The Vertical Structure of TOGA COARE Convection

1. Introduction and Objectives
2. Data and Analysis Methods
3. Results of Convective Feature Analysis
   • Feature distributions by cruise
   • Rainfall production by feature characteristics
   • Temporal variation of feature structure
   • Diurnal variations of convective vertical structure
4. Relation of Vertical Structure to Passive Microwave Signals
   • Comparison of 30 dBZ contour heights to 85 GHz brightness temperature
   • Feature characteristics vs. TRMM Microwave Imager (TMI) footprints
5. Vertical Structure and Rainfall vs. "Large Scale" Variables
   • Vertical structure and dry intrusions
   • Rainfall and variables measured by the sounding network
6. Convective Vertical Structure and Convective Heating Profiles
7. References

• Back to Charlotte DeMott's home page
• View my TOGA COARE slide show

Introduction and Objectives

Tropical precipitation plays important roles in providing an energy source for global circulations and in maintaining global radiative equilibrium. Global circulation models have improved our understanding of how precipitation influences both the global circulation and radiative balance, but progress in this area is limited by our knowledge of just how much precipitation falls in the tropics, how the rainfall amounts vary over time, and the three-dimensional structure of precipitating systems. Improving our understanding of tropical precipitation is hampered by the facts that precipitation is highly variable in both space and time and that most of the tropics are covered by ocean.

To overcome these difficulties, the Tropical Rainfall Measuring Mission TRMM is in the process of building a satellite to measure tropical rainfall. The TRMM satellite will pass over a given 5-degree longitude-latitude square twice per day, at different times each day, such that over the course of about a month, precipitation will have been measured at all hours of the day. The sampling strategy of the TRMM satellite and spatial resolution of the payload instruments requires that ground-based measurements of precipitation at high spatial and temporal resolutions be made in order to improve interpretation of the satellite data. The recent TOGA COARE provided an excellent opportunity to collect such ground verification data.

The intensive observing period (IOP) of TOGA COARE, (Tropical Ocean, Global Atmosphere Coupled Atmosphere-Ocean Response Experiment) was conducted in the equatorial western Pacific from November 1992 through February 1993. While the main goal of TOGA COARE was to investigate the interaction of the tropical ocean and atmosphere, one objective was to quantify rainfall and its variability. Since the TRMM payload. The vertical structure of precipitation is also an important scientific issue since it is linked to the vertical structure of atmospheric heating and ensuing large-scale circulations.

With these needs in mind, I have come up with the following thesis objectives:

1. What is the vertical structure of conveciton during TOGA COARE? How does the vertical structure vary with time?
2. What "large scale" variables correlate with precipitation and vertical structure?
3. How do variations in system vertical structure impact resulting heating profiles?

Data and Analysis Methods

Data Processing

Raw data was stored in the IRIS data format (a format developed by SIGMET, Inc.) and then converted to Universal Format using an in-house software package. A similar package developed at NASA/GSFC is called Nsig2uf. Data were then interpolated to a 2 x 2 x 0.5 km Cartesian grid using the Reorder package developed by the Research Data Program of NCAR's Atmospheric Technology Division. For the purposes of expediency and data storage concerns, every other volume collected during the COARE IOP (20-minute resolution) was gridded according to the process described above. This resulted in a nearly continuous time series of over 6200 full volume scans.

Analysis Procedures

• convective, characterized by horizontally inhomogenous "columns" of hydrometeors being carried aloft by strong vertical motions
• and stratiform, characterized by horizontally homogenous vertical motions less than the terminal fallspeed of hydrometeors, which often results in the presense of a radar "bright band" near the freezing level.

Once a volume has been partitioned, individual convective features are identified and assigned a unique number. A convective feature differs from a convective cell in that a feature may be composed of several convective cells. Multi-cell features arise when several individual features are too close together for the algorithm to distinguish them as individual elements. This situation often arises at the leading line of squall line type convection. A database of all convective feature characteristics is then constructed. The database includes information such as feature height, mean rainfall rate, total rainflux, height of the 30 dBZ contour, area, percent of total rainfall, percent of total convective area, and mean reflectivity profile.

