This study characterizes convective clouds that occurred during the Tropical Composition, Clouds and Climate Coupling Experiment as observed within GOES imagery. Overshooting deep convective cloud tops (OT) that penetrate through the tropical tropopause layer and into the stratosphere are of particular interest in this study. The results show that there were clear differences in the areal coverage of anvil cloud, deep convection, and OT activity over land and water and also throughout the diurnal cycle. The offshore waters of Panama, northwest Colombia, and El Salvador were the most active regions for OT-producing convection. A cloud object tracking system is used to monitor the duration and areal coverage of convective cloud complexes as well as the time evolution of their cloud-top microphysical properties. The mean lifetime for these complexes is 5 hours, with some existing for longer than 16 hours. Deep convection is found within the anvil cloud during 60% of the storm lifetime and covered 24% of the anvil cloud. The cloud-top height and optical depth at the storm core followed a reasonable pattern, with maximum values occurring 20% into the storm lifetime. The values in the surrounding anvil cloud peaked at a relative age of 20%–50% before decreasing as the convective cloud complex decayed. Ice particle diameter decreased with distance from the core but generally increased with storm age. These results, which characterize the average convective system during the experiment, should be valuable for formulating and validating convective cloud process models.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Recent observations have shown that stratospheric water vapor (WV) has been increasing over at least the last half century [Oltmans et al., 2000; Rosenlof et al., 2001], so understanding its sources and sinks is crucial for climate change studies. The 2007 Tropical Composition, Clouds and Climate Coupling Experiment (TC4) in Costa Rica was designed to address a number of questions related to the interactions among convection, clouds, and humidity in the tropical tropopause layer (TTL) and lower stratosphere [Toon et al., 2010]. Changes in water vapor within the TTL can play an important role in modulating the climate since water vapor is the most powerful greenhouse gas in the atmosphere. Understanding how water behaves in the TTL is one key to better understanding the impacts of greenhouse gases on global climate change.
 Overshooting deep convective cloud tops that penetrate through the TTL and into the stratosphere have been recognized as a significant source of lower stratospheric water vapor. For example, by analyzing aircraft measurements, Dessler  demonstrated that up to 60% of the water vapor crossing the 380 K potential temperature surface at ∼17 km was detrained above 15 km. In a later study employing airborne measurements, Corti et al.  showed that ice particles from overshooting tops reached as high as 18.8 km. Gettelman et al.  used satellite data to estimate that overshooting tops cover ∼0.5% of the tropics and penetrate up to 1.5 km into the stratosphere. Setvak et al.  employed satellite radiances to show that midlatitude deep convective clouds also inject some of their water vapor into stratosphere. These and other empirical results are consistent with a variety of modeling studies [e.g., Wang, 2003; Jensen et al., 2007; Chemel et al., 2009] that estimate the moisture balance of the TTL and lower stratosphere in the presence of overshooting convection.
 The water vapor and ice crystals introduced into the TTL by overshooting convection are thought to be responsible for the thin, often subvisible, cirrus above 14 km in the tropics [e.g., Wang et al., 1996; Liu, 2007]. These clouds, which are characterized as having very small ice crystals [e.g., Wang et al., 1996], are thought to form from a combination of effects including the direct injection of ice crystals and, more indirectly, convective generated gravity wave pulses clouds above the main convective cloud tops [Garrett et al., 2006]. Fujita  described these clouds as “above anvil cirrus plumes” and/or “jumping cirrus.” Wang  used a cloud model to show that breaking gravity waves atop a deep convective storm can cause some water vapor to detach from the storm cloud and remain in the stratosphere. This water vapor can condense to form a cloud at levels up to 3 km above the anvil [Levizzani and Setvák, 1996]. The above anvil cirrus plumes can extend over 100 km away from the overshooting top (OT) source region (see Figure 1). However, the mechanisms that maintain these thin TTL clouds remain elusive.
