (Fig. 15c). Of the split cells, the left mover (C3) lasts only for about 30 min while the right mover (C2) lasts until the end of the simulation. Evidently, the stronger vertical wind shear in this sim- ulation favors longer-lived cells that sustain themselves within the strongly convectively inhibited flow down- wind of the terrain. Strong midlevel vertical vorticity and anomalously low pressure within cell C2 between 2100 and 2200 UTC (not shown) suggests that it is a su- percell over that time. As in case 1, divergent storm outflow interferes with the convergent upslope flow driving the convection, ceasing cell initiation over the ridge after 2100 UTC. d. Further discussion Although thermodynamic conditions (and their spa- tial gradients) are surely important, the simulations re- veal that variations in the background winds alone may largely explain the observed variations in the motion and longevity of convective cells initiated by the Black Hills. Prescribed changes to the vertical wind shear aloft and/or the low-level wind direction allowed for a full sampling of all three classes of storm evolution (STSD, STLD, and LTLD). Consistent with the observational analysis and well-established theory (e.g., Weisman and Klemp 1982), the stronger vertical wind shear in case 3 favored longer-lived cells that, subsequent to their initiation over the mountain, sustained themselves as they propagated into a hostile (large CIN) environment downwind. Similarly, consistent with expectation and case study analyses, weaker winds aloft favored quasi- stationary STLD events (Maddox et al. 1978). However, this was just one of two requirements for quasi-stationary convection-the background low-level winds also needed to parallel the long terrain axis. Because this finding is novel and holds implications for mountain flood fore- casting, it merits further investigation. In our conceptual diagram in Fig. 10, we argued that the along-ridge winds minimized the negative feedbacks of storm outflow on the thermally driven storm inflow, allowing for persistent convection on the downwind side of the ridge crest. To quantitatively evaluate this argu- ment, we compute the mountain-scale convergence (MSC), defined by Geerts et al. (2008) as the loop in- tegral of the surface normal wind component yn around the perimeter of an orographic feature: ... (3) where A is the area within the perimeter and ds is an incremental distance along the curve. We take the pe- rimeter to be the 250-m-height contour of the terrain, which encompasses the majority of the ridge. The MSC provides a measure of the integrated surface-based up- slope flow around the obstacle and is thus sensitive to the formation of cold pools that may temporarily diminish (or even reverse) this circulation (Demko and Geerts 2010). Based on our hypothesis from Fig. 10, the cases with cross-barrier low-level winds should experience a major reduction in MSC after the onset of deep con- vection while the cases with along-barrier low-level winds should see a much weaker effect. The comparison of MSC time series from cases 1, 3, and 7 in Fig. 16a indicates that, following model spinup, a divergent signature of upstream flow blocking emerges due to the initially stable boundary layer. As the surface heating increases, the MSC becomes positive and rea- ches a peak at around 1800 UTC (noon local time). Shortly thereafter convection initiation occurs, which abruptly diminishes the MSC in cases 1 and 3 but has little effect on case 7. Despite reduced sensible heating, the MSC in case 7 actually increases slightly over 1930- 2200 UTC, likely as a result of enhanced inflow into the quasi-stationary cell. Case 3 also experiences a resurgence in MSC at around 2300 UTC due to strong convective outflow from the supercell to the east (see Fig. 15c). However, this cold and stable surface-based flow does not initiate any new convection over the ridge. Overall, the development of deep convection severely curtails upslope flow in cases 1 and 3 but not in case 7. To isolate the role of latent heating on the MSC profiles, we compare MSC time series from dry versions of the same three simulations in Fig. 16b, revealing a diurnal MSC evolution very similar to the moist case until just after 1800 UTC. Thereafter the profiles con- tinue their sinusoidal evolution, decreasing back toward zero. This contrasts with the moist versions of the three cases, where the development of convective outflow causes the MSC to rapidly decrease in cases 1-3 and increase in case 7. Another interesting finding is that, despite possessing stronger initial vertical shear than case 3, case 6 does not develop any long-track cells. The main difference be- tween these