Volume 125, Issue 5 e2019JD031164
Research Article
Free Access

Retrievals of Convective Detrainment Heights Using Ground-Based Radar Observations

M. Starzec

Corresponding Author

M. Starzec

Department of Atmospheric Sciences, University of North Dakota, Grand Forks, ND, USA

Correspondence to: M. Starzec,

[email protected]

Search for more papers by this author
G. L. Mullendore

G. L. Mullendore

Department of Atmospheric Sciences, University of North Dakota, Grand Forks, ND, USA

Search for more papers by this author
C. R. Homeyer

C. R. Homeyer

School of Meteorology, University of Oklahoma, Norman, OK, USA

Search for more papers by this author
First published: 14 February 2020
Citations: 2


To better constrain model simulations, more observations of convective detrainment heights are needed. For the first time, ground-based S band radar observations are utilized to create a comprehensive view of irreversible convective transport over a 7-year period for the months of May and July across the United States. The radar observations are coupled with a volumetric radar echo classification scheme and a methodology that uses the convective anvil as proxy for convective detrainment to determine the level of maximum detrainment (LMD) for deep moist convection. The LMD height retrievals are subset by month (i.e., May and July), by morphology (i.e., mesoscale convective system, MCS, and quasi-isolated strong convection, QISC), and region (i.e., northcentral, southcentral, northeast, and southeast). Overall, 135,890 deep convective storms were successfully sampled and had a mean LMD height of 8.6 km or tropopause-relative mean LMD height of urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00014.3 km; however, LMD heights were found to extend up to 2 km above the tropopause. May storms had higher mean tropopause-relative LMD heights, but July storms contained the highest overall LMD heights that more commonly extended above the tropopause. QISC had higher mean tropopause-relative LMD heights and more commonly had LMD heights above the tropopause while only a few MCSs had LMD heights above the tropopause. The regional analysis showed that northern regions have higher mean LMD heights due to large amounts of diurnally driven convection being sampled in the southern regions. By using the anvil top, the highest possible convective detrainment heights extended up to 6 km above the tropopause.

Key Points

  • Quasi-isolated strong convection transports mass to higher altitudes than mesoscale convective systems
  • May convection typically detrains higher relative to the tropopause than July, but July contains cases of the deepest transport
  • Convection in northern CONUS has higher mean detrainment altitudes than southern CONUS on average

1 Introduction

Mass exchange between the upper troposphere and lower stratosphere (UTLS) changes the chemical composition, radiative properties, and distribution of greenhouse gases in this layer, which has significant implications for climate studies (e.g., Holton et al., 1995; Ramaswamy et al., 1992; Stohl et al., 2003). It is generally accepted that large-scale transport mechanisms such as isentropic transport from the tropical troposphere and Rossby wave breaking are responsible for the majority of mass exchange between the stratosphere and troposphere, with small-scale features such as deep convection contributing substantially less mass. Nevertheless, deep convection has the ability to rapidly and efficiently transport boundary layer mass to the upper atmosphere (e.g., Mullendore et al., 2005; Pickering et al., 1988). Convection can transport mass in minutes to hours, as compared to days for extratropical cyclones, and weeks or months for turbulent diffusive processes (Dickerson et al., 1987; Sigmond & Siegmund, 2000). Boundary layer mass can be transported into the upper troposphere, and in many cases, convection has been observed to penetrate the tropopause and directly inject mass into the stratosphere (e.g., Fischer et al., 2003; Hegglin et al., 2004; Homeyer et al., 2014; Poulida et al., 1996; Smith et al., 2017).

Polluted air is transported upward via the updraft until it dynamically detrains out of the updraft or mixes with environmental air to become neutrally buoyant. Overshooting tops have recently received a lot of attention with regard to convective transport due to the proximity of polluted air to pristine stratospheric environment (e.g., Cooney et al., 2018; Homeyer & Kumjian, 2015; Solomon et al., 2016). Nevertheless, the air within the overshooting top is negatively buoyant and relies solely on mixing processes (such as gravity wave breaking) to eject mass into UTLS; otherwise, even though the overshoot penetrates the tropopause it may not be ejecting a large amount of mass into the stratosphere. The greatest amount of convective mass detrainment occurs in the convectively generated anvil and thus cannot be neglected (e.g., Carletta et al., 2016; Mullendore et al., 2013). Air from the anvil can also be directly injected into the lower stratosphere (e.g., Hegglin et al., 2004; Mullendore et al., 2005; Poulida et al., 1996) or can be mixed across the tropopause via processes such as gravity wave breaking aloft (e.g., Hassim & Lane, 2010).

Both model simulations and in situ measurements reveal that air transported by convection can be relatively undiluted (e.g., Mullendore et al., 2005; Ström et al., 1999). Injection of polluted boundary layer air directly into the upper troposphere and lower stratosphere can have significant chemical impact as free tropospheric air has a much different chemical composition than the boundary layer air. Important yet different radiative, climatological, and/or health impacts exist depending on whether boundary layer air is transported into the upper troposphere or is able to reach the lower stratosphere (e.g., Anderson et al., 2012; Dickerson et al., 1987; Kiehl et al., 1999; Thompson et al., 1994).

Quantitatively identifying the altitudes where boundary layer mass is transported to has been challenging. Retrievals of the height of mass transport have been determined from aircraft measurements (e.g., Pickering et al., 1996), satellite measurements (Takahashi & Luo, 2012; Takahashi et al., 2017), multi-Doppler observations (e.g., Mullendore et al., 2013), and modeling studies (e.g., Barth et al., 2007; Bigelbach et al., 2014). While invaluable, aircraft measurements are limited temporally, cannot continuously sample every region of storm outflow, and generally are only available from field campaigns. Satellite measurements are too coarse spatially and/or temporally to be used reliably for assessing three-dimensional cloud-scale properties and generally require thermodynamic assumptions about the vertical structure of the atmosphere. Multi-Doppler analyses require rigorous processing techniques and also typically rely on field campaigns as only a few locations are set up for such observations. While modeling efforts are useful in understanding how mass is transported and are able to provide three-dimensional detail into transport processes, chemical transport models are largely unconstrained by observations and have a large uncertainty regarding the height of transport. The uncertainty in these models is a combination of uncertainty in both the chemical processes and convective dynamics.