 It has been shown by many researchers [e.g., Short and Wallace, 1980; Minnis and Harrison, 1984; Alcala and Dessler, 2002; Liu and Zipser, 2005; Liu et al., 2008] that the deepest convection over tropical land areas peaks during the late afternoon and, over ocean, during the 6 hours after local midnight. Because TC4 was designed to examine the interactions of convection and the TTL, but was limited logistically to flights only during daytime, the aircraft measurements did not sample the complete diurnal cycle. Most flights ended at or before 1600 LT, with only one mission extending to 1700 LT [Toon et al., 2010]. In addition to missing a large portion of the diurnal cycle of convection, the aircraft also sampled only a small portion of each storm system that was encountered during the flights. To fill in the diurnal cycle and provide a large-scale characterization of the convection over the TC4 domain, it is necessary to employ geostationary satellite measurements. Although they do not provide the fine detail available from the in situ and remote sensing measurements aboard the aircraft, satellite measurements can be used to infer much about the context and broader implications of the aircraft data. P. Minnis et al. (Cloud properties determined from GOES and MODIS data during TC4, submitted to Journal of Geophysical Research, 2010a) provided a general overview of the clouds observed from geostationary satellites during TC4, but did not focus specifically on the properties of the convective systems (e.g., areal coverage, OT activity, cloud-top height and microphysical variability) over the diurnal cycles or lifetimes of deep convection. Such information is important for obtaining a more comprehensive picture of TTL-convection interactions and for guiding modeling studies of tropical convection.
 The purpose of this paper is to characterize convective clouds that developed during the TC4 experiment as observed by the Twelfth Geostationary Operational Environmental Satellite (GOES 12). We seek to determine (1) the locations of frequent OT activity throughout the diurnal cycle, (2) the diurnal variability of anvil cloud, deep convective cloud, and OT areal coverage, (3) the characteristic size and duration of individual thunderstorm complexes, and (4) the temporal evolution and spatial variability of satellite-derived cloud-top properties throughout the lifetime of a convective cloud complex. A technique exploiting the water vapor and infrared window bands of GOES 12 [Setvak et al., 2007] is used here to detect OTs as this method provides a direct inference of moisture present in the TTL. The following sections describe the data sets and methodology used to conduct this study in addition to the primary results.
2. Data and Methodology
 Signatures in multispectral weather satellite imagery indicate the presence of OTs and moisture transport into the TTL. OTs exhibit a lumpy or “cauliflower” textured appearance in visible and near-infrared channel imagery as they can be up to 2 km higher than the surrounding anvil cloud [Heymsfield et al., 1991]. OTs are also inferred through the presence of a small cluster of very cold brightness temperatures (BTs) in the ∼11 μm infrared window (IRW) region. OTs continue to cool at a rate of 7–9 K km−1 as they ascend into the lower stratosphere, causing them to be significantly colder than the surrounding anvil cloud temperature [Negri, 1982; Adler et al., 1983; Bedka et al., 2010].
 The WV-IRW brightness temperature difference (BTD) technique, which employs the difference between the 6 and 7 μm WV absorption channel BT minus the 11 μm IRW channel BT, has been described extensively for objective detection of OT clouds [Fritz and Laszlo, 1993; Ackerman, 1996; Schmetz et al., 1997; Setvak et al., 2007; Martin et al., 2008]. Generally, the BTD between these two channels results in a value below zero, since the WV channel weighting function usually peaks at higher altitudes and at lower temperatures than that of the IRW channel. Positive values of this BTD have been shown to identify OTs. The reasons for this correlation are that (1) the atmospheric temperature profile warms with height in the lower stratosphere, (2) water vapor is forced into the lower stratosphere at levels above the physical cloud top by the overshooting storm updraft, (3) this water vapor emits at the warmer stratospheric temperature whereas emission in the IR window channel originates from the colder physical cloud top, and (4) positive differences between the warmer WV and colder IRW BTs can therefore reveal where overshooting is occurring. The aforementioned literature describing the WV-IRW BTD method indicates that the required WV-IRW BTD threshold for OT detection can vary depending upon satellite instrument spatial resolution and spectral channel coverage, intensity of the convective updraft, stratospheric lapse rate, and water vapor residence time in the stratosphere. For 4 km GOES imagery, a BTD value ≥1 K is shown to relate to the presence of overshooting [Martin et al., 2008], whereas a larger value (2–3 K) is a better indicator of overshooting for higher-resolution imagers [Setvak et al., 2007]. The 1 K criterion is used in this study for objective OT identification.
