1. Introduction

Summertime deep convection over the continental United States exhibits multiple robust signatures at different scales. On the continental scale, the dominant feature is afternoon convection initiation over the Rocky Mountains organizing into mesoscale convection systems that propagate eastward at night across the Great Plains (e.g., Carbone et al. 2002; Carbone and Tuttle 2008). Superposed on this pattern are afternoon mesoscale hotspots over Florida and the Gulf of Mexico coastline in the southeastern United States, where daytime sea breezes create organized zones of horizontal convergence and vertical ascent (Byers and Rodebush 1948; Carbone and Tuttle 2008). Afternoon convection over the Rockies also exhibits mesoscale character, with hotspots over and downwind of larger mountain ridges (e.g., Kuo and Orville 1973; Banta and Schaaf 1987; Damiani et al. 2008).

Beyond the first-order features mentioned above, subtler regional convection extrema occur in response to complex terrain, varying land use, proximity to water bodies, etc. Because such comparatively modest effects are often masked on shorter time scales by synoptic and/or interannual variability, they are best exposed through long-term climatologies that consider hundreds (if not thousands) of convection events. To address this need, Fabry et al. (2013) studied the diurnal cycle of convection over the conterminous United States using 15 years (1996–2007 and 2011–13) of radar-composite maps from the WSR-88D network. Among other novel findings, they observed a robust afternoon minimum in convection occurrence along the interior lower Mississippi Valley on summer afternoons, relative to surrounding areas. In this brief article, we review those observations and perform numerical simulations to gain insight into the cause of this regional convection signature.

2. Observations

Ready-made radar composites over the continental United States were obtained from two sources: (i) the National Operational Weather radar (NOWrad) composites from the WSI Corporation, which are similar to the data used by Carbone et al. (2002) and Carbone and Tuttle (2008) (from 1996 to 2007), and (ii) the low-altitude U.S. composite made by the Warning Decision Support System–Integrated Information (WDSS-II; Lakshmanan et al. 2006) (from 2011 to 2014). Both datasets attempt to characterize the echo strength and coverage in the lower troposphere. Using these data, Fabry et al. (2013) mapped the probability of radar echoes at or exceeding 40 dBZ (or ), where 40 dBZ is a common threshold for distinguishing convective from stratiform precipitation in the midlatitudes (e.g., Johnson et al. 1998; Soderholm et al. 2014). Amidst many expected phenomena, a subtle and unexpected signature emerged: the interior lower Mississippi Valley was less convectively active than its immediate surroundings. On summer (July–August) afternoons [1900–2300 UTC, or 1300–1700 local solar time (LST)], the frequency of convection incidence in the valley was about half that of surrounding areas, with the daily afternoon maxima much weaker and occurring later (Figs. 1c,d).

View Full Size

Observations of the regional minimum of convection occurrence within the Mississippi Valley: (a) satellite image of the Mississippi Valley (courtesy of Google Maps), (b) terrain height, (c) in midafternoon (1900–2300 UTC or 1300–1700 LST) in July and August derived from radar composites over the years 1996–2007 and 2011–14, and (d) as a function of time of day and day of the year in two zones identified in (b) and (c), zone 1 (above) in the valley and zone 2 (below) surrounding the valley. The time–day plot was smoothed by a 10-day Hamming window to remove small-scale variability, and uses the same scale as in (c).

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

Given that the regional convection minimum in Fig. 1c possesses a very similar shape as the Mississippi Valley itself (Fig. 1a), one might expect that the valley topography (specifically, orography and/or land surface variations) is somehow responsible for it. Such an association was demonstrated in the satellite analysis of Gambill and Mecikalski (2011), who found that cumulus convection over the southeastern United States was most common over heterogeneous topography, particularly orographic regions. The pattern in Fig. 1c is indeed reminiscent of mountain–valley contrasts in convection initiation driven by differential heating, such as those in the western United States and other orographic regions (e.g., Banta 1990; Weckwerth et al. 2011; Kirshbaum 2011). However, the terrain relief over the Mississippi Valley is unremarkable; tucked between the Ozark Plateau to the west and the east Gulf coastal plains to the east, the valley lies on average just a few hundred meters below its surroundings (Fig. 1b). On visible satellite imagery, the land-use variations are more striking: lighter-colored partially irrigated fields cover the land, contrasting with the darker surrounding forests (Fig. 1a). Could variations in land use explain the regional differences in convection (e.g., Rozoff et al. 2003)? Are the modest terrain variations between the valley and its surroundings to blame? Or are we being misled by a simple coincidence?