Due to the limitations in height retrievals, the level of neutral buoyancy (LNB) from parcel theory has been commonly utilized to determine the height of convective mass detrainment. Parcel theory describes the ascent path and available potential energy of an air parcel initially chosen using the atmospheric properties near the surface. The LNB is defined as the level at which the density of a positively buoyant air parcel is equal to that of the environment. However, parcel theory does not encompass parcel variability (spatial or temporal), ignores hydrometeor loading, perturbation pressure forces, and most importantly does not include effects owing to entrainment and mixing. Entrainment dilutes the parcel by mixing in cooler, drier environmental air resulting in reduced parcel buoyancy. Therefore, deep convection typically does not detrain at the LNB but in actuality detrains below it. In certain cases, the actual detrainment level can be lower than the LNB by 5 km or more (Mullendore et al., 2013). The height of detrainment as determined by parcel theory is typically accepted as the theoretical maximum height a parcel can reach.

To avoid the limitations in identifying the convective mass detrainment height with previous methods, Mullendore et al. (2009) showed that profiles of bulk dynamic detrainment retrieved from dual-Doppler analyses coincided well with the location of the anvil visible in the radar reflectivity field. Furthermore, Mullendore et al. (2009) noted that the peak in total ice water content (IWC) of the anvil coincided well with the level of maximum horizontal divergence (i.e., vertical convergence), which they termed the level of maximum detrainment (LMD). Carletta et al. (2016) used this knowledge to develop a procedure that could objectively identify the LMD for S band radars and coupled a radar classification scheme from (based on Feng et al., 2011; Steiner et al., 1995) to identify suitable anvil near convection. After analyzing several dual-Doppler cases, they concluded that radar-based detection of the LMD is promising and typically accurate within 2 km when compared to dual-Doppler-derived vertical divergence of the updraft; however, they noted several limitations: developing convection lacks adequate anvil, detection can be limited in complex storm clusters where convective elements are embedded in deep stratiform cloud, and sufficient radar coverage is required aloft. Furthermore, they used a strict threshold radar echo classification scheme that utilized only low-level echo at one height, which resulted in misclassified echo aloft.

In order to improve our understanding of convective mass detrainment and constrain model simulations, more frequent and easily obtainable observations are needed. In this study, the methodology of Mullendore et al. (2009) and Carletta et al. (2016) is built upon and expanded from case studies to 7 years of hourly high-resolution volumetric radar composites across the CONUS east of the Rocky Mountains for the months of May and July. To enable a statistical quantification of LMD heights, the radar data are coupled with a three-dimensional radar-echo classification algorithm in order to objectively identify both convection and convectively generated anvil. The months of May and July are chosen as both months are characterized by different environments. May is commonly a transition period and is typically associated with more strongly forced convective systems due to enhanced shear and stronger upper level dynamical support, in part due to the location of the jet stream (e.g., Bigelbach et al., 2014), while July typically contains a warmer, more subtropical environmental in the southern analysis regions. The differences between May and July are visible in frequencies of overshooting storms and frequencies of extreme precipitation rates (e.g., Cooney et al., 2018; Hitchens et al., 2012; Solomon et al., 2016). May and July also exhibit differences in tropopause heights, where May tropopause heights are typically lower and more variable than in July (e.g., Randel et al., 2000; Wong & Wang, 2000). Differences in tropopause height and convective depth are also dependent on location. For example, the southeast United States typically has consistently higher tropopause heights due to the location of the subtropical jet, while the northern and central CONUS can have comparatively lower tropopause heights, particularly in spring in the presence of a large-scale trough (e.g., Hoerling et al., 1991; Nastrom et al., 1989). Likewise, convection in the central plains was found to be more likely to overshoot the tropopause, which may have a impact on the LMD heights (e.g., Cooney et al., 2018; Solomon et al., 2016). Therefore, to determine whether there is a regional or geographical dependence on observed LMD heights, the analysis is also subdivided across four regions in the United States: Northern Plains, Southern Plains, Northeast, and Southeast. Lastly, the retrieved LMD heights for different convective morphologies are examined to determine if there is a height dependence on convective mode, by classifying convection as either (quasi-isolated strong convection) QISC or (mesoscale convective system) MCS as previous studies have shown that, for a limited number of cases investigated, supercells tended to detrain mass higher and closer to the LNB than MCSs (e.g., Carletta et al., 2016; Mullendore et al., 2013).

2 Data

2.1 Radar Composites

Radar observations used in this study are merged hourly volumes of NEXRAD WSR-88D S band radars across the continental United States and east of the Rocky Mountains, known as Gridded NEXRAD WSR-88D Radar (GridRad) data (Bowman & Homeyer, 2017). GridRad data are created via space and time binning of data from individual radars with vertical sampling intervals that are frequently less than 750 m and contain contributions from over four radars. The resultant three-dimensional grid has urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00022 km horizontal grid spacing and 1 km vertical grid spacing. During binning, a Gaussian function is used to weight individual radar observations out to 300 km in range and within 5 min of the analysis time. The minimum detectable signal at 300 km is  7.5 dBZ; however, the minimum reflectivity threshold of GridRad data is 0 dBZ. Following the recommendations outlined in Homeyer and Bowman (2017), only grid volumes with a high echo fraction and large cumulative bin weight are retained for analysis. During the binning of individual radar beams, there is no correction for anomalous propagation in the GridRad data and binning assumes standard atmospheric refraction. Anomalous propagation of the radar beam may occur when the index of refraction changes vertically, resulting in either overestimation or underestimation of the height of the beam. Other potential sources of error in the GridRad data include sidelobe contamination, sunlobes (i.e., detection of solar radiation), and technical faults in the operation of radar systems. We use hourly GridRad data for the months of May and July, from 2004 to 2010. Radar observations are analyzed between 25°N and 49°N latitude and between 105°W and 70°W longitude (Figure 1). The analysis domain is split into four regions: North Central (NC), South Central (SC), Northeast (NE), and Southeast (SE). The 37°N parallel and 87.5°W meridian demarcate the North/South and Central/East regions, respectively.