 An example of the WV-IRW BTD OT detection product is provided by Figure 2 for a storm that occurred during the TC4 experiment at 1324 UTC on 24 July 2007. An OT is evident in the GOES 12 visible channel image (Figure 2a) at the center of the convective cloud (∼6.9°N, 86.5°W) and is detected by the WV-IRW BTD method (Figure 2b). The Cloud Physics Lidar (CPL) instrument aboard the NASA ER-2 aircraft observed this cloud at approximately the same time of the GOES image. The CPL cloud-top height retrieval verifies the presence of this OT which has a peak height that extends ∼0.6 km above the surrounding thick cirrus anvil cloud (Figure 2c). After review of all aircraft observations during the entire TC4 experiment, it has been determined that this was the only OT simultaneously observed by both GOES and research aircraft so it is not possible validate OT detection accuracy in a comprehensive manner during TC4.
 To characterize the diurnal variability of the convective clouds, data from half hourly, 4 km GOES 12 imagery are analyzed over a subregion of the greater TC4 experiment satellite domain (Minnis et al., submitted manuscript, 2010a) extending from 3°N to 18°N latitude and 77°W to 92°W longitude. The 22 day time period, 18 July–8 August 2007, analyzed in this study encompasses all of the TC4 tropical flight days. The 10.7 μm IRW channel pixel BTs are used to (1) determine anvil cloud and deep convective cloud domain fractional coverage and (2) define cloud objects that are tracked from the time of first storm detection until decay. The corresponding 6.5 μm WV channel BTs are used to compute the WV-IRW BTD, which defines OT locations for this study. The GOES 12 data used in this study were provided by the University of Wisconsin-Madison Space Science and Engineering Center.
 Cloud properties retrieved from the daytime GOES 12 imagery are used to characterize the micro- and macrophysical variations of convection during TC4. The parameters of interest are the cloud-top height Zt, optical depth COD, ice crystal effective diameter De, and the ice water path IWP. Their values were retrieved by Minnis et al. (submitted manuscript, 2010a) using the visible-infrared-shortwave-infrared technique (VIST), a variant of the four-channel method of P. Minnis et al. (CERES Edition-2 cloud property retrievals using TRMM VIRS and Terra and Aqua MODIS data, part I: Algorithms, submitted to IEEE Transactions on Geoscience and Remote Sensing, 2010b). The postexperiment retrievals were used here. Since the VIST uses the 3.9 μm channel to retrieve De, the values represent the ice crystal sizes only in the upper part of the clouds. The IWP was computed as a function of De and COD. The method of Minnis et al.  was used to estimate cloud-top height from IRW BT for optically thick clouds.
 It is beyond the scope of this paper to validate the VIST microphysical products used in this study, but previous studies have investigated the relative accuracy of VIST relative to aircraft and spaceborne observations. Garrett et al.  show good agreement between VIST De and aircraft observations within a thick convective cirrus anvil cloud. Waliser et al.  show that VIST retrievals of IWP are comparable to those retrieved via CloudSat Cloud Profiling Radar data. Mace et al.  and Minnis et al. (submitted manuscript, 2010b) show good agreement in microphysical properties for thin cirrus. More details of the retrievals and their errors can be found in the work of Smith et al. , Chang et al. , Yost et al. , and Minnis et al. (submitted manuscript, 2010a).
 Before we begin the discussion of the methodology, we will review and introduce acronyms and terminology that will be used throughout the remainder of this paper. Figure 3 provides a schematic that illustrates the cloud types that would be identified through the use of the following criteria: (1) anvil cloud (AC), pixel with an IRW BT ≤ 230 and IRW BT > 215 K; (2) anvil cloud object (AC object), a contiguous area of at least 50 pixels with IRW BTs ≤ 225 K; (3) deep convection (DC), pixel with an IRW BT ≤ 215 K; (4) deep convection object (DC object), a contiguous area of at least 25 pixels with IRW BTs ≤ 215 K; (5) extremely deep convection (extreme DC), pixel with an IRW BT ≤ 200 K; (6) overshooting top (OT), pixels with a WV-IRW BTD ≥ 1 K; and (7) convective cloud complex, a generic term used to refer to a region of deep convective cloud in a satellite image.