3. Numerical simulations

To gain insight into the topographic (specifically, orographic and land surface related) controls on the Mississippi Valley signature, we perform convection-permitting numerical simulations with the Weather Research and Forecasting (WRF) Model. Because real-case simulations of all 1000 or so convection events in the 15-yr climatology would have been computationally prohibitive, a reduced set of experiments was sought. While real-case simulations of one- or two-years’ worth of convection events would have been affordable, such short time periods contain too few events to reproduce the long-term signature in Fig. 1c.

As a more affordable alternative, we use quasi-idealized simulations with real topography but horizontally homogeneous initial conditions and no synoptic-scale forcing. This approach follows from our hypothesis that the topography of the Mississippi Valley region is largely responsible for the regional variations in deep-convection occurrence. Although the neglect of synoptic forcing is a major simplification, it provides a simple and direct evaluation of topographic effects in the absence of other mitigating factors. Moreover, given that the horizontal scales of synoptic forcing are typically larger than the width of the Mississippi Valley, and that such forcing is climatologically weak during midsummer, it is unlikely to single handedly produce the detailed mesoscale variations seen in Fig. 1c.

In environments lacking synoptic-scale forcing for ascent, afternoon convection initiation is promoted by various surface-based processes including (i) broad destabilization from diurnal heating, (ii) mechanically and thermally forced orographic ascent, and (iii) thermal circulations and frictional convergence driven by variations in land use and land cover. Mechanical orographic ascent is facilitated by the weak stability of the convective boundary layer, which submerges all of the regional mountains and allows impinging flow to easily surmount them. This weak stability also maximizes the strength of thermal circulations driven by differential surface heating (e.g., Kirshbaum 2013).

A single model domain is used with a horizontal grid spacing of km, centered in southeastern Arkansas (34°N and 91.5°W) with 480 by 480 grid points in the x and y directions and 81 stretched vertical levels. Subgrid parameterizations include a five-layer, thermal-diffusion land surface scheme, with land-use data from the U.S. Geological Survey (USGS). Although the simplicity of this scheme renders it prone to error, it attractively does not require the prescription of various uncertain, time-varying parameters (e.g., soil moisture, canopy water, etc.) that are utilized by more sophisticated schemes. Instead, the values of such parameters are held fixed to the USGS climatological values throughout the simulation. This simplified land surface treatment is consistent with other simplifications in our model setup.

Other subgrid parameterizations include the Mellor–Yamada–Janjić surface and boundary layer schemes, a Smagorinsky-type horizontal turbulent-mixing scheme, the Goddard shortwave and Rapid Radiative Transfer Model longwave radiation schemes, and the Morrison two-moment cloud microphysics scheme. The initial flow (described below) is assumed to be in geostrophic balance, hence the Coriolis force is applied only to flow perturbations using an f-plane approximation with a characteristic midlatitude value of f = 10⁻⁴ s⁻¹. The surface skin temperature is initialized to 292 K at sea level, with a 5 K km⁻¹ lapse rate over terrain to mimic the mean vertical gradient of air temperature. While this idealization neglects initial horizontal variations in and its lapse rate over varying land surfaces, sensitivity experiments showed minimal sensitivity to the initialization of , within reasonable bounds (not shown). The lateral boundary conditions are open (radiative) and the upper boundary is closed with a 5-km-deep sponge layer below the model top at 20 km.

Because km is similar in scale to individual convective-storm elements (~10 km), it is insufficient to properly resolve their dynamics and microphysics (e.g., Bryan et al. 2003). However, numerous recent modeling studies have found km sufficient to broadly capture the timing and location of continental deep convection (e.g., Clark et al. 2012; Hanley et al. 2013). Although it is prone to err in the details of individual storms, this value of still resolves mesoscale envelopes of moist instability along with some (but not all) of the local circulations that help to release it. Thus, km is deemed acceptable for our purposes.