Details are in the caption following the image
The four analysis regions demarcated by the 37°N parallel and 87.5°W meridian.

2.2 Model Reanalysis

The European Centre for Medium-range Weather Forecasts Interim Reanalysis (ERA-Interim; Dee et al., 2011) is used to identify the tropopause height across the analysis region (further described in section 8). The native ERA-Interim data has a horizontal grid spacing of urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-000380 km on a vertical grid containing 60 levels stretching from the surface to 0.1 hPa, with a vertical grid spacing of 250 m near the surface, decreasing to 800 m at 10 km, and 1,200 m at 16 km (Fujiwara et al., 2017).

3 Determining Radar-Based LMD

3.1 Radar Echo Stratification

The Storm Labeling in Three-Dimensions (SL3D; Starzec et al., 2017) algorithm is used to objectively classify radar echo into four mutually exclusive categories: convection, precipitating stratiform, nonprecipitating stratiform, and anvil. SL3D utilizes volumetric radar data to stratify three-dimensional radar echo primarily based on storm height, depth, and intensity. In short, the convective classification encompasses precipitation that is directly generated by convective motions (i.e., updrafts) and is identified by columns of enhanced reflectivity extending vertically in the atmosphere. The precipitating (nonprecipitating) stratiform category consists of mixed-phased cloud that likely has (does not have) precipitation reaching the surface and is identified by the presence (lack) of radar echo near the surface. The anvil category is defined as nonprecipitating radar echo above the freezing level and above 5 km that consists of cloud assumed to contain only ice hydrometeors. For more detailed stratification criteria, see Table 1 in Starzec et al. (2017).

Table 1. The Total Number of Storms Sampled (i.e., LMD Retrievals Performed) for Each Year of the Analysis Period Categorized by Region, Month, and Morphology
Region Month Morphology
2004 4,347 5,537 3,162 6,273 5,712 13,607 2,269 17,050
2005 3,839 6,506 2,538 5,737 3,763 14,857 1,924 16,696
2006 4,927 6,969 2,751 6,144 5,825 14,966 2,002 18,789
2007 5,002 9,033 2,094 6,114 6,198 16,045 2,332 19,911
2008 4,730 5,638 1,758 5,724 3,897 13,953 2,471 15,379
2009 3,724 6,037 1,708 6,022 6,768 10,723 2,708 14,783
2010 4,496 6,997 2,360 5,723 6,714 12,862 2,570 17,006
Total Sampled 31,065 46,717 16,371 41,737 38,877 97,013 16,276 119,614
% Sampled 22.9 34.4 12.0 30.7 28.6 71.4 12.0 88.0

There are instances where low-level data are missing in the reflectivity composites. In regions with poor radar coverage this can result in no data up to 6 km above ground level, which adversely affects the SL3D algorithm. For instance, the SL3D anvil category is defined as reflectivity data present above 5 km with no radar echo below. When a large layer of data up to 5 km is missing, SL3D will still be able to identify convection regions but will stratify all echo around the convection as anvil even though the echo is most likely a stratiform rain region. Including this layer of missing data would result in very low LMD retrievals of urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00046 km. To filter out the erroneous classifications due to missing data, convective cores are disregarded from the analysis (and do not have an LMD retrieved) if no reflectivity data are present within SL3D convective cores up to 4 km but reflectivity data are present at 5 km or more. Since the layer of missing data may only be present in a portion of the convective core, convective cores are disregarded if over 20% of the original core is deemed to have missing data. Additionally, if the first anvil pixel is present at 5 or 6 km, but has a reflectivity magnitude of urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-000515 dBZ (similar to the method used in Carletta et al., 2016), the entire anvil column is disregarded as it is assumed to have missing data below that would change the classification to precipitating or nonprecipitating stratiform.

3.2 Anvil-Based Methodology

To identify the LMD, the convectively generated anvil is used as a proxy for the dynamic detrainment envelope as in Mullendore et al. (2009) and Carletta et al. (2016). The radar reflectivity of the anvil is used to determine the IWC by
where urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0007 is the equivalent radar reflectivity factor (Leary & Houze, 1979). The IWC is horizontally integrated across a 20 km radius from identified convective regions (as discussed below) to determine the total IWC present at each altitude. The height of the maximum in total IWC corresponds to the LMD.

The LMD is found for every precipitating storm cell for each hour, if the precipitating cell contains an identified deep convective core and if adequate anvil is sampled near the convective source. Only anvil within 20 km of an identified convective core or updraft is used for the analysis. The 20-km threshold follows Mullendore et al. (2009) and was verified by a sensitivity study (not shown) to perform well in representing a balance between accuracy of LMD retrieval and storm sample size when applied to the radar data utilized by this study. For example, a 40-km threshold will increase the sample size by including more anvil, but may also start to include edges of the stratiform anvil or anvil from other sources, while a 10-km threshold will reduce the sample size but improve accuracy due to sampling only the closest anvil regions (further discussed in section 13). For an LMD retrieval to occur, there must be at least five anvil grid squares present at one height and at least 25 total anvil grid squares sampled within the 20-km radius to ensure adequate anvil is sampled. Isolated anvil pixels that are not adjacent to any other anvil pixels (horizontally and vertically) are ignored. Identification of precipitating storm cells, convective cores, and anvil is determined by the SL3D algorithm (discussed in section 5).

