4.1. Comparisons of Upper Troposphere CTHs
 The CPL and GOES-12 matched data are analyzed from 9 selected ER-2 flight days during the July–August 2007 TC4 experiment. Data from four other days (July 14, 25, 29 and August 9) are not included here because they were taken during transit flights or flights dedicated to measuring boundary layer clouds and/or aerosols. The CPL uppermost CTHs were averaged every 10 s. The averaging time of 10 s implies a ground track of ∼2 km since the ER-2 traveled at a speed of ∼200 m/s. Each 10-s averaged CPL CTH was matched with collocated GOES-12 pixel data from the two closest imagery scan times, one scanned before and another scanned after the CPL time. Since the GOES-12 imager scans at 30-min intervals, the collocated GOES-12-retrieved CTHs from the two images scanned before and after were then linearly interpolated in time to match the CPL CTH observation. However, when only one image pixel had retrieved CTH, that pixel CTH was treated as a match to the CPL data if the observing time difference between the image pixel and CPL data was less than 3 min. It is noted that the different times and horizontal resolutions of the GOES-12 and CPL cloud data make the comparisons of CTHs from the two measurements somewhat problematic, for example, a cloud could appear or disappear between the 30-min intervals or it may only occur in part of the pixel. To reduce the impacts of cloud breaks and inhomogeneous CTHs, the comparisons between matched CPL and GOES-12 CTH data are restricted by two conditions: the 10-s CPL data detect 100% cloud coverage and the 10-s CPL averaged uppermost CTH is above the 500-hPa level.
 Table 1 shows the numbers of matched data points obtained for the CPL and GOES-12 and those with CTHs above the 500-hPa level retrieved by the CPL, CO2ATs, and VISSTs from the nine flights. In each flight, the total numbers of matched data (Nmatch) are divided into three categories (under NCPL), denoted by h, m and l, according to the statistics of each 10-s CP CTH data. The h category denotes those that satisfy the aforementioned two conditions, i.e., the 10-s CPL had 100% cloud coverage and the 10-s mean uppermost CTH is above the 500-hPa level. The m category denotes those having a few CPL CTHs above 500 hPa, but their 10-s mean CTH is below the 500-hPa level. The l category then denotes the remaining matched data points which had either lower CTHs or no cloud retrieved by the CPL. In the three categories, the number of matched data having a valid CTP < 500 hPa inferred by the CO2ATs, the old and new VISST are denoted in Table 1 by NCO2AT, NVISST-old and NVISST-new. Only those matched CPL and GOES-12 data in the h category are compared in this study. Numbers in the m and l categories may be less reliable and could indicate data mismatches or overestimations by individual satellite techniques.
Table 1. ER-2 Flight Dates, Time Periods, Numbers of Matched CPL and GOES-12 Data, and the Numbers of Data Having Retrieved CTP < 500 hPa From the CPL, CO2ATs, Old VISST, and New VISSTa
|Jul. 17||12:59:25–16:44:09||1348||1262 h||963||806||890|
|Jul. 19||12:55:21–17:51:41||1777||1053 h||513||450||528|
|Jul. 22||12:29:23–17:15:45||1717||1628 h||1475||1259||1417|
|Jul. 24||12:11:31–18:14:42||2179||1745 h||1292||1225||1312|
|Jul. 31||13:15:56–17:19:40||1462||1462 h||1435||1379||1396|
|Aug. 3||13:49:16–17:51:17||1452||1452 h||1349||1113||1213|
|Aug. 5||13:21:29–16:58:11||1298||1298 h||1244||1143||1218|
|Aug. 6||12:40:47–18:14:03||1999||1694 h||230||191||242|
|Aug. 8||12:40:45–17:40:16||1796||1793 h||1724||1568||1667|
 In general, from comparisons of Nmatch and NCPL, the CPL detected large percentages of CTP < 500 hPa (four days had ∼100%). Based on NCO2AT, the CO2ATs retrieved large percentages (75–98%) of those upper troposphere clouds (CTP < 500 hPa), except for July 19 (∼49%) and August 6 (∼14%). The two versions of VISST also retrieved consistently large percentages of CTP < 500 hPa. The new VISST showed good agreement with the CO2ATs while the old VISST retrieved about 10% fewer than those from the new VISST. More than 2% of the matched data had some scattered CPL CTPs < 500 hPa within 10-s average CTPs that are greater than 500 hPa (the m category). Because of the inhomogeneous cloud top fluctuations and/or broken cloud fields for this category, large discrepancies between the CTP and GOES-12 retrieved CTPs are expected and are therefore excluded from the comparisons. About 9% of the matched data had no CPL < 500 hPa retrieved by CPL. Less than 0.2% of the pixels have no CTP < 500 hPa from the CPL, while CTPs < 500 hPa were retrieved by the CO2ATs and VISSTs (the l category).
