a. Description of the radar and antenna scanning methodology

The Doppler radar installed on N42 and N43 is described in detail in Jorgensen et al. (1983). The radar is a vertically scanning X-band (3.2-cm wavelength) Doppler radar mounted in the aircraft’s tail. The basic scanning methodology is termed the Fore–Aft Scanning Technique (FAST). This scanning procedure consists of collecting two scans of data, one pointing ∼25° forward from a plane normal to the flight track and one scan pointing aft ∼25°. As the aircraft moves in a quasi-straight flight path, FAST scans sweep out a three-dimensional region of space surrounding the aircraft’s track. If the antenna is rotated at its maximum rate of 10 rpm (60° s⁻¹), at typical P-3 ground speeds (∼120 m s⁻¹) adjacent beams intersect with horizontal spacing of ∼1.4 km. An improvement in the horizontal data spacing to ∼750 m can be obtained if the antenna is scanned in a sector confined to one side of the track. The increase in horizontal data density comes at the expense of the areal coverage on both sides of the track.¹ Since the start times of most flight legs were synchronized to within 2 min (a few of the legs were within 30 s) most of the time delay between beams arises because of the time delay between intersecting forward- and aft-looking beams (Jorgensen et al.1996). For the P-3 antenna this delay is roughly 1 min for each 10-km range from the aircraft’s track. For two flight tracks separated by 40 km, the range to a midpoint would imply a time delay of 2 min. Whatever system evolution occurs during this time delay (plus any delay arising because of different flight leg start times) would produce an error inherent in the analysis. System propagation is handled through advecting the grid with the system motion vector (12 m s⁻¹ to the east).

b. Determination of horizontal and vertical velocities

Where two radar beams intersect, an estimate of the horizontal wind velocity can be obtained using the traditional “dual-Doppler” overdetermined synthesis methods described in Ray et al. (1985). Vertical velocity can then obtained by downward (or upward) integration of the equation of continuity subject to boundary conditions near the surface and echo top, using an O’Brien (1970) correction to the divergence profile. This traditional method of vertical velocity determination using vertical divergence integration uses a zero vertical velocity assumption at echo top. This procedure is known to be in error in regions of active convection because often what is detected at echo top is not really updraft top due to radar insensitivity (Jorgensen et al. 1991), and therefore the echo-top vertical motion could be substantially different from zero. Should echo-top vertical motion be near zero, such as within the stratiform region of an MCS, the dual-Doppler approach should work well. To alleviate this problem in TOGA COARE, where we expected to see deep convective verical velocities, however, a special flight strategy was designed to utilize two Doppler-equipped aircraft to gather two sets of FAST data nearly simultaneously. This strategy (called “quad-Doppler” since up to four Doppler observations are made at each point) required that the aircraft fly parallel flight legs bracketing the area of interest, spaced by 30–50 km. The echo-top vertical air velocity can then be estimated from a “triple-Doppler” solution for the wind field (Jorgensen et al. 1996) using an empirical relationship to remove ice particle terminal velocity from radar reflectivity. The echo-top vertical velocity is then used as a boundary condition to start the downward integration of horizontal divergence to get the column vertical velocity, followed by the O’Brien (1970) correction to zero the surface vertical motion. This method of vertical integration is called a hybrid analysis method since it combines the traditional dual-Doppler scheme with a triple-Doppler solution. The hybrid technique produces about 50% stronger updrafts in the convective line region in TOGA COARE compared to those found applying the traditional assumption of zero vertical velocity at echo top. A comparison with NASA DC-8 in situ vertical velocities obtained during a penetration of the 22 February squall line coincident with quad-Doppler measurements reveals much closer agreement between the in situ vertical velocities and the hybrid technique compared to velocities derived from the dual-Doppler methodology (Jorgensen et al. 1996), lending credence to the validity of using the quad-Doppler methodology in TOGA COARE. To filter noise in the vertical velocities, the horizontal wind fields and the vertical winds derived from the triple Doppler solution at echo top are lightly smoothed with a two-pass Leise filter (Leise 1981) prior to divergence calculation and vertical integration. This procedure dampens features with wavelengths less than about 3–4 times the horizontal grid spacing (1.5 km for this study), which implies that features with wavelengths smaller than 4.5–6.0 km are removed. Theanalysisregions are generally 75 × 75 km on a horizontal grid of 1.5 km and vertical grid of 500 m, centered on the reflectivity maximum of the squall line. The lowest analysis level is 500 m. To partially compensate for the effects of attenuation of the radar beam by intervening precipitation, radar reflectivity is calculated by maximizing the estimates from all the radar gates that pass through a given Cartesian grid point (i.e., up to four beams). Attenuation will mostly affect the estimation of water loading in the buoyancy calculation, with the consequence that the water loading would be underestimated and the buoyancy fields (e.g., Fig. 8) would be too weak. In the course of determining radar biases for TOGA COARE, extensive intercomparison of point values of maximum reflectivity were made for all COARE flights between the aircraft (tail and lower fuselage) and ship radars. The result of this intercomparison indicated that maximum reflectivity values generally differed by less than 2–3 dBZ and could generally be explained by target distances from each radar. If the attenuation were as large as 3 dBZ, the reduction in estimated water content and hence buoyancy, according to Eq. (1), would be 2^4/7 or about 50%. Reflectivity is also used in the wind synthesis to remove terminal fallspeeds from radial velocities. Attenuation is not felt to be a serious problem for wind determination since the aircraft were flying at very low altitudes and attenuation would be greatest at very flat radar elevation angles, where the fallspeed contamination is minimal anyway.

