Volume 107, Issue D24 p. AAC 17-1-AAC 17-7
Aerosols and Clouds
Free Access

Observations of deep convection in the tropics using the Tropical Rainfall Measuring Mission (TRMM) precipitation radar

C. M. Alcala

C. M. Alcala

Earth System Sciences Interdisciplinary Center, University of Maryland, College Park, Maryland, USA

Search for more papers by this author
A. E. Dessler

A. E. Dessler

Earth System Sciences Interdisciplinary Center, University of Maryland, College Park, Maryland, USA

Department of Meteorology, University of Maryland, College Park, Maryland, USA

Search for more papers by this author
First published: 26 December 2002
Citations: 103

Abstract

[1] A satellite-borne precipitation radar is used to study the penetration of convection bearing large particles to altitudes around the tropical tropopause, a region now known as the tropical tropopause layer (TTL). This overshooting convection has been identified as potentially important in the dehydration of air entering the stratosphere. The global distribution of the radar reflectivity tops in the TTL follows the interseasonal and interannual patterns of the surface precipitation rates. The amount of overshooting convection is ∼5% of total deep convection and ∼1.5% of the total convective rain. In agreement with previous studies, the radar observations show that continental convection typically extends to higher altitudes than oceanic convection.

1. Introduction

[2] The transition from the troposphere to stratosphere in the tropics has come to be understood as a gradual one, rather than a sudden change at a sharp material boundary [Highwood and Hoskins, 1998; Sherwood and Dessler, 2000; Thuburn and Craig, 2002]. Sherwood and Dessler [2000, 2001] have called this region the tropical tropopause layer (TTL) and define it to extend from the level of zero net radiative heating (∼14 km, 150 hPa) to the highest level of overshooting convection (∼18 km, 70 hPa).

[3] The physical processes occurring in the TTL are of great interest. Air entering the stratosphere must transit the TTL, where final dehydration to stratospheric abundances of ∼3.85 ppmv [Dessler and Kim, 1999] occurs. The amount of water vapor in the stratosphere is important because it can affect stratospheric chemistry [Kirk-Davidoff et al., 1999] and the troposphere-surface radiative balance [Forster and Shine, 1999]. Brewer [1949] first proposed that the water is frozen out as the air ascends through the tropical tropopause. However, the mechanism responsible for the dehydration of the air to stratospheric abundances is still an open question [Sherwood and Dessler, 2000; Kley et al., 2000].

[4] The maximum level of neutral buoyancy (LNB) of tropical deep convection is around 14 km, coincident with the base of the TTL. Thus, convection penetrating into the TTL is likely overshooting its LNB. Johnston and Solomon [1979] noted that such convection would experience temperatures far colder than the environment. Growth of ice crystals in these convecting air masses, followed by sedimentation, could effectively remove much of their water. Sherwood and Dessler [2000, 2001] incorporated this process into their model of stratospheric dehydration.

[5] In order to characterize this dehydration, it is necessary to quantify the frequency, heights, and locations of convection penetrating into the TTL. Recent studies of deep convection have used passive measurements of visible and infrared radiance to map the geographical extent of overshooting convection. These studies include Hall and Vonder Haar [1999], Liu et al. [1995], and Mapes and Houze [1993] in the tropical western Pacific warm pool region, Roca and Ramanathan [2000] in the Indian Ocean, Soden [2000] in the Americas, and Hendon and Woodberry [1993] and Gettelman et al. [2002] over the entire tropics.

[6] In order for dehydration to take place in these overshooting events, however, the particles must grow large enough to have significant fall velocities. Infrared radiation is sensitive to the emissivity of the clouds. Once the clouds become optically thick, as in the convective core regions, the brightness temperature is largely independent of particle size. In this paper, we will examine data from the Tropical Rainfall Measuring Mission (TRMM) Precipitation Radar (PR) to identify these regions containing large particles in the overshooting convection. Since the radar was originally designed to measure larger particles, such as raindrops, it is ideally suited for this task.