3.3 Classification of Storm Morphology

Precipitating storms are also objectively classified into one of three distinct categories: weak convection, quasi-isolated strong convection (QISC), and mesoscale convective system (MCS) following criteria similar to Bigelbach et al. (2014). Bigelbach et al. (2014) define QISC and MCSs as precipitating storms having a maximum reflectivity urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-000840 dBZ, containing a deep convective point, and covering urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00097,000 km urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0010 and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00117,000 km urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0012, respectively. Weak convection is defined as precipitating storms with maximum reflectivity urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-001340 dBZ regardless of size. In Bigelbach et al. (2014), a deep convective point is defined as a column containing vertical velocities of at least 2 m s urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0014 at 4 km and 5 m s urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0015 at 8 km. Since no vertical velocity data are available in this study, the deep convective point definition has been altered to follow the SL3D convective definition that focuses on deep convection and uses the criteria of 25 dBZ echo tops urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-001610 km.

3.4 Identification of Tropopause Height

While it is important to know the absolute heights that convection detrains mass to, the height of the LMD in relation to the tropopause height provides more insight into whether mass is potentially being mixed into the stratosphere and allows for better comparison of storms from different regions. Several methods have been previously used to identify the tropopause height in relation to deep convection, including temperature lapse rate, stability, and potential vorticity-based (PV) definitions (e.g., Maddox & Mullendore, 2018, and references therein). While temperature lapse rate and stability definitions typically perform well in identifying the tropopause for sounding data, there are occasional occurrences where both methods are unable to capture the initial tropopause but instead capture the “cold-point” tropopause. This misidentification can result in a high tropopause height bias of up to 6 km or more, which introduces a significant low bias in tropopause-relative LMD height retrievals of deep convection. These misidentifications were also seen by Homeyer et al. (2010) and Solomon et al. (2016) who noted that such misidentifications occur most commonly near the subtropical jet or regions of tropopause folding when the lower vertical resolution of the reanalysis data resulted in an apparent “smoothed” temperature field (see Figure 4 in Homeyer et al., 2010 and Figure 2 in Solomon et al., 2016, and discussion therein).

To avoid introducing such large errors in tropopause heights resulting from lapse rate definitions missing the initial tropopause in the ERA-Interim data, the tropopause is instead identified using Ertel's PV Theorem
where urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0018 is density, urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0019 and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0020 are the horizontal velocity components in the urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0021 and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0022 directions, respectively, urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0023 is the Coriolis parameter, urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0024 is potential temperature, and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0025 is height. The first height with a PVU of urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0026 2.0 with no heights having a PVU urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0027 1.5 above it is identified as the tropopause height (where 1 PVU is 10 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0028 K urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0029m urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0030k urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0031s urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0032 of PV). PV is calculated on the native ERA-Interim grid; however, because of the coarse grid spacing aloft, the PV field is linearly interpolated to a vertical grid with 250 m spacing ranging from 0 to 20 km to more precisely identify the height that crosses the 2.0 PVU threshold. The representative tropopause height used for each storm is the mean tropopause height found within the convective pixels for that storm as determined by the SL3D algorithm. While convection strongly perturbs the PV field at convective scales (i.e., 1 to 10 km; Maddox & Mullendore, 2018), the ERA-Interim data do not resolve convective-scale perturbations due to the coarse grid spacing ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-003380 km), so PV remains a usable field for this study.

4 Results

4.1 Applications of Anvil-Based Methodology

An example application of the anvil-based methodology is presented for an intense supercell located over northeastern Nebraska on 13 July 2004 at 00 UTC (Figure 2). The main storm has reflectivity magnitudes reaching urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-003465 dBZ with the highest echo tops extending above 18 km (Figures 2a and 2c). A few less intense convective cells are visible around the main storm. The SL3D algorithm identifies five convective regions during this time (light brown; Figure 2b) and an extensive anvil emanating from the convective complex with the vast majority being generated by the largest convective region (dark blue). Several precipitating regions are identified (yellow contours); however, only the precipitating regions containing convective cores (thick yellow contours) are considered for sampling following the criteria in section 6. A cross-section through the main storm depicts the long anvil present between 7 and 14 km downstream of the storm and a forward, upstream anvil extending up to 16 km (Figure 2c). The vertical profile of horizontally integrated IWC of all anvil pixels within 20 km of largest convective core (i.e., storm that is intersected by the magenta line in Figure 2b) is shown in Figure 2d. The peak in IWC (and therefore, the radar-derived LMD) is located at 13 km, around 2 km above the tropopause height. The LMD defines the height where the most mass is detrained, but it is clear that the forward anvil has a much wider envelope (Figure 2c); indicating the large vertical spread of detrainment altitudes observed for different parcels.

Details are in the caption following the image
The (a) composite reflectivity, (b) SL3D classification, (c) vertical cross section of reflectivity following A to B, and (d) horizontally integrated IWC for a supercell located in Nebraska on 13 July 2004 at 00 UTC. The yellow lines in (b) denote identified precipitating objects, where the thick yellow lines are the objects that match the LMD retrieval criteria. The black line in (c) and (d) denotes the height of the radar-derived LMD using the anvil-proxy method and the red line denotes the tropopause height.

The anvil-based methodology is also demonstrated for a large MCS located over southeastern Iowa and northwestern Illinois on 18 July 2006 at 01 UTC (Figure 3). The MCS is being forced by a cold front associated with a surface low-pressure center and upper-level shortwave trough located to the northwest of the storm complex (not shown). The majority of the convective activity is located on the southwestern quarter of the storm complex, with additional convective activity found on the northern edge of the complex (Figure 3a). The tropopause heights are found to vary between 9 and 11 km across the region containing the MCS, with lower heights found to the northwest in correlation with the location of low pressure and the incoming trough. The SL3D algorithm identifies a large anvil emanating from the storm complex, with a large region of stratiform precipitation present in the center of the complex (Figure 3b). Anvil (dark blue) used for the LMD retrievals is contoured by a cyan line, demonstrating the extent of the 20 km radius from convective cores. Vertical cross sections taken across and along the anvil next to the large convective cell on the southwestern edge of the complex reveal that the anvil region is urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00357–8 km deep, ranging between 9 and 17 km (Figures 3c and 3d). The LMD for this convective cell is found to be 12 km (black line in panels c and d). Figure 3d illustrates the change in height of the anvil and the slope of the anvil with range away from the convective source. For this case, using the anvil height near point D would result in a low LMD retrieval; however, the 20 km radius results in only anvil at point C to the midpoint point C and D being used for the LMD retrieval.