It generally took about 5–8 min to complete each set of flight legs. Over this time, the system is assumed to be stationary.

c. Aircraft positions and ground speeds

Accurate navigation of each aircraft (to an absolute position within ±250 m) is required to properly combine the four radial velocity estimates. In addition, removal of the aircraft motion from the Doppler radial velocities requires that the aircraft’s ground speed be known to an accuracy of better than 0.5 m s⁻¹. The standard aircraft navigation (inertial navigation system, INS) in TOGA COARE was augmented by Global Positioning System (GPS) information. Although normally the GPS positions are sufficiently accurate to properly position the aircraft (the GPS does not suffer from accelerometer drift, or the Shuler oscillation, as does the INS), during TOGA COARE the GPS observations exhibited data gaps and slow updating that resulted in spikes in the ground speed time series. Dropouts most commonly occurred during aircraft turns; it would often take as much as 5 min after the turn was completed before the system locked on to the required number of satellites and navigation was restored. A procedure was developed to merge the accurate ground speed measurements of the INS with the accurate position information of the GPS using a variational adjustment procedure to interpolate over the GPS position gaps (Matejka and Lewis 1997). This procedure produced an accurate dataset of aircraft positions and ground speeds that are mutually consistent; that is, differentiation of the positions produces the ground speeds.

d. Calculation of pressure and buoyancy perturbations

Interpretation of dynamic circulation features of mesoscale convective systems is aided through the determination of perturbation pressure and buoyancy fields. Here the term “perturbation” refers to deviations from horizontal averages. The method used to calculate these fields is termed dynamic retrieval and is based originally on the work of Gal-Chen (1978) as applied to deep convection by Hane et al. (1981). The method has been used successfully to diagnose kinematic structure and inthe calculation of momentum budgets of mesoscale systems (e.g., Jorgensen et al. 1991; LeMone and Jorgensen 1991; Hane and Jorgensen 1995). Although the methodology allows for the use of local time derivatives of the three wind components, for this case steady state is assumed due to the lack of apparent evolution in mesoscale organization and line-averaged flow structure during the highly linear stage of the system during the early phase of the aircraft’s investigation. There could be, however, substantial time variability associated with individual convective cells (Sun and Houze 1992) that might impact retrieval results. To evaluate those impacts, an evaluation of temporal changes in the wind fields and their effect on the retrieved pressure fields was made using a least-squares approach in estimating time derivatives described in Hane and Jorgensen (1995) for the first three coordinated flight legs (scans roughly every 10 min). The differences in the deduced pressure values at each grid point between the steady state fields and those calculated with estimates of time derivatives for the linear part of the line were less than about 10% with very similar spatial patterns. Therefore the steady state assumption for the calculation of retrieved mesoscale fields is felt to be good and is not surprising given the degree of filtering employed and the coarseness of the time resolution. However, the system clearly evolved greatly following the appearance of the line-end vortices, and retrieval calculations will not be discussed here for those later times.

The “most optimal” frame of reference was calculated according to the method of Hane (1993) in order to move the reference frame with the most dynamically active features in the flow. Probably because of the highly organized and consistent motion of the updrafts associated with the leading convective line, the most optimal reference frame determined by the Hane method was virtually identical to the motion of the line determined by tracking the leading edge radar reflectivity pattern (eastward at 12 m s⁻¹). A measure of the relative error in the retrieved pressure gradients is E_r (Gal-Chen and Kropfli 1984). The 22 February retrievals had E_r values of 0.331–0.358, which is in the same range as other Doppler radar studies of mesoscale convective systems (Hane and Jorgensen 1995).