Other data used in this work

SSM/I (Special Sensor Microwave/Imager) data were also used to obtain information relating to the vertical structure of convection.

Results of Convective Feature Analysis

Feature distributions by cruise

Studies of convective vertical structure have usually focused on convective feature height. However, there are "internal' variations in convective structure that occur independent of feature height and may impact passive microwave measurements as well as the shape of convective heating profiles. In this work, the height of the 30 dBZ reflectivity contour is used as a second measure of convective vertical structure in addition to feature height. Although the absolute value of the reflectivity threshold used as the second indicator of convective vertical structure is arbitrary, we chose 30 dBZ since this reflectivity value has been correlated with lightning production when it extends above the freezing level. The correlation between 30 dBZ contour height and lightning production implies the presence of an active mixed-phase microphysical process above the freezing level.

The following schematic illustration depicts how two convective features of similar height but different internal structure--30 dBZ contour heights (blue shading)--impact upwelling microwave radiation at 85 GHz (radiant energy proportional to red arrow width). The primary emitter of radiation at this wavelength is liquid water below the freezing level. As the radiation passes through the ice and graupel region above the freezing level, it is scattered in many directions, so the amount of radiation penetrating through the cloud top is reduced. The amount of reduction (scattering) is dependent on the ice mass in the cloud.

The following three panels summarize the distribution of convective features as defined by both cloud top height and 30 dBZ contour height. The distribution of features by cloud top height is much more broad than the distribution by 30 dBZ contour height. Note that these number distributions are fairly constant for all three cruises. Although features with reflectivities in excess of 30 dBZ make up only about 10% of all identified convective features, these features account for over 90% of all convective rainfall.

Rainfall production by feature characteristics

Temporal variations of convective vertical structure

The above analysis suggests that rainfall production by feature height may vary over periods at least as short as one month. However, it is possible that convective vertical structure and the ensuing rainfall production varies over shorter time scales. To investigate this possibility, plots of the temporal evolution of feature distributions have been constructed.

Recalling the bimodal distribution of rainflux by feature height in Cruise 1, it can be seen from the right hand figure that the bimodal distribution resulted from two separate height distributions that occurred at different times. During most of the first cruise, features with heights between 9 and 18 km produced most of the rainfall. However, during the two periods where the majority of feature heights were suppressed (Julian days 318-324 and 332-340 in the left hand figure), most of the rainfall was produced by features with cloud top heights between 5 and 10 km. Short periods of rainfall production by more shallow features occurred during the second and third cruises, but apparently did not last long enough to produce a bimodal rainflux distribution.

The following two plots are identical to those above, except convective features have been classified by the height of their 30 dBZ contour, rather than cloud top height. The distribution of 30 dBZ heights varies substantially during Cruises 1 and 3, and least during Cruise 2. Interestingly, the production of rainfall by 30 dBZ heights does not show as much variability as the number distribution. Especially during Cruises 2 and 3, most of the rainfall is produced by features with 30 dBZ heights in the 6-7 km range. The height of maximum rainfall production during Cruise 1 drops slightly to 3-5 km during the aforementioned "suppressed" periods. A brief example of how variations in convective vertical structure are manifested in passive microwave measurements is the topic of the next section.

Relation of vertical structure to passive microwave signals

Comparison of 30 dBZ contour heights to 85 GHz brightness temperatures

As the schematic figure early in the document indicates, variations in convective vertical structure impact upwelling microwave radiation. We focus here on the effects of 85 GHz radiation. The motivation for this analysis lies with the fact that the MIT radar observes only a small area of precipitation, and we wanted to know if the temporal variations in vertical structure observed over the small area were representative of variations over a larger area. Also, because the SSM/I sensor does not view the same location on the globe with each pass, there would not have been enough times where the ship-area was observed to construct a time series of microwave brightness temperatures.