Details are in the caption following the image
The (a) composite reflectivity, (b) SL3D classification, and vertical cross sections of reflectivity following (c) A to B and (d) C to D in panels (a) and (b) for an MCS located primarily over southeastern Iowa and northwest Illinois on 18 July 2006 at 01 UTC. The gradient fill in panel (a) denotes the tropopause height and the black line in panels (c) and (d) denotes the LMD height retrieved for the very southwestern convective core.

4.2 LMD Height Retrievals

Applying the radar-based LMD retrieval methodology results in a total of 3,203,716 precipitating cells being detected, of which 946,381 (29.5% of cells) contain SL3D convective cores. Of these SL3D cores, 157,230 (4.9% of cells or 16.6% of SL3D cores) are considered deep. Application of the anvil-based method results in a final sample size containing 135,890 storms, meaning 86.4% of deep convective cores meet the required anvil areal thresholds (discussed in section 6) within a 20 km radius to enable a successful LMD retrieval. Table 1 illustrates a breakdown of the number of storms sampled by region, month, and morphology across each year. The total number of storms sampled is greater in the Southern regions, with 65.1% of sampled storms being in either the SC or SE region. Of all sampled storms, 71.4% occurred in July and 88.0% are of QISC morphology. These sample statistics showcase the large amounts of diurnally driven convection that occur over the southern regions during the summer months. Overall, the anvil-based method has a successful application rate ranging from 84.3% to 89.0% of usable cores when looking at different regions, months, or morphologies, implying that the method is not biased toward any of those variables.

After the initial LMD identification, the data undergoes a last round of quality assurance checks to make sure the retrievals are physically consistent. For any particular convective core, if there is only a single height that contains anvil or if the LMD height is found at 6 km (the lowest height that is considered anvil), the LMD retrieval is removed as this is likely due to missing radar data aloft. LMD retrievals are also disregarded if the range between the LMD and anvil top is greater than 5 km or if the anvil top is above 18 km as these were found to be commonly caused by false radar echoes. These thresholds were subjectively determined by looking at a cluster of cases that had unrealistically high LMD heights and anvil top heights, which in many cases extended to nearly the top of the data set. The anvil top is considered the highest height that contains at least 10% of the maximum horizontally-integrated IWC (i.e., 10% of the LMD IWC). This weighted IWC threshold is used to avoid using a strict IWC or reflectivity threshold, which may vary from case to case or between regions, but still eliminates weak reflectivity artifacts that are occasionally found above the anvil region and would result in a biased anvil top height.

The probability density functions (PDFs) of aggregated absolute LMD height and boxplots of tropopause-relative LMD height retrievals for all convection categorized by the months of May and July are shown in Figure 4. When comparing the absolute LMD height distributions (Figure 4a), July storms have the highest probability of LMD heights at 9 km, 1 km higher than May, denoting that convection detrains mass higher in July. July storms also have the highest LMD heights, with outliers reaching up to 15 km. Even though July storms have a higher probability of higher absolute LMD heights, on average, May storms detrain closer to tropopause (Figure 4b). The mean tropopause-relative LMD heights for May and July are urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00364.08 and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00374.31 km, respectively (Table 2). There are storms in both May and July that have LMD heights at and above the tropopause, with July containing a wider distribution of LMD heights that reach up to 2 km above the tropopause. Overall, convection tends to detrain the maximum amount of mass urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00384.25 km below the tropopause. The aggregated LMD height retrievals are also categorized by morphology to investigate the differences between MCSs and QISC (Figure 5). The distributions of absolute and tropopause-relative LMD heights are higher for QISC than MCSs, indicating the QISC have a higher probability to detrain to higher altitudes and closer to the tropopause. The QISC distribution also contains a larger spread in tropopause-relative LMD heights and has more storms that breach the tropopause. The absolute LMD retrievals for July and QISC look similar as the month of July is dominated by QISC type convection (Table 1).

Details are in the caption following the image
The (a) probability density function (PDF) of retrieved LMD heights and (b) boxplots of tropopause-relative LMD heights across the entire analysis period for the months of May and July. For the PDFs, data are present every 1 km. For the boxplots, the black asterisk is the mean, the horizontal red line is the median, top and bottom of the boxes are the 75th and 25th percentiles, respectively, the whiskers are 3 standard deviations from the mean, and the red crosses are the outliers.
Table 2. The Mean Tropopause-Relative LMD and Anvil Top Heights (in km) Categorized by Region, Month, and Morphology
Retrieval Categorization NC SC NE SE All Regions
LMD May urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00393.72 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00404.18 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00414.06 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00424.29 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00434.08
July urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00443.51 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00454.66 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00463.65 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00474.77 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00484.31
MCS urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00493.94 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00504.84 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00514.29 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00524.96 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00534.46
QISC urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00543.50 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00554.46 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00563.66 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00574.64 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00584.22
All convection urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00593.58 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00604.50 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00613.75 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00624.66 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00634.25
Anvil top May urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00641.52 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00651.98 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00661.92 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00672.18 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00681.90
July urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00691.18 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00702.24 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00711.43 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00722.33 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00731.93
MCS urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00741.34 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00752.14 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00761.86 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00772.36 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00781.86
QISC urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00791.28 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00802.15 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00811.49 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00822.29 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00831.93
All convection urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00841.29 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00852.15 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00861.55 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00872.30 urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00881.92
Details are in the caption following the image
As in Figure 4, except categorized by morphology into either MCS or QISC.

The aggregated results across all storm types are subdivided by region and are shown in Figure 6. While all regions contain LMD heights that reach and surpass the height of the tropopause, the distributions for the northern regions are shifted urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00891 km higher relative to the southern regions and contain LMD heights that have a higher probability of reaching the tropopause. The northern regions also have mean tropopause-relative LMD heights that are over 0.75 km higher than the southern regions (Table 2). The regional results are further broken down by month and morphology to identify whether the differences are driven by thermodynamic (i.e., seasonal) changes or by differences in convective dynamics. Figure 7 shows the distributions of tropopause-relative LMD heights aggregated across all storm types for each region for May and July. For May, all regions have relatively similar distributions with the NC region having the highest tropopause-relative LMD heights (Figure 7a). In July, the distributions of tropopause-relative LMD heights for both northern regions shift to higher altitudes while the distributions of heights for the southern regions shift to lower altitudes relative to May (Figure 7b). This shift in the distribution is also visible in the mean tropopause-relative LMD heights when comparing the May and July mean values in all regions (Table 2). The medians show that 50th percentile of storms detrain at an average urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00903.60 km below the tropopause for the northern regions and urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00914.81 km for the southern regions. Only 20% of storms (i.e., the 80th percentile) detrain higher than urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00923.60 km in the southern regions in July, further indicating that May storms commonly transport mass to higher altitudes.

Details are in the caption following the image
The tropopause-relative LMD height categorized by the four analysis regions: North Central (NC), South Central (SC), Northeast (NE), and Southeast (SE). The black asterisk is the mean, the horizontal red line is the median, top and bottom of the boxes are the 75th and 25th percentiles, respectively, the whiskers are 3 standard deviations from the mean, and the red crosses are the outliers.
Details are in the caption following the image
The tropopause-relative LMD height categorized by the months of (a) May and (b) July for the four analysis regions. The black asterisk is the mean, the horizontal red line is the median, top and bottom of the boxplots are the 75th and 25th percentiles, respectively, the whiskers are 3 standard deviations from the mean, and the red crosses are the outliers.

In order to investigate why the southern regions have lower mean tropopause-relative detrainment heights as compared to the northern regions, a subjective analysis is performed by selecting over 100 storms occuring on different days in each region. The selected storms had tropopause-relative LMD heights at either the 10th percentile, median, or the 90th percentile. The analysis reveals that the decrease in mean LMD heights for the southern regions is driven by a large increase in weakly forced, diurnally driven convection that has low tropopause-relative LMD heights and occurs in the summer months. The SC and SE regions have 97% and 304% more LMD retrievals in July than May, respectively. Similar trends are visible when comparing differences between MCSs and QISC by region (Figure 8). Overall, MCSs have a lower distribution of LMD heights than QISC for all regions, and lower mean tropopause-relative LMD heights (Table 2). Only seven MCSs (out of 16,276; Table 1) across the years analyzed have LMD heights at or above the tropopause, while 206 cases of QISC (out of 119,614; Table 1) LMD heights breaching the tropopause are found. A subjective analysis into 48 cases of QISC with the highest LMD heights reveals that most of these storms are supercells based on radar characteristics such as hook echoes, bounded weak echo regions, propagation direction of the storm, and presence of midlevel rotation. Since the GridRad composites only contain hourly reflectivity data, Level II data from the NEXRAD radar closest to each of the 48 storms of interest was used for this analysis. This result follows the findings of previous studies, which showed that supercells detrained to higher altitudes than storms of other morphologies in both models (Bigelbach et al., 2014; Mullendore et al., 2005) and observations (Mullendore et al., 2013). The storm that has the highest tropopause-relative LMD is the supercell storm depicted in Figure 2.

Details are in the caption following the image
As in Figure 7, except categorized by morphology.

While the LMD shows the height of maximum mass detrainment for an individual storm, mass is still being detrained above the LMD within the most unstable (i.e, less mixed) parcels atop the anvil (Carletta et al., 2016); therefore, investigating anvil tops helps identify the maximum height that mass detrains to and provides additional information for when the LMD heights are biased low (discussed in the next section). Figure 9 shows the tropopause-relative heights subdivided by region and month for the height of anvil tops. Anvil tops are defined as in section 11, where tops are denoted as the highest altitude with at least 10% of the maximum horizontally integrated IWC. Overall, the trends between mean anvil top detrainment heights follow similar trends when using the LMD heights (Figure 7). The detrainment heights in May are relatively similar between regions and July detrainment heights increase for the north regions and decrease for the south regions as with the tropopause-relative LMD heights. The mean anvil top detrainment heights are urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00931.92 km below the tropopause in May, as compared to urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00944.25 km below the tropopause for the LMD heights (Table 2). Ten percent of all storms have anvil tops above the tropopause, reaching up to 6 km above the tropopause for the deepest convective cores sampled in July (Figure 9).

Details are in the caption following the image
As in Figure 7, except for the anvil-top heights and not the LMD heights.

A similar anvil top analysis focusing on morphology is shown in Figure 10. The MCS and QISC distributions are more similar when using anvil top heights instead of tropopause-relative LMD heights. The variability in mean anvil top heights between all regions is also smaller than the variability in the tropopause-relative LMD heights (Table 2). The mean anvil top height is closer to the tropopause for MCSs ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00951.86 km) than QISC ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00961.93 km) when accounting for storms in all regions even though the mean tropopause-relative LMD height was consistently higher for QISC (Figure 2). These results need to be further investigated as they imply that while on average the greatest amount of mass is detrained lower in MCSs than QISC, MCSs may have a wider detrainment envelope. These results may also be capturing the large variability observed in several cases of MCS dual-Doppler derived detrainment heights by Mullendore et al. (2013). QISC still has a larger spread of anvil-top detrainment heights and contains the highest detrainment heights overall, which are not surprisingly the same storms that have the highest tropopause-relative LMD heights. The differences between MCSs and QISC also show the importance of morphology on the LMD and detrainment heights. Further detailed analysis needs to be done on why supercells have higher detrainment heights relative to other QISC and MCSs, and why a larger variability in MCS detrainment heights is present. Only seven MCSs have LMD heights at or above the tropopause and a subjective analysis of these MCSs reveals that they were either supercells or other QISC that were previously discrete but began clustering together and transitioning to a linear convective system that met the size criteria of an MCS.