The buoyancy deviation from horizontal averages is represented by

View Expanded

where θ is potential temperature, q′_υ the deviation from base-state water vapor mixing ratio, and q_r the condensate mixing ratio (cloud and rain). Buoyancies are presented as perturbation virtual potential temperatures and do not include water loading, which has been removed with an empirical formula based on radar reflectivity derived by Hauser and Amayenc (1986) from data collected by an optical spectropluviometer (Hauser et al. 1984) in tropical convection:

View Expanded

where ρ is air density and Z the radar reflectivity (mm⁶ m⁻³). Since water vapor and cloud water cannot be independently estimated from radar data, their effects are included in the retrieved temperatures. Although horizontal gradients of retrieved temperatures are accurately portrayed (to the extent the velocity field allows), vertical gradients of temperature should not be interpreted as departures from a base state because active convection covers asubstantial portion of the domain. This caution applies most readily to direct comparisons of retrieved buoyancies (temperatures) to model simulations. In addition, horizontal gradients of retrieved fields of perturbation pressure and perturbation buoyancy are considered more accurate in vertical cross sections than vertical gradients since the calculations contain an unknown integration constant at each vertical level.

e. Data collected

The flight patterns of the two P-3 aircraft consisted of seven quad-Doppler patterns parallel to and on either side of the convective leading edge from 2055 UTC to 2200 UTC at altitudes of 150 and 300 m MSL. Flight leg duration was generally 5–8 min, which equates to leg lengths of 40–64 km. The low-level flight altitudes and proximity to the line gain two advantages: accurate echo-top precipitation vertical motions can be derived because of the high antenna elevation angles, and in situ measurements in the boundary layer could be simultaneously made. Since it takes about 2 min for the aircraft to reverse its heading, a complete Doppler volume scan was repeated every 7–10 min. Up until 2200 UTC, the foci of the flight patterns was on the leading convective line and its change from a highly linear orientation to one that had a pronounced bow in the leading edge. After 2200 UTC both aircraft began flight tracks oriented perpendicular to the convective line (termed “whole system” patterns since the leg lengths were ∼100–150 km to encompass most of the stratiform rainfall region of the system).

a. Horizontal structure

At the lowest levels during the highly linear phase of the squall line the circulation was predominantly two-dimensional. Fairly uniform and strong convergence of air at 1.5 km MSL is indicated in Fig. 4, near the leading reflectivity maxima. At 1.5 km, the system-relative flow behind the line was from the rear to the front of the system. A slight augmentation of this rear inflow of about 4–5 m s⁻¹ be seen near the location in the line where the bow as to take place (within the box region of Fig. 4a). This rear-to-front flow was stronger than the westerlies shown at 800 mb (about 1.5 m s⁻¹ system relative) in the environmental sounding (Fig. 3) indicating some acceleration of the westerlies by the system. Approximately 12–15 km to the rear of the convective line was a relative minimum in radar reflectivity of 25–30 dBZ. The minimum in radar reflectivity is much closer to the convective line than the usual position of the “transition zone” seen in other squall line systems (e.g., Biggerstaff and Houze 1993; Smull and Houze 1987) and is associated with the updraft minimum (to be shown later) between the leading edge updraft and the strong updraft aloft ∼20 km to the rear of the convective line. The effects of attenuation of the X-band radar beam by intervening rainfall cannot by completely discarded since the reflectivity minimum is nearly equidistant from the two flight tracks.² However, the double-banded reflectivity structure is also evident in the model results (Trier et al. 1997; Fig. 2).

Updrafts at 1.5 km MSL (Fig. 4b) were oriented in aribbon approximately 4–7 km wide with maxima of 2–3 m s⁻¹ just ahead of the reflectivity maxima, although the entire width of the low-level updraft channel is not resolved due to the close proximity of the aircraft during this pass. These low-level Doppler-derived vertical motions are much lower than those simulated by Trier et al. (1997) and weaker than vertical air velocities measured by the aircraft during direct penetration of the line (up to 13.5 m s⁻¹ updraft). Since the majority of the low-level inflow occurred below 1 km MSL, it is likely that the Doppler radar suffered from significant sea-clutter contamination and hence underestimation of this low-level inflow, which reduced the magnitudes of the low-level updrafts. Downward motion up to 1 m s⁻¹ dominated the areas farther west, indicating a descending rear-to-front flow. At 9.5 km, the reflectivity maximum has shifted westward, indicative of a westward sloping updraft (Fig. 5) with the dominant flow being front to rear. The updraft character also changed from the narrow ribbon of updraft at low levels to a much more cellular structure at 9.5 km with maximum updraft velocities of 7–8 m s⁻¹.