 The satellite-observed characteristics of convection present during the TC4 experiment are investigated in three ways. The first analysis involves determining the fractional coverage of AC, DC, and OT over the TC4 subdomain described in the previous section. The purpose of this analysis is to characterize the spatial extent of DC and AC as well as to examine the diurnal and spatial cloud variability during TC4. The total numbers of AC, DC, and OT pixels are computed for each of the 48 images and the fractional coverage is computed by dividing these values by the total number of GOES 12 pixels in the domain. The areal coverage of extreme DC pixels is used to help interpret the AC and DC areal coverage results. A 1 km terrain map is interpolated to the 4 km GOES 12 resolution and navigation and is used to separate land and ocean to understand differences in convective activity over these two surfaces. The 215 K BT criterion used to define DC was also used by Fu et al. , Zhang , and Martin et al. . Rickenbach et al.  experimented with IRW BT values ranging from 240 to 220 K to define AC using GOES 12 observations collected during the Cirrus Regional Study of Tropical Anvils and Cirrus Layers-Florida Area Cirrus Experiment (CRYSTAL-FACE). They show that the areal coverage of clouds that meet a given BT criterion increases with increasing BT, but the timing of maximum areal coverage does not change significantly when the BT criteria were adjusted by ±5 K. Nasiri et al.  used a 230 K threshold as the basis for radiative transfer calculations used to model the microphysical properties of cirrus clouds.
 For the second analysis, convective cloud complexes are defined as coherent “objects” within the IRW imagery and are objectively tracked to determine cloud complex lifetime, duration of DC, and maximum cloud areal coverage. The Warning Decision and Support System, Integrated Information (WDSS-II [Lakshmanan et al., 2007]) is the tool used here to define and track these objects. The component of WDSS-II used in this study employs a hierarchical K means clustering method to identify objects at a user specified minimum spatial scale and BT range. Motion estimates are obtained by maximizing the correlation of translated objects to the images at previous time periods, and smoothed over time using a Kalman filter [Lakshmanan et al., 2003]. Objects are matched across frames by greedy optimization of projected position error and longevity. Lakshmanan et al.  and Lakshmanan and Smith  provide detailed descriptions of the WDSS-II object definition and tracking methodology.
 Objects are assigned an identification (ID) number by WDSS-II and object properties associated with this ID number are tracked from the time of first object detection until decay. It is possible that within the same convective cloud complex, one region of DC may be decaying and warm while another is developing nearby. Since a 215 K threshold is used to define DC objects, the WDSS-II system would define these two DC regions as separate objects with differing ID numbers, even though the two objects may have some relation to and interaction with each other. We handle this by tracking the entire AC as an object, which helps to maintain the time series throughout the lifetime of the convective cloud complex. Though the definition of an AC object may seem inconsistent with the AC definition (215–230 K) used in the aforementioned domain areal coverage analysis, the authors' experience with satellite imagery of convective clouds and use of the WDSS-II indicates that a warmer threshold would often cause AC from distant convection to merge, producing unnaturally large objects that do not represent a single convective cloud complex. When two or more DC objects are found within one larger AC object at the same time, the areal coverage and duration of the two DC objects are combined.
 The third component of this analysis involves analysis of the temporal evolution and spatial variability of the VIST cloud properties for individual convective cloud complexes. The analysis begins with finding the location of the minimum IRW BT in the AC object, which is considered the “core” of the convection. Only the coldest object BT is considered if more than one DC object is found within an AC object. The mean cloud-top height, ice crystal effective diameter, ice water path, and optical depth are computed in a 3 × 3 pixel box surrounding the core. These values, rather than single pixel values at the core areas, are recorded for the 0 km radius data point. The next step is to analyze the cloud properties at increasing radii from the core. The mean cloud properties are computed within 10 km wide concentric rings at 10 km radius intervals from the core out to a 100 km radius. For example, the first ring would include pixels at a distance between 5 and 15 km, the second would include those from 15 to 25 km, continuing out to 100 km.
