References

Abstract. Data from the Atmospheric Infrared Sounder (AIRS) on the EOS Aqua spacecraft each day show tens of thousands of Cold Clouds (CC) in the tropical oceans with 10 μm window channel brightness temperatures colder than 225 K. These clouds represent a mix of cold anvil clouds and Deep Convective Clouds (DCC). This mix can be separated by computing the difference between two channels, a window channel and a channel with strong CO 2 absorption: for some cold clouds this difference is negative, i.e. the spectra for some cold clouds are inverted. We refer to cold clouds with spectra which are more than 2 K inverted as DCCi2. Associated with DCCi2 is a very high rain rate and a local upward displacement of the tropopause, a cold "bulge", which can be seen directly in the brightness temperatures of AIRS and Advanced Microwave Sounding Unit (AMSU) temperature sounding channels in the lower stratosphere. The very high rain rate and the local distortion of the tropopause indicate that DCCi2 objects are associated with severe storms. Significant long-term trends in the statistical properties of DCCi2 could be interesting indicators of climate change. While the analysis of the nature and physical conditions related to DCCi2 requires hyperspectral infrared and microwave data, the identification of DCCi2 requires only one good window channel and one strong CO 2 sounding channel. This suggests that improved identification of severe storms with future advanced geostationary satellites could be accomplished with the addition of one or two narrow band channels.


Introduction
Inspection of Advanced Infrared Sounder (AIRS, Aumann et al., 2003) spectra shows the expected absorption features of water, carbon dioxide and ozone relative to spectral regions of low atmospheric opacity (window areas) in almost all spectra, including those with heavy cloud cover. However, for about 1% of the spectra in the trop-5 ics (30 S-30 N) the spectra are inverted: the strong CO 2 absorption near 14 µm, the strong water vapor absorption near 6 µm and ozone absorption in the 10 µm area are seen in emission relative to cold cloud top in atmospheric window channels. Negative brightness temperature differences between the 11 µm window and 6.7 µm water channel were first noted by Ackerman (1996). Schmetz et al. (1997) noted their as-10 sociation with severe thunderstorms and first attributed it to Overshooting Convection (OC). Conceptually, OC forces water vapor into the lower stratosphere where it emits at the warmer stratospheric temperature. Romps and Kuang (2009) defined overshooting cloud tops as clouds with brightness temperature colder than the monthly tropopause temperature climatology at that location, as derived from the NCEP reanalysis (Kalnay 15 et al., 1996). Others have used the reanalysis pertaining to the time and location of the observation to infer a number of reference levels. The relatively coarse temporal, spatial and vertical resolution of the reanalysis make the accuracy of the calculated height of the tropopause uncertain (Liu and Zipser, 2005), particularly in the presence of strong convection. What appears as OC and penetration into the stratosphere, may 20 actually be related to the difference between the height of the tropopause under average conditions and strong convection conditions. The objective of our paper is to use AIRS hyperspectral data to gain new insights into properties of these cold cloud tops.
The utility of using the inversion of strong atmospheric absorption lines for the definition of the tropopause in the presence of deep convection relies on having a large and Introduction spheric window channel, bt1231. However, water vapor absorption poses a potential problem, since bt1419 is a function of temperature and water vapor mixing ratio in the atmospheric column. Observations that strong convection "moistens" the upper troposphere (Ray and Rosenlof, 2007) and speculations that strong convection may result in injection of water vapor into the stratosphere complicate the interpretation of inverted 10 water vapor spectra. For this reason we use CO 2 absorption. The weighting function of the strong CO 2 line near 712 cm −1 also peaks near 200 hPa, but this peak is almost independent of the water vapor profile. The difference between window channels at 961 cm −1 and 790 cm −1 (bt961 and bt790) channels provides additional insight into the cloud optical depth and the location of the cloud top relative to the tropopause.