A total of 15 initial flows are considered from July to August 2013, all taken from the mornings of “airmass” convection events where clear skies transitioned to scattered deep convection over the Mississippi Valley region in the afternoon. Consideration of weakly forced events is consistent with our quasi-idealized model setup, in which large-scale forcing is neglected. The sounding for each case is obtained by averaging (in height) the corresponding 1200 UTC radiosondes from Little Rock, Arkansas; Jackson, Mississippi; and Shreveport, Louisiana (see Fig. 1a for city locations). Examples of three such soundings (Fig. 2a) indicate surface-based nocturnal inversions overlaid by deep layers of conditional instability and high relative humidity. Statistics of equivalent potential temperature , relative humidity, zonal wind U, and meridional wind V over the entire 15-member ensemble in Figs. 2b–e reveal potential instability up to about 5 km and gradually decreasing relative humidity up to 7 km, with light south-southwesterly low-level winds and varying winds aloft.

View Full Size

(a) Examples of soundings from three events: 10 Jul (blue), 9 Aug (green), and 23 Aug 2013 (red). Short (long) wind barbs correspond to 5 (10) m s⁻¹. Statistics of soundings over the full 15 events, including (b) equivalent potential temperature, (c) relative humidity, and (d) zonal and (e) meridional wind components, respectively. The mean profile is shown in the thick line, the mean profile 1 standard deviation is shown in dashed lines, and the full ensemble range is in gray.

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

The simulations are integrated from 1200 to 0000 UTC the next day (or 0600–1800 LST), thus capturing half the diurnal cycle and encompassing the 1300–1700 LST afternoon period. Three simulations are conducted for each event: one with both land use/cover and terrain variations (FULL), one with terrain variations but uniform land use/cover (set to cropland/woodland mosaic, a common land-use type within the Mississippi Valley, as seen in Fig. 3a) (TERRAIN), and one with land use/cover variations but flat terrain (LANDUSE). Note that the TERRAIN simulations prescribe the same land surface index at all model grid points, which removes all land-use variation and land–water contrasts. Note also that the name “LANDUSE” herein refers to both land-use variations (e.g., from agriculture, urbanization, etc.) and land-cover variations (e.g., from land–water contrasts).

View Full Size

Topography for the numerical simulations, including (a) USGS land-use index, (b) surface albedo, (c) volumetric soil moisture fraction, and (d) terrain height. The asterisk and box in (d) correspond to the domain center point and the interior region with full terrain (outside of which the terrain was gradually decayed to zero). The lines A–A′ and B–B′ in (d) correspond to the two vertical cross sections shown in Fig. 6.

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

In initial experiments using the real orography over the entire model domain, large-amplitude inertia–gravity waves developed at the lateral boundaries that propagated into the interior, corrupting the solution everywhere. To eliminate these spurious waves, we decay the orography to zero outside of the Mississippi Valley region. The real terrain is retained over a box centered at 34°N, 91.5°W with a radius of 500 km, outside of which it is gradually reduced to zero using a Gaussian function with a half-width of 200 km along both horizontal directions (Fig. 3d). Experiments increasing the box size and the outer decay rate indicated virtually no sensitivity of the Mississippi Valley convection to these settings (not shown). The USGS land surface classifications are retained over the entire domain (Fig. 3a), the albedo and volumetric soil-moisture content of which are shown in Figs. 3b and 3c.

4. Results

a. Description

To enable direct comparison between observations and simulations, the simulated reflectivity is calculated at the lowest model grid level at each model output time of each simulation, then processed to obtain a simulated distribution for the FULL, TERRAIN, and LANDUSE simulations. In the TERRAIN simulations this distribution is strongly tied to the underlying orography, with a maximum over the Ozarks and Ouachita Mountains in Arkansas, a pronounced minimum over the flat Mississippi Valley, and a secondary maximum north of the Gulf Coast where the coastal plain transitions to rolling hills (Fig. 4b). In the LANDUSE simulations, by contrast, is maximized along the Gulf Coast owing to the sharp land–sea contrast and the consequent development of daytime sea breezes there (Fig. 4c). The LANDUSE simulations also exhibit a broken west–east strip of enhanced extending from eastern Oklahoma through the Arkansas River valley and across the Mississippi Valley to western Tennessee (the last part connecting Little Rock and Memphis, Tennessee). In west-central Arkansas, this strip is flanked by minima to the north and south, coinciding with the Ozarks and Ouachita maxima from the TERRAIN simulations.

View Full Size

Frequency of simulated radar reflectivity exceeding 40 dBZ over 1300–1700 LST (or ) in the 15 simulations for each topography type (FULL, TERRAIN, and LANDUSE). Black lines show terrain-height contours at 100, 250, and 500 m.