[7] Other radar systems have previously been used to investigate deep convection. Several of these studies show evidence of convection penetrating into the TTL [Williams et al., 1992; DeMott and Rutledge, 1998; Rickenbach and Rutledge, 1998]. However, these measurements were confined to local geographic regions and were not able to provide observations of overshooting convection over the entire tropics. Simpson et al. [1998] provides the first TRMM PR observations of deep convection extending into the tropical tropopause layer during the Pacific typhoon Paka in early December 1997. Nesbitt et al. [2000] and Toracinta et al. [2002] also use PR data to study precipitating systems in several tropical regions. Both show deep convection extending into the TTL. They also found that continental convection typically displays greater magnitudes of reflectivity extending to higher altitudes than oceanic convection in the tropics.

2. Data and Methodology

[8] In this paper, we will use spaceborne radar measurements from the TRMM satellite to identify and to characterize strong convection extending into the TTL over the entire tropics. The TRMM satellite was launched in 1997 to study rainfall and energy exchange over the tropics. The 35° inclination allows it to sample the diurnal cycle over this entire region using a variety of instruments. Kummerow et al. [1998] provides the detailed specifications on the three primary instruments on the satellite.

[9] TRMM is the first meteorological satellite to carry an active radar system. The 13.8-GHz precipitation radar (PR) scans a 215-km swath every 0.6 seconds, measuring the radar reflectivity factor profiles. For this study, we use the 1C21 data product which contains a horizontal resolution of 4.4 km and a vertical resolution of 250 m at nadir.

[10] These radar data are also compared with cloud brightness temperature measurements from the Visible and Infrared Scanner (VIRS) instrument on TRMM. This instrument measures radiance at 5 different wavelengths. For this study, we will use 10.8-μm radiances from the 1B01 data product for comparison with the PR observations to look at the correlation between the altitudes of convection in the TTL and the cloud brightness temperatures. This instrument records data with a horizontal resolution of 2 km at nadir.

[11] In Figure 1, we show a particularly impressive example of TRMM observations from a thunderstorm over northern Australia during the late afternoon on 2 January 1998. The top panel shows the profiles of radar reflectivity factor along the PR scan. The cross range values are the distances along the surface of the Earth relative to the nadir-pointing location. The bottom panel contains the colocated 10.8-μm brightness temperature measurements from VIRS. Since low brightness temperatures correspond to high altitude clouds, we have inverted the temperature scale for comparison with the radar observations.

Details are in the caption following the image
Colocated PR and VIRS measurements of deep convection penetrating into the TTL over northern Australia at 1817 h local solar time on 2 January 1998 (nadir location: 124°E, 17.81°S). The dashed line in the bottom VIRS panel shows the approximate brightness temperature of the 14-km level. Note that the VIRS temperature scale is inverted for comparison.

[12] The PR radar scan in Figure 1 demonstrates many interesting features about this data set. The convection penetrates well into the TTL, peaking at slightly over 19 km. In fact, this event was one of the highest cloud top altitudes observed during the present study. The signal in the TTL is also exceptionally strong with reflectivity factor values as high as 35–40 dBZ, as is more likely for continental than oceanic convection. In fact, Toracinta et al. [2002] show that for ocean systems, the 40-dBZ contour occurs mostly below 7 km with nearly one-third having no 40-dBZ echo.

[13] Comparison of the PR and VIRS observations in Figure 1 also demonstrates the differences in the sensitivities of these two instruments. Although there is good correlation between the location of the radar maximum altitude and the minimum brightness temperature, the VIRS indicates the presence of a much broader cloud within the TTL. Even at −90 km, where the PR sees no cloud, the VIRS still measures a cloud temperature consistent with the base of the TTL. This demonstrates clearly the sensitivity of VIRS to smaller cloud particles that the PR cannot see.

[14] We determine the maximum altitudes of convection containing large particles from the structure of the radar reflectivity profiles. According to Kummerow et al. [1998], the average noise level for the TRMM PR is ∼15 dBZ. However, examination of the individual profiles shows the noise level can fluctuate from 10 to 20 dBZ, even between profiles within the same scan. Because of these fluctuations, use of a simple threshold value is not applicable. Instead, we will consider both the vertical and horizontal structure of the measured profiles for determination of the radar reflectivity tops. We first require that the profile contain a top layer with a thickness of at least 6 range gates (∼1.5 km) where all the reflectivities exceed a minimum threshold of 12 dBZ. We then compare the high reflectivity tops in the TTL with their neighboring profiles to remove noise-induced artifacts. Although individual convective towers can have widths equal to the horizontal resolution of the PR [Shenk, 1974; Roach, 1967], they are embedded in a much larger cloud system. To establish this horizontal coherence, we further require that 3 of the 8 neighboring reflectivity tops contain altitudes within at least 4 km. This requirement assumes that the noise will not exhibit a horizontal correlation between different scans. We will discuss the sensitivity of our results to the noise threshold in later sections.