 We repeat the above process for each image where the AC object is present in the imagery. As an AC object can exist for a wide range of time periods (1–12 or more hours), the time of each image where the object is present is normalized to a number between 0 and 10, with 0 corresponding to the time of first detection and 10 corresponding to the last image before convection decay. For an object with a lifetime of 6 images, the VIST properties from the first image are assigned to time bin 0 and those from the last image are assigned bin 10. The four remaining images are assigned to bins, 2, 4, 6, and 8 as the object had lived 20%, 40%, 60%, and 80% of its lifetime at each of these intervals, respectively. If an object is present for longer than eleven 30 min images (i.e., 5 hours), then VIST properties from multiple images could be assigned to the same bin. In this case, the data are averaged so that one data point resides in each bin. Included in this analysis are AC objects with a duration of ≥3 hours that existed at solar zenith angles ≤76°. The 76° criteria limits this to a “daytime-only” analysis and is found to minimize biases in low-light conditions within the VIST algorithm. These criteria limit our sample size to 127 out of the 877 total (to be described later) AC objects. After the object lifetime is normalized, the average properties of all 127 objects are computed at each of the 11 time intervals and the 11 radii to form a composite that shows cloud temporal evolution and spatial variability throughout the convective complex lifetime.
3. Results and Discussion
3.1. Distribution and Areal Coverage of Deep Convection
 The average diurnal variations in the domainwide fractional coverage of AC, DC, extreme DC, and OT are shown in Figure 4. Over land, the AC, DC, and OT pixels occupy the maximum area within an hour of 1800 LT. Extreme DC and OT activity peaks at 1715 LT and the areal expansion of DC continues for an hour until 1815 LT. As the DC weakens, the cloud tops warm, but convectively induced momentum still causes the AC to expand further until it reaches its peak extent 1 hour later, on average. The pattern of OT activity mirrors that of the DC rather than the extreme DC, as an OT can have an IRW BT > 200 K. A secondary maximum in extreme DC occurs between 1900 and 2100 LT while DC is seemingly dissipating. This will be examined in more detail below.
 Over water, there is a clear maximum in extreme DC and OT activity during the night and early morning hours. The pattern shown by the OT line would suggest that development of strong convection occurs near 0000 LT and increases in coverage through 0600 LT. As the convection continues to develop, the DC expands to peak area near 0900 LT. After the DC dissipates after 0900 LT, the convectively induced momentum continues to expand the cloud while it decays and warms into the BT range classified as AC. The AC area peak lags that of the DC by 5 hours, which is longer the lag observed over land (2 hours). The increase in convection activity in coastal regions over land during the early afternoon is likely producing AC over the offshore waters. This fact, in combination with the still expanding and dissipating AC over water, may be biasing the peak toward a later time than would occur in the absence of a convectively active land region.
 Though the criteria chosen to identify AC and DC in this study have been used to analyze these cloud types in previous studies, Figure 5 shows the impact of varied BT thresholds on the AC and DC areal coverage results during TC4. As expected, when colder BT criteria are used, the areal coverage of DC and AC generally decreases. The clear separation in time of peak AC and DC coverage suggests that two different cloud types are being identified. The timing of peak coverage remains very consistent regardless of BT criteria over land (Figure 5, top). Over water, peak coverage of AC and DC is earlier when colder BT criteria are used (Figure 5, bottom). This result is similar to that of Rickenbach et al.  for CRYSTAL-FACE who found that clouds with warmer BTs reached their maximum area 1–2 hours after the peak areal coverage of colder cloud regions due to anvil cloud expansion. Nevertheless, despite this variability, the results produced by a change in the AC and DC BT criteria do not contradict the primary results from Figure 4, namely the existence of a daytime DC peak over land and nighttime DC peak over water.
Figure 6 shows the locations of OT detections during the afternoon, evening, early morning, and late morning hours, which facilitates interpretation of the patterns shown in Figure 4. During the afternoon, Figure 6a shows that OT activity is most concentrated throughout the entirety of the Panama landmass and over interior regions of Costa Rica and Nicaragua. A significant concentration of OT activity is also present in the waters north of Honduras. This is caused by a series of tropical waves that moved from east to west across the northern portion of this domain during TC4. During the evening hours, OT activity was concentrated over northwest Nicaragua and the interior of Panama. The afternoon convection over Costa Rica and Panama mostly dissipated or moved westward along their Pacific coasts. The presence of convection with large areas of cold BT along the coastlines causes the areal coverage of extremely DC over land to increase during the 1900–2100 LT time period though the total number OTs decreased (see Figure 4).