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

While the FULL simulations reproduce the sea-breeze-driven convection enhancement along the Gulf Coast seen in the LANDUSE simulations, their patterns over the interior Mississippi Valley are more in line with the TERRAIN simulations: a broad minimum along the valley is flanked by maxima over the surrounding hills (Fig. 4). However, the contrast between the valley and its surroundings is weaker than in the TERRAIN simulations due to the impacts of land-use heterogeneities. In particular, the Ozarks/Ouachita maxima are weakened and the valley contains isolated maxima coinciding with those in the LANDUSE simulations, the most pronounced of which is the aforementioned west–east strip of enhanced across central Arkansas. Thus, while the simulated minimum along the Mississippi Valley appears broadly dictated by the regional orography, it is significantly modulated by land-use variations.

b. Interpretation

The sensitivity of simulated to terrain height owes to the combination of increased forcing for ascent and decreased convective inhibition (CIN) over higher terrain. Figure 5 compares the mean surface horizontal convergence () and mean-layer [0–500-m above ground level (AGL)] CIN at 1100 LST (during the mainly preconvective period). In quasi-incompressible flow, stronger surface convergence gives rise to stronger boundary layer ascent and thus increases the likelihood of convection initiation. In the FULL and TERRAIN simulations, the amplitude of the convergence/divergence is minimized over the Mississippi Valley and maximized over the surrounding terrain, particularly the Ozarks/Ouachita ridges in Arkansas (Figs. 5a,b). The zones of maximum convergence in Arkansas form to the north of the ridge crests, or on the lee side given the mean southerly low-level winds (Figs. 2d,e).

View Full Size

Averaged (a)–(c) horizontal convergence at 10 m AGL and (d)–(f) mean-layer (0–500 m AGL) CIN, of the 15 simulations for each topography type (FULL, TERRAIN, and LANDUSE) at 1100 LST. Black lines show terrain-height contours at 100, 250, and 500 m.

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

The reduced CIN over the high terrain in Figs. 5d,e and 5g,h arises primarily from the protrusion of the mountainous terrain above the nocturnal inversions (e.g., Fig. 2). In ascending from sea level to 600 m (the approximate height of Ozarks ridgeline in Arkansas), the CIN of the ensemble-averaged initial sounding decreases from 205 to 46 J kg⁻¹ while the CAPE increases from 595 to 1713 J kg⁻¹. The vertical variations in CIN translate into large spatial differences in mean-layer CIN over varying terrain: the initial CIN within the valley is generally more than 100 J kg⁻¹ larger than that over surrounding areas. These spatial differences in CIN gradually diminish throughout the simulations until about 1400 LST, when CIN reduces to zero over the entire domain (not shown). The reduced CIN over the mountains allows convection to initiate more readily there than within the Mississippi Valley, at least early into the analysis period.

Other notable features in Figs. 4 and 5 include the sea-breeze-driven convergence just inland of the Gulf Coast in the LANDUSE and FULL simulations, which gives rise to the aforementioned maxima there (Figs. 5d and 5f). The strip of enhanced between Little Rock and Memphis coincides with several patches of savanna land that possess a much lower volumetric soil water fraction () than the majority of the valley () (Figs. 3a and 3c). The consequently reduced surface evaporation fosters increased sensible heating, which locally reduces CIN and drives convergent thermal circulations (Fig. 5). Similarly, the FULL and LANDUSE simulations exhibit hotspots over urban areas (e.g., Memphis and Nashville, Tennessee, and Birmingham, Alabama) stemming from the low albedo and soil moisture () of paved surfaces. Consistent with various recent studies (e.g., Orville et al. 2001; Rozoff et al. 2003; Niyogi et al. 2011), these urban heat islands are associated with reduced CIN, increased convergence, and enhanced convection occurrence downwind.

To gain insight into the vertical structure of the regional flow, Fig. 6 presents vertical cross sections of ensemble- and time-averaged (over the preconvective 1000–1100 LST period) , θ, and plane-parallel wind vectors for the 15 FULL simulations. These cross sections are taken along lines A–A′ across the Mississippi Valley and B–B′ parallel to the valley over the high Arkansas terrain (see Fig. 3d). Both cross sections show decreasing with height (as in Fig. 2b), with lower values over the mountains than in the valleys, and θ increasing with height, reflecting stable stratification above the convective boundary layer. Section A–A′ (Fig. 6a) shows very weak cross-valley flow at lower levels, indicating that the modest gradients in orography and land use across the valley are unable to drive a detectable valley-wide thermal circulation. Thus, such circulations likely cannot explain the strong cross-valley gradients in in Figs. 4a and 4b.

View Full Size

Vertical cross sections of (color filled), θ (contours), and plane-parallel wind vectors along lines A–A′ and B–B′ (see Fig. 3d), averaged over the 15-member FULL ensemble over 1000–1100 LST.

Citation: Monthly Weather Review 144, 1; 10.1175/MWR-D-15-0238.1

• Download Figure
• Download figure as PowerPoint slide

Cross-section B–B′, by contrast, reveals substantial vertical motion as the mean south-southwesterly low-level flow parallels the complex Arkansas terrain (Fig. 6b). This mechanically forced ascent is consistent with a mountain Froude number (where is the crest height and and are the averaged cross-barrier wind speed and dry Brunt–Väisälä frequency, respectively, over the subcrest layer). Performing this calculation at point B, with m to represent the Ozarks ridgeline, we obtain , suggesting that the majority of the subcrest flow manages to surmount the tallest ridges. The boundary layer updrafts extend through the convective boundary layer and into the lower free troposphere, where they exhibit the slight upstream tilt with height that is characteristic of mountain waves in a stably stratified environment (e.g., Smith 1979). Low-level convergence forms in the lee of the ridges due to wave-induced pressure troughs there, which explains the strong leeside convergence zones in Figs. 5d and 5e. However, the largest forced ascent occurs over the windward (southern) slopes where the flow directly surmounts the terrain. Combined with the locally low CIN, such ascent helps to initiate deep convection and enhance over the ridge tops (as seen in Figs. 4a and 4b).

5. Discussion

Because the FULL simulations include both land-use and terrain variability, they are nominally the most realistic and thus the most comparable with observations. These simulations broadly reproduce the Mississippi Valley convection minimum over 32°–37°N (Figs. 4a and 1c), and hence likely capture at least some of the key physical processes underlying it. The similarity of this regional minimum in the FULL and TERRAIN simulations (Figs. 4a,b) suggests that terrain elevation, rather than variable land use, is largely responsible for it, through the mechanisms discussed in section 4b. This finding is consistent with the satellite observations of Gambill and Mecikalski (2011), which suggested that elevation gradients were the most important forcing mechanism for convective clouds during weakly forced convection events over the southeastern United States. The FULL simulations also successfully reproduce the strong maximum along the Gulf Coast due to the sharp land–sea contrast and associated sea breezes that initiate deep convection just inland.

On the other hand, the distribution in the FULL simulations differs from the observations on multiple fronts. The magnitudes of the simulated maxima surrounding the Mississippi Valley are around 2–3 times larger than those in the observations, a discrepancy that may arise from our exclusive simulation of widespread convection events. The observations, in contrast, include many nonprecipitating and isolated convection events, which lower the overall magnitude as well as its contrast between the valley and its surroundings. Other potential sources of bias include the Morrison microphysics parameterization, which may overestimate the prevalence of hail and thus the values of [such a bias was found in the simulations of Robinson et al. (2011)] and uncertainties in the forward model used to obtain simulated reflectivity.

The simulated convection minimum along the Mississippi Valley in the FULL simulations is also less coherent than that in the long-term observations (or in the TERRAIN simulations, for that matter), which appears to stem from an oversensitivity to land-use heterogeneities. Although many of the patterns in the LANDUSE simulations are diminished in the FULL simulations due to the competing effects of orography, the Gulf Coast maximum, the urban enhancements, and the broken strip of enhanced convection between Little Rock and Memphis, persist (Figs. 4a and 4c). With the exception of the Gulf Coast maximum, however, all of these features appear much stronger in the simulations than in the long-term observations. Urban enhancements are observed around Memphis and Nashville, but to a lesser degree than those simulated (Figs. 1c and 4a). Moreover, the scattered maxima within the Mississippi Valley, in particular the strip from Little Rock to Memphis, are virtually nonexistent in the observations.

The simulations’ oversensitivity to land-use variations within the Mississippi Valley may stem from their highly simplified land surface scheme and/or the small numerical sample size, the latter of which may allow a few strong events to exert disproportionate influence. But an even stronger contribution to this bias may arise from our exclusive consideration of airmass convection events under weak synoptic forcing. The light winds and small CIN of such events renders the convection highly responsive to mesoscale thermal anomalies over topographic features, due to their local destabilizing effects and strong thermal forcing. In contrast, the broader cloud cover and stronger winds of synoptically disturbed events would tend to suppress such thermal anomalies by weakening insolation and increasing heat ventilation. Because land-use heterogeneities influence deep convection primarily through thermal forcing (e.g., Souza et al. 2000; Rozoff et al. 2003; Niyogi et al. 2011), their impacts would thus tend to weaken under stronger synoptic forcing. By the same logic, mountain thermal forcing would also diminish in such events. However, because the orographic forcing is largely mechanical in nature (Fig. 6), it would thus tend to increase under stronger winds (e.g., Kirshbaum and Wang 2014). Therefore, consideration of the full spectrum of events would likely produce a more coherent, and thus more realistic, convection minimum within the valley.

6. Conclusions

As shown by a recent 15-yr radar climatology over the continental United States, a mesoscale minimum of convection occurrence exists over the Mississippi Valley during midsummer (July–August) afternoons (1900–2300 UTC or 1300–1700 LST), with the valley experiencing roughly half the occurrence of its surroundings (Fabry et al. 2013). While the strong correlation between this convection minimum and the regional topography suggests a causal connection between the two, the topography itself is unremarkable, with terrain relief of a few hundred meters and modest land-use heterogeneities. Nonetheless, we hypothesized that these modest topographic variations, through their mechanical and thermal forcing for ascent, are responsible for the valley convection minimum. To test this hypothesis, we performed quasi-idealized, convection-permitting numerical simulations with the Weather Research and Forecasting (WRF) Model. The simulations used the real orography and land use of the region but were initialized horizontally homogeneously from morning soundings on days of afternoon airmass convection under weak synoptic forcing and light ambient winds.

The simulations broadly reproduced the valley convection minimum and thus supported the hypothesis that the regional topography was behind it. Sensitivity simulations further revealed that the regional orography, rather than land-use heterogeneities, was the dominant cause of this feature. Thus, even modest terrain relief (here less than 1 km) can exert strong control over regional convection occurrence. Over the higher terrain, reduced convective inhibition coincided with increased surface-based ascent to promote convective initiation. The reduced inhibition was associated with weaker initial static stability as the terrain protruded above the nocturnal inversion. The increased ascent was mainly associated with the mechanical lifting of impinging flow over the terrain (rather than mountain thermal circulations), which created vertically coherent updrafts in the convective boundary layer that transitioned into mountain waves in the free troposphere.

The numerical simulations only considered a small subset of convection events within the Mississippi Valley: synoptically undisturbed, airmass convection events. As a possible result, the simulated convection minimum was less coherent than the observed minimum, with isolated maxima forming in the valley in response to mesoscale land-use variations. While such discrepancies between the observations and simulations may relate to limitations of the model setup (coarse grid resolution, a simple land surface scheme, insufficient sampling of events, etc.), they may also stem from our neglect of synoptic forcing. Although such forcing is typically weak in the southeast United States during midsummer, it still significantly influences the summer regional precipitation climatology (e.g., Diem 2006). We speculate that, because the stronger winds and broader cloud cover of synoptically disturbed events would tend to suppress thermal circulations within the valley while enhancing mechanical lifting by the orography, the inclusion of such events would give rise to a sharper valley convection minimum.

The current findings apply strictly to the Mississippi Valley in midsummer and may not extend to other geographic regions or seasons. For example, the valley convection minimum during these months is much more pronounced than that in May and June (not shown). Such seasonal variation may stem from the stronger upper-level forcing during the late spring, which limits the relative impact of the regional topography. Similarly, river valleys in different geographic regions may exhibit different regional convection patterns than that observed herein. For example, within southern Québec and northern New York state, summer convection appears to be diminished, rather than enhanced, over higher terrain (Wasula et al. 2002; Bellon and Zawadzki 2003; Kovacs and Kirshbaum 2015, manuscript submitted to J. Appl. Meteor. Climatol.). Physically interpreting these patterns, and contrasting them with those over the Mississippi Valley, is a topic of ongoing study.