[15] The solid line in the top panel of Figure 1 shows the reflectivity top determined by the algorithm for this example. The method appears to work quite well, rejecting the noise features located at cross range values of −12 km, +30 km, +60 km, and +68 km. The simultaneous VIRS IR data also do not show any evidence of a physical signal at these locations. One limitation of our method however is that it cannot identify high layers thinner than 1.5 km.

[16] We can illustrate how the measured reflectivity relates to the particle size using the model of Matrosov et al. [1992, 1994] for order-of-magnitude calculations. For solid ice spheres, with number densities of 400 l−1, similar to the in situ measurements of McFarquhar and Heymsfield [1996], a reflectivity of 12 dBZ would correspond to a mean mass diameter of ∼275 μm. For graupel or snow particles with densities much less than solid ice, even larger particles are required to produce the same reflectivity, assuming the same concentration [Battan, 1973]. For a 50% reduction in density, the median mass diameter would increase by 25%. Particles of this size have terminal fall speeds of ∼1–3 m/s and will sediment rapidly once they exit the updraft, dehydrating the air.

3. Distribution of Deep Convection

[17] For the present study, we have collected the altitudes of the reflectivity tops in each tropical radar profile during January 1998, July 1998, January 1999, and July 1999. We refer to those reflectivity tops exceeding 10 km as “deep convection” and 14 km as “overshooting convection.” Figure 2 shows the fraction of deep convection that is also overshooting. To create this plot, we sort the data into 5° longitude-latitude grid boxes. The average number of PR observations in each box exceeds 5000 per day. Figure 2 also includes the contour line enclosing all surface precipitation rate values greater than 6 mm/day from the monthly averages of the CPC Merged Analysis of Precipitation (CMAP) [Xie and Arkin, 1997, 1996]. The CMAP data uses a 2.5° longitude-latitude grid.

Details are in the caption following the image
The percent of deep convection penetrating the 14-km lower boundary of the tropical tropopause layer within each 5° longitude–latitude box. We define deep convection as PR observations of reflectivity profiles extending above 10 km. The contour in red encloses the surface precipitation values greater than 6 mm/day from the monthly average of the CMAP data for comparison. A logarithmic color scale is used to enhance the details.

[18] The geographical distribution of overshooting convection shows a good correspondence with the CMAP contour in Figure 2, indicating that the method appears to work satisfactorily. The number of pixels containing overshooting convection is quite small, accounting for only ∼0.01% of all the TRMM PR observations, corresponding to 12,000–16,000 pixels in each of the four months. The distributions from both the radar and the CMAP data track the seasonal migration of the regions of maximum convective activity toward the summer hemisphere. In addition, both distributions also show similar results for the location and extent of the intertropical convergence zone (ITCZ) in the Pacific and Atlantic Oceans and the south Pacific convergence zone (SPCZ) in each of the four months.

[19] Figure 2 also illustrates the interannual variations over these years. Both the PR and the CMAP distributions show an expansion of the regions of the most intense convection into the central and western Pacific Ocean during the strong El Niño of January 1998. The moderate La Niña event in the following January causes a contraction of these regions back into the western Pacific.

[20] The definition of deep convection (reflectivity tops exceeding 10 km) is an arbitrary one. To investigate the impact this choice has on our analysis, we use a different normalization in Figure 3. In this figure, we plot the fraction of pixels containing convective rain that also include overshooting convection. The number of convective rain pixels is determined from the number of “convective certain” pixels in the TRMM 2A23 data product. We expect this method to yield a lower fraction of overshooting convection than Figure 2 because the number of convective rain pixels will certainly be bigger than the number of deep convection pixels. For each of the four months, ∼0.7% of the PR observations contain convective rain.