Data
We use the data from AIRS, supported by data from the Advanced Microwave Sounder Unit (AMSU, Lambrigtsen et al., 2003)  AIRS is a hyperspectral infrared sounder which covers the 650 to 2665 cm −1 region of the thermal infrared spectrum with 2378 spectral channels. The AIRS footprint subtends an angle of 1.1 degree (full width 1/2 peak), corresponding to a 15 km footprint at nadir. The footprints are scanned ±49.5 degrees cross-track, resulting in a 1650 km wide swath with 90 footprints. The absolute calibration at the 200 mK level 5 and stability at the better than 10 mK/year level have been validated using the sea surface temperatures (Aumann et al., 2006). The validation of the radiometric accuracy and stability at the extremely cold temperatures relevant for the study of cold cloud tops used Dome Concordia data in Antarctica (Walden et al., 2006;Elliott et al., 2007). AMSU is a microwave radiometer with 14 channels centered on the 57 GHz 10 oxygen line with a 3.3 degree (full width at 1/2 peak), which is synchronized with AIRS and covers the same crosstrack swath. AMSRE is a conical scanning passive microwave imaging radiometer. The rain rates, measured within minutes of the AIRS data, are available each day averaged on a 0.25 degree grid, i.e., approximately on a 28 km scale. Data from AIRS and AMSU since September 2002 are available from 15 the GDIS at GSFC. The AMSRE rain rates (Wilheit et al., 2003) were obtained from ftp://rain.atmos.colostate.edu/RAINMAP/data/amsre/. Unlike AIRS and AMSU data, which are calibrated radiances and brightness temperatures (level 1b), the AMSRE rain rates are based on a level 3 (gridded data) product. Data for our study were collected in three sets. This allows us to contrast clear, 20 random, and extremely cloudy conditions. For the first set we identified each day typically 30 000 cloud-free spectra. In the second set we collected typically 20 000 random spectra, selected within one degree of nadir, regardless of land/ocean or cloud status. For the third set we collected all objects with brightness temperatures colder than 225 K in the 1231 cm −1 atmospheric window channel between latitude 40 S and Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | brightness temperatures measured in these channels as bt712, bt790, etc. and define DW=bt1231−bt1419, DT=bt1231−bt712, and DC=bt961−bt790. The measurements at these cold temperatures require very high instrument sensitivity. The Noise equivalent Delta Temperature (NeDT) is a measure of the uncertainty in the brightness temperature in a single measurement. It is commonly quoted at 280 K. For ex-5 ample for the 1231 cm −1 channel, NeDT280=0.05 K; however for a much colder scene, NeDT200=0.31 K. The effective noise for difference measurements is further amplified, e.g. stdev(DT) is close to 0.5 K. Figure 1a shows the DCC for the night tropical oceans as a scatter diagram of 10 DT vs. bt1231 from the night overpasses from 6 September 2010. The red line is the scatter diagram ridge line. This line highlights the most likely functional dependence of the two variables in the scatter diagram. It is generated by dividing the x-axis into equal width bins and connecting the mean value of each bin (red circles). Figure 1b shows the scatter diagram derived from simulations, which will be described in Sect. 3.

15
The difference between two window channels which straddle the 11 µm window, DC, contains additional information to characterize the cloud. Figure 2a shows a scatter diagram of DC vs. DT for the same data as Fig. 1a. Figure 2b shows the corresponding scatter diagram derived from the simulations. 20 The data base created by the random nadir spectra was used to calculate mean properties of cold clouds for various thresholds. The frequency of occurrence is defined as the number of spectra which pass a threshold divided by the total number of spectra in a given area, expressed as percent. There are large regional differences between the Western Tropical Pacific (WTP) and tropical ocean (TO) exclusive of the WTP. We also ACPD 10, 2010 Deep convective clouds at the tropopause

