Assessing the benefits of Imaging Infrared Radiometer observations to the CALIOP version 4 cloud and aerosol discrimination algorithm

The features detected in monolayer atmospheric columns sounded by the Cloud and Aerosol Lidar with Orthogonal Polarization (CALIOP) and classified as cloud or aerosol layers by the CALIOP version 4 (V4) cloud and aerosol discrimination (CAD) algorithm are reassessed using perfectly collocated brightness temperatures measured by the Imaging Infrared Radiometer (IIR) onboard the same satellite. Using the IIR’s three wavelength measurements of layers that are confidently classified by the CALIOP CAD algorithm, we calculate two-dimensional (2-D) probability distribution functions (PDFs) of 5 IIR brightness temperature differences (BTDs) for different cloud and aerosol types. We then compare these PDFs with 1-D radiative transfer simulations for ice and water clouds and dust and marine aerosols. Using these IIR 2-D BTD signature PDFs, we develop and deploy a new IIR-based CAD algorithm and compare the classifications obtained to the results reported by the CALIOP-only V4 CAD algorithm. IIR observations are shown to be able to identify clouds with a good accuracy. The IIR cloud identifications agree very well with layers classified as confident clouds by the V4 CAD algorithm (88 %). More 10 importantly, simultaneous use of IIR information reduces the ambiguity in a notable fraction of "not confident" V4 cloud classifications. 28 % and 14 % of the ambiguous V4 cloud classifications are confirmed thanks to ::::::::: reclassified ::::: more ::::::::::: appropriately :: as ::::::: confident ::::: cloud :::::: layers :::::: through :::: use :: of the IIR observations in the tropics and in the midlatitudes respectively. IIR observations are of relatively little help in deriving high confidence classifications for most aerosols, as the low altitudes and small optical depths of aerosol layers yield IIR signatures that are similar to those from clear skies. However, misclassifications of aerosol 15 layers, such as dense dust or elevated smoke layers, by the V4 CAD algorithm can be corrected to cloud layer classification by including IIR information. 10 %, 16 %, and 6 % of the ambiguous V4 dust, polluted dust, and tropospheric elevated smoke respectively are found to be misclassified cloud layers by the IIR measurements.

Section 2 presents the IIR and CALIOP data, and the 1-D radiative transfer model. Section 3 presents the IIR 2-D BTD signatures in cloud and aerosol monolayer atmospheric columns using radiative transfer IIR simulations with a 1-D radiative transfer model and quantify the uncertainty in the observed clear-sky IIR signatures. Section 4 describes the new IIR-based 60 CAD algorithm. Section 5 compares the IIR cloud-aerosol classifications to the results reported by the CALIOP V4 algorithm.
Section 6 summarizes the main conclusions.