 During the early morning, OTs were abundant in the offshore and coastal regions of Colombia, Panama, and El Salvador. Analysis of animated GOES 12 IRW imagery (not shown; available online at http://angler.larc.nasa.gov/tc4/) reveals that strong convection that initially formed over northwest Colombia moved westward throughout the morning hours, triggering new development off the coasts of Panama. A similar trend is observed near El Salvador where convection that moves westward off of northwest Nicaragua helps to initiate vigorous new development in the offshore waters. During late morning, the convective clouds near Panama continue to move westward and maintain their intensity while those off of El Salvador mostly dissipate.
 Analysis shows that 54.4% (68.1%) of land (water) OTs was present during the 9 P.M. to 9 A.M. LT period. The OT occurrences are proportionately distributed over land and ocean; 17.8% of the total OT activity occurred over land, which covers 20.4% of the domain. The greatest areal coverage of OTs over the study domain was present on 4 August and the lowest coverage was present on 24 July.
 These results are generally consistent with previous studies that show, over tropical oceans, a broad midafternoon maximum in relatively weak convective cloud tops over ocean is accompanied by a second broad maximum in the most intense convection during the early morning hours [e.g., Minnis and Harrison, 1984; Liu and Zipser, 2005]. The peaks in the AC, DC, and OT over land between 1600 and 1900 LT are typical of most land areas [e.g., Minnis and Harrison, 1984; Liu and Zipser, 2005]. However, the late night maximum in extreme DC and secondary maxima in OT and extreme DC over land during the early morning hours are atypical. According to Figure 6, the greatest concentration over land, during early morning, is found over the Panama-Columbia border region. Liu et al.  found an early morning peak in precipitation over the same region indicating the extreme DC maximum at that time is not unusual.
3.2. Storm Object Analysis
 The analysis will now transition to the characterization of cloud objects defined and tracked by the WDSS-II. Figure 7 shows GOES 12 visible imagery from 31 July 2007 for a convective cloud complex west of Costa Rica. This complex is of special interest to the TC4 experiment as the eastern portion of the cirrus anvil was well sampled by the NASA DC-8 and ER-2 aircraft. In the central portion of the domain in Figure 7a, an OT is apparent in a region of enhanced texture surrounded by a smoother anvil cloud. Extending to the east of the OT is an above-anvil cirrus plume. As noted in the Introduction, an above-anvil cirrus plume indicates the presence of water vapor in the TTL or stratosphere that has condensed to form a cirrus cloud. The plume extended ∼100 km across the top of the convective cloud complex by 1345 UTC (see Figure 7c). The plume could still be seen in the 1415 UTC image (not shown) but no later due to a combination of plume dissipation and a reduction in visible channel image texture with decreasing solar zenith angle.
Figure 8 (bottom) shows that the OT signatures seen in Figure 7, including the plume-producing system, are well captured by the WV-IRW texture method. Though Bedka et al.  and other aforementioned WV-IRW BTD references show that this method can overdiagnose the spatial coverage of OTs, the results shown here and in Figure 2b indicate that detections are isolated to the OT regions.
 AC and DC cloud objects are also well defined in this example. The 1245 UTC panel shows two distinct DC areas in the northern object separated by a narrow area of warmer cloud (see Figure 8, middle). By 1315 UTC, the two DC areas merge, though the area defined by WDSS-II does not extend as far west as the cold cloud area shown in the IRW imagery. At 1345 UTC, the two anvil cloud objects come closer together and later merge into one larger object by 1415 UTC (not shown).
 This AC object was tracked by WDSS-II for a 16 hour period. Figure 9 shows the minimum IRW BT within this object throughout its lifetime. After the first 2 hours where the convective cloud complex intensified, the minimum BT hovered near 195 K for a period of 7 hours. OTs were detected within this object for 10 consecutive hours. During the first 8 hours of existence, the areal coverage of the AC and DC continued to expand, reaching a peak near 1500 UTC. The merger between the northern and southern AC objects in the 1345–1415 UTC time frame is evident in this plot. The results here would indicate that two other mergers later occurred at 1645 and 1815 UTC. Animated IRW imagery (not shown) indicates that the increases in AC and DC areas are also caused by new convective initiation within the bounds of the AC object, which provided a new influx of cold cloud tops. Warming cloud tops and decreasing areal coverage after 1900 UTC indicate that the convective cloud complex is decaying rapidly.