Temperature structure related to DCC
In the presence of high cloud overcast conditions it is difficult to measure the temperature structure near the tropopause in the thermal IR or with 57 GHz microwave chan-5 nels. However, we can use AIRS and AMSU temperature sounding channels to evaluate the lapse rate in the lower stratosphere to infer conditions near the tropopause. The 668.2 cm −1 (bt668) and 679.9 cm −1 (bt679) AIRS channels measure the brightness temperature in broad layers near 2 and 40 hPa, respectively. Figure 3a shows the scatter diagram of the temperature at 2 hPa and 40 hPa as function of bt1231 from the 6 September 2006. Only the scatter diagram ridge lines are shown. As bt1231 becomes colder than 210 K the temperature at 2 hPa gets warmer, while the temperature at 40 hPa gets colder. Since bt679 at 40 hPa may be effected by the presence of clouds, we checked with AMSU data, which are much less effected by clouds than IR channels. There are no exact equivalent channels, but AMSU#14 and AMSU#9 measure the tem- 15 perature in broad layers near 2 and 60 hPa, respectively. Figure 3b shows an overlay of the scatter diagrams of AMSU#14-AMSU#9 and bt669-bt679.9. Only the scatter diagram ridge lines are shown. The 7 year mean temperature at 60 hPa (AMSU#9) under DT< −2 K conditions is 1.7 K colder during the day, 2.4 K colder at night than under random conditions. The temperature at 2 hPa (AMSU#14) is 0.

Rain rate
We use the rain rate measured by AMSRE to evaluate the correlation between rain rate and DCC. Since AMSRE and AIRS are on the same spacecraft, the DCC identification with AIRS and the rain rates measured with AMSRE daily gridded product are time coincident within minutes. The measurements are not exactly simultaneous because 5 AIRS is cross-track scanning within 0.1 degree of nadir, while AMSRE uses a forward looking conical scan. Figure 4 is a scatter diagram of the rain rate as function of DT. There is large scatter, but the ridge lines show the rapid increase in the rain rate for DT<0 K, and even faster at DT< −2 K. Since the AMSRE rain rate refers to the average in a 28 km grid, the rain rate in the 15 km AIRS FOV could be considerably larger.
10 Table 2 includes the rain rate.

Simulations
The observed brightness temperatures allow a physical interpretation of the height of the cloud tops relative to the height of the tropopause. We used model calculations for 34 tropical ocean profiles from a combination of two data sets: the set of regression 15 profiles (including the AFGL profiles) used by Strow et al. (2003) to make the AIRS Fast Model Radiative Transfer Algorithm (RTA), and the TIGR3 (1999)  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | above the tropopause pressure, but no models were used with cloud top pressure less than 30 hPa. The amount of cirrus ice was distributed in the RTA per the layer average pressure. The scattering part of the RTA used the Parametrization of Cloud Longwave Scattering for Atmospheric Modeling (PCLSAM) algorithm by Chou et al. (1999). Ice aggregate scattering parameters were based on Baran (2003). The integrated Ice

5
Water Content (IWC) of 5 g/m 3 was selected to approximate the mean observed DC under DT< −2 K conditions. This number is larger than that determined by Iwasaki et al. (2010), as they studied the sub visible cirrus blown off the top of a DCC cloud, while in this paper we are more interested in studying the overall characteristics of optically thick DCC. The cloud bottom pressure in the model was always 150 hPa larger than the 10 cloud top pressure. This assumption is not critical since the clouds become optically thick within 20 hPa of the cloud top in all cases. Figure 1b shows the results of the model calculation DT vs. bt1231 while Fig. 2b shows DC vs. DT. In both cases the calculations use 30 micron particles. Model spectra for 77 cloud tops more than 5 hPa above, 30 within 5 hPa of, and 204 more than 15 5 hPa below the tropopause are shown in blue (x), red (+), and green (o), respectively. Results from the same model calculations shown as function of bt1419 instead of bt712 look almost identical. The models with 20 and 50 micron effective particle sizes also look very similar. With 20, 30, and 50 micron effective sizes bt961−bt790 for cloud tops below the tropopause (green) increases from 1.6 K to 2.0 K to 2.5 K, but the clear 20 separation between the blue, red and green cloud tops is unchanged.