 Figure 2 illustrates the matched CTHs inferred by the new VISST (blue), old VISST (green), MCO2AT (red) and SCO2AT (purple) overlaid on the ER-2 CPL vertical cloud mask data for 4 flight days. Each figure shows a 3-h period of matched data obtained during the ER-2 flights on August 8 (Figure 2a), July 31 (Figure 2b), July 17 (Figure 2c) and July 19 (Figure 2d), which were selected to demonstrate different cloud scenarios. An example shown in Figure 3 illustrates the GOES-12 imagery 0.65- (Figure 3a) and 10.8-μm (Figure 3b) data and the MCO2AT (Figure 3c) and new-VISST (Figure 3d) inferred CTPs for the data obtained at 14:45 UTC, August 8, 2007 for the TC4 region. The ER-2 aircraft trajectories (flying at 20-km altitude) for the 3-h time period shown in Figure 2a are also plotted in Figure 3. Note that the aircraft trajectory is for the flight time between 12:40:45 and 15:40:45 (UTC) whereas the GOES-12 images resemble the snapshot at 14:45 (UTC).
Figure 2. Comparisons of the different CTHs inferred from the GOES-12 imager data using the new-VISST (blue), old-VISST (green), SCO2AT (purple), and MCO2AT (red). The CPL cloud vertical mask is shown in gray. (a) August 8 between 12:40:45–15:40:45 UTC. (b) July 31 between 13:15:56–16:15:56 UTC. (c) July 17 between 12:59:25–15:59:25 UTC. (d) July 19 between 12:55:21–15:55:21 UTC.
Download figure to PowerPoint
Figure 3. GOES-12 (a) 0.65-μm and (b) 10.8-μm images and associated (c) MCO2AT and (d) new-VISST derived CTHs for 14:45 UTC 8 August 2007 over the TC4 area with overlaid ER-2 flight tracks between 12:40:45 and 13:40:45 (cyan), 13:40:45 and 14:40:45 (blue), and 14:40:45 and 15:40:45 (yellow).
Download figure to PowerPoint
 During August 8 (Figure 2a), the ER-2 flew over several convective cores and anvils. Comparing the data from this flight (12:40:45–17:40:16) when the CO2ATs had valid CTH retrievals (CTP < 500 hPa), the CPL measured a mean (±standard deviations) CTH of 13.9 ± 1.4 km whereas the MCO2AT, SCO2AT, new VISST, and old VISST inferred 12.3 ± 1.1, 10.7 ± 1.8, 11.4 ± 2.5, and 9.7 ± 2.4 km, respectively. Generally, good agreement among the CPL, MCO2AT, and new-VISST CTHs was found near the convective cores, but away from the cores their CTH differences increased as the anvil cloud optical depths decreased. The MCO2AT CTHs were sometimes a few kilometers lower and the new-VISST CTHs were sometimes much lower than the CPL heights. On average, when compared with the MCO2AT, the new-VISST CTHs were lower by 0.9 km, the old-VISST CTHs were lower by 2.6 km, and the SCO2AT CTHs were lower by 1.6 km.
