Tropical thin cirrus and relative humidity observed by the Atmospheric Infrared Sounder

Global observations of cloud and humidity distri- butions in the upper troposphere within all geophysical con- ditions are critically important in order to monitor the present climate and to provide necessary data for validation of cli- mate models to project future climate change. Towards this end, tropical oceanic distributions of thin cirrus optical depth ( ), effective diameter (De), and relative humidity with re- spect to ice (RHi) within cirrus (RHic) are simultaneously derived from the Atmospheric Infrared Sounder (AIRS). Cor- responding increases in De and cloud temperature are shown for cirrus with > 0.25 that demonstrate quantitative consis- tency to other surface-based, in situ and satellite retrievals. However, inferred cirrus properties are shown to be less cer- tain for increasingly tenuous cirrus. In-cloud supersaturation is observed for 8-12% of thin cirrus and is several factors higher than all-sky conditions; even higher frequencies are shown for the coldest and thinnest cirrus. Spatial and tem- poral variations in RHic correspond to cloud frequency while regional variability in RHic is observed to be most promi- nent over the N. Indian Ocean basin. The largest cloud/clear sky RHi anomalies tend to occur in dry regions associated with vertical descent in the sub-tropics, while the smallest occur in moist ascending regions in the tropics. The char- acteristics of RHic frequency distributions depend on and a peak frequency is located between 60-80% that illustrates RHic is on average biased dry. The geometrical thickness of cirrus is typically less than the vertical resolution of AIRS temperature and specific humidity profiles and thus leads to the observed dry bias, shown with coincident cloud ver- tical structure obtained from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO). The joint distributions of thin cirrus microphysics and humidity derived from AIRS provide unique and important regional and global-scale insights on upper tropospheric processes not available from surface, in situ, and other contemporary satel- lite observing platforms.

significantly larger than the radiative effects of anthropogenic agents such as aerosols and changes in greenhouse gas concentrations (Forster et al., 2007). Geometrically and optically thin cirrus is most frequent within and below the tropical tropopause layer (TTL) and is generated and maintained to various degrees by convective detrainment (Lilly, 1988), in situ forma- 15 tion from mesoscale gravity waves (Kärcher and Ström, 2003), ascent associated with synoptic-and planetary-scale waves (Boehm and Verlinde, 2000;Jensen et al., 2001;Peter et al., 2003), and radiative cooling above cold convective anvils (Hartmann et al., 2001). By tracking cirrus evolution, Massie et al. (2002) calculated that in situ and convective detrainment processes contribute approximately 50% each towards the ob-to exceed a nominal value that depends on the water activity of the solute or the composition and nature of the IN. For homogeneous nucleation to occur, S must typically exceed 1.4-1.7 for water ice to spontaneously form (Koop et al., 2000). For heterogeneous nucleation to occur, S is generally considered to be lower than that required for homogeneous nucleation. Some recent in situ observations of S>1.7 over a limited spatial and temporal extent (Jensen et al., 2005) suggest a possible role by organic coatings on aerosols in suppressing ice nucleation. Furthermore, recent laboratory measurements of low ice deposition coefficients may explain in part the existence of high S within and adjacent to cold cirrus clouds composed of small ice particles (Magee et al., 2006). Observations of S are also important for contrail formation, maintenance 15 and impacts on cirrus frequency and microphysics (Minnis et al., 2004). Additionally, trends of increasing UT water vapor have been observed (Soden et al., 2005) and its relevance to climate sensitivity also suggests the importance of long-term monitoring of the UT. These and other studies demonstrate that the understanding of the UT hydrological cycle is far from complete (Peter et al., 2006). Therefore, improved measure-20 ments (greater precision, higher spatial and temporal sampling) of cirrus microphysics and humidity are critical to the investigation of the UT region.
A new generation of coordinated satellite platforms collectively known as the "A-train" is providing global observations of atmospheric temperature, water vapor, aerosol, cloud, surface, and trace gas properties with unprecedented capabilities (Stephens 25 et al., 2002). For instance, UT cloud properties and relative humidity (RH) are measured by the Atmospheric Infrared Sounder (AIRS), an infrared/microwave sounder on EOS Aqua that simultaneously observes cloudiness, temperature, and water vapor (Aumann et al., 2003). Using AIRS retrievals of temperature and specific humidity,  Gettelman et al. (2006) showed that ice supersaturation is frequent throughout the UT with latitudinal, regional and seasonal dependences. The frequency of supersaturation derived from AIRS is several times higher than observed by the Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (TOVS) , attributable to improved spectral resolution, sensitivity, noise and calibration character-5 istics. Numerous recent studies have proven the utility of AIRS radiances for cloud detection, amount and height retrievals (Kahn et al., 2007a, b, c), and for the inference of cirrus optical depth (τ) and effective diameter (D e ) (e.g. Kahn et al. 2003;Wei et al., 2004;Yue et al., 2007). Herein, simultaneous observations of thin cirrus τ and D e and in-cloud RH with respect to ice (RH i c ) are derived from AIRS and their correlative 10 relationships are quantified. Previous studies emphasized the relationships between cirrus frequency and humidity. For instance, Sandor et al. (2000) and Clark (2005) used limb-viewing satellites to show that UT RH i and cirrus frequency are positively correlated. Many in situ measurements of RH i within and adjacent to cirrus (e.g. Heymsfield et al., 1998;Ovarlez et 15 al., 2002;Spichtinger et al., 2004;Jensen et al., 2005) and those derived from surfacebased Raman lidar observations (Comstock et al., 2004) exist as well. They illustrate the high spatial and temporal detail of ice-supersaturated regions within cirrus, and in particular in situ instrumentation provides a more diverse and precise measurement capability over satellite observations though it is highly limited in spatial and temporal 20 extent. Many parcel modeling studies of cirrus properties have shown that the evolution of RH i c is controlled to various extents by ice crystal number concentration, ice water content, size distribution, and other factors (e.g. Khvorostyanov et al., 2006;and references therein). Furthermore, Haag et al. (2003) has suggested that heterogeneous and homogeneous ice nucleation can be inferred from observed distributions of RH i 25 originating within and adjacent to cirrus, although making this distinction from satellite observations has not been conclusively demonstrated to date. Therefore, AIRS retrievals of τ, D e , and in-cloud and adjacent clear sky RH i should be investigated for a sensitivity to cirrus formation and maintenance mechanisms, and to evaluate and 16189 their error characteristics, and the propagation of the input errors through the cirrus radiative transfer model (RTM). The sensitivity of τ and D e to errors in RTM inputs is quantified for three representative thin cirrus cases. In Sect. 3, histograms and spatial distributions of cirrus quantities are shown and are then compared to distributions derived from other retrieval platforms. Furthermore, we present joint histograms of τ, D e and RH i c , illustrate spatial distributions of RH i and in-cloud/clear sky RH i anomalies, and relationships between cirrus cloud geometrical thickness and RH i c . In Sect. 4, the primary findings are summarized and potential applications of joint cirrus and humidity distributions are discussed.