Model spectra and observation comparison
The label "Deep Convection" has been given to a wide range of objects identified by various thresholds, from cloud tops colder than 235 K, to 1 × 1 degree areas where Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | the clouds associated with deep convection is wavelength dependant and, since the clouds are likely to by spatially inhomogeneous, the label is sensitive to the size of the FOV. Table 1 shows that 3% of the spectra from the tropical oceans, almost 7% in the WTP, are identified as DCC based on bt1231<225 K. Most of the DCC with cloud tops between 210 K and 225 K in the tropical oceans are far below the tropopause. In 5 the TO 0.8% of the spectra are inverted (DT<0 K or DW<0 K), 0.4% are identified with DT< −2 K or DW< −2 K. DDC which satisfy the DT< −2 K condition, referred to in the following as DCCi, reach very close to the tropopause. The near equivalence of DT and DW seen in the model calculations, is also seen in the observations summarized in Table 1. The table also allow a comparison with other work, although this is complicated 10 by differences in the FOV used for other observations and possible calibration effects at cold temperatures. In the following we use CO 2 absorption to sense the height of the temperature inversion, which define as the height of the tropopause. The issue of possible water vapor enhancement above or below this inversion in the presence of DCC will be dealt with in another paper.
15 Figures 1b and 2b show that the DT< −2 K threshold fairly cleanly separates the cloud tops more than 5 hPa below the tropopause from the higher clouds. The additional DC>0.5 K test separates the clouds more than 5 hPa above (blue) from those more than 5 hPa below the tropopause (green). For the nominal tropical tropopause at 100 hPa, 5 hPa corresponds to 0.4 km altitude difference. The clear separation be-20 tween cloud tops above and below the tropopause is provided by the window channel difference DC=bt961−bt790: the opacity of the ice clouds at 790 cm −1 is larger than the opacity at 961 cm −1 , i.e. the weighting function of the 790 cm −1 channel penetrates into the cloud for a small distance before the cloud become optically thick. If the cloud top is below the tropopause, bt790 is colder than bt961, i.e. DC>0 K, if the cloud top is 25 above the tropopause then DC<0 K. The comparison of Fig. 1a and b shows similarities in the slope as function of bt1231. However, the observations come to a sharp point at bt1231=180 K, DT=−12 K, while the blunt tip of the scatter diagram for the models is filled with cloud tops near and above the tropopause. This indicates that the observations contain few cloud tops more than 5 hPa above the tropopause. This is confirmed in the comparison of the data in Fig. 2a and the model in Fig. 2b: Very few of the observed cloud tops have DC<0. The mean DC for clouds which satisfy the (DT< −2) conditions is +1 K with standard deviation of 0.65 K (Table 2), but is close to 2 K for lower cloud tops. The noise in DC due to 5 detector noise alone is 0.5 K, i.e. the few observations with DC<0 are consistent with random noise. Figures 1b and 2b show model calculation using 30 micron effective particle size. For lower (warmer) cloud tops (0<DT<5 K) the models show little change in DC (DC=+2 K), while the observations show a gradient (from DC=1.5 K to DC=2 K). We attribute this difference to the simplification of using a fixed IWC, particle size and 10 ice crystal habit for all models, including clouds well below the tropopause. Particle sizes are expected to diminish with height, as well as change structure.