Details are in the caption following the image
As in Figure 8, except for the anvil-top heights and not the LMD heights.

5 Discussion

5.1 Statistical Significance

The findings show that on average, storms in May detrain most of their mass closer to the tropopause than storms in July, and QISC storms have higher tropopause-relative detrainment heights than MCSs. These observational results match the general conclusions of the Bigelbach et al. (2014) modeling study, who used high-resolution Weather Research and Forecasting (WRF) model forecasts of convection in the Southern Great Plains to determine the detrainment heights by using simulated vertical velocity values. There are, however, some differences between the details of both studies. Bigelbach et al. (2014) had almost no storm LMD heights surpassing the tropopause while many such storms were sampled in this study (particularly the north regions). Overall, the mean tropopause-relative LMD heights for May were higher than in July but the highest tropopause-relative LMD heights were more commonly observed in July (Figure 5), which did not match Bigelbach et al. (2014). However, if results are constrained to only the SC region (better resembling the modeling domain used in Bigelbach et al., 2014), the highest tropopause-relative LMD heights occur in May rather than July (Figure 7) matching the Bigelbach et al. (2014) results. The decrease in mean LMD heights from May to July is also visible if focusing on the SC region, albeit the magnitude of decrease is much greater in Bigelbach et al. (2014) ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00972 km decrease vs. urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-00980.5 km in this study; Table 2). Comparing these studies showcases the importance in investigating the regional differences in storm detrainment heights and in using both observations and models to study storms, as observational analyses of storm detrainment heights can constrain model simulations, which can then provide more accurate depictions of the impacts of chemical transport on the composition of the UTLS.

The detrainment envelope depicts the altitudes where parcels of air are detraining from a convective system, but also provides information on the buoyancy and relative dilution experienced by parcels rising through the updraft. A parcel that may be near the updraft edge experiences more mixing and entrainment of environmental air, resulting in a reduction of the parcel's buoyancy and hence, reduces the parcel's detrainment height; however, mixing also dilutes the parcel. Parcels that experience less entrainment (so called “lucky” parcels) are more buoyant and likely detrain to higher altitudes, but are also less diluted. Therefore, although most mass is commonly detrained below the tropopause as visible by the mean LMD heights, the parcels that reach the highest altitudes are likely less diluted and contain a higher concentration of polluted boundary layer air than the parcels that detrain at the LMD and below it (i.e., where most mass is detrained). This implies that although fewer parcels reach or penetrate the tropopause, their effect on altering the chemistry of the UTLS may still be substantial relative to the large number of (more diluted) parcels detraining at the LMD. More research is required to investigate the relationship between detrainment altitudes and parcel dilution to determine the relative impact of the most unstable and undiluted parcels.

Two-sample urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-0099 tests performed comparing the tropopause-relative LMD height and anvil top height distributions of May to July, MCSs to QISC, and intercomparing all regions found the differences to be statistically significant at 99% level. The statistical significance is due to the large sample size; however, statistical significance does not necessarily mean practical significance. For example, is the difference between mean tropopause-relative LMD heights for QISC ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-01004.22 km) and MCSs ( urn:x-wiley:jgrd:media:jgrd56065:jgrd56065-math-01014.46 km) practically important or should more emphasis instead be placed on the outliers? Supercells have been shown to detrain close to or at their LNBs (Mullendore et al., 2013) and make up the outliers in Figure 5, which implies the parcels reaching and breaching the tropopause are relatively undiluted and likely have a greater impact on the chemistry of the UTLS (following the discussion in the previous paragraph). Supercells also transport more mass per updraft element as compared to MCSs, but overall, MCSs transport substantially more mass (Bigelbach et al., 2014). For the sake of discussion, if we were to assume that all QISC have the updraft flux of supercells, then QISC would more commonly transport the maximum amount of mass to higher altitudes. However, the comparison of anvil tops revealed that the distributions (excluding outliers) and averages between QISC and MCSs were very similar (Figure 10), meaning both storm types have “lucky” parcels that can reach similar heights. Therefore, MCSs are also capable of transporting relatively undiluted mass to the same heights as QISC, although less commonly, but because they transport more mass overall, are the differences averaged out? To determine practical significance and to put more context on the results of this study, more work needs to be performed to investigate how much impact these differing distributions would have on the chemical composition of the UTLS and to determine whether differences in mean or median tropopause-relative heights between storm types, regions, or seasons carry any practical importance.

5.2 Limitations

There are some inherent limitations to utilizing the anvil-based methodology. Only storms that are actively producing a notable anvil can be sampled, which excludes developing convection and storms that are embedded within an extensive stratiform rain region. Storms that are developing into preexisting anvil will have their LMD retrieval incorrectly attributed to the height of the preexisting anvil (e.g., the cell north of the large supercell in Figure 2), although this is somewhat limited by focusing only on the deepest convection as weaker convection is generally excluded from the analysis. Lastly, when several deep convective clusters are near each other, anvil may be sampled from multiple convective sources and not just from the convective core being analyzed, which can be sensitive to the anvil search radius used.

Of particular note is the low bias in LMD estimates. The LMD heights in this study likely contain a low bias that is dependent on the anvil search radius. The 20 km anvil search radius was chosen in order to both maximize the accuracy of the retrieved LMD height and minimize the number of convective objects that contribute to a given LMD retrieval. Due to varying environmental conditions and internal dynamic processes, convection may produce limited forward anvil or anvil that spans up to 20 km or more away from the convective source. For the anvil-based LMD retrieval to work properly, adequate anvil near the convective source must be sampled; however, sampling anvil that is too far from the convective source may be detrimental. For example, if the distance threshold is shorter (e.g., 10 km), the accuracy of the LMD retrieval should be improved since only the anvil closest to the convective core is being sampled; however, there may be no or limited anvil present in the sampling distance (i.e., five grid squares), thereby reducing the overall detection rate. Conversely, if the distance threshold is larger (e.g., 40 km) the amount of anvil sampled is increased and a larger sample size of anvil pixels is attained; however, at larger distances the anvil may begin to slope downward due to sedimentation and aggregation of ice hydrometeors (see ice fall speed adjustment in Mullendore et al., 2009), potential for sampling anvil from differing convective source increases, and downstream stratiform anvil becomes included which leads to less accurate (i.e., low-biased) LMD retrievals. An example of anvil slope is visible in Figure 3d. An example of a low-biased LMD retrieval is visible in Figure 2c, where the forward anvil is clearly seen extending at least 1 to 2 km higher than the retrieved LMD height (black line).