A time series of polarization-corrected brightness temperatures (PCT) (Spencer et al., 1989) was constructed from the AIP-3 (Algorithm Intercomparison Project) dataset provided by Dr. Beth Ebert of BMRC. The above figure shows the relative sizes of the Intensive Flux Array (IFA), the ship area, and a sample SSM/I swath. In this case, some of the SSM/I measurements are co-located with the ship area, but this is more often the exception than the rule and accounts for some of the discrepancies between the 30 dBZ contour height and PCT time series.

The time series of 30 dBZ contour heights and 85 GHz brightness temperature relative frequencies are shown below. 1 December 1992 is indicated by a vertical black line for reference. Although the SSM/I does not always sample the same area as the radar, more cold brightness temperatures are present during the times when 30 dBZ heights are elevated. Although the agreement between the two time series is generally good, there are some differences. Once such difference can be seen at 16 November. The 85 GHz brightness temperature is very cold, but the radar data indicate relatively shallow 30 dBZ contour heights. Examining the SSM/I data for this day reveals that the SSM/I sensor was sampling intense ITCZ convection to the north of the radar.

Feature characteristics vs. TRMM Microwave Imager (TMI) footprints

Until now, the vertical structure of convection has been classified by either cloud top height or 30 dBZ contour height. In this section, convective features are described using both variables. This analysis greatly reduces the number of features being studied since only about 10% of all identified features have reflectivities in excess of 30 dBZ. However, as has been discussed, these features account for over 90% of the convective rainfall during the COARE IOP.

The following three images depict the frequency, relative rainflux, and mean equivalent diameter distributions of convective features when classified by the two variables (height and 30 dBZ contour height) for each cruise. Mean equivalent diameter was calculated by taking the mean of all feature equivalent diamters for each height-30 dBZ height bin. The contour values correspond to the mean footprint dimensions (a + b) / 2) for each of the five microwave frequencies aboard the TRMM Microwave Imager (TMI).

During Cruise 1, most features were shallow (tops less than about 8 km) with 30 dBZ contour heights less than about 4 km. The distribution of features during Cruises 2 and 3 is somewhat bimodal, with peaks in the same "shallow" region as Cruise 1, but with an additional peak for features with heights of 11 km and 30 dBZ contour heights of 5-6 km. Despite the concentration of shallow features, most of the convective rainfall was produced by deep intense features in all three cruises. However, the abundance of shallow cells present during Cruise 1 (and to a lesser degree during Cruise 2) has produced a "tail" of maximum rainfall production into the shallow region of the contour plot.

The two left-most plots above illustrate that rainfall is produced by convective features with a variety of vertical structures. Since the passive microwave sensors aboard the TMI have fixed horizontal resolutions, it is of interest to know how well these convective features can be resolved by the TMI. The figure on the right illustrates the mean equivalent diamters of convective features as classified by their vertical structure characteristics, with contours corresponding the the mean resolution of each TMI sensor. Overall, features were the smallest during Cruise 1, presumably due to the lesser degree of mesoscale organization during this cruise. Furthermore, the taller and more intense features were also the larger features. For each equivalent diamter bin, the percent of the convective rainfall that can be resolved by each channel is presented below.

Vertical Structure and Rainfall vs. "Large Scale" Variables

Vertical structure and dry intrusions

The previous section documented variations in convective vertical structure over time scales of days to weeks. Since these variations apparently occur over spatial scales larger than the radar field of view, this suggests that some type of large-scale control is acting to modulate convective vertical structure. Since modulation of vertical structure is so well defined during the first cruise, we focus on this time period.

Lin (1995) has identified "disturbed" and "supressed" periods during Cruises 1 and 2 based on mean IR brightness temperature, surface wind speed, and mean vertical motion. According to his paper, the supressed phase runs from 12 November to 10 December, with the exception of 24 and 25 November, which are disturbed. The frequency distribution of 30 dBZ heights (above) generally support this partitioning, although it could be argued that November 21-23 should be included in the short disturbed period.