two cases is their initial low-level wind di- rection, which in case 6 is rotated to align with the terrain. Prior to convection initiation, jDUj1 is similar between the two cases but jDUj2 is significantly different (Table 3). This is caused by the more strongly perturbed low-level winds in case 3, which recirculate toward the crest due to the adverse lee pressure gradient (Fig. 15). The orographically enhanced vertical shear in case 3 thus provides a more favorable environment for multi- cells and/or supercells. Case 3 also has a more supportive directional-shear profile for long-lived cells than case 6, which is demon- strated by the wind hodographs of Fig. 17. These profiles are taken from the grid point nearest to, and averaged in time over the 30 min prior to, first-cell initiation in each case. Although the case-3 hodograph is initially straight, it develops low-level clockwise curvature due to mountain- induced flow perturbations at later times (Fig. 17a). Consistent with Houze et al. (1993), the orographic en- hancement of directional shear in the lee enhances streamwise vorticity, which in this case favors right-moving supercells (Davies-Jones 1984). By contrast, case 6 exhibits hodograph curvature in opposite directions at different levels: the 0-3-km wind rotates counterclockwise while the 3-6-km wind rotates clockwise (Fig. 17b). Such a profile is less favorable for cell survival than that of case 3 because it favors vertically stacked internal pressure perturbations of alternating sign, producing less co- herent dynamical ascent. 5. Summary and conclusions In this study we have conducted a 10-yr observational climatology and numerical investigation of diurnally forced convective storms over the Black Hills (United States), an isolated mountain ridge with a peak height of 1.2 km above the surrounding terrain. The observational climatology used automated storm tracking to classify events into three categories based on their potential hydrometeorological impact: short-track short-duration (STSD), short-track long-duration (STLD), and long- track long-duration (LTLD). While STSD events pose little threat to life and property, the quasi-stationary cells in STLD events carry flash-flooding potential and the long-track cells in the LTLD events may threaten populated areas downwind. Of the 130 events consid- ered, the vast majority (77) were classified as STSD, with smaller but significant amounts in the STLD (28) and LTLD (25) categories. Analysis of radiosonde and model-analysis data revealed very subtle differences in thermodynamic pro- files and convective parameters between the three event types, suggesting that the vertical wind profile exerted more control over cell evolution. The LTLD events had the strongest mean vertical wind shear (21.7 m s21 over 0- 6 km AGL) and lowest mean bulk Richardson number (BRN) (20), while the STLD events had the weakest mean shear (13.8 m s21) and largest mean BRN (100). This is consistent with the expectation that, for a given moist instability, stronger shear favors better organized and longer-lived cells (e.g., Markowski and Richardson 2010) and that weak winds aloft favor quasi-stationary convection (e.g., Maddox et al. 1978). Although the mean shear in the STSD events was weaker than that for the LTLD events, the mean BRN (45) of these events was still low enough to support long-lived cell structures (Weisman and Klemp 1982). The inability of cells to survive downwind in these cases may have been tied to the strong convective inhibition there (CIN ; 130 J kg21). In such hostile environments, the cells may have re- quired a heightened level of internal organization (and thus stronger vertical shear) to survive. A new insight to emerge from the observations is that quasi-stationary STLD events are favored by low-level winds paralleling the long axis of the ridge. A conceptual hypothesis for this association (Fig. 10) focused on the interaction between the thermally driven upslope flow initiating the convection and the divergent (and typi- cally downslope) outflow created by the convective cells. When the low-level flow is directed across the barrier the outflow cuts off the upslope inflow ascending the lee slope, weakening the elevated convergence over the ridge and thus preventing new cells from initiating there. However, when the low-level flow parallels the barrier the cells and their outflow develop downwind of the elevated convergence giving rise to the convection, allowing cells to repeatedly form in the same location. A series of quasi-idealized, convection-permitting numerical simulations was conducted to evaluate the hypotheses derived from the observational analysis. These experiments, which