 On July 31 (Figure 2b), the ER-2 flew over some geometrically thick anvils formed by a large mesoscale complex in the Pacific just off the coast of Costa Rica. The data from this flight (13:15:56–17:19:40) show that when the CO2ATs had valid CTH retrievals, the CPL measured a mean CTH of 16.3 ± 0.3 km whereas the MCO2AT, SCO2AT, new VISST and old VISST inferred mean CTHs of 12.8 ± 1.7, 12.2 ± 2.0, 13.0 ± 2.7, and 11.7 ± 2.5 km, respectively. While all four techniques underestimated the optically thin anvil CTHs by more than 3 km, differences between their mean CTHs were generally quite small (within 1.3 km) with the new VISST being the highest and the old VISST being the lowest. It was also found that the new VISST had better agreement with the CPL for optically thicker anvils (see Figure 2b) and convective cores (see Figure 2a). This day also had the highest percentages of CTP < 500 hPa retrieved by all four techniques (CO2ATs ∼98%, new VISST ∼95% and old VISST ∼94%).
 On July 17 (Figure 2c), the ER-2 flew over a large mesoscale complex off the Pacific coast of Costa Rica. Many optically thin cirrus clouds were missed by the four techniques at the beginning of this flight. The CPL-measured CTHs showed large fluctuations over the mesoscale complex causing problems in collocating the CPL and GOES-12 imager data. The CPL detected CTP < 500 hPa ∼94% of the time, compared to about 71, 66, and 60% for the CO2ATs, the new VISST and the old VISST, respectively. For the period 12:59:25–16:44:09 UTC, when CO2ATs retrieved CTP < 500 hPa, the associated mean CTHs were 12.8 ± 1.8 km (CPL), 12.0 ± 1.5 km (MCO2AT), 10.3 ± 2.2 km (SCO2AT), 10.3 ± 3.1 km (new VISST), and 8.8 ± 3.0 km (old VISST).
 On July 19 (Figure 2d), the ER-2 flew over the cores of several convective systems in the Pacific and then over the Caribbean to measure Sahara dust and low-lying clouds. There were high-altitude sub-visible thin-cirrus clouds above the convective systems during the first couple of flight hours. The sub-visible, thin cirrus clouds were generally not well retrieved by the four satellite techniques, but the new VISST showed significant improvement in the CTH retrievals relative to the old VISST. Comparing the data when CO2ATs had valid CTP < 500 hPa, the mean CTHs inferred on this day were 14.5 ± 1.3 km (CPL), 12.2 ± 1.2 km (MCO2AT), 10.5 ± 1.9 km (SCO2AT), 11.7 ± 2.4 km (new VISST) and 9.2 ± 3.0 km (old VISST). The later periods of this flight were mainly over low-lying stratocumulus clouds [Toon et al., 2010]. Overall, the CPL detected ∼59% of CTP < 500 hPa during the flight as compared to only ∼29%, ∼30% and ∼25%, detected by the CO2ATs, new VISST, and old VISST, respectively.
 On August 6 (Table 1), the CPL detected an extensive, thin layer of sub-visible high-altitude (∼15 km) cirrus clouds that occurred high above a deck of low-altitude (∼1 km) boundary layer clouds [Toon et al., 2010]. The sub-visible cirrus clouds were generally missed by the four satellite techniques, leading to the largest differences in Table 1 between NCPL (1694), NCO2AT (230), NVISST-old (191) and NVISST-new (242). The sub-visible cirrus clouds on this day are responsible for most of the undetected upper troposphere clouds in the passive retrieval results.
 Overall, there were a total of 15,028 matched data points as shown in Table 1. Out of these, ∼89% or 13,387 pixels (NCPL) had CPL-retrieved CTHs above 500 hPa. There were ∼68% (NCO2AT) having CO2AT-retrieved CTHs above 500 hPa (i.e., CTP < 500 hPa). The CO2ATs retrieved CTHs above 500 hPa only 0.5% of the time when the CPL did not retrieve a valid CTP < 500 hPa. The new VISST (NVISST-new) retrieved CTPs < 500 hPa for ∼66% of the pixels in contrast to 61% for the old VISST (NVISST-old). The rates of overestimation by both new and old VISSTs are smaller than the 0.5% by the CO2ATs. Relatively speaking, when the CPL retrieved upper tropospheric clouds (CTP < 500 hPa), the CO2ATs retrieved ∼76%, the new VISST retrieved ∼74% and the old VISST retrieved ∼69% of such upper tropospheric clouds. The findings that large percentages (24–31%) of upper tropospheric clouds were not retrieved by the satellite techniques are reasonable considering the large fractions of optically very thin cirrus clouds that occurred during the TC4 experiment [Toon et al., 2010]. The lidar system is much more sensitive to optically thin clouds than the passive sensors on the GOES-12 imager, which results in more detection of high clouds by the CPL.