15
Here we provide a brief overview of the RTM and thin cirrus retrieval approach, the atmospheric and surface RTM inputs with a summary of uncertainties, and a quantification of the bias and variability in τ and D e caused by the various input uncertainties.

Thin cirrus retrieval approach
The inference of τ and D e employs a modified approach of Yue et al. (2007). A 20 rapid, clear-sky RTM, the Optical Path Transmittance (OPTRAN) model (McMillin et al., 1995), is coupled to a thin cirrus parameterization. The simultaneous retrieval uses a minimization method that searches for the optimal fit between observed and simulated radiance spectra. A total of 14 channels are used in the 8-12 µm window region (Fig. 1). Little to no sensitivity of τ and D e is observed in comparisons between sim-Introduction EGU ulations that use the channel list in Fig. 1 and the one used in Yue et al. (2007). This is consistent with studies of Moderate Resolution Imaging Spectroradiometer (MODIS) radiances that suggest little additional information content of cloud properties is available beyond a set of 4-5 channels (L'Ecuyer et al., 2006). Yue et al. (2007) define the ice crystal size and habit distributions as free parameters. In this work, the bulk cir-5 rus scattering models developed by Baum et al. (2007) are used for consistency with MODIS Collection 5 and to reduce the dimensionality of the retrieval parameter space for increased computational efficiency. To further justify the use of Baum et al. (2007), the frequency of the optimal fit for particular size and habit distribution combinations used in Yue et al. (2007) is shown for an illustrative AIRS granule in the tropics (Fig. 2). 10 Three combinations occur most frequently: (1) the McFarquhar et al. (1999) habit distribution and D e ≈24 µm fit the thinnest cirrus best (τ≤0.3), (2) a 100% solid column distribution and D e ≈92 µm fit slightly thicker cirrus (τ≥0.4-0.5) more often, and (3) the size-dependent habit mixture of Baum et al. (2007) and D e ≈26 µm fit similar cirrus as in (2). In this work, we allow D e to be a free parameter and fix the habit mixture to that 15 used in Baum et al. (2007). For calculated clear sky radiance, temperature T (z), specific humidity q(z), and O 3 (z) profiles are required along with surface temperature (T S ), IR emissivity (ε), IR reflectivity (ρ), and the viewing geometry. AIRS Version 5 (V5) Level 2 (L2) Standard and Support products provide the RTM with these necessary inputs. For the cloudy radi-Introduction Several validation studies of AIRS Version 4 (V4) and Version 5 (V5) T (z) and q(z) profiles in the presence of varying amounts of clouds and clear sky have quantified their accuracy and precision throughout a significant variety of geophysical conditions (Divakarla et al. 2006;Tobin et al. 2006;. Comparisons of AIRS T (z) and q(z) 5 using Vaisala RS-90 radiosondes, determined to be of sufficient precision in and near UT ice clouds by Miloshevich et al. (2006), were launched at the ARM program sites and are discussed in Tobin et al. (2006). For the tropical Western Pacific site of Nauru Island, the mean bias of AIRS V4-ARM T (z) from 100-400 hPa is approximately 0.0 K, although at a particular level the bias can be as large as ±0.5 K. The root mean 10 square (RMS) is 0.5-1.0 K from the surface to 200 hPa and increases to 1.0-2.0 K from 200 hPa to the tropical tropopause in V4, but V5 is improved with reductions of 0.1-0.2 K. Mean biases in q(z) from the surface to 400 hPa are near 0%, although below 400 hPa, a dry bias of approximately 10% is shown for AIRS V4; in V5 this bias is near 0%. RMS values for q(z) are near 10% in the lower and middle troposphere 15 and increase to 20-30% between 300 hPa and the tropopause in both V4 and V5. In summary, ARM comparisons with V5 show slight improvements over V4 for T (z) and q(z) in the tropical UT (Tobin et al., 2006. The precision of cloud top height derived from AIRS was quantified with coincident cloud observations from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite 20 Observation (CALIPSO) platform in Kahn et al. (2007c