were based on the mean conditions from three years of observed convection events, sys- tematically varied the initial wind profile while holding the initial thermodynamic profile and the diurnally varying surface fluxes fixed. The simulations indicated that variations in the background winds alone could explain the observed variations in cell-track length and duration. They also reinforced the main hypotheses drawn from the observational analysis: stronger vertical shear favors LTLD events while weaker vertical shear and along-barrier low-level winds favor STLD events. The mechanisms by which cold-air outflow and ther- mally driven upslope inflow interact to control cell du- ration were consistent with the hypotheses drawn from the observations. The simulations also revealed a sensitivity of storm evolution to localized, terrain-induced modifications to vertical wind shear. In a simulation with strong vertical shear and cross-barrier low-level flow (case 3), low-level dynamical perturbations induced by the terrain en- hanced the leeside vertical shear and created clockwise hodograph curvature that aided the development of a supercell. Such leeside enhancement of supercell ro- tation was also found in the numerical simulations of Markowski and Dotzek (2011), where supercells imping- ing on a mountain ridge were temporarily invigorated as they interacted with leeside mesoscale vorticity anomalies. Because the present study focuses on diurnally forced mountain convection, it complements previous studies of high-impact events driven primarily by mechanical orographic forcing. These studies have documented floods promoted by moist, conditionally unstable, and persis- tent cross-barrier flow (e.g., Maddox et al. 1978; Nair et al. 1997; Landel et al. 1999). Because they involve anomalously strong low-level winds, such events are often associated with strong synoptic-scale forcing (frontal sys- tems, upper-level cutoff lows, etc.). However, as shown by the STLD events described herein, localized heavy pre- cipitation events may also occur under more quiescent, lighter-wind conditions. Awareness of the specific condi- tions that favor these events may thus serve to improve severe-weather forecasting in mountainous regions. Acknowledgments. This research was funded by the Canadian Natural Science and Engineering Research Council (NSERC) Grant NSERC/RGPIN 418372-12 and NSERC's Undergraduate Student Research Awards program. The authors are grateful for the insightful comments and suggestions of Matthew Bunkers and Rich Rotunno, as well as two anonymous reviewers. The nu- merical simulations were performed on the Guillimin supercomputer at McGill University, under the auspices of Calcul Qu^ebec and Compute Canada. REFERENCES Akaeda,K., J. Reisner,and D.Parsons, 1995: The role of mesoscale and topographically induced circulations in initiating a flash flood observed during the TAMEX project. Mon. Wea. Rev., 123, 1720-1739. Banta, R. M., 1984: Daytime boundary-layer evolution over mountainous terrain. Part I: Observations of the dry circula- tions. Mon. Wea. Rev., 112, 340-356. -, 1990: Atmospheric Processes over Complex Terrain. Meteor. Monogr., No. 45, Amer. Meteor. Soc., 323 pp. -, and C. B. Schaaf, 1987: Thunderstorm genesis zones in the Colorado Rocky Mountains as determined by traceback of geosynchronoussatelliteimages.Mon.Wea.Rev.,115,463-476. Bougeault, P., and Coauthors, 2001: The MAP special observing period. Bull. Amer. Meteor. Soc., 82, 433-462. Bryan, G. H., and J. M. Fritsch, 2002: A benchmark simulation for moist nonhydrostatic models. Mon. Wea. Rev., 130, 2917- 2928. Bunkers, M. J., B. A. Klimowski, J. W. Zeitler, R. L. Thompson, and M. L. Weisman, 2000: Predicting supercell motion using a new hodograph technique. Wea. Forecasting, 15, 61-79. -, M. R. Hjelmfelt, and P. L. Smith, 2006a: An observational examination of long-lived supercells. Part I: Characteristics, evolution, and demise. Wea. Forecasting, 21, 673-688. -, J. S. Johnson, L. J. Czephya, J. M. Grzywacz, B. A. Klimowski, and M. R. Hjelmfelt, 2006b: An observational examination of long-lived supercells. Part II: Environmental conditions and forecasting. Wea. Forecasting, 21, 689-714. Chu, C.-M., and Y.