 Since the results have shown that the WDSS-II can accurately monitor the evolution of convective cloud complexes for long time periods, WDSS-II is applied to 30 min imagery over the full duration of the TC4 experiment. We focus on AC objects that exist longer than 30 min and contain a DC object at least once during their lifetime. A total of 877 AC objects met this criteria and are included in the following analyses. Figure 10 shows that the lifetime of the 31 July object described above was in the 95th percentile of all objects tracked by WDSS-II. The mean lifetime of AC objects is 5 hours and those objects contained DC objects for 3 hours (60%) of this time (not shown). Figure 10 also shows that the 31 July convective cloud complex was quite large relative to other complexes that developed during TC4. The mean of the maximum areal coverage of AC objects is 13,900 km2 and the mean of the maximum coverage of DC objects is 3345 km2, indicating that DC covers 24% of a mature convective cloud complex (not shown). Based upon the areal coverage of the WV-IRW BTD OT detections, it is determined that OT pixels represent only 0.15% of a convective cloud complex. It is important to note here that these results are based on the anvil cloud being defined by a 225 K BT. In reality, semitransparent anvil cloud along the periphery of a storm complex can spread and add a significant amount of areal coverage to the anvil cloud statistics, thereby reducing the mean coverage of deep convection and OT pixels.
3.3. Deep Convective Cloud and Anvil Microphysics
 Areal coverage of the convective cloud complexes constitutes only one aspect of the convective cloud lifetime. The ice water budget of the clouds and its impact on the radiation budget are important characteristics of the convection that need to be understood if they are to be properly modeled. The influence on the radiation budget can computed using the De, COD, and the cloud-top temperature or Zt, while the ice water variations can be monitored using IWP.
Figure 11 shows the mean variations of each of those parameters for 127 daytime-only AC objects as functions of the relative age of the system and the distance from the core of the storm using the approach described earlier. The cloud-top height (Figure 11a) of the core drops by ∼0.8 km during the average complex's lifetime, from 15.8 km to 15.0 km. This decrease with age extends 100 km from the storm center, but diminishes to a magnitude of less than 0.3 km at ∼75 km from the storm center. The decrease is not monotonic with the peak heights occurring around 20% into life of the system.
 The mean COD (Figure 11b) of the core is at a maximum of ∼95 for the first half of its existence and drops by ∼40% as it decays. Overall, the average COD varies nonmonotonically, with mean maximum values occurring during the first 50% of the lifetime to a distance of 70 km from the core. Beyond 70 km, the youngest and oldest clouds have the greatest COD, which are about half of the core values. The mean De (Figure 11c) at the core is smallest early in the storm life but generally increases with age. Within the first 20 km radius, De drops by 1–3 μm regardless of age. At distances between 30 and 60 km from the center, De is a maximum about 60% through the average storm's life. It is lowest for the new anvil at this distance range. IWP peaks ∼40% though the storm life when COD and De are near or at their peak values (Figure 11d). IWP drops by a factor of 2 at 100 km from the center at the time of peak storm intensity. When the storm decays, the IWP is lowest and the drop with distance is much less pronounced due to the decreased COD and the homogeneity of De.
 When cores penetrate the TTL, the cloud properties tend to be more extreme. For OT clouds, the mean values of COD, De, and Zt are 104.6, 84.8 μm, and 16.3 km, respectively (not shown). These values can be compared to 83.2, 80.5 μm, and 15.7 km for cores that are not classified as having OTs, respectively. By definition, the cloud top is higher within an OT than in a typical non-OT anvil cloud. The increased COD is due in part to the greater thickness and reflects the intensity of the convection, but it also reflects the age of the convection since the cores are less vigorous with age as seen in Figure 8b.
 When an object can be identified with the WDSS-II, the core is still growing, reaching its peak at 20%–40% into its lifetime. This peak is defined by maxima in all parameters except De. The rise and fall of Zt follows the growth and decay of the convective cloud complex (Figure 11a). The anvil continues to build vertically until the system is in middle age. Its top then drops to lower altitudes at a faster pace than that of the growth stage. The initially smaller particles at the top of the core may be due to the rapid rise of the younger particles in the relatively dry environment above the cloud, whereas, the droplets and ice crystals ascending within the already formed convective core are in a moister environment where they can grow to larger sizes (Figure 11c). This might explain why the anvil contains significantly smaller particles early in its existence, but rapidly accumulates large ice crystals as the convection proceeds. The slower rate of decrease in De with increasing radius as the system decays could also be due to the greater abundance of moisture in the surrounding environment compared to the initial stages. It is not clear why De increases with distance beyond ∼60 km for the younger clouds. Perhaps, there is some influence of other systems overlapping the younger anvils.