Details are in the caption following the image
The percent of total convection penetrating in the tropical tropopause layer. The number of observations of rain classified as convective (raintype flag values 20–22) in the TRMM 2A23 data product is used to determine the total amount of convection. A logarithmic color scale is used to enhance the details.

[21] The shape of the geographical distributions in the two figures are very similar. Both figures show the same types of interseasonal and interannual variability. Using “convective rain” yields a fraction of overshooting convection 2–3 times smaller than that from our definition of deep convection (>10 km). Thus, our analysis suggests that 1.2–1.8% of convective pixels penetrates into the TTL.

[22] It should also be noted that the number of pixels is determined by the sensitivity of the TRMM PR. It is clear that a more sensitive radar would observe convection penetrating into the TTL more frequently as shown in the comparisons of Heymsfield et al. [2000]. In that sense, our results can be considered a lower limit for the real occurrence.

[23] Figure 4 shows the altitude distribution of deep convection over the tropics. In all of these four months, the fraction decreases approximately exponentially with altitude, with few reflectivity tops observed above 18.5 km. We have also overlaid a line on the distributions showing the percent of deep convection occurring over land for each altitude bin. At lower altitudes, oceanic convection contributes the majority of the reflectivity tops with ∼70% measured at 10 km altitude. As the altitude increases, the fraction of continental deep convection also increases. At the base of the TTL, continental convection accounts for ∼40–50% of the observations. In three of the four months, over 50% of the reflectivity tops observed in the upper TTL occur over land. Although the July 1999 distribution follows the same patterns as the other three, oceanic clouds dominate at all altitudes.

Details are in the caption following the image
Distribution of deep convection over the Tropics (20°S–20°N) as a percent of the total radar-observed cloud tops greater than 10 km. The results are plotted for each 500-m bin of altitude. The line with triangles shows the fraction of cloud tops occurring over continental regions for each altitude bin.

[24] DeMott and Rutledge [1998] have also measured the cloud top (defined by the 0-dBZ level) distribution using ship-based radar data from oceanic convection in the western Pacific. Their results show that below ∼8 km the number of cloud tops increases with increasing altitude. Above this altitude, the dependence of the distribution with altitude is inverted, showing the same behavior as our distributions.

[25] The higher fraction of continental clouds in the TTL is in agreement with the results of Nesbitt et al. [2000]. Using TRMM data from August, September, and October of 1998, they determined from measurements of the maximum altitude of the 30-dBZ contour occurring over mesoscale convective systems (MCS) that continental convection was much more likely to penetrate into the TTL than oceanic convection. They present cumulative distribution functions showing that ∼85% of the identified MCSs over Africa and ∼30% over south America contain at least one 30-dBZ reflectivity measurement greater than 14 km. Their results from the regions in the eastern and western Pacific show almost negligible penetration into the TTL. The 30-dBZ contour height is frequently studied because lightning production is often associated with measurements of this height above the freezing level [Zipser, 1994; Peterson et al., 1996].

[26] We have compiled the reflectivity top data from Figure 4 to determine that ∼5% of the deep convection, defined by the >10-km reflectivity tops, penetrates into the TTL. The fractions are similar in both winter and summer. For comparison, the fraction of deep convection extending over 16.5 km, the typical altitude of the tropopause, is ∼0.44% during winter and ∼0.16% during summer, a significant seasonal difference. Although these value are small, they still indicate that a significant amount of convection reaches the tropopause level and above. This result is consistent with recent work that suggests that transport by overshooting convection well above the 14-km level is required to explain the abundance of ozone and CO in the TTL [Dessler, 2002].

[27] Sensitivity tests indicate that increasing the reflectivity threshold from 12 dBZ to 17 dBZ would increase the tropical average by about a factor of two. This result occurs because the number of reflectivity tops above 10 km falls off faster with increasing threshold than the number above 14 km. Thus, the ratio increases. Therefore, our conclusions that large-particle convection is penetrating into the TTL is not affected by our choice of reflectivity threshold. And the variation of our results with changing threshold can be thought of as a rough measure of the uncertainty of our calculation. Raising the threshold to 17 dBZ will also cause a 40% decrease in the number of rain pixels containing overshooting convection.

[28] It is instructive to compare the simultaneous TRMM VIRS measurements to the PR results. Figure 5 shows the global distribution of the IR brightness temperature observations for each 5° × 5° grid box for the four months. For consistency with Figure 2, the figure contains the observations of temperatures colder than 207.2 K, corresponding to the 14-km lower boundary of the TTL, normalized by all the observations of deep convection colder than 243.9 K, corresponding to an altitude of 10 km.

Details are in the caption following the image
The percent of IR brightness temperature measurements of all deep convective clouds colder than 207.2 K, the approximate temperature at 14 km, the lower boundary of the TTL. A logarithmic color scale is used to enhance the details.

[29] The general shapes of the distributions in Figure 2 and in Figure 5 are quite similar. Both show the same interseasonal and interannual patterns, including the behavior of the ITCZ and the SPCZ. However, quantitative differences exist, primarily due to different sensitivities of the instruments. The PR signal is dominated by large particles lofted into the TTL by strong convective updrafts. While VIRS observes these core regions, it is also sensitive to the surrounding cirrus outflow because its signal is not dominated by the large particles. In addition, because convection is less intense over oceans [Toracinta et al., 2002], VIRS should also measure more high clouds than PR.

[30] The diurnal cycles of the PR and VIRS data are plotted in Figure 6. These plots show the hourly data collected from the two instruments of convection penetrating the lower boundary of the TTL (Z >14 km, T < 207.2 K) in the tropics (20°N–20°S). We have further partitioned these data between the continental and oceanic cases based on the terrain beneath the observation pixel.

Details are in the caption following the image
Comparison of the diurnal cycle information for the PR and VIRS instruments in the Tropics, separated between observations of continental and oceanic convection, for each of the four months of study. The PR data are shown as a solid line and the VIRS data as a dotted line.

[31] The continental data from both instruments maximize in the late afternoon in response to the daytime solar forcing. Using TRMM PR data, Nesbitt et al. [2000] also found a late afternoon maximum in the diurnal cycle of MCSs over Africa and South America. Brightness temperature measurements of Soden [2000], Hendon and Woodberry [1993], and Gettelman et al. [2002] show a continental diurnal cycle in agreement with the VIRS data. In general, the peak in the diurnal cycle is stronger in the PR data than in the VIRS data. In addition, the radar diurnal cycle has a seasonal dependence with a stronger peak occurring in January than in July.

[32] The oceanic diurnal cycle is a much weaker signal, as has been noted by a number of authors [Nesbitt et al., 2000; Hendon and Woodberry, 1993]. The VIRS data show evidence of a morning peak in all of the months, except for January 1999. By examining statistics of the area covered by clouds colder than 208 K, Hall and Vonder Haar [1999] found an early morning peak in the diurnal cycle of the western Pacific during both summer and winter. The PR data generally do not show any clear signal.

4. Conclusion

[33] The PR observations show that several percent of deep convection penetrates into the tropical tropopause layer (TTL), the bottom of which is located at 14-km altitude. Other studies also suggest that this overshooting convection can play an important role in determining the composition of air entering the stratosphere [Dessler, 2002].

[34] As we have discussed, the PR is primarily sensitive to big particles (several hundred microns in diameter). These particles have high fall speeds and therefore short lifetimes in the TTL. Sherwood and Dessler [2000, 2001] have suggested that growth of these particles followed by sedimentation is dehydrating air entering the stratosphere. These data show that such large particles are indeed transported by convection into the TTL. Future analysis will hopefully be able to determine what role such convection plays in the water vapor budget of the stratosphere.

Acknowledgments

[35] This work was supported by a NASA New Investigator Program in Earth Science grant and an EOS/IDS grant, both to the University of Maryland. The data used in this study were acquired as part of the Tropical Rainfall Measuring Mission (TRMM). The algorithms were developed by the TRMM Science Team. The data were processed by the TRMM Science Data and Information System (TSDIS) and the TRMM office; they are archived and distributed by the Goddard Distributed Active Archive Center. TRMM is an international project jointly sponsored by the Japan National Space Development Agency (NASDA) and the US National Aeronautics and Space Administration (NASA) Office of Earth Sciences.