-L. Lin, 2000: Effects of orography on the generation and propagation of mesoscale convective systems in a two-dimensional conditionally unstable flow. J. Atmos. Sci., 57, 3817-3837. Damiani, R., and Coauthors,2008: The cumulus, photogrammetric, in situ, and Doppler observations experiment of 2006. Bull. Amer. Meteor. Soc., 89, 57-73. Davies-Jones, R., 1984: Streamwise vorticity: The origin of updraft rotation in supercell storms. J. Atmos. Sci., 41, 2991-3006. Demko, J. C., and B. Geerts, 2010: A numerical study of the evolving convective boundary layer and orographic circula- tion around the Santa Catalina Mountains in Arizona. Part II: Interaction with deep convection. Mon. Wea. Rev., 138, 3603- 3622. Dixon, M., and G. Weiner, 1993: TITAN: Thunderstorm Identifi- cation, Tracking, Analyzing and Nowcasting-A radar-based methodology. J. Atmos. Oceanic Technol., 10, 785-797. Doswell, C. A. I., 2001: Severe convective storms-An overview. Severe Convective Storms, Meteor. Monogr., No. 50, Amer. Meteor. Soc., 1-26. Ducrocq, V., O. Nuissier, D. Ricard, C. Lebeaupin, and T. Thouvenin, 2008: A numerical study of three catastrophic precipitating events over southern France. II: Mesoscale triggering and stationarity factors. Quart. J. Roy. Meteor. Soc., 134, 131-145. Geerts, B., Q. Miao, and J. C. Demko, 2008: Pressure perturbations and upslope flow over a heated, isolated mountain. Mon. Wea. Rev., 136, 4272-4288. Houze, R. A., 1993: Cloud Dynamics. Academic Press, 573 pp. -, W. Schmid, R. G. Fovell, and H.-H. Schiesser, 1993: Hail- storms in Switzerland: Left movers, right movers, and false hooks. Mon. Wea. Rev., 121, 3345-3370. Johnson, J. T., P. L. MacKeen, A. Witt, E. D. Mitchell, G. J. Stumpf, M. D. Eilts, and K. W. Thomas, 1998: The storm cell identification and tracking algorithm: An enhanced WSR-88D algorithm. Wea. Forecasting, 13, 263-276. Kirshbaum, D. J., 2011: Cloud-resolving simulations of deep con- vection over a heated mountain. J. Atmos. Sci., 68, 361-378. -, 2013: On thermally forced circulations over heated terrain. J. Atmos. Sci., 70, 1690-1709. Kuo, J.-T., and H. D. Orville, 1973: A radar climatology of sum- mertime convective clouds in the Black Hills. J. Appl. Meteor., 12, 359-368. Landel, G., J. A. Smith, M. L. Baeck, M. Steiner, and F. L. Ogden, 1999: Radar studies of heavy convective rainfall in moun- tainous terrain. J. Geophys. Res., 104 (D24), 31 451-31 465. Maddox, R. A., L. R. Hoxit, C. F. Chappell, and F. R. Caracen~a, 1978: Comparison of meteorological aspects of the Big Thompson and Rapid City flash floods. Mon. Wea. Rev., 106, 375-389. Markowski, P., and Y. Richardson, 2010: Mesoscale Meteorology in Midlatitudes. Wiley, 430 pp. -, and N. Dotzek, 2011: A numerical study of the effects of orography on supercells. Atmos. Res., 100, 457-478. McCaul,E.W., and C. Cohen, 2002: The impact on simulated storm structure and intensity of variations in the mixed layer and moist layer depths. Mon. Wea. Rev., 130, 1722-1748. Miglietta, M. M., and R. Rotunno, 2009: Numerical simulations of conditionally unstable flows over a mountain ridge. J. Atmos. Sci., 66, 1865-1885. -, and -, 2012: Application of theory to simulations of ob- served cases of orographically forced convective rainfall. Mon. Wea. Rev., 140, 3039-3053. Nair, U. S., M. R. Hjelmfelt, and R. A. Pielke, 1997: Numerical simulations of the 9-10 June 1972 Black Hills storm using CSU RAMS. Mon. Wea. Rev., 125, 1753-1766. Petersen, W. A., and Coauthors, 1999: Mesoscale and radar ob- servations of the Fort Collins flash flood of 28 July 1997. Bull. Amer. Meteor. Soc., 80, 191-216. Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline clima- tology of sounding-derived supercell and tornado forecast parameters. Wea. Forecasting, 13, 1148-1164. Smith, R. B., and Coauthors, 2012: Orographic precipitation in the tropics: The Dominica Experiment. Bull. Amer. Meteor. Soc., 93, 1567-1579. Soderholm, B., 2013: The evolution of convective storms initiated by an isolated mountain range. M.S. thesis, Department of Atmospheric and Oceanic Sciences, McGill University, Montr^eal, Quebec, Canada, 91 pp. Weisman, M. L., and J. B. Klemp, 1982: The dependence of nu- merically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110, 504-520. -, and -, 1984: The structure and classification of numeri- cally simulated convective storms in directionally varying wind shears. Mon. Wea. Rev., 112, 2479-2498. Wulfmeyer, V., and Coauthors, 2008: The convective and oro- graphically induced precipitation study: A research and development project of the World Weather Research Pro- gram for improving quantitative precipitation forecasting in low-mountain regions. Bull. Amer. Meteor. Soc., 89, 1477- 1486. BRETT SODERHOLM,BRYN RONALDS, AND DANIEL J. KIRSHBAUM Department of Atmospheric and Oceanic Sciences, McGill University, Montre^al, Que^bec, Canada (Manuscript received 4 September 2013, in final form 2 December 2013) Corresponding author address: Daniel J. Kirshbaum, Depart- ment of Atmospheric and Oceanic Sciences, McGill University, Room 945, Burnside Hall, 805 Sherbrooke St. West, Montr^eal, QC H3A 0B9, Canada. E-mail: [email protected] (c) 2014 American Meteorological Society |