 Figure 4 shows scatterplots comparing the CTHs retrieved from the four satellite techniques to those from the CPL for all 9 flight days when the CO2ATs retrieved CTPs < 500 hPa. The mean CTHs are 14.2 ± 2.1, 10.7 ± 2.1, 12.1 ± 1.6, 9.7 ± 2.9, and 11.4 ± 2.8 km for the CPL, SCO2AT (Figure 4a), MCO2AT (Figure 4b), the old VISST (Figure 4c), and the new VISST (Figure 4d), respectively. The corresponding overall mean biases relative to the CPL are −3.5, −2.1, −4.5 km, and −2.8 km. The MCO2AT reduced the mean biases of the SCO2AT by 1.4 km whereas the new VISST reduced the mean biases of the old VISST by 1.7 km. Note that much better agreement between the new VISST and CPL are found for CTH > 14 km. Unlike the new VISST, all of the SCO2AT (Figure 4a), MCO2AT (Figure 4b) and old VISST (Figure 4c) have generally underestimated the CTHs between 14 and 16.5 km.
Figure 4. Comparisons of CTHs inferred from the GOES-12 imager and the CPL data. (a) SCO2AT versus CPL. (b) MCO2AT versus CPL. (c) Old VISST versus CPL. (d) New VISST versus CPL.
Download figure to PowerPoint
4.2. Cloud Emissivities and Multilayer Clouds
 Figure 5 shows the CTH differences (dzc) between the CPL and the four passive methods as a function of the MCO2AT-inferred cloud 10.8-mm effective emissivity (ɛc11). Results in the figure were obtained from the 9-day data shown in Figure 4. Overall mean biases, assuming that CPL CTHs were the truth, are −3.5 ± 2.3 (SCO2AT), −2.1 ± 2.0 (MCO2AT), −4.5 ± 2.9 (old VISST), −2.8 ± 2.8 (new VISST) km, as given in each sub-panel. For more opaque and likely optically thick clouds with ɛc11 > 0.95, the mean dzc were found to be −1.9, −1.4, −2.4, and −0.2 km for the SCO2AT (Figure 5a), MCO2AT (Figure 5b), old VISST (Figure 5c), and new VISST (Figure 5d), respectively. The underestimation of CTH by 1.4–2.4 km for those nearly opaque clouds (except for the new VISST case) are consistent with earlier results found by Sherwood et al.,  , who showed that the satellite infrared-derived CTHs were 1–2 km below the physical cloud tops detected by lidar instruments. This underestimation appeared to have been largely corrected for optically thick clouds, using the method of Minnis et al. [2008b] in the new-VISST algorithm (Figure 5d).
Figure 5. CTH difference dzc as a function of the 10.8-μm cloud effective emissivity ɛc11. (a) SCO2AT minus CPL. (b) MCO2AT minus CPL. (c) Old VISST minus CPL. (d) New VISST minus CPL. Thick gray lines represent the running means.
Download figure to PowerPoint
 For less opaque clouds with ɛc11 < 0.95, the absolute differences increase progressively with decreasing ɛc11. For instance, for semitransparent clouds at ɛc11 ∼ 0.3, the mean dzc were found to be −5.1 km (SCO2AT − CPL), −2.8 km (MCO2AT − CPL), −5.7 km (old VISST − CPL) and −3.9 km (new VISST − CPL). Note that the MCO2AT appeared to have more overestimated CTHs for less opaque clouds (ɛc11 < 0.8) and have the overall smallest mean dzc compared to the SCO2AT (Figure 5a) and two VISSTs (Figures 5c and 5d).
 To examine the impact of multilayer clouds on the retrievals, Figure 6 shows the CTH differences from Figure 5 plotted as a function of the 10-s averaged number of cloud layers (Nlayer) retrieved by the CPL. In general, the absolute mean dzc of all four techniques increase with increasing Nlayer, except that the MCO2AT shows the smallest mean biases for all single- and multilayered clouds and it systematically reduces the SCO2AT mean biases by ∼40%. The increased dzc with increasing Nlayer as revealed in Figures 6a, 6c and 6d may be attributed to the single-layer cloud assumption used by the SCO2AT and the old and new VISSTs in multilayered cases. It is also possible that the CPL retrieved more cloud layers when the uppermost or upper cloud layers were optically thinner in those cases. This may imply that the increase in Nlayer is related to the decrease in ɛc11.
Figure 6. CTH difference dzc as a function of the number of cloud layers Nlayer. (a) SCO2AT minus CPL. (b) MCO2AT minus CPL. (c) Old VISST minus CPL. (d) New VISST minus CPL. Thick gray lines represent the running means.
Download figure to PowerPoint
 Figures 7–9 present dzc as a function of ɛc11 by separating the single-layered (Figure 7), two-layered (Figure 8), and multilayered (Figure 9) clouds. For the single-layered cases, the mean dzc is fairly constant within each technique until ɛc11 falls below 0.5. For 0.5 < ɛc11 < 0.95, the MCO2AT has the smallest mean dzc (−0.5 to −1.0 km) and it reduces the absolute mean biases of the SCO2AT by ∼1 km from −1.88 (Figure 7a) to −0.92 km (Figure 7b). The new VISST also reduces the absolute mean biases of the old VISST significantly toward larger ɛc11.
 For the two-layered (Figure 8) and multilayered (Figure 9) cases, their mean differences behaved like those discussed in Figure 5 because the majority of TC4 clouds mainly consist of more than one cloud layer. Nonetheless, for two-layered and multilayered clouds, the absolute mean biases in all four techniques and in every bin of ɛc11 are generally twice as large as those from the single-layered conditions. The respective mean biases for SCO2AT, MCO2AT, old VISST, and new VISST are −1.88, −0.92, −2.52 and −0.84 km for single-layered cases (Figure 7), −3.27, −1.93, −4.12 and −2.47 km for two-layered (Figure 8), and −4.88, −3.03, −6.24 and −4.64 km for multilayered cases (Figure 9). Among the four techniques, the MCO2AT has the smallest absolute mean biases when ɛc11 < 0.9. Even though the MCO2AT was developed to account for multilayer cloud conditions, the mean biases of MCO2AT also increased significantly from −0.92 for single-layered to −3.03 for multilayered cases. This increase is likely caused by multiple transmissive upper level layers. In those instances, the MCO2AT infers an average height for the multiple transmissive layers. Additionally, many of the upper layer clouds in these cases are clouds that cannot be detected by the CO2AT even in single-layer conditions. Hence, there is not enough change in the radiances for the MCO2AT to account for the small optical depth of the uppermost cloud. Finally, it is worth noting that about 21% (89% – 68%) of the matched data had the CPL-retrieved CTP < 500 hPa, but had no CO2ATs CTP retrieval. Among these data, nearly half of them had VISST-retrieved CTHs and these are plotted in Figure 10a (old VISST) and Figure 10b (new VISST) as compared with the CPL CTHs. Since such cases were very optically thin clouds, it is not surprising to see that most of the VISST CTHs are much too low, especially since there were no MCO2AT/SCO2AT retrievals available. The mean CTHs in Figure 10 are 3.3 km for the old VISST and improved to 3.9 km for the new VISST as versus 13.2 km for the CPL. The ODs of these clouds were on the order of ∼0.1. The accuracy of their retrieved CTHs is, thus, limited by both the sensitivity and horizontal resolution of the passive satellite instruments like GOES-12.
Figure 10. Comparisons of the (a) old-VISST and (b) new-VISST CTHs with the CPL CTH when there was no SCO2AT/MCO2AT retrieval.
Download figure to PowerPoint