CALIOP observations
The CALIOP L2 5 km Merged Layer Data Product V4 reports tropospheric and stratospheric cloud and aerosol detection information on a 5-km-horizontal grid. However, the amount of horizontal averaging required to detect a layer may exceed 5 km and hence the search for features is also 20 km and 80 km averaging intervals [Vaughan et al., 2009]. Here, we do not 80 retain feature layers detected with a horizontal averaging of 80 km because their optical depths are typically very small and are therefore considered as transparent for the IIR (see Appendix A). This means that a 5-km atmospheric column containing, for example, a cloud layer detected with a horizontal averaging of 5 km and another cloud layer detected with a horizontal averaging of 80 km is considered as a cloud monolayer column here. 5-km columns are composed by 15 lidar single shots (every 333 m) averaged together. When a layer is detected at 5-km-horizontal resolution in the planetary boundary layer 85 (PBL), the CALIOP L2 processing searches within the layer to identify those especially dense regions that can be confidently detected at single-shot resolution. In those cases where clouds are detected at single-shot resolution, the original 5-km profile is reaveraged to a nominal 5-km resolution, with the data from all single shot layer detections being excluded from the newly averaged profile. This new profile is once again searched for layers whose presence would previously have been obscured by the very strong backscatter from boundary layer clouds. This allows, for example, the detection of aerosol layers at a nominal 90 5-km horizontal resolution with embedded small cumulus clouds detected at single-shot resolution. 5-km columns containing layers detected at single-shot resolution that were cleared from the original profile are not considered in this study. Indeed, the IIR observations can be strongly influenced by these large optical depth features.
The CALIOP V4 dataset also reports top altitude z top , the estimated optical depth τ , and the CAD score calculated by the CAD algorithm for each detected layer. Nominal CAD scores range between -100 to 100. The layer is classified as cloud when 95 the CAD score is positive and as aerosol when it is negative. The absolute value of the CAD score provides a confidence level for the classification. Here, we consider confident layers as those for which 70 ≤ |CAD score | ≤ 100 following the "high" confidence definition of Liu et al. [2009], and ambiguous layers as those for which 0 ≤ |CAD score | < 70. There are also several "special" CAD score values that represent classification results that are based on additional information beyond that normally considered in the standard CAD algorithm. For example, weakly scattering features detected along the edges of ice clouds that 100 are initially classified as depolarizing aerosol layers are subsequently reexamined using spatial proximity analysis. As a result, the vast majority of these layers are reclassified as "cirrus fringes" and assigned a special CAD score of 106 [Liu et al., 2019]. Figure 1 shows the occurrence of the 5-km column types derived from the CALIOP Level 2 5  cloud and aerosol multilayer. First, we note that clear-sky columns are more frequent during daytime than nighttime. This is mainly due to the fact that the lidar detection sensitivity is much lower during daytime due to background solar noise making it more difficult to detect faint features [e.g., Thorsen et al., 2017;Toth et al., 2018]. For the same reason, more monolayer columns and fewer multilayer columns are found during daytime compared to nighttime. In the tropics, monolayer columns represent 54 % of daytime observations and 37 % of nighttime observations. In the midlatitudes, monolayer columns represent 110 57 % of daytime observations and 45 % of nighttime observations. High aerosol monolayers are very rare (0.1-0.4 %). Note that approximately half of the aerosol columns (difference between solid and transparent bars) will not be studied using the IIR due to the presence of dense clouds detected at full resolution (333 m) which have been cleared during the CALIOP L2 data processing. Cloud monolayer and multilayer columns (solid bars) represent 76 % of the midlatitudes CALIOP observations in which no clouds were cleared at single-shot resolution, and 59 % in the tropics. In contrast, aerosol monolayer and multilayer 115 columns are much more common in the tropics (25 %) than in the midlatitudes (13 %). We see that the proportion of features classified with an ambiguous V4 CAD score (hatched part of the solid bars) is very low (0.5-5 % for cloud monolayer; 3-12 % for low aerosol monolayer). This means that the V4 CAD algorithm is quite confident in its ability to correctly discriminate aerosol from cloud layers. We note that the proportions of ambiguous features (0 ≤ |CAD score,V4 | < 70) in this study is lower than those found by Liu et al. [2019] for the year 2008 (≈ 9 % for cloud layers; ≈ 20 % for aerosol layers). This disparity arises due to the different column types being examined. In particular, cloud-aerosol discrimination is more challenging in multilayer scenes, as wavelength-dependent signal attenuation by overlying layers introduces additional uncertainties (e.g., lower signalto-noise ratios) when classifying lower layers. Similarly, classification uncertainties are higher when dealing layers detected over land, layers detected at 80-km-horizontal resolution, and/or layers from which clouds detected at single-shot resolution have been cleared. Unfortunately, IIR will not provide any help for ::: For those column typesand : , :::::::: extracting :::::: useful :::::::::: information 125 :::: from ::: the ::: IIR :::::::::::: measurements ::: for ::: the ::::: cloud ::: and :::::: aerosol :::::::::::: discrimination :: is :::: very :::::::::: challenging, ::: so they are not studied here. Columns containing a monolayer with a special CAD score value are not shown in this figure. Note that monolayer columns with CAD score of 106 are more common during daytime that nighttime and represent between 0.01 % and 0.05 % of the total occurrences. Note that when IIR observes a clear-sky profile or a profile containing only infrared-transparent aerosol layers, the IIR signature is zero. Therefore, if the IIR signature is non zero, the observed atmospheric profile contains a layer with a nonnegligible absorption at IIR wavelengths. The accuracy of this assertion is bounded by the joint uncertainties in the measured 155 and computed clear-sky brightness temperatures. brightness temperature more pronounced at 12.05 µm than 8.65 µm. However, the distribution is well centered in the y-axis, because biases at 10.60 µm and 12.05 µm cancel each other out, consistent with Garnier et al. [2021]. The red and blue ellipses represent the 95 % and 50 % confidence intervals of the 2-D gaussian PDF estimated from those observations. If a monolayer IIR signature falls into this clear-sky uncertainty region, its identification will be difficult. However, far from this region, an IIR signature can be confidently attributed to a cloud or an absorbing aerosol layer. Reliable discrimination between cloud and 165 aerosol will be possible where their expected signature regions do not overlap.