The retrieved pressure perturbation field at 1.5 km MSL and the retrieved perturbation thermal buoyancy (potential temperature) field at 1.0 km MSL (lowest information level) during the linear stage are shown in Fig. 6. The lowest perturbation pressures (approximately 0.7–0.8 mb) lies just behind (to the west) of the convective line near the north end of the domain near the center of counterclockwise curvature in the low-level flow shown in Fig. 4a. The region of coolest air is also located immediately behind the high reflectivity zone at the north end of the domain. The difference in retrieved temperatures between the air ahead of the system and the rain-cooled air in the wake of the convective line is only about 0.6°C. In situ data collected during a low-level penetration of the leading edge gust front (shown later) shows about a 3°C cooling as the gust front is penetrated at 150 m MSL. This discrepancy is probably explained by the shallowness of the cold pool, which sounding data (to be shown later) reveals is confined mostly below 0.5 km MSL. Such cold pool shallowness is not unexpected; previous studies by Parsons et al. (1994) and Jabouille et al. (1996) documented such structure for COARE convective systems.

b. Line-average vertical cross section through the convective line

The highly two-dimensional structure of the airflow and vertical circulation during the early, linear stage of the convective line can be better characterized by alongline averages that are approximately parallel to the principal part of the leading convective line, which were well observed by the two aircraft radars. The domain of the average is indicated by the rectangles in Figs. 4 through 6. Line-averaged cross sections of radar reflectivity and vertical velocity are shown in Fig. 7. A pronounced characteristic of this squall line is the rearward tilt (east to west, which is relative to the direction of propagation toward the east) with height of the leading edge and airflow. A tilt of roughly 35° from the horizontal is seen in both the vertical airflow and reflectivity contours in the cross section. This pronounced tilt was evident in the early, linear stage of the convective line as well as the bow stage and is very similar to that seen in earlier studies (LeMone et al. 1984a). This tilt was also replicated in the numerical simulations of this system by Trier et al.(1996) and Trier et al. (1997). The cross section depicts many of the common characteristics of tropical squall lines, including a relatively narrow leading edge reflectivity maximum of 10–15 km just to the rear of the leading edge updraft, a rapid reflectivity reduction with height above the melting level (∼4.5 km), and a rear-to-front flow below 4.5 km penetrating to the convective line. Pronounced reflectivity decreases above the melting level are commonly observed in tropical convection, even in convective regions, in distinct contrast to midlatitude convection (Jorgensen and LeMone 1989; Szoke et al. 1986; Zipser and Lutz 1994) and has been associated with much weaker updraft velocities in tropical convection (Zipser and LeMone 1980).

The vertical motion cross section (Fig. 7b) reveals distinct updraft maxima at successively higher heights from the leading edge rearward. Located at distance of 46 km (just ahead of the high reflectivity leading edge) was a 4–5 m s⁻¹ updraft extending upward from about 2.5 km MSL. Since this low-level maximum was ahead of the leading edge it was probably associated with new convective development forced by lifting at the gust front. Just rearward of the leading edge (around point “D”) was a relative minimum in vertical velocity at midlevels (4–9 km MSL), followed farther rearward (at point “B”) by another relative maximum in updraft strength at very high levels (9–15 km MSL). This double-peaked updraft structure with height has been seen in previous Doppler-derived vertical motion fields of tropical squall lines that were embedded in strong lower-tropospheric sheared environments (Chong et al. 1987). The double-peaked structure has also been documented in several numerical simulation studies of multicellular squall lines over a broad range of idealized environmental wind profiles (e.g., Fovell and Ogura 1989) as well as the simulations of this system by Trier et al. (1997), although the secondary updraft maximum seemed to be somewhat transitory in nature compared with these observations. Associated with the double-peaked updraft structure were twin maxima in reflectivity at distances of 42 and 28 km from the left edge of the plot. Farther rearward (distances >20 km from the left edge) there is a faint bright band in reflectivity indicative of stratiform precipitation.

The line-averaged cross section also shows that the region rearward from the active leading edge is dominated by weak descent (between 0 and −1 m s⁻¹) below the sloping updraft and below about 3 km MSL (point “A”). As in earlier squall line studies, this weak descent is associated with the system-relative rear-to-front flow, while the strong updraft zone extending rearward from the leading edge is predominantly front-to-rear flow. The shallow near-surface front-to-rear flow often seen in squall lines [and the simulations of this system by Trier et al. (1996)] just rearward from the leading edge was not seen by the airborne Doppler radar, probably due to sea clutter contamination. Near the echo top, a zone of relatively strong downdraft (up to about −4 m s⁻¹) was seen rearward of the decelerating principal updraft above about 11 km MSL, although the equilibrium level of a undiluted parcel is ∼15 km. Roux (1988) diagnosed strong downdrafts in Doppler-derived vertical motion fields near echo top in a West African squall line. Yuter and Houze (1995) have also documented upper-level downdrafts near deep convective updrafts that are apparently “mechanically” forced by pressure gradientforces that arise near the top of buoyant updrafts. According to their hypothesis, near the tops of these rising updrafts air is being pushed laterally, producing local convergence and downward motion through mass continuity.