Penetrating convection
The observation of inverted spectra has been taken as evidence of Overshooting Convection (OC, e.g. Schmetz et al., 1997), with the implication of tropopause penetration 15 (Wang, 2007). The model calculations show that the inversion of spectra starts with cloud tops well below the tropopause. Figure 3 shows that the temperature structure near the tropopause in the presence of DCCi is distorted. The gradient between 2 and 40 hPa is 0.3 to 2 K steeper under DCCi conditions than at bt1231>225 K and warmer conditions. The steeper lapse rate means an unusually cold tropopause. This cold 20 (upward) bulge in the tropopause causes the cloud top to be colder than, i.e. appear to overshoot, the temperature of the tropopause cold point derived from the analysis or reanalysis. However, water vapor, cirrus ice and pollutants collected in bulges may be left in the lower stratosphere, as soon as the strong convection subsides and the bulges disappear. This is consistent with observations of transport of pollutants into 25 the lower stratosphere in the presence of strong convection, e.g. Randel et al. (2010). There is evidence in the literature that GOES brightness temperature images of cold cloud tops show fine structures with 5-10 km spatial scale, e.g. Bedka et al. (2010) finds very cold clouds just resolved with GOES and MODIS data protruding above large cold anvils. These clouds appear to be buoyant convective bubbles, referred to in the following as Protruding Convective Bubbles (PCBs), several kilometer higher than the surrounding cold anvil. With the assumption that the cold anvil reaches the tropopause, the top of a several km high PCB above the cloud top implies stratosphere 5 penetration. The linear combination of model data shows that cloud tops well below the tropopause with an imbedded PCB can have an inverted spectrum. Assume the cloud top is at 150 hPa (approximately 210 K), i.e. about 50 hPa and 3 km below the tropopause, without a PCB. This is seen by AIRS as bt1231=210 K and DT=0 K. Assume a 5 km diameter imbedded PCB, with the same composition as the anvil, reaches 10 the tropopause. With a 5 km FOV this would be detected as a DT=−10 K inverted spectrum. Averaged in the 15 km AIRS FOV it would appear as a bt1231=201.5 K cloud, with a DT=−1 K. If the PCB were to penetrate several km into the stratosphere, it would have to cross the level of neutral buoyancy (LNB) and the spectral inversion decreases. For the DT< −4 K conditions the entire anvil has to reach 200 K or several imbedded 15 PCBs are required. We conclude that AIRS data are consistent with PCBs that reach the tropopause, and what is referred to as OC appear to be PCBs. A model where the AIRS 15 km FOV includes one or more PCBs which reach or cross the LNB helps to explain what appears to be an inconsistencies related to the rain rate associated with DCC: The rain rate under DCC conditions is a factor of 40 higher 20 than under average conditions and Fig. 4 shows the rapid increase in the rain rate with DT−2 K, but even there the average 4 mm/hr rain rate is not particularly high. The association between cold cloud tops and rain rate in the form of the GOES Precipitation Index (e.g. Arkin, 1979) is based on the 5 km GOES FOV. In the GOES data the rain rate starts to increase steeply for cloud top temperatures colder than 230 K, but with 25 the AIRS 15 km footprint the rain rate does not increase steeply until DT=−2 K, roughly equivalent to bt1231=205 K. This inconsistency can be explained by imbedded PCBs. If the rain rate is confined to a 5 km area associated with a BB, the 4 mm/hr average rate associated with DT< −2 K spectra would be more than a factor of 10 higher locally. The rain rate should increase steeply as the PCB imbedded in the anvil is pushed to the LNB. The signal contributed by one PCB, although averaged in the AIRS 15 km FOV, causes a detectable inversion in the AIRS spectra and the steep increase in the rain rate. 5 Various proxies have been proposed for the detection of severe storms using remote sensing from Earth orbit or geostationary positions. The association between cloud top temperatures colder than 210 K in the 11 µm GOES window channel and 5 km FOV with severe storms, including torrential rain and hail over land is relatively secure and goes back to Reynolds (1980). A number of papers (e.g. Adler et al., 1985;Brunner et al., 2007)

Conclusions
Strong convection causes cloud tops to rise close to the tropopause. Protruding above these cloud tops, but not spatially resolved by the AIRS 15 km FOV, are convective bubbles (PCBs) which reach the tropopause. The PCBs are detected as inverted AIRS spectra. Tropopause penetration is not required. The intense convection creates cold 5 bulges in the tropopause, which are not present in the reanalysis. This gives the appearance of clouds overshooting the tropopause and penetrating into the stratosphere. The contents of the cold bulges may be left in the lower stratosphere, when the convection supporting the bulges subsides. Footprints with spectra with more than 2 K inversion (TD< −2 K), which are found for 0.4% of the AIRS spectra in the tropical 10 oceans, are associated with a rapid increase in the rain rate. The high rain rate and the likely presence of PCBs suggests that AIRS footprints identified with DT< −2 K are associated with significant, but not necessarily severe storms. Introduction