Simulated cloud and aerosol IIR signatures
In presence of a cloud or an aerosol layer, the brightness temperatures in the IIR channels depend on the layer altitude, the layer optical depth, the microphysics of the layer, the atmospheric profile, and the surface temperature and emissivity. We briefly present here how the layer parameters affect the IIR signature using the radiative transfer simulations presented in Sect. 2.3.  Figure 3 shows how the IIR signature from an ice or liquid cloud layer varies with the cloud optical depth, the cloud top altitude, and the particle equivalent ::::::: effective diameter in tropical and midlatitudes atmospheres. Ice cloud particle equivalent ::::::: effective diameter of 20 µm, 40 µm, and 90 µm are used. We note that as the cloud optical depth or cloud altitude is changing, all the rest being the same, the simulated IIR signature describes arches on this representation, which is consistent with previous studies using the split-window technique [e.g.;Baum et al., 1994;Giraud et al., 1997;Dubuisson et al., 2008;Hong et al., 2010]. The arches converge toward the clear-sky IIR signature (zero) as the cloud optical depth and cloud altitude decrease and toward the top-of-atmosphere black body IIR signature (red cross) as the cloud optical depth and cloud altitude increase. Indeed, if the cloud is dense enough ::::::: optically :::: thick, its emissivity is close to one in each channel and then their brightness temperatures are the same. Moreover, if the cloud is high enough ::: top :: is ::::::::: sufficiently :::: high ::: (> : 8 :::: km), ::: i.e. :::::: above ::: the ::::: lower ::::: levels :: of ::: the :::::::::: atmosphere 180 :::::

Aerosols
are : then of no help for the classification of such layers.

IIR CAD score
The IIR signature 2-D PDFs derived from confident observations of a specific z top -τ class, as illustrated by their 95 % and 275 50 % confidence intervals (solid lines) in Figs. B1, B2, C1, C2, 7, and 8, are used to derive an IIR CAD score.
The IIR signature PDFs are derived for several z top -τ ranges in order to keep ::: that :::::::: minimize the PDF widths as small as possible as we saw in previous section that :::::: because : the IIR signatures are mainly ::::::: primarily : dependent on layer altitude and optical depth ::::::: thickness. These narrower PDFs increase the likelihood that the PDFs of individual cloud and aerosol classes are well separated. Probabilities are then computed on a z top -τ grid, with z top boundaries from 0-4 km, 4-8 km, and above 280 8 km and τ spanning ranges from 0-0.2, 0.2-0.6, 0.6-1.5, 1.5-3, and above 3. The PDFs p i , where i represents all cloud types (liquid, ice, oriented crystals) and all aerosol types (dust, smoke, marine, ...) of the troposphere and stratosphere of the CALIPSO V4 classification, are then defined for each z top -τ grid cell. PDFs characterizing specific layer types are derived whenever there are at least 500 confident occurrences of layer type i in a z top -τ grid cell. Then, we derive the IIR CAD score according to: with CAD noCS = 100 (P C + P bkg ) − (P A + P bkg ) (P C + P bkg ) + (P A + P bkg ) (1 + 2P bkg ) representing the CAD score if there were no clear-sky atmospheric columns (or no uncertainty in the computed clear-sky brightness temperatures),

290
CAD Cloud/CS = 100 (P C + P bkg ) − (kP CS + P bkg ) (P C + P bkg ) + (kP CS + P bkg ) (1 + 2P bkg ) representing the CAD score of clouds if there were only cloud and clear-sky atmospheric columns, and representing the CAD score of aerosols if there were only aerosols and clear-sky atmospheric columns. In those equations, P CS corresponds to the clear-sky PDF weighted by a coefficient k = 2 in order to decrease the absolute value of IIR CAD score in 295 the clear-sky uncertainty region. The cloud and aerosol PDFs are given by P C (X 1 , X 2 , ..., X m ) = max i∈cloud types p i (X 1 , X 2 , ..., X m ) and P A (X 1 , X 2 , ..., X m ) = max i∈aerosol types p i (X 1 , X 2 , ..., X m ) where p i are the multidimensional PDFs for cloud and aerosol types as a function of attributes X 1 , X 2 , ..., X m . A background 300 PDF P bkg = 0.05 is added to the equations in order to avoid unreasonably large CAD values in the regions that both cloud and aerosol would not present by nature. The CAD score equations are then renormalized by multiplying them by (1 + 2P bkg ).
Attributes X 1 , X 2 , ..., X m are both components of the IIR signature (i.e., 8.65 µm-12.05 µm and 10.60 µm-12.05 µm), the top altitude z top and optical depth τ of the monolayer inferred from lidar observables, and the latitude to determine the region (tropics or midlatitudes). Unlike the CAD score derived solely from lidar observations [Liu et al., 2009], the CAD score 305 from IIR observations does not account a priori for the relative occurrence frequencies of different layer types within a z top -τ grid cell. We then consider that the probability of occurrence of an ith type of cloud or aerosol with a given IIR signature is independent of the probability of occurrence of another type. The maximum of PDFs for the different cloud or aerosol types are then considered to compute the CAD score instead of merging the different PDFs when comparing P C and P A .

IIR CAD score masks
310 Figures 10 and 11 show the IIR CAD scores derived from CALIPSO observations for the tropics and the midlatitudes respectively. Aerosol classifications are shown in red and cloud classifications are shown in blue. Each pattern is due to a specific layer type, some of them being annotated in Fig. 10. Color intensity varies according to classification confidence, with fainter colors representing lower confidence, which decreases with the distance to the PDF centers. The yellow lines represent the |CAD score,IIR | = 70 isocontours, separating ambiguous and confident IIR classifications.

315
As expected according to its definition, the IIR CAD score is very close to 0 in the clear-sky uncertainty region. In the tropics, we note that for low altitude layers (z top < 4 km; left column), the intensity of the colors is quite faint (unless the optical depth is very large), meaning the IIR CAD score is almost never confident. Indeed, the discrimination between cloud and aerosol is difficult for low layers due to the lack of contrast in their brightness temperatures compared to the surface.
Mid and high-altitude layers (z top > 4 km; middle and right columns) are easier to discriminate both in the tropics and the 320 midlatitudes. Layers with an IIR signature falling in the red regions are almost always classified as aerosol without confidence due to overlap with cloud PDFs in the IIR signature regions where we found them. The only exception arises in the midlatitudes for layers at mid-altitudes and with optical depths between 0.2 and 1.5.

Results
5.1 IIR CAD score vs V4 CAD score 325 Figure 12 compares CAD score of the V4 algorithm with those derived from IIR BTD observations for all monolayer columns observed by CALIPSO during the 2008-2019 period. Transparent white lines show the limit between confident and ambiguous cases (|CAD score | = 70) and between cloud and aerosol CALIOP V4 classification (|CAD score,V4 | = 0). For IIR classification, we consider CAD scores very close to zero (|CAD score,IIR | < 10) as undefined classification. Therefore, cloud and aerosol ambiguous layers have CAD scores of 10 < |CAD score,IIR | < 70. Tables below the plots summarize the fractions in these 330 CAD score classes. We first note that the CALIOP V4 algorithm is generally very confident in its ability to discriminate cloud and aerosol, as seen by the many values very close to 100 and -100, consistent with Liu et al. [2019]. When the V4 detection is ambiguous (|CAD score,V4 | < 70), the CAD score is mainly very close to 0 (peak around 0). A large majority of the CALIOP V4 confidently classified clouds are also confidently classified as clouds (ambiguous or confident) by the IIR (91 % in the tropics, 86 % in the midlatitudes). Very few V4 confidently classified clouds are classified as ambiguous aerosols by the IIR 335 CAD algorithm (≈ 0.15 %) and virtually none as confident aerosols. Some of the V4 ambiguous clouds can be confirmed thanks to IIR observations ::::::::: reclassified :::: more :::::::::::: appropriately :::: with ::: the ::: aid :: of ::: IIR :::::::::::: measurements : (28 % in the tropics and 14 % in the midlatitudes) as they received a confident IIR CAD score. The V4 confident aerosols are mainly undefined by the IIR CAD algorithm (84 % in the tropics and 87 % in the midlatitudes). A few occurrences of V4 aerosols with an IIR confident cloud classification are found in both the tropics (0.11 %) and the midlatitudes (0.22 %).