The pressure and buoyancy perturbation fields (from virtual temperature and cloud water only) for the same line-average cross section are shown in Fig. 8. The perturbation buoyancy (Fig. 8b) for the principal updraft was positive between 1.5 and 12 km MSL, consistent with active, and perhaps growing, convection. The largest positive buoyancies (∼0.8 K) of the principal updraft were located just above the freezing level between about 5 and 8 km MSL (point “D”) near the location of relative updraft minimum (Fig. 7b). These positive buoyancies in the updraft region are still low compared to an undiluted parcel (∼3°C), which indicates probable mixing of the updraft with environmental air.

The perturbation buoyancy calculated from Doppler-radar wind retrievals are lower by a factor of 2–3 from those calculated by Trier et al. (1997) from simulations, in part because the model outputs temperature (the major contributor to the buoyancy in this case) as deviations from the initial model base state rather than a departure from a horizontal average. In the Doppler retrievals convection covers a moderate percentage of the domain, which would reduce perturbation buoyancy compared to deviations from a base state. Although it may be tempting to attribute the vertical velocity minimum to water loading (especially since it occurs within the maximum of thermal buoyancy), Trier et al. (1997) showed in the simulation that it is the dynamic pressure gradient forces within a few kilometers of the leading edge that act to vertically decelerate upward-moving air parcels immediately rearward of the leading-edge updraft. In addition to the dynamic pressure force, there is a downward-directed pressure force associated with the buoyancy-generated mesolow located farther to the rear (point “A” in Fig. 8a) that nearly compensates the upward-directed buoyancy force (e.g., Trier et al. 1997). Once air moves upward and rearward of this adverse pressure gradient associated with the mesolow, its positive buoyancy results in upward accelerations leading to the secondary updraft maximum.

Near the echo top existed substantial deceleration and cooling, although the updraft region shows negative buoyancy as low as 11–13 km MSL, which is about 4 km below the tropopause. Such strong negative buoyancy this far below the tropopause in Fig. 8b may be partially an artifact of the line-averaging methodology as Fig. 5b shows only two strong updraft cores in the line-average domain. A cross section through one of those cores shows rapid vertical deceleration probably associated with the updraft penetrating beyond its equilibrium level.

Lack of scatterers prevents complete documentation of the buoyancy perturbation ahead of the line. Cool air is evident to the rear of the line below about 3 km MSL, although warming is suggested farther to the rear. The region above 2 km MSL <20 km rearward from the line was weakly descending and was also weakly cool relative to the surroundings, perhaps indicative of melting and evaporative cooling. The simulations by Trier et al. (1997) are consistent with these observations and show this region from 1 to 2 km MSL is warmer than initial conditions, indicative of subsidence-induced adiabatic warming exceeding cooling by evaporation. In all likelihood, the low- to midlevel warming or cooling of the rearward region of squall lines is a delicate balance between rearinflow–produced temperature and humidity advection, hydrometeor supply from the overhead anvil, and mesoscale downdraft strength. The “onion” sounding in the wake region of squall lines (Zipser 1977) is a common observation indicating descending air and subsidence warming exceeding cooling by microphysical processes.

The most striking feature of the perturbation pressure cross section is a mesolow of about −0.3 mb located beneath the sloping updraft in the reflectivity minimum region (distance of 36 km at point “A”). The mesolow is commonly observed in convective lines that exhibit considerable leading-edge tilt (LeMone et al. 1984b), which results in lowering of pressure due to the presence of warm updraft air above and rain-cooled near surface air below. The commonly observed shallow mesohigh pressure ridge beneath the mesolow is not seen in the Doppler-derived retrieval fields, probably either due to the shallow cold pool or because of inadequate resolution of near-surface winds or sea clutter contamination. In situ data from the low-level flight legs, however, show a well-defined mesohigh that is well replicated in the simulations of Trier et al. (1996). The ascending front-to-rear flow is dominated by high perturbation pressure, reminiscent of interaction of the vertical shear of the horizontal wind with the vertical motion (Rotunno and Klemp 1982). Another possible explanation for the high perturbation pressure seen in the ascending flow is that the buoyancy decreases upward between 6 and 9 km MSL.