4. Summary and Conclusions
 This study characterizes the convective cloud complexes that occurred during the TC4 field program as observed within GOES 12 imagery. There were clear differences in the areal coverage of AC, DC, and OT activity over land and water and also throughout the diurnal cycle. OT detections from the WV-IRW BTD method were used to determine where this activity is most frequent across a subset of the TC4 domain.
 As would be expected, convection over interior land regions increased in areal coverage during the 1315–1815 LT period in conjunction with strong solar heating of a humid tropical air mass. The northern coast and offshore waters of Honduras were particularly active during the 1315–2345 LT period. Northwest Nicaragua and El Salvador also showed a convective maximum within the 1845–2345 LT period. Convection developed and moved across the offshore waters and coastal areas of these two nations during the early morning hours, producing a distinct regional maximum in OT activity. The offshore waters of Panama, northwest Colombia, and El Salvador were the most active regions for OT-producing clouds. The greatest areal coverage of OT activity over the study domain was present on 4 August.
 A large convective cloud complex that occurred on 31 July was examined in detail in this paper as the AC from this complex was well sampled by TC4 research aircraft. The 1 km visible channel imagery showed the presence of an above-anvil cirrus plume that was evident for a 1.5 hour period. This plume was connected to a persistent OT area that served to inject water vapor into the TTL and/or stratosphere that condensed to form a cloud. This large convective cloud complex was detected and tracked for 16 hours by the WDSS-II system. The AC from this complex covered a ∼65,000 km2 area, making it one of the largest and most long-lived of those detected during the TC4 experiment. OTs were found within this complex for 10 consecutive hours.
 A total of 877 AC objects were detected and tracked by WDSS-II throughout the duration of the TC4 experiment. The mean lifetime for these objects is 5 hours with AC at its greatest horizontal extent covering a mean area of 13,900 km2. DC is found within the AC during 60% of the storm lifetime and covers 24% of the total AC area at the time of storm maturity. As most observed OTs are ≤15 km in diameter, they only occupy 0.15% of the AC, on average. The areal coverage proportions of DC and OT are likely a bit too high because a relatively cold 225 K BT threshold is used to define the spatial extent of the anvil cloud.
 The microphysical properties of convective cloud complexes were determined only for clouds that developed and decayed during sunlit conditions. On average, they followed a reasonable pattern with maximum values of Zt and COD occurring at the core roughly at a time corresponding to 20% of the system's lifetime. The values in the AC surrounding the core peaked at relative age of 20%–50% before decreasing as the convective system decayed. De decreased in size with distance from the center of the core but generally increased in size with cloud age. These results, which characterize the average convective cloud complex during the experiment, should be valuable for formulating and validating convective cloud process models.
 The results presented here indicate that the diurnal cycles of convective clouds and OTs are quite consistent with climatology, not only over the experiment area, but also over the Tropics, in general. Thus, the TC4 aircraft measurements, when taken as a whole, should be representative of tropical convective clouds. However, because they were confined to daylight flights, they did not sample most of the clouds that produce OTs, the main source for TTL moisture. Nevertheless, at least, two daytime convective cloud complexes (24 July and 31 July) were sampled by the aircraft and should provide the basis for good case studies.
 It is clear that the satellite imagery is critical for learning the behavior of these convective cloud complexes. By bringing the knowledge gained from the satellite analyses together with the TC4 in situ data and numerical cloud process models, it should be possible to make some important strides in understanding deep convection over the tropics and its interaction with the TTL.
 This research was supported by the NASA Radiation Science Program TC4 project, the NASA Applied Sciences Program, and the Department of Energy Atmospheric Radiation Measurement Program through interagency agreement DE-AI02-07ER64546. The authors thank Kirk Ayers for processing the TC4 VIST microphysical retrievals and the University of Wisconsin Space Science and Engineering Center for providing the GOES 12 data and McIDAS software support.