340
We now compare the IIR and V4 CAD scores for each cloud and aerosol type. Figure 13 shows the confusion matrices between confident and ambiguous cloud and aerosol IIR and V4 CAD scores for each cloud and aerosol type observed in the tropics. The ambiguous IIR CAD score gathers the ambiguous cloud, aerosol, and undefined layers. For the aerosol observations, we are mainly interested in ambiguous cases that could be misclassified by the CALIOP CAD algorithm but subsequently reclassified correctly as confident clouds by the IIR CAD score. A small fraction of ambiguous dust (10.1 %), polluted dust (25 %), and tropospheric elevated smoke (6.4 %) seem to be misclassified clouds. Even a small fraction 360 of V4 confident aerosols seems to be misclassified clouds: 0.3 % for dust, 1 % for polluted dust, and 0.4 % for tropospheric elevated smoke. Figure 14 shows the confusion matrices obtained in the midlatitudes. Some of the clouds that are ambiguously classified by the V4 CAD can be confirmed using the IIR observations in this region: 33 % for ice clouds, 25 % for liquid clouds, 64 % for oriented ice crystals and 5 % for unknown phase clouds. A small fraction of the aerosol layers identified as ambiguous by the V4 CAD would be reclassified as clouds by the IIR observations in this region: 10 % for dust, 6.8 % for polluted dust, and 4.8 % for tropospheric elevated smoke. Some V4 confident aerosols seems also to be misclassified clouds: 0.5 % for dust, 0.2 % for dusty marine, 0.7 % for polluted dust, 0.1 % for clean marine, 0.1 % for polluted continental smoke, and 0.6 % for tropospheric elevated smoke. A very few confirmations of ambiguous polluted dust (0.6 %) and ambiguous tropospheric elevated smoke (1.6 %) layers occur in the midlatitudes. :::: Note ::: that ::: for :::: both ::: the :::::: tropics ::: and ::: the ::::::::::: midlatitudes, :: 95 :: % ::: of :: the ::::::: aerosol 370 ::::: layers ::::::::: reclassified :: as ::::: cloud :::::: layers ::: are ::: dust ::: or ::::::: polluted :::: dust.
Since :: the : IIR confident cloud CAD score is used either to confirm V4 ambiguous clouds or to reclassify V4 ambiguous dust or smoke layers, the IIR signature dependency with z top and τ must be analyzed. Figure 15 shows the results obtained for V4 ambiguously classified (0 ≤ |CAD score,V4 | < 70) ice clouds in the tropics. The blue line contour shows the regions where the layer get a confident IIR CAD score (|CAD score,IIR | ≥ 70). We note that many of those V4 ambiguous cloud classification are 375 confirmed by the IIR CAD score, especially at altitudes above 8 km where almost 100 000 observations can be confirmed (blue values in the bottom-right corner), where more than half of them has τ > 3. Figure 16 shows the characteristics of the V4 ambiguous dust layers reclassified as clouds in the tropics. The largest number (a few thousands) of ambiguous dust layers detected as confident cloud layers by the IIR is found at high altitude for τ < 0.6.
For τ > 1.5 most of the ambiguous dust layers are detected as clouds by the IIR. thus IIR cannot provide a confirmation that they are clouds on an individual basis. However, we note that the centroid of the distribution is discernibly shifted to the right-top from the clear-sky uncertainty region, which is consistent with high thin cirrus ( Fig. B1). In total, 11 % of the cirrus fringes are classified as confident clouds by the IIR algorithm. These IIR observations seem to provide independent global evidence :::::: suggest : that the feature type reclassifications made by the CALIOP V4 "cirrus fringe amelioration" algorithm are indeed correct. Figure 18 shows an example of a V4 ambiguously classified (CAD score,V4 = −36) dust layer confidently reclassified as a cloud (CAD score,IIR = 99) using the IIR observations. The layer is located at about 12 km altitude in the 5-km atmospheric column indicated by the dash line and located at 30.6°S. Its optical depth is estimated at 0.63 in the CALIPSO product. It then falls in the z top > 8 km and 0.6 < τ < 1.5 grid cell of the midlatitudes region (Fig. 11). Its IIR signature is quite large: corresponds to an IIR CAD score of 99. Note that the marine aerosol layer detected below in the boundary layer (Fig. 18d) is detected with 80-km-horizontal averaging and are then not seen by IIR.

Example of cloud misclassified as dust by the V4
The spatial structure and magnitudes of the attenuated backscatter signal (Fig. 18a) is reasonably consistent with what we would expect from a cirrus cloud. In fact, most of the feature is correctly classified as cloud by the V4 CAD algorithm 400 (Fig. 18b). Furthermore, Navy Aerosol Analysis and Predictions System (NAAPS) simulations [Lynch et al., 2016] show there is no dust transport from Australia for this day. Then, we are very confident the section layer classified as aerosol by the V4 algorithm (Fig. 18c) can be reclassified as cloud layer.
This case study confirms that for monolayer columns IIR observations can be useful to correct misclassified aerosol layer by the V4 CAD algorithm. Comparison between V4 and IIR CAD scores of V4 clouds shows a very good agreement with 88 % of V4 confident clouds classified as clouds by IIR. A small fraction (≈ 1.5 %) of the ice, liquid, and oriented ice crystal clouds are ambiguous in the V4 classification. About 28 % and 14 % of these cases would be confidently classified as clouds using IIR observations in the 415 tropics and the midlatitudes respectively. Unknown phase cloud layers are mainly ambiguous in the V4 and 12.3 % and 5.4 % of them, in the tropics and the midlatitudes respectively, are confirmed by the IIR observations. They mainly correspond to liquid clouds according to their IIR signature.
Comparison between V4 and IIR CAD scores of V4 aerosols show less good ::::: lower : agreement with most of V4 aerosol layers classified as ambiguous undefined layers by IIR. This is due to a larger uncertainty in the discrimination of aerosols 420 and clouds by IIR at low altitude and for optically thin layers. 10 %, 16 %, and 6 % of the ambiguous V4 dust, polluted dust, and tropospheric elevated smoke respectively are found to be misclassified cloud layers by the IIR measurements. A specific analysis of a case study of the misclassified dust layers observed by CALIOP have ::: was : shown to be consistent with the classification as a confident cloud proposed by the IIR CAD score. Confirmation of aerosol layers or reclassification of cloud layers in aerosol layers virtually never occur.

425
Layers classified as "cirrus fringes" by the V4 CAD algorithm are optically very thin layers and have IIR signatures very similar to clear sky. However, compare to :::::::: compared :: to ::: the : clear-sky IIR signature PDF, there is an offset of the "cirrus fringes" IIR signature PDF toward the cirrus IIR signatures and 11 % of the cirrus fringes even have a confident IIR cloud CAD score, suggesting the "cirrus fringes amelioration" algorithm in the CALIOP V4 CAD processing make correct cloud classifications.
We retain all uppermost layers (first layer from top detected in an atmospheric column) detected at 80-km horizontal aver-450 aging with a top layer altitude above 8 km for the 2008 year. Figure A1 shows the cumulative distribution function (CDF) of their integrated attenuated backscatter at 532 nm γ 532 . We note that 99 % of these layers have γ 532 ≤ 0.00207. We will use this value to consider the worst possible case.
We now derive the optical depth τ of an ice cloud at 8, 12, and 16 km and a liquid cloud at 8 km with γ 532 = 0.00207. In a tropical atmosphere [McClatchey, 1972], the temperature at 8, 12, and 16 km is -23, -49, and -76°C respectively. According Finally, we simulate the IIR signature (see Sect. 3) of these worst case scenarios in a 1-D radiative transfer code. Figure A2 shows the results. We note that even for these worst case scenarios, the IIR signature stays in the clear-sky uncertainty region (see Sect. 3.1). Then, we can confidently remove the layers at 80-km horizontal averaging since their effect to the IIR 465 measurements is negligible.
Note that a layer found at 80-km horizontal averaging below another layer could have been detected with higher resolution if the top layer was not present. It means that such layer could have an optical depth larger than those found here. However, if such a layer is detected at 80-km horizontal averaging and not at a higher resolution, it means that the layer above is quite thick (in addition to be colder) and then clearly dominates the IIR signature. and a liquid cloud (right column) with 1-km-geometrical thickness in a tropical atmosphere (top row), a midlatitudes summer atmosphere (middle row), and a midlatitudes winter atmosphere (bottom row). The ice cloud particle size distribution is a general habit mixed particle distribution [ :::::::::::: Baum et al., 2011] with an equivalent :::::: effective diameter Deq = 90 µm. The liquid cloud particle size distribution is typical of a stratus cloud [ ::::::::::: Stephens, 1979] : . :::::: Surface ::: and ::::::::: atmospheric :::::: profile ::::::: properties ::: for :::: both :: the :::::: tropics ::: and :::::::::: midlatitudes :: are :::: from ::: the ::::::: standard :::::::::                  Figure A2. Radiative transfer simulation of the IIR signature of layers detected at 80-km horizontal averaging with the largest possible integrated attenuated backscatter. They represent the very worst possible cases that can affect the IIR measurments. We note than even those worst cases do not escape from the clear-sky uncertainty region.