c. Vortex structure associated with the bow in the line

The flight paths of the two aircraft from 2100 to 2245 UTC were focused on that part of the convective line exhibiting rapid eastward bowing motion. The N42 and N43 aircrafts flew seven consecutive coordinated quad-Doppler flight legs during this period. Low-level flow (1.5 km MSL) obtained from the quad-Doppler wind synthesis is shown in Fig. 9 to illustrate the rapid evolution in circulation as the bow developed. Coincident with the rapid eastward surge in the convective line was the development of a mesoscale vortex at the northern end of the bow. Vortex development (Figs. 9a and 9b) occurred at the northern end of the convective line, where low-level flow was able to turn southward behind the line. The apex of the bow (in Fig. 4a at y = ∼58 km) coincided with the location of strongest rear inflow. At 1.5 km MSL, both the horizontal scale and the magnitude of the relative vertical vorticity increased dramatically near the northern end of the bow within 1 h (from ∼1 × 10⁻³ s⁻¹ in Fig. 9a to ∼2.5 × 10⁻³ s⁻¹ in Fig. 9d). Although initially confined to the lowest levels, the vortex eventually extended to near 5 km MSL. At the later times (Figs. 9e and 9f) the strong low-level north winds produced convergence and new convective line development that was oriented approximately east–west. As the line passed over Guadalcanal Island (Fig. 9g), the convection rapidly decayed (see Fig. 2).

d. Structure of the “transverse bands”

As noted previously in the low-level radar depictions (Fig. 2), several bands of enhanced reflectivity were noted that were oriented approximately transverse to the principal convective line that, in its early stages, was oriented north–northwest to south–southeast. The band farthest south (labeled “A” inFig. 2b) resembles the structure simulated in Trier et al. (1997; their Fig. 10). Upper-level enhanced front-to-rear flow near the southern end of the line apparently transported hydrometeors rearward creating the appearance on radar of a transverse band of reflectivity, although in situ generation of precipitation may have also played a role as the model’s bands formed in moist ascending flow. Such enhanced flow at the extreme southern part of the Doppler analysis domain along y = 5 km is seen in Fig. 10, although the location is too high (4.5 km MSL) to show much enhanced reflectivity. At low levels (Fig. 9g) the band is much better defined.

A notable transformation in the orientation of the enhanced reflectivity region associated with the upper-level updraft and secondary reflectivity maximum to the rear of the convective line occurred as the convective line bowed eastward. During the highly linear stage of the line, this band was primarily oriented parallel to the leading line (Fig. 9a). As the central part of the leading line surged eastward, the portion of the line to the north of the bow became aligned east–west, its inflow being enhanced by southward flow on the west side of the northern vortex. A secondary east–west band appeared immediately to the south of this segment (Fig. 9f along y = 45 km), but it simply marks the location of the upper-level updrafts associated with the east–west portion of the leading edge (cf. Fig. 5). This conclusion is supported by the vertical cross section in Fig. 11, which is oriented approximately NNE to SSW through the line segment. In the figure (which resembles the average east–west cross sections in Fig. 7), the strong front-to-rear flow is provided by the southward flow on the west side of the vortex. Farther south the precipitation structure was more stratiform in nature, as indicated by the bright band near the melting level and the mesoscale downdraft below it.

e. Low-level structure from in situ data

A comparison of the dynamic retrieval values to in situ data is possible at the lowest levels, although the in situ data were collected at a different location and at later times than the retrieved pressures provided by the Doppler data. Following the quad-Doppler patterns parallel to the line, both aircraft performed east–west passes through the line and rearward 100–150 km at various altitudes from 2218 UTC to about 2319 UTC. Cross sections of perturbation (i.e., horizontal means removed) D value (the difference between pressure altitude and radar altitude, i.e., the departure from a “standard” atmosphere defined by the environment) for both aircraft are presented in Fig. 12. A D value of 1 m is about 0.1 mb at altitudes <500 m, so there is qualitatively good comparison between the retrieved mesolow strength (Fig. 8a) and N43’s cross section (Fig. 12a), which went approximately through the apex of the bow (the southernmost track in Figs. 9f and 9g), although by the time of the cross section the convective line was weakening rapidly, probably due to its proximity to Guadalcanal Island. The perturbation high pressure evident at the lowest levels from −20 km to about −90 km from the leading edge in the in situ data is not well represented in the Doppler-derived perturbation pressure, probably due to inadequate resolution of the winds due to sea cluttercontamination. The low-level high pressure is likely due to the cold near-surface outflow from the downdrafts in the high rainfall region of the convective line. The cross section by N42, taken along the northern boundary of the vortex (Fig. 12b) and about 35 km north of N43’s, shows a much narrower and slightly weaker mesolow than N43’s, probably because N43’s was taken underneath the broad, positively buoyant, front-to-rear flow. The N42 winds between 0 and 20 km are influenced by the mesolow; to the west of −20 km to about −90 km the aircraft passes into the cold pool, which is marked by southerly flow and low θ_e air (not shown).