Numaguti et al. (1995) have studied 4-5 day disturbances in the TOGA COARE region. The 4-5 day period variation is explained by westward-propagating mixed Rossby-gravity waves. Two such disturbances resulted in the advection of dry low-level sub-tropical air into the COARE IFA. The first "dry tongue" advected into the IFA from 10-15 November, while the second made its way into the IFA from 23-28 November. These periods correspond very closely to the periods where the radar observed both suppressed cloud top heights and 30 dBZ contour heights. The intrusion of the first dry tongue into the IFA was well-resolved by the COARE gridded sounding array and is shown in the figure to the right. The northerly advection of the dry tongue as determined by the gridded sounding data agrees quite well with the time series of SSM/I-derived precipitable water maps presented in Numaguti et al. (1995).

The following two figures are profiles of theta (black), theta-e (blue), and saturated theta-e (red) for 13 November (a "supressed" day) and 24 November (a "disturbed" day). Although the base of the stable dry tongue on 13 November is located at about 3 km, the time series of feature height distributions (click here for quick look back at these figures) indicate that the most frequently occurring feature top height was about 5 km, while most of the rainfall on this day was produced by features with cloud top height between 6 and 9 km. Apparently, there was sufficient low-level forcing to allow convective updrafts to break through the stable layer, but the low humidity in this layer quickly evaporated the hydrometeors, therefore limiting the convection's vertical extent. The more moist conditions over the entire depth of the atmosphere on 24 November allowed the features to extend to greater heights.

Rainfall and variables measured by the sounding network

During GATE (GARP Atlantic Tropical Experiment), the time series of radar-derived rainfall was found to be well correlated to the time series of sounding-derived vertical motion. Furthermore, rainfall production was found to be strongly modulated by the diurnal cycle and the passage of easterly waves with 3-5 periods emanating from the African continent. Such comparisons are usually discussed in the context of "large scale modulation of rainfall production." However, this terminology is somewhat misleading when applied to tropical situations since variations in "large-scale" flow and rainfall production are both associated with (usually) transient disturbances.

Perhaps a more accurate interpretation of the correlations between radar-derived rainfall and sounding-derived kinematic variables is obtained by considering the spatial scale of the disturbances. We will focus on correlations between rainfall and vertical motion, or omega. The production of rainfall requires upward motion, so we expect the two quantities to be well-correlated, as they were during GATE. The radar is able to measure rainfall over very small spatial scales, while the scale of kinematic quantities resolved by the sounding network is determined by the distance between soundings. During GATE, the spatial scale of the circulations associated with the easterly waves was large enough to be well resolved by the sounding network. As we shall see, the same cannot be said for disturbance scales during TOGA COARE.

The following figure is a time series of 7-day correlations between radar-derived rainfall and sounding-derived kinematic variables. Correlations are calculated over a 7-day period, and the result is plotted on day four of each run. Focusing on the correlation between rainfall and omega (green curve), note that there are extended periods where the correlation is quite low (note the scale on the right axis). This does not imply that rainfall is being produced in the absence of upward motion. Rather, one explanation is that the periods of low correlation indicate times when the horizontal scale of the circulation responsible for the rainfall production are too small to be resolved by the sounding network. Alternatively, these periods also correspond to low rainfall amounts and weak vertical motion, so excessive noise may play a role in lowering the correlations during these periods.

As an example, consider the period from about Julian day 316-336 (the first 2/3 of the first cruise). The mean correlation during this period is about -0.2, suggesting that the dominant scale of motion during this time is too small to be resolved by the sounding network. It is possible that during this relatively quiet period, the driving mechanism for rain-producing circulations was differential heating of the ocean surface arising from fresh water "footprints" depositied by previous rain cells. In contrast, the correlation between rainfall and omega is approximately -0.7 during the second cruise. Rainfall production during this cruise was dominated by the passage of the Madden-Julian Oscillation (MJO), a disturbance with a large spatial scale that is easily resolved by the sounding network.