When all the limitations and biases are combined, the anvil proxy method likely depicts detrainment heights that are lower than the actual detrainment height, especially when considering that S band radars cannot detect the entire anvil; however, it provides a way to collect a statistically robust observational data set of midlatitude convective detrainment heights that does not currently exist and can be used to help constrain model simulations. The methodology may be further improved by incorporating an improved IWC calculation scheme. The IWC calculation scheme utilized by this study was simplistic but represents a good first-order estimation of IWC that when used as the LMD proxy matched the dual-Doppler divergence profiles (Mullendore et al., 2009). There has been a significant amount of research into ice-phase microphysics and further development of complex IWC calculation schemes and their evaluation (e.g., Fridlind et al., 2015; Hogan et al., 2006; Leroy et al., 2016, 2017; Sayres et al., 2008). If the vertical spacing of the radar data is improved, if data from different wavelength radars is incorporated, or if more parameters are included, more complex IWC calculation schemes can be incorporated to improve the identification of the LMD. An improved IWC calculation scheme can be coupled with dual-polarized radar data, which would enable the use of SL3D's objective updraft identification and provide a way to estimate hydrometeor fall speeds to help account for the low bias.

6 Summary and Conclusions

Seven years of hourly ground-based radar observations for the months of May and July were evaluated to determine the level of maximum (mass) detrainment (LMD) by convection for four different regions: north central, south central, northeast, and southeast United States. Radar observations are objectively stratified by the SL3D algorithm to identify regions of deep convection and convectively generated anvil. Anvil near active deep convection is sampled and used as a proxy for the detrainment envelope, where the maximum in the radar-derived ice water content of the anvil represents the LMD.

Analysis of mean LMD heights shows that July storms tend to have a higher absolute detrainment height than May storms; however, May storms have a higher tropopause-relative detrainment height indicating that on average, May storms are more likely to detrain closer to the tropopause than July storms. Both months contain individual storms that have LMD heights above the tropopause, but July contains the highest LMD heights extending up to 2 km above the tropopause and a higher frequency of storms reaching the tropopause. When categorized by morphology into mesoscale convective system (MCS) or quasi-isolated strong convection (QISC), QISC are found to have both higher absolute and tropopause-relative LMD heights and more commonly have storm LMD heights above the tropopause. Subjective evaluation of the cases with the highest LMD heights reveals that most exhibited the radar characteristics of supercells, supporting the findings of Mullendore et al. (2013).

A regional categorization found that both northern regions had higher mean tropopause-relative LMD heights than the southern regions, regardless of month or morphology. The finding that QISC have higher mean tropopause-relative LMD heights than MCSs was true for all regions; however, there were regional differences between May and July. The mean tropopause-relative LMD heights increased in July for the northern regions but decreased for the southern regions, relative to May. The decrease in the LMD heights in the southern regions is due to a large amount of diurnally driven convection present in the summer months and a seasonal increase in tropopause heights, decreasing the mean LMD heights.

Lastly, while the LMD showcases the height where the maximum mass is being transported to, there is still mass being transported above the LMD. To account for mass transport above the LMD, the anvil top is used as a proxy for the top of the detrainment envelope. Results showed that there are a substantial number of storms with mass being transported well above the unperturbed tropopause (up to 6 km). The mean tropopause-relative anvil top heights for MCSs were found to also be slightly higher than QISC when accounting for storms in all regions, even though the mean LMD height for MCSs was lower, which suggests that MCSs may have a wider detrainment envelope. More research needs to done investigating morphological differences between MCSs, supercells, and diurnally driven convection and their impacts on detrainment altitudes.

This study built a large data set of observed convective detrainment heights in the midlatitudes that was previously absent to provide an idea of aggregated detrainment altitudes and to help constrain model simulations. Although the detrainment heights are likely systematically biased low, the shape of the detrainment profile should still be similarly represented in model simulations and both seasonal and morphological differences should be visible. Furthermore, the low systematic bias implies that the detrainment height results act as a lower bound on the model data. If combined with mass flux estimates, this statistical database can be used to approximate the amount of mass of different chemical species that is being transported into the troposphere and stratosphere. The observed variability present in convective detrainment altitudes for differing morphology has important implications on parameterizations and modeling of mass transport as all convective modes cannot be treated equally. For example, supercells are able to penetrate and transport mass deeper into the UTLS, but MCSs have a wider range of mass detrainment heights, which changes the vertical distribution of transported mass. The observed regional variability in LMD heights further demonstrates the importance of convective morphology, as the diurnally-driven convection notably reduced the mean LMD heights in the southern regions of the United States. The difference in detrainment heights across May and July also indicates that seasonal differences need to be accounted for when estimating long-term transport statistics or influences of transport on climate.


Support for this project was provided by the National Science Foundation (NSF) under Grant AGS-1432930. Mullendore was partially supported by NASA Grant 80NSSC19K0343. Homeyer was partially supported by NASA Grant 80NSSC19K0347. Individual radar data were obtained from the NEXRAD Data Archive at National Climatic Data Center (https://www.ncdc.noaa.gov/nexradinv/) and composited into GridRad data (https://rda.ucar.edu/datasets/ds841.0/). ERA-Interim data were provided by the European Centre for Medium-Range Weather Forecasts (https://www.ecmwf.int/en/forecasts/datasets/archive-datasets/reanalysis-datasets/era-interim).