a. Characteristics of the air behind the convective line

In this section we quantify the rate at which the boundary layer can “recover” from squall line passage through enhanced surface fluxes of energy from the ocean. The character of the gust front air following passage of the 22 February squall line is shown in Fig. 13, which displays plots of equivalent potential temperature, potential temperature, mixing ratio, liquid water content, and ground-relative wind for the N42 flight leg segment flown at 150 m from 2148 to 2156 UTC. The flight track shown is the eastern track in Fig. 9d; the aircraft clearly penetrated the heavy rain region of the convective line roughly between the end points of the arrow on the θ_e plot. Note the increased westerlies, colder temperature, and lower mixing ratios; these values correspond to penetrations of the outflow air, while the higher values of θ and q and the northerly winds at the ends of the leg appear to be those of the environment. Although the cloud water content measured by the Johnson–Williams (JW) hot-wire sensor measured around 0.5 g kg ⁻¹ during the penetration, this amount seems high for 150 m MSL, which is below cloud base. Probably rain splashed on the JW sensor giving a false high reading. Within the outflowing air at 150 m, marked by the shift in wind to strong westerlies (recall the system motion was east at 12 m s⁻¹), the potential temperature dropped about 3°C and the air became saturated as the aircraft penetrated rain. The abrupt drop in θ and q as the aircraft penetrated the gust front air corresponds to a drop in θ_e from about 356 K in the inflowing air to about 346 K behind the gust front. Associated with this θ_e drop was a downdraft (not shown) of ∼1.5 m s⁻¹.

The data obtained at 150 m by N43 flying parallel to N42 but to the west of the convective line (western track in Fig. 9c, ∼50 km behind the leading edge) indicate that the aircraft was penetrating air that was a mixture of convective downdraft air and mesoscale downdraft air (Fig. 14). Depending on the convective downdraft intensity and the timing of the downdraft penetration, the demarcation between the two air masses is quite distinct. The penetration of the convective downdraft outflow is marked by a sharp increase in θ_e and q, and a sharp decrease in θ and u (shown in Fig. 14 between the arrow on the θ plot). At this time the back edge of the rain-cooled convective downdraft air was moving eastward at 10 m s⁻¹. Thus, the cool and moist high θ_e air spreads rearward from below the convective band, and the warm, drier, lower θ_eair is associated with the mesoscale downdraft air coming in from the west. A penetration of the convective band about 0.5 h later indicated considerable low-level divergence normal to the band, confirming the association of such air with the convection. The time series in Fig. 14 also suggests that the aircraft was very near the top of the outflow layer; just after the θ drop between the arrows on the q time series, the aircraft passes through about 20 s of cooler air, followed by alternating warm-dry and moist-cool events.

The high degree of correlation between thermodynamic variables and east–west wind u from all quad-Doppler flight legs at 150 m MSL of N43 behind the convective line is shown in Fig. 15. As in the time series plot of Fig. 14, the air encroaching from the west has lower mixing ratios, higher values of θ, and lower values of θ_e. Both the time series and scatterplots suggest that considerable mixing is occurring as well as variation in the thermodynamic character of both convective-scale downdraft and frontward flowing air.

Following the quad-Doppler legs that focused on the convective line from 2100 to 2219 UTC, the two P-3 aircraft began flight legs oriented approximately normal to the convective line and extending >100 km toward the rear of the system. An example of these patterns for the time period 2336 to 2356 UTC is shown in Fig. 16. The southern aircraft (N43) flew at ∼150 m MSL while the northern track was at ∼325 m to avoid some small islands near the middle of the flight leg. Even though the squall line had locally reformed into a quasi-linear north–northwest to south–southeast line, there remained in the rear evidence of a large-scale counterclockwise vortex (anticyclonic in the Southern Hemisphere) in the in situ wind field. The line motion at this time was toward 55° at 12 m s ⁻¹.

Time series of thermodynamic quantities, liquid water, ground-relative wind velocity, and storm-relative wind components along and normal to the flight track (u_s and υ_s, in right-handed, storm-relative coordinates with x axis aligned toward 55° at 12 m s⁻¹) from data obtained by N43 from (the southernmost flight track in Fig. 16) appears in Fig. 17. Two minima in q and θ_e are evident, at ∼20 and ∼80 km behind the leading edge. These minima correspond to divergence in u_s, probably representing maximum penetration of the low θ_e air. The minimum θ_e of 343 K along the flight leg near 80 km approaches but is significantly less than the minimum (338 K) in the environmental sounding, its higher value may indicate some mixing during descent as well as some modification from surface fluxes. The θ_e minimum near the leading edge (∼20 km) is probably associated with a convective-scale downdraft. Rearward of both θ_e minima, the boundary layer seems to be recovering fairly rapidly but not quite to the θ_e values (i.e., 355 K) seen in the environment ahead of the convective line. Cloud or rain, indicated by the JW liquid water meter, are not significant at or below flight level between 20 and 100 km behind the line. Above flight level, however, a deep stratiform cloud was evident on the radar display and in photographs from N43.

The hypothesized air motion associated with the minima is idealized in two dimensions in Fig. 18. At theminimum θ_e point the descending mesoscale downdraft depresses the top of the boundary layer, which grows in depth as the diverging air recovers. The radar-determined vertical velocity (Fig. 19), averaged over the 105 km × 105 km box in Fig. 16, indicates that the air below 2 km is descending on average.

b. Boundary layer recovery from convective line passage

The profiles of θ and q obtained by both P-3 aircraft based on “minisoundings” (quick ascents/descents through the boundary layer) and flight-level data characteristic of the environment ahead of the squall line (Fig. 20a), at the θ_e minimum point ∼20 km to the rear of the convective line (Fig. 20b), and in a more “recovered” region (∼120 km) farther to the rear (Fig. 20c). Also shown in Fig. 20 is an environmental q profile “adjusted” for bias between the two aircraft’s dewpoint sensors.³

We refer to the well-mixed lower layer of air as the boundary layer (BL). The modified BL is about 200 m deep, and the environmental BL is about 500 m deep. We estimate the BL recovery time from observed data using the simplified two-dimensional equation for 〈q̄〉 and 〈θ̄〉. The equation for the time rate of change of 〈q̄〉 is written as

View Expanded

where the first term on the right-hand side of (2) represents horizontal advection, the second term vertical advection, and the third term represents the contribution of water-vapor fluxes from the surface and through entrainment at the top of the BL. The equation for 〈θ̄〉 takes similar form. Entrainment at the top of the air column is assumed zero, since vertical averaging (〈 〉) is over the depth of the fairweather BL (depth h). In other words, we assume that there is no turbulent mixing at h = 500 m in the modified air, something that is certainly valid until the BL gets close to h in depth. Overbars and primes represent local horizontal means and deviations, respectively. The BL is considered recovered when 〈θ̄〉 and 〈q̄〉 reach their environmental values. The effects of rainfall are assumed negligible.

To calculate how quickly the BL air recovers as it moves away from the cold air source, advection is set to zero and the time change of q is approximated by

View Expanded

where t_r is the approximate recovery time assumed the same for q and θ. The two resulting recovery equations for 〈θ̄〉 and 〈q̄〉 have two unknowns, t_r and w̄, and t_r is given by

View Expanded

where Δ_z denotes a change over a fixed depth (here, 100 m) and w̄ is found by substituting t_r in the finite-difference form of (2).

The terms in (3) are estimated from the profiles of θ̄ and q̄ from the bias-adjusted N43dataand N42 data in Fig. 20. Estimates of latent and sensible heat fluxes are computed using the flight-level data using the TOGA COARE bulk aerodynamic technique (Fairall et al. 1996) in the recovering area (Fig. 21) as

View Expanded

With these values, solving the two simultaneous equations yields t_r ∼ 3 h with a mean 500-m MSL subsidence w̄ of 0.02 m s⁻¹. The flight-level data suggests that at the θ_e minimum point the air was receding from the line at about 4 m s⁻¹, implying, if no other processes were acting, a BL recovery within about 45 km from this point. Figure 17 shows similar rates of recovery, and the profiles in Fig. 20 show that air that is about 120 km from the convective line is significantly warmer and moister, although not quite to environmental values. Figure 20 also shows warming and drying above the BL at this location, apparently due to subsidence, which produces an “onion” sounding (Zipser 1977). However, the inversion is not much stronger than that for the environmental sounding in Fig. 20.

To estimate the recovery of the air in squall line relative coordinates, the effects of advection must be added, here idealized as only along the flight leg (NE–SW). From Fig. 17, ∂θ̄/∂x ≈ −2.6 × 10⁻⁵ K m⁻¹. Adding this advection to the surface flux contribution produces a net cooling with time. Although these estimates are a gross simplification, the implication is that this squall line cold dome continued to cool or deepen, since advection of cooler and dryer air from the convective line was greater than the effects of surface fluxes near the θ_e minimum point, unless it is counteracted by the neglected y-advection term.

Since surface fluxes are not significantly different that those computed from earlier field projects (Johnson and Nicholls 1983), we suspect that Eulerian recovery times (at a point) are also similar (∼10 h).