Cloudy and clear-sky relative humidity in the upper troposphere observed by the A-train



[1] Cloudy and clear-sky distributions of relative humidity with respect to ice (RHI) are derived from the Atmospheric Infrared Sounder and cloud profiles from CloudSat and the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation satellites. A peak frequency in RHI of 10–20% exists for clear sky, increasing to 75–95% within clouds detected by CloudSat. The range of values depends on the season, altitude, latitude, cloud type, and ice water content. Global distributions of RHI reveal persistent supersaturation in the clear and cloudy tropical upper troposphere and lower tropospheric polar regions, large but variable RHI in the cloudy midlatitude storm tracks, and small yet variable RHI in the clear-sky subtropics. Previous studies using satellite and in situ observations have shown a greater frequency of midlatitude supersaturation in the Southern Hemisphere (SH). Seasonal and interannual variations in RHI are also demonstrated with larger frequencies of supersaturation generally present in the winter hemisphere. An analytical method quantifies the impacts of mean temperature (equation image) and temperature variance (σT) on the distribution characteristics of RHI. The seasonal variation of RHI in the Northern Hemisphere is primarily modulated by seasonal changes in σT, whereas in the SH, both equation image and σT modulate variations in RHI. However, variations in equation image and σT do not explain the presence of slightly greater supersaturation frequency in the SH, suggesting a link between anthropogenic aerosol and ice cloud processes.

1. Introduction

[2] Partitioning satellite-observed relative humidity distributions into clear and cloudy components has been a challenge because of ambiguous cloud detection and characterization, as well as instrument, algorithm, and sampling limitations. Since 2006, the A-train constellation has made unprecedented geophysical observations of Earth's atmosphere and surface using multiple active and passive sensors and platforms closely collocated in space and time. The sensors most relevant to cloudy/clear-sky humidity discrimination are CloudSat, a 94-GHz cloud profiler [Stephens et al., 2008], the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), a dual wavelength lidar (532 and 1064 nm) capable of profiling optically thin cloud and aerosol structures located onboard the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) satellite [Winker et al., 2007], and the vertical profiles of temperature and water vapor from the Atmospheric Infrared Sounder (AIRS)/Advanced Microwave Sounding Unit (AMSU) [Aumann et al., 2003]. Previously, AIRS and CALIPSO were combined to quantify relative humidity with respect to ice (RHI) within tropical thin cirrus [Kahn et al., 2008b; Lamquin et al., 2008].

[3] A multisensor observational synergy of clouds, temperature, and water vapor is necessary to investigate observations of free tropospheric humidity structures, trends, and feedbacks [Held and Soden, 2000; Soden et al., 2005; Bony et al., 2006]. The Intergovernmental Panel on Climate Change (IPCC) has concluded that cloud feedbacks are the biggest source of uncertainty in future climate model projections [Randall et al., 2007]. As clouds, temperature and water vapor are closely coupled, a robust partitioning of cloud-type and clear-sky humidity distributions is necessary for climate model evaluation and parameterization improvements to advance climate prediction.

[4] Determination of the precision and accuracy of RHI is necessary to quantify the structure and behavior of the upper troposphere [e.g., Peter et al., 2006]. Numerous studies have shown the existence of supersaturation within and outside of ice clouds from satellite platforms [e.g., Gierens et al., 2004; Gettelman et al., 2006a, 2006b; Read et al., 2007], surface-based Raman lidar [Comstock et al., 2004], radiosondes [e.g., Miloshevich et al., 2006, and references therein] and in situ observations [e.g., Jensen et al., 2001, 2005; Ovarlez et al., 2002; Haag et al., 2003; Krämer et al., 2009]. Efforts to include ice supersaturation in some climate models have demonstrated that it significantly impacts the atmospheric structure and circulation [Lohmann et al., 2004; Gettelman and Kinnison, 2007; Tompkins et al., 2007; Liu et al., 2007]. A recent survey of cloud ice amount calculated by several climate models shows order of magnitude differences between some of the models [Waliser et al., 2009], much of it attributable to the differences in the details of cloud parameterizations [e.g., Cusack et al., 1999; Boville et al., 2006]. However, there are also ongoing discrepancies of several hundred percent of ice water content and/or path among several satellite instruments (e.g., CloudSat, Moderate Resolution Imaging Spectroradiometer, Microwave Limb Sounder) driven primarily by the sensitivity and sampling to different types of ice hydrometeors such as small ice crystals, graupel, and snow [Waliser et al., 2009, and references therein]. Another recent survey of water vapor vertical structure from climate model output revealed a systematic model dry bias in the lower tropical troposphere and a moist bias in the extratropical upper troposphere relative to AIRS [Pierce et al., 2006]. The aforementioned model and observational discrepancies, and the strong coupling between cloud, temperature and water vapor fields through radiative, thermodynamic, and hydrologic processes, demands a robust multisensor observational approach to observing clouds and humidity.

[5] In situ observations suggest that interhemispheric asymmetries in RHI distributions are caused by anthropogenic aerosol [Ovarlez et al., 2002; Gayet et al., 2004]. Results obtained from numerical simulations assuming various mixtures of heterogeneous and homogeneous cloud nuclei also imply large differences of RHI within and near ice clouds [e.g., Haag et al., 2003]. Although in situ observations are made on spatial and temporal scales that resolve the fine structure of clouds and their environment, they cannot provide continuous global-scale sampling. Therefore, satellite-based retrievals via a multisensor approach are complementary and necessary despite inherent limitations in sampling, vertical and horizontal resolution, and precision. Using RHI retrieved from AIRS, Gettelman et al. [2006a] (hereinafter referred to as G06) suggested that interhemispheric differences in temperature variance, also obtained from AIRS, may explain the higher frequency of ice supersaturation observed in the SH rather than aerosol differences alone.

[6] This paper is (in part) motivated by the work of G06. Cloud hydrometeor profiles from CloudSat and CALIOP are used to separate cloud and clear-sky components of RHI. Then, AIRS temperature retrievals are used to test the temperature variance hypothesis. Section 2 discusses the AIRS, CloudSat and CALIOP instruments and data sets, sampling and precision issues, and illustrates an example vertical cross section of cloud geometry and RHI. Section 3 summarizes the results of a global climatology of cloudy and clear-sky RHI partitioned by latitude, height, a few cloud types, and the sensor (radar and/or lidar) used for cloud detection. In section 4, a discussion on RHI seasonality and its interhemispheric differences is presented, and the effect of temperature variance on RHI is tested with the analytical approach developed by Kärcher and Haag [2004]. Section 5 summarizes the findings of this work.

2. Data and Methodology

[7] The cloud profiles observed by CloudSat and CALIOP determine the cloud vertical structure within collocated AIRS profiles of temperature and water vapor. Although recent studies have quantified the accuracy and precision of AIRS-derived cloud top properties [e.g., Kahn et al., 2008a], they are not capable of providing information on the vertical structure required to partition layers of cloud and clear sky.

[8] Both CloudSat and CALIPSO have been in operation since the summer of 2006 and AIRS since September 2002. In sections 2.12.3, results of cloudy and clear-sky RHI are presented for 1 year of matched observations, beginning 1 September 2006 and ending 31 August 2007.

2.1. AIRS Temperature and Water Vapor Profiles

[9] The AIRS/AMSU instruments onboard EOS Aqua have been operating in tandem since September 2002 [Aumann et al., 2003], and AIRS observes Earth's IR spectrum across 2378 channels between 3.7 and 15.4 μm. The AIRS footprint size is ∼13.5 km at nadir, while AMSU is ∼40 km at nadir and is the native resolution of temperature (T) and specific humidity (q). The sounding tandem scans ±48.95° off nadir for near-global coverage with approximately 2.9 million IR spectra and 324,000 Level 2 (L2) retrievals daily. The T and q operational retrievals are based on the “cloud-clearing” methodology [Susskind et al., 2003]. The Level 2 (L2) Standard product is used to calculate RHI following G06 and Kahn et al. [2008b]. Portions of retrieval profiles that are retained herein are restricted to pressure levels lower than reported in the variable PGood, the pressure level below which geophysical retrievals are considered “good” quality [Susskind et al., 2006]. Frequently, profile pressure levels greater than PGood are found within precipitating or opaque clouds and are not considered further (Figure 1, bottom, vertical white stripes). Additional work is required to investigate microwave-only derived RHI within the cloud types that AIRS cannot sample, such as the RHI climatology compiled by Buehler et al. [2008].

Figure 1.

Vertical cross sections of (top) radar-lidar cloud mask (2B-GEOPROF-LIDAR), (middle) radar-only cloud type (2B-CLDCLASS), and (bottom) AIRS-derived (L2 Standard retrieval product) relative humidity with respect to ice (RHI) for an illustrative ascending orbit on 4 June 2007 using CloudSat granule 5852 that contains AIRS granules 52–58. RHI is only calculated in layers where a given level is below “PGood,” T < 243 K, and q > 15 ppmv. The black outline in the bottom plot denotes RHI = 100%.

[10] The vertical profiles of RHI are overinterpolated to the vertical resolution of the CloudSat/CALIPSO 2B-GEOPROF-LIDAR vertical cloud mask (see section 2.2) using log linear interpolation [Kahn et al., 2008b]. The AIRS mission was designed for an accuracy of 1.0 K in 1 km layers for T and 20% in 2 km layers for q [Aumann et al., 2003]. Tobin et al. [2006] showed that root-mean squared differences of AIRS-derived T and q matched to dedicated radiosondes generally meet or exceed the design requirements described by Aumann et al. [2003]. However, calculations of AIRS averaging kernels, a measure of the vertical resolution, demonstrate that the vertical resolution requirements are generally not met for both T and q and have significant variability depending on atmospheric state and altitude [Maddy and Barnet, 2008]. In the case of T (or q), the full width at half maximum of the averaging kernels can range anywhere from 2.5 to 7 km (2.7–4.3 km) from the surface to the tropopause. The height-dependent nature of the vertical resolution estimates is a consequence of variable temperature lapse rate, water vapor burden, and other scene-dependent quantities [Maddy and Barnet, 2008].

[11] A majority of the clouds that AIRS can “penetrate” and also that simultaneously meet the quality control standards described above have geometrical thicknesses less than the vertical resolutions of T and q [e.g., Kahn et al., 2008b]. On the one hand, significant smoothing of RHI within vertically stacked and geometrically thin cloud and clear-sky layers is expected to reduce the magnitude of RHI maxima (and increase minima) [Kahn et al., 2008b; Lamquin et al., 2008]. On the other hand, retrieval noise in T and q will widen the distributions of RHI, usually more so in the case of T [Buehler and Courcoux, 2003; Kärcher and Haag, 2004; Gettelman et al., 2006a]. The minimum sensitivity of q is estimated to be around 15–20 ppmv [Gettelman et al., 2004; Read et al., 2007; Fetzer et al., 2008a] because the radiative signature in water vapor channels approaches radiance noise levels in these very dry conditions. Mean climatological water vapor mixing ratios at 150 hPa are on the order of 5–15 ppmv [Fueglistaler et al., 2009], but natural variability implies that some values of q exceed the nominal AIRS threshold (see section 3). For further discussion on instrument characteristics, refer to Aumann et al. [2003], and additional quality control and sampling discussions are given by Susskind et al. [2006], Fetzer et al. [2008a], and Kahn and Teixeira [2009].

2.2. CloudSat/CALIPSO Cloud Profiles

[12] CloudSat is a 94-GHz cloud profiling radar orbiting within ∼55 s of EOS Aqua, or within a few hundred kilometers of the AIRS nadir view [Kahn et al., 2008a]. The temporal and spatial synergy between AIRS and CloudSat not only provides vertically resolved hydrometeor detection collocated with T and q, but also cloud ice and liquid water content (IWC/LWC), precipitation, cloud classification, radiative fluxes and heating rates [Mace et al., 2007; Austin et al., 2009; Sassen and Wang, 2008; Stephens et al., 2008]. The CloudSat vertical resolution is 480 m with 240 m oversampling, and the horizontal resolution is approximately 1.4 km (cross track) × 2.5 km (along track) with profiles sampled roughly every 1 km.

[13] The CALIPSO platform is orbiting within ∼15 s of CloudSat and consists of three instruments: CALIOP, the imaging infrared radiometer (IIR), and the wide field camera (WFC) [Winker et al., 2007]. The instrument suite of CALIPSO is designed to observe a range of cloud and aerosol quantities such as vertically resolved cloud and aerosol detection, cloud and aerosol type, cloud water phase, extinction, optical depth, and ice particle size and shape. Tenuous cloud and aerosol features are associated with relatively weak backscatter that approaches the limits of feature detection with CALIOP. As a result, an adjustable horizontal averaging approach (333 m, 1, 5, 20, or 80 km, depending on the feature) is required for noise reduction to detect tenuous cloud and aerosol features. The vertical resolution is 30 m from the surface to an altitude of 8.2 km, and is 60 m above 8.2 km, much finer than the vertical resolution for both AIRS and CloudSat.

[14] Three cloud-related CloudSat/CALIPSO products are used to characterize cloud vertical structure: (1) the Release 4 (R04) combined radar+lidar vertical cloud mask (2B-GEOPROF-LIDAR), (2) the R04 radar-only cloud type classification (2B-CLDCLASS) [Sassen and Wang, 2008], and (3) the R04 radar-only cloud water content (2B-CWC-RO) [Austin et al., 2009]. All three products are reported at the native 240 m vertical and 1 km horizontal resolution of CloudSat. At the time of writing, no version of products 2 or 3 from combined radar+lidar observations was publicly available. For the radar-only cloud classification product, clouds are classified into Altocumulus (Ac), Altostratus (As), Cumulonimbus (Cb), Cirrus (Ci), Cumulus (Cu), Nimbostratus (Ns), Stratus (St), and Stratocumulus (Sc); further discussions of this product and its applications are given by Sassen and Wang [2008] and Kahn et al. [2008a]. The focus of this paper is on upper tropospheric ice clouds and humidity, and AIRS cannot sample through opaque cloud types (see section 2.3) like Cb and Ns; thus only distinctions between Ci and “All” cloud types will be made henceforth. The sampling rates of valid temperature and water vapor retrievals as a function of cloud type, optical depth, cloud fraction and other geophysical quantities will be published elsewhere.

2.3. Collocation and Sampling Considerations

[15] In Figure 1, an illustrative vertical cross section of humidity and cloud structure is shown along the CloudSat/CALIPSO track ±70° latitude. Figure 1 shows the combined radar-lidar cloud mask, the radar-only cloud type, and RHI. In the calculation of RHI, T < 243 K is required to eliminate the vast majority of water and mixed phase clouds. This is done in absence of a satisfactory cloud phase assessment methodology [Kahn et al., 2008b; Nasiri and Kahn, 2008], which explains the latitude dependence of RHI contours (higher altitudes in warmer low latitudes). Furthermore, q > 15 ppmv is required following estimates of the lower sensitivity limit of AIRS [Gettelman et al., 2004; Read et al., 2007; Fetzer et al., 2008a], which explains the upper reach of the RHI contours (higher altitudes in moist low latitudes). Note that many of the thin Ci detected in the radar+lidar cloud mask, but missing in Figure 1 (middle), are greater in altitude than the highest altitude containing values of q > 15 ppmv. Only portions of profiles below PGood (in pressure) are used and are shown in Figure 1.

[16] Although the profiles in Figure 1 are nearly coincident in space and time, the different vertical and horizontal resolutions of the three instruments must be considered. This work uses the “nearest neighbor” collocation approach described by Kahn et al. [2008a]. The RHI profile has a nominal horizontal resolution of ∼45 km, thus approximately 45–50 radar-lidar cloud mask profiles are matched per RHI profile. Furthermore, the vertical gridding of RHI is coarser than the radar-lidar cloud mask (∼1–2 km versus 0.24 km, respectively). The CloudSat/CALIPSO ground track only samples a vertical cross section through the AMSU footprint; see Kahn et al. [2008a] for sampling differences in the presence of inhomogeneous cloud fields.

[17] For optically thick and overcast clouds such as Cb and Ns, the AIRS retrieval usually fails or obtains a valid retrieval only above the cloud top (see Figure 1, middle and bottom). In the case of optically thin and overcast clouds such as Ci, the AIRS retrieval frequently samples above, within and below the cloud layer. In the case of optically thick and broken clouds with horizontal dimensions smaller than the AMSU footprint such as Ac, Cu and Sc, the AIRS retrieval samples around (but not within) the clouds. Therefore, the view of cloudy and clear-sky RHI presented in the following sections should be considered a “blurred” view at the scale of ∼45 km and is not intended to address RHI near ice cloud edges [e.g., Haag et al., 2003; Ström et al., 2003]. Future work will address subfootprint-scale cloud heterogeneity and its impact on the interpretation of T and q retrievals.

[18] In summary, the inferences drawn from Figure 1 and additional sampling analyses [Fetzer et al., 2008b] show the importance of considering cloud-type and its impact on the precision and accuracy of T and q retrievals. A much higher percentage of failed retrievals exists in opaque and precipitating clouds [Soden and Lanzante, 1996] like Cb and Ns, as well as over areas of St [Fetzer et al., 2006], whereas a much higher proportion of successful retrievals are found above, within and below tenuous clouds like Ci [Kahn et al., 2008a]. Furthermore, AIRS retrieval failures occur over some land surfaces (e.g., 40°N–48°N in Figure 1) and Sc. Qualitatively, patterns of humidity correspond with patterns of clouds and clear sky. Higher RHI is found near the tropical tropopause in and near thin Ci and in the midlatitude storm tracks. Also, lower RHI is found in regions of subsidence and clear-sky in the tropics and subtropics [e.g., Cau et al., 2007].

3. Cloudy and Clear-Sky RHI

[19] In this section, RHI climatologies within cloudy and clear-sky scenes are presented in histograms and in zonal means. Histograms for clear sky (radar+lidar), Ci (radar only), and “All” cloud types (radar only) are presented below for 1 year of collocated profiles.

3.1. Histograms of RHI

[20] Figure 2 presents global, seasonal variations of RHI and for three latitude bands: 40°S–60°S, 20°S–20°N, and 40°N–60°N. One clear-sky and two cloudy categories are shown: radar-detected Ci, all radar-detected cloud types combined (“All”), and clear sky according to the 2B-GEOPROF-LIDAR feature mask. RHI within Ci has a peak frequency near ∼90% in the midlatitudes but is ∼10% less in the tropics and in the global mean. The maxima of the RHI distributions, even in the tropics, are notably higher than reported by Kahn et al. [2008b] and Lamquin et al. [2008] (∼60–80%). Both studies used CALIOP-detected Ci layers that are optically and geometrically thinner than those detected by CloudSat. Furthermore, the number of low RHI values is significantly reduced in the tropics when using radar-detected Ci [Kahn et al., 2008b, Figure 7a]. The differences between RHI from these studies and the present one are consistent with the dry bias introduced from the vertical smoothing of stacked in-cloud (high RHI) and clear-sky (low or high RHI) layers. Geometrically thicker cloud layers detected by CloudSat are similar to the vertical resolution of AIRS T and q. Similar behavior between latitudes exists for “All” clouds, although the difference in RHI peak frequency between the tropics and midlatitudes is greater at ∼20%. Some seasonality in the frequency of supersaturation is observed for “All” clouds, with a stronger signal in the NH than SH, and somewhat larger variability for Ci is observed owing to the smaller sample size.

Figure 2.

The RHI histograms obtained using the criterion specified in Figure 1 for September–November (SON), December–February (DJF), March–May (MAM), and June–August (JJA) in 2006–2007. (right) “Clear sky” is defined as clear portions of profiles by 2B-GEOPROF-LIDAR. Clouds are defined two different ways using the radar-only 2B-CLDCLASS: (middle) for all cloud types together and (left) for the cloud type “Cirrus” only. RHI histograms are shown for (a) the northern midlatitudes (40°N–60°N), (b) the tropics (20°S–20°N), (c) the southern midlatitudes (40°S–60°S), and (d) all latitude bands.

[21] For RHI within radar-detected Ci, a notable dropoff in the frequency of highly supersaturated values (>150%) is observed. The precision of AIRS T and q decreases in the presence of tenuous and broken clouds compared to clear scenes [Susskind et al., 2006]. Thus, if the frequency of supersaturation is driven primarily by noise in the AIRS retrievals, it would be larger in cloudy RHI distributions with all else constant [e.g., Buehler and Courcoux, 2003; Gettelman et al., 2006a]. Hence, the reduction in the frequency of supersaturation in Ci suggests a physical reduction of RHI when compared to “All” clouds and clear-sky RHI as opposed to retrieval or instrumental effects. Simulations of cirrus evolution have shown that RHI relaxes toward saturation some time after cloud formation, but supersaturated layers may persist with characteristics controlled by variations in ice nucleation or other mechanisms [e.g., Khvorostyanov et al., 2006; Comstock et al., 2008; Spichtinger and Gierens, 2009; Krämer et al., 2009]. Therefore, the reduction of RHI within Ci is consistent with their physical evolution, namely, the uptake of water vapor onto growing ice crystals [Lamquin et al., 2008]. For the “All” cloud RHI histograms, the tails are more prominent, but could indicate the presence of high values of RHI around the edges of broken and opaque clouds.

[22] Clear-sky RHI peaks near 10–20% but is highly skewed and has a large tail of supersaturation. These distributions are most similar to in-cloud RHI for ice cloud optical depth <0.1 from Kahn et al. [2008b] and are similar to G06. The high frequency of values >150–160% suggests the effects of retrieval noise, more so in T than q [Kärcher and Haag, 2004]. However, published values of retrieval noise are not able to quantitatively explain the existence of most supersaturated values (G06). A stronger seasonal signal in the frequency of supersaturation is observed in clear-sky rather than cloudy RHI. Peak frequencies are found in the local wintertime of each midlatitude band, and December–February (DJF) in the tropics.

[23] As noted above, RHI has a peak frequency in the range of 85–90% in Ci (Figure 2). Further insight is obtained when sorting by ice water content (IWC) derived from CloudSat [Austin et al., 2009]. For the lowest values of IWC (1–30 mg m−3), RHI peaks more closely to 90–95% with a further reduction in frequencies of RHI < 50% (Figure 3), which resemble RHI distributions observed inside Ci during the Measurement of Ozone by Airbus In-service airCraft (MOZAIC) project [Spichtinger et al., 2004]. These distributions are also consistent with subsaturated air (RHI < 70%) found in very tenuous ice clouds observed during the INCA experiment [Ström et al., 2003] and other field campaigns [Krämer et al., 2009]. However, when IWC = 30–100 mg m−3, peak RHI decreases by 5–10%, and the frequency of RHI < 50% increases. For IWC = 300–1000 mg m−3, RHI peaks near 55–65% with very few values of RHI > 130%. Manual inspection (not shown) of randomly selected vertical cross sections of IWC and RHI show an anticorrelation for some geometrically thick clouds (high RHI/low IWC near cloud top, and vice versa). Ice crystals grow at the expense of water vapor, descend to the base of the cloud and precipitate into the subsaturated environment, producing a vertical sorting of IWC and ice crystal size [Heymsfield and McFarquhar, 2002]. Given that CloudSat is insensitive to small ice particles, a combined radar-lidar cloud water content product is necessary for a more complete view of quantifying RHI–IWC relationships and investigating aspects of the ice cloud lifecycle. Furthermore, quantifying the causes of the histogram structure in Figure 3 from the spectrum of small- and large-scale processes is best attempted with a combined observational/modeling approach [e.g., Jensen et al., 2001].

Figure 3.

Global RHI histograms sorted by ice water content (IWC) (2B-CWC-RO) for January 2007.

[24] The histograms in Figures 2 and 3 show a preponderance of supersaturation modulated by latitude, the presence of clouds or clear sky, cloud type and IWC. Next, zonal and height structures are shown for cloudy and clear-sky RHI.

3.2. Zonal Cross Sections of RHI

[25] The zonal annual-averaged mean RHI (equation image), standard deviation (σRHI) and total counts in clear and cloudy skies (radar only and radar+lidar) are shown in Figure 4. Clear-sky equation image shows extensive dry air in the subtropics and midlatitudes, the middle troposphere of the tropics, and high altitudes of the middle and high latitudes (equation image < 40%). Persistent supersaturation is observed in the clear-sky tropical upper troposphere and the lower troposphere of the Arctic and Antarctic (equation image > 100%) [Gettelman et al., 2006c]. The highest variability occurs in the lower troposphere of the Antarctic (σRHI > 40%) with weaker maxima in the Arctic and throughout the drier portions of the subtropics and midlatitudes (σRHI = 20–30%). Interestingly, the clear-sky variability in the tropical upper troposphere on either side of the equator is especially small (σRHI = 10–20%), indicating that the mean clear sky is consistently supersaturated with respect to ice in the deep tropics near the base of the tropical tropopause layer. The total counts are also presented in Figure 4 and show the altitudes and latitudes where the sample sizes are smallest (tropopause and lower/middle troposphere) and greatest (upper troposphere). The seasonally varying zonal cross sections that follow reveal the importance of differences in temporal sampling.

Figure 4.

Zonal-averaged cross sections of (left) the mean RHI, (middle) σRHI, and (right) total counts of cloudy and clear-sky RHI for September 2006 to August 2007. (a) Clear-sky and (b) cloudy sky from the radar-only are defined the same as in Figure 2. (c) Cloudy sky from the radar+lidar uses clouds detected in 2B-GEOPROF-LIDAR. The red line indicates the 100% RHI contour.

[26] The radar-only view of cloudy sky equation image shows a moister atmosphere than clear sky for almost all altitude and latitude bins except in the tropical upper troposphere (equation image = 100%). The lower values in the tropical upper troposphere are a result of the inherent sampling of radar-detectible clouds (around 13–15 km with larger hydrometeors and IWC), which have less frequent supersaturation (Figures 2 and 3). In the midlatitude storm tracks and subtropics, equation image is much larger than clear sky in most altitude and latitude bins. However, equation image < 100% in much of the tropics and subtropics, suggesting a preponderance of geometrically thin cloud layers and concomitant low biases of equation image (see section 2). In comparison to clear sky, σRHI is lower in most altitude and latitude bins for radar-only cloudy scenes, especially in the midlatitudes and subtropics, as well as the lower tropospheric Arctic and Antarctic regions. Supersaturation is persistent in and around polar clouds. Last, the cloudy radar-only counts are approximately an order of magnitude less than the clear-sky counts. Although numerous cloud climatologies based on imager and sounder data have observed a frequency of cloudiness ∼70% globally [e.g., Rossow and Schiffer, 1999], clouds usually occupy thin vertical layers of the troposphere [Comstock et al., 2002]. Furthermore, as discussed previously, AIRS retrievals fail in the presence of precipitating and opaque clouds, further exacerbating the difference in the cloudy and clear-sky sample sizes.

[27] The cloudy radar+lidar equation image patterns are similar to the radar-only view, with a few important differences. The equation image in the cloudy tropical upper troposphere is similar to equation image in clear sky, a result of including clouds with small hydrometeors at higher altitudes that contain much more frequent supersaturation [Jensen et al., 2001; Khvorostyanov et al., 2006]. Furthermore, equation image is somewhat lower near the tropopause in the subtropics and in the middle troposphere of the tropics, a result of the lidar detecting geometrically thinner cloud layers than the radar. The magnitude of σRHI is similar between the radar-only and radar+lidar climatologies, except that σRHI in the radar+lidar view more closely resembles clear sky in the lower troposphere of the Arctic and Antarctic. The total counts are somewhat higher than the radar-only counts, with a greater frequency of sampling just below the tropopause, especially in the tropics.

[28] The seasonal variations of the three types of equation image climatologies shown in Figure 4 (clear sky, radar-only clouds, and radar+lidar clouds) are presented in Figures 5 and 6. At least three patterns of seasonal variability emerge. First, the latitudes and altitudes for which the conditions T < 243 K and q > 15 ppmv are met vary over the course of the year for both clear and cloudy skies (Figures 5 and 6). Thus, the total counts shown in Figure 4 are not evenly sampled throughout the year, suggesting a seasonal sampling bias of equation image in a few of these regions. An example of seasonal sampling biases in clear sky is seen in the SH high latitudes > 8 km (Figure 5). Note that in Figure 4, the annual equation image ∼ 20–70% from 12 to 8 km. In Figure 5, equation image ≥ 100% in June–August (JJA) and September–November (SON) (local winter and spring); however, these atmospheres are very cold and dry and there are fewer occurrences of q > 15 ppmv compared to warm season DJF and March–May (MAM) in the SH. Therefore, the annual equation image is more consistent with DJF and MAM values.

Figure 5.

Seasonal variations of zonal-averaged cross sections of the mean clear-sky RHI for September 2006 to August 2007.

Figure 6.

Seasonal variation (DJF and JJA) as in Figure 5 except for (top) radar-only and (bottom) radar+lidar only cloudy sky.

[29] The second pattern of seasonal variability is that equation image in the clear-sky subtropics is somewhat lower during the local wintertime, showing broad consistency to humidity climatologies derived from infrared and microwave sounders [e.g., Randel et al., 1996; Sandor et al., 1998; Stone et al., 2000; Gettelman et al., 2006a, 2006b; Buehler et al., 2008]. Similar variability and consistency between these data sets is also observed in the moist midlatitudes and tropical upper troposphere. The third pattern is equation image in the tropical upper troposphere peaks in DJF and MAM and reaches a minimum in SON, consistent with the annual variation in thin Ci frequency [e.g., Sandor et al., 2000; Sassen et al., 2008]. Furthermore, in JJA the peak equation image shifts to 5–25°N in the radar+lidar view of equation image in the bottom row of Figure 6, also very similar to the latitudinal migration of thin Ci [Sassen et al., 2008]. This latitudinal shift is not as prominent in the radar-only view in Figure 6 (top), as CloudSat does not detect the vast majority of thin Ci. Therefore, subtle differences in the seasonal cycle are observed when comparing cloudy equation image determined by radar- or lidar-detected clouds.

3.3. Constant Pressure Maps of RHI

[30] Global distributions of cloudy (lidar-only) and clear-sky equation image at 300 and 150 hPa during SON are shown in Figure 7. equation image is significantly higher in clouds although with a low bias due to the vertical geometry as described by Kahn et al. [2008b] and Lamquin et al. [2008] from vertical smoothing of clear (dry) and cloudy (moist) layers. Similar to Sandor et al. [1998], Stone et al. [2000] and others, the highest equation image at 150 hPa occurs in the tropics, whereas at 300 hPa, the lowest equation image occurs there. Observe that the clear and cloudy equation image at 150 hPa are very similar, perhaps equation image is 10–20% higher in clouds. At 300 hPa, cloudy equation image is 30–60% higher than clear sky and largest in the midlatitudes. Any clear-sky “bias” at 150 hPa is small since cloudy and clear equation image distributions are very similar. At 300 hPa, the number of cloudy equation image counts is several factors smaller than clear sky, therefore the “all sky” equation image (not shown) will only be slightly greater than clear sky as pointed out by Soden and Lanzante [1996], who suggested a clear-sky bias of 10% or less relative to all sky conditions. Furthermore, in Figure 7 observe the absence of cloudy equation image values in regions of strong descent west of Africa, South America and Australia caused by the lack of clouds. For cloudy equation image, the midlatitudes are noisier because of reduced sampling from increased opaque cloudiness and AIRS retrieval failures (see section 2.3). This sampling effect highlights the importance of quantifying the precision and accuracy of temperature and humidity retrievals from microwave sounders within opaque cloud scenes [e.g., Randel et al., 1996; Buehler et al., 2008].

Figure 7.

(left) Clear-sky and (right) cloudy RHI defined by the radar+lidar at (top) 300 hPa and (bottom) 150 hPa for SON 2006.

4. Interhemispheric and Seasonal Differences in RHI

[31] The influence of anthropogenic and natural aerosol on ice cloud formation, evolution, and microphysical structure is poorly understood. In situ observations [Ovarlez et al., 2002; Gayet et al., 2004] and numerical experiments [e.g., Haag et al., 2003; Kärcher and Lohmann, 2003; Comstock et al., 2008] have provided strong evidence that anthropogenic aerosol may cause interhemispheric differences in RHI from increased contributions of heterogeneous nucleation in the NH. In contrast, G06 suggested that interhemispheric differences in temperature variance (σT), which can serve as a proxy for vertical velocity variance, may also explain the higher frequency of AIRS-derived ice supersaturation observed in the SH. It is important to consider that the temporal/spatial resolution and sampling, precision and accuracy of in situ and satellite observations are vastly different such that (1) the signal of interhemispheric RHI differences from in situ observations may be effectively “smeared out” if viewed by the satellite multisensor approach advocated here and (2) the limited number of in situ campaigns does not capture a globally representative temporal and spatial sample of cloudy and clear-sky RHI variability. The remainder of this paper investigates the nature of RHI derived from AIRS in the context of temperature variability using the analytical method developed by Kärcher and Haag [2004] (hereinafter referred to as KH04).

4.1. AIRS Upper Tropospheric T, q, and RHI

[32] Significant spatial and temporal variability of T and q variance is observed in AIRS retrievals [Kahn and Teixeira, 2009]. Clear-sky σT for DJF and JJA at 300 hPa is shown in Figure 8 on a 4° × 4° grid. Note that σT < 0.5 K in the tropics but is frequently >1.0 K in the midlatitude storm tracks. Higher values of σT are also found over elevated topographical features. The patterns of cloudy sky σT are similar to clear sky except the magnitudes are greater by 0.5–1.0 K. The magnitude of σT is proportional to the grid size used for averaging and typically follows power law scaling relationships that were quantified by Kahn and Teixeira [2009].

Figure 8.

Spatial distribution of σT (K) at 300 hPa limited to clear sky for both (left) JJA and (right) DJF at a gridded resolution of 4° × 4°. “Clear” AIRS pixels are defined as those with an effective cloud fraction <0.05 (averaged over the AMSU field of view).

[33] Histograms of T and q in DJF and JJA of 2007 are shown in Figure 9 but are restricted to 400–200 hPa, and 40°N–60°N and 40°S–60°S in clear sky; all sky distributions are similar (not shown). A strong seasonal variation of mean temperature (equation image) is observed in both latitude bands with “all sky” equation image and σT summarized in Tables 1 and 2, respectively, for DJF and JJA separately for the years 2003–2008. (Clear and cloudy sky equation image and σT have been derived separately and show similar seasonal and interhemispheric variability, and are not shown for sake of brevity.) For instance in 2007, equation image during local winter is very similar in the NH and SH, but the NH is warmer by 3.4 K in summer compared to the SH; the same pattern holds true in other years (see Table 1). Interestingly, σT is only 0.05 K higher in the SH winter than summer, consistent with a more spatially and seasonally invariant storm track than the NH, while σT is 1.5 K higher in the NH winter when compared to summer. The magnitudes of σT in Table 2 are much larger than those shown in Figure 8 because they are for the entire latitude-pressure range discussed in Figure 9 and are several factors higher at larger scales [Kahn and Teixeira, 2009]. Last, Figure 9 shows histograms of q for the same seasons and latitude bands. The histograms of q during local winter are nearly identical, whereas the NH has a higher frequency of q > 100 ppmv compared to the SH, consistent with the hemispheric differences of equation image.

Figure 9.

(top) T and (bottom) q distributions of clear-sky AIRS pixels for the NH and SH latitude bands separately during 2007 DJF and JJA and between 200 and 400 hPa.

Table 1. Seasonal and Hemispheric (40°–60° South and North) Mean Temperature for All Sky Restricted to Retrieval Layers Between 200 and 400 hPa That Reside Above PGood and Have q > 15 ppmva
Latitude BandSeason2002–20032003–20042004–20052005–20062006–20072007–20082002–2008 Average
  • a

    Temperature is given in Kelvins.

Table 2. Same as Table 1 Except Showing Temperature Variancea
Latitude BandSeason2002–20032003–20042004–20052005–20062006–20072007–20082002–2008 Average
  • a

    Temperature variance (σT) is calculated on each pressure level separately and then is averaged with the other levels and weighted by the total number of counts.


[34] Cloudy and clear-sky RHI histograms in the NH and SH 400–200 hPa layer during DJF and JJA are shown in Figure 10. If the distributions for the SH and NH are averaged into a DJF+JJA histogram (not shown), the differences between the NH and SH (for both cloud and clear sky) are small unlike that shown by G06. In that study, 4 years of AIRS retrievals (SON for the NH, and MAM for the SH) were used to form the RHI distributions that showed a greater frequency of supersaturation in the SH. In this work, only values along the CloudSat/CALIPSO track are used (∼1/30th of all retrievals) for a 1-year period, for a data volume about ∼1/120th of G06. Thus, it is not surprising that a much smaller sample shows differences compared to G06. Another possible cause of the differences is that G06 used Version 3 (v3) AIRS retrievals and here Version 5 (v5) is used. There are algorithm and quality assessment differences between v5 and previous versions (see section 2.1 and references therein), but sampling differences arising from clouds also exist (see section 3 and Soden and Lanzante [1996]). Furthermore, Tables 1 and 2 show significant interannual variations in equation image and σT. Large variations in σT are especially crucial to the sensitivity of RHI (KH04). The control of seasonal variations of RHI by equation image and σT, both simultaneously and individually, will be addressed in section 4.2.

Figure 10.

(top) Radar+lidar defined clear-sky and (bottom) radar-only cloudy RHI distributions for the same latitude bands, pressure levels, and seasons as in top plot of Figure 9.

4.2. Temperature Controls of RHI

[35] The sensitivity of RHI to seasonal variations of T and q is assessed using the analytical approach of KH04. The method of KH04 was developed to quantify the probability distribution of RHI from combinations of saturation vapor pressure and vapor pressure over ice. KH04 does not consider the temporally varying aspects of transport, radiation, and microphysical processes on redistributing T and q. Instead, RHI histograms are diagnosed given Gaussian distributions of T and q based on observed in situ values for the upper troposphere. The distributions of T and q shown in Figure 9 are not perfect Gaussian distributions. Although they may be skewed to some extent, KH04 shows that, given the known limitations of this assumption, assumed Gaussian distributions are able to faithfully reproduce the magnitudes and slopes of RHI. KH04 shows that both equation image and σT modulate the distribution characteristics of RHI, much more so than variations in q, consistent with other studies [e.g., Buehler and Courcoux, 2003; Gettelman et al., 2006a]. According to KH04's temperature dependence formulation, three physical quantities control the functional behavior of RHI distributions: the mode of RHI, equation image, and σT (see KH04, Figure 2). For an arbitrary, fixed mode of RHI and constant σT, a variable equation image acts to slightly broaden (narrow) the RHI distribution for lower (higher) equation image. For an arbitrary, fixed mode of RHI and constant equation image, a variable σT acts to strongly skew the RHI distribution toward greater (lesser) supersaturation for higher (lower) σT. More concisely, higher frequencies of supersaturation are obtained for a fixed mode of RHI by decreasing equation image and increasing σT, with a greater sensitivity from adjustments in σT.

[36] Tables 1 and 2 show a general decrease in equation image and increase in σT from local summer to winter in all years in the NH, with a smaller seasonal variation of σT in the SH. These factors contribute to an increasing skewness of RHI > 100% during winter. Values of equation image and σT from DJF and JJA of 2007 are tested with KH04 for a mode of RHI = 30%, and the results are presented in Figure 11. RHI is calculated using variable values of equation image and σT simultaneously for different seasons (Figure 11a), variables values of equation image for different seasons and setting σT = 4.8 K (to isolate impacts of equation image) (Figure 11b), and variable values of σT for different seasons and setting equation image = 225.8 K (to isolate impacts of σT; Figure 11c). Note that the NH (black curves) in Figures 11a and 11c have much different seasonal slopes and magnitudes, namely, that σT explains almost all of the difference in RHI between DJF and JJA than equation image (Figure 11b). In the SH (gray curves), equation image dominates the contributions to RHI seasonality (Figures 11a and 11b), although the effects of equation image and σT are more similar in other years. Thus, seasonal variations in σT are generally more important for controlling RHI in the NH, while both equation image and σT control RHI in the SH, with some interannual variation in the relative contributions of equation image and σT. Last, the frequency of supersaturation in Figure 11 is highly sensitive to the choice of the RHI mode; however, the relative order and spacing between the different curves does not change as the fixed modal value of RHI = 30% is adjusted upward (not shown). Thus, the inferences drawn from Figure 11 are generally insensitive to the RHI mode chosen. This remains true if multiple modes are treated individually, weighted by the total counts, and combined a posteriori.

Figure 11.

These curves are calculated from the analytical expressions of Kärcher and Haag [2004] and use AIRS values from Tables 1 and 2 unless noted. RHI distributions for 2007 NH (black) and SH (gray) bands using (a) values of mean temperature (equation image) and temperature variance (σT), (b) variable equation image and but fixing σT = 4.8 K, and (c) variable σT but fixing equation image = 225.8 K.

[37] To more clearly quantify the impacts of interannual variations of equation image and σT on RHI, all of the years in Tables 1 and 2 (2003–2008) are calculated individually and are shown in Figure 12. The NH has a factor of 2–3 greater difference in the spread of RHI > 100% compared to the SH, emphasizing the importance of interannual variations of σT on RHI. The SH has a somewhat smaller spread in equation image from DJF to JJA than the NH, but is much less important in impacting the skewness of RHI when compared to σT.

Figure 12.

As with Figure 11, these curves are calculated from the analytical expressions of Kärcher and Haag [2004] and use AIRS values from Tables 1 and 2 for each year individually. Annual variations of RHI for (a) NH DJF, (b) NH JJA, (c) SH DJF, and (d) SH JJA.

[38] Figure 13a shows calculations of RHI using values from Tables 1 and 2 and the method of KH04 but is restricted to the NH during SON and SH during MAM (2002–2008 individually), approximating the time periods and latitude bands observed during INCA [Ovarlez et al., 2002; Gayet et al., 2004]. Observe that the tail of large RHI in the NH is slightly greater than RHI in the SH in all years, demonstrating that variations of equation image and σT are unable to quantitatively explain the interhemispheric differences as suggested by G06. Figure 13b shows the observed “all sky” RHI distributions for the same latitude bands and seasons from v5 AIRS retrievals, implying the presence of other cause(s) for the observed characteristics of RHI. Also, the relative difference between RHI in the NH and SH is substantially reduced compared to work by G06, discussed further below.

Figure 13.

(a) Modeled (using data from KH04 and Tables 1 and 2) and (b) observed RHI for SON and MAM in the NH and SH, respectively, for individual years between 2002 and 2008. The dashed curves are taken from Gettelman et al. [2006a].

[39] With regard to the variance of water vapor, Figure 9 shows that in the winter hemisphere, the SH and NH water vapor distributions are nearly identical. However, the NH summer is slightly skewed toward higher values compared to the SH summer. KH04 show that a higher water vapor variance produces a broader RHI distribution (see KH04, Figure 3), causing a small increase in supersaturation frequency. Thus, if the water vapor variance effects shown in Figure 9 were included, the supersaturated tail of RHI in the NH would likely broaden more so than the SH, and thus would widen the discrepancy between modeled and observed RHI.

[40] As discussed earlier, interhemispheric asymmetries of the concentration and composition of ice nucleating aerosols are possible contributors to interhemispheric asymmetries in RHI. On the one hand, the presence of heterogeneous ice nuclei may reduce the critical RHI necessary for ice particle nucleation, thus reducing the frequency of highly supersaturated values. On the other hand, some recent studies suggest that heterogeneous ice nuclei may increase RHI through the presence of aerosols composed of organics that increase the critical RHI for nucleation [Jensen et al., 2005] or help to reduce the total number concentration of aerosol nucleated ice [Krämer et al., 2009], which can lead to fewer nucleated ice crystals that subsequently grow to larger sizes, precipitate more rapidly, and leave a residual layer of high RHI in their wake [Comstock et al., 2008]. This study does not attempt to reconcile the various proposed ice nucleation mechanisms that ultimately control (in part) supersaturation frequency. However, the slightly reduced skewness of the RHI distributions in the NH (in the zonal averaged mean) offers some evidence that anthropogenic aerosols may reduce RHI “in bulk,” similar to the findings of Ovarlez et al. [2002], Haag et al. [2003], and others.

[41] However, interhemispheric variability in ice nucleating aerosol concentration and composition are not the only possible causes of AIRS-derived RHI differences. There is evidence of sampling differences between v3 (G06) and v5 (this study) (E. Fishbein, personal communication, 2008), namely that the slope of RHI <100% is slightly steeper and the frequency of highly supersaturated values is more similar between the NH and SH than shown by G06. This suggests that some of the cloudier (and more moist) v5 retrievals are eliminated using the quality control methodology discussed in section 2 that is appropriate for v5 AIRS retrievals (the quality control methodology was different for v3). Therefore, some combination of quality control, retrieval algorithm changes and/or sampling differences between v3 and v5 cannot be ruled out as contributing to the reduction of interhemispheric differences in supersaturation frequency (Figure 13b) compared to work by G06. Further work is necessary to quantify these effects on the AIRS geophysical retrievals.

5. Summary and Conclusions

[42] Cloudy and clear-sky distributions of relative humidity with respect to ice (RHI) are obtained from a combination of spatially and temporally coincident temperature and water vapor profiles from the Atmospheric Infrared Sounder (AIRS)/Advanced Microwave Sounding Unit (AMSU) suite, and cloud hydrometeor profiles from CloudSat and the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) satellites using the collocation and interpolation methodology of Kahn et al. [2008a, 2008b]. A multisensor observational synergy of the vertical structure of clouds, temperature, and water vapor is used to partition clear and cloudy components of RHI as a function of season, altitude, latitude, ice water content (IWC) and cloud type. A multisensor effort is expected to provide new and useful information for the new generation of climate models that are configured for the existence of ice supersaturation [e.g., Tompkins et al., 2007; Liu et al., 2007].

[43] Clear-sky RHI shows a peak frequency of 10–20% that increases to 75–95% within clouds detected by CloudSat, larger than that obtained with CALIPSO alone [Kahn et al., 2008b; Lamquin et al., 2008]. The ranges in values depend on the season, latitude, IWC and cloud type. The frequency of supersaturation is also similarly controlled by these quantities; primarily greater frequency and magnitude are found in clear sky, DJF in the tropics, and the midlatitude winter hemisphere (cloudy and clear). Less frequent occurrences of high supersaturation are observed in cirrus compared to clear sky. Global distributions of the mean, variance and seasonal variations of cloudy and clear-sky RHI reveal persistent supersaturation in the clear and cloudy tropical upper troposphere and lower tropospheric polar regions, high but variable RHI in the cloudy midlatitude storm tracks, and small yet variable RHI in the clear-sky subtropics. Last, since the AIRS retrievals are less precise in the presence of clouds and this additional retrieval noise causes increased perceived supersaturation, the lower frequency of observed supersaturation offers satellite-based evidence for the microphysical evolution of cirrus.

[44] Previous works have shown greater midlatitude RHI in the SH from satellite [Gettelman et al., 2006a] and in situ [Ovarlez et al., 2002; Haag et al., 2003; Gayet et al., 2004] observations. In particular, the in situ observations have provided some evidence that anthropogenic aerosol may cause interhemispheric differences in RHI by the increased importance of heterogeneous ice nucleation in the NH. In contrast, Gettelman et al. [2006a] suggested that interhemispheric differences in temperature variance (σT) could modulate the differences in RHI. In this work, a strong seasonality in RHI is also shown, with significantly larger RHI in both midlatitude winter hemispheres. The analytical method of Kärcher and Haag [2004] is applied to quantify the impacts of mean temperature (equation image) and σT on the interhemispheric and seasonal differences in RHI. The larger seasonal variation of RHI in the NH is shown to be modulated primarily by seasonal changes in σT, whereas both equation image and σT modulate RHI in the SH.

[45] The slightly higher frequency of supersaturation in the SH midlatitudes compared to the NH midlatitudes is not explained by interhemispheric differences in equation image and σT. This result suggests a link between anthropogenic aerosol and ice cloud processes in controlling the characteristics of RHI. However, the possibility of changes in the AIRS algorithm and quality control methodology, and their impacts on cloud sampling among different retrieval versions, is also discussed in the context of these differences. This study illustrates the usefulness of a multiple sensor approach for observing spatially and temporally simultaneous cloud and humidity properties in the upper troposphere.


[46] During the course of this study, a NASA Postdoctoral Program fellowship and the AIRS and CloudSat projects at JPL provided funding support for B.H.K. E.J.F. was supported by the AIRS project at JPL and by the NASA and Energy and Water-cycle Study (NEWS) project. The authors gratefully acknowledge the constructive comments of the three anonymous reviewers. AIRS data were acquired as part of the NASA's Earth-Sun System Division and archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC) Distributed Active Archive Center (DAAC) ( CloudSat data were obtained through the CloudSat Data Processing Center ( CALIPSO data were obtained through the Atmospheric Sciences Data Center (ASDC) at NASA Langley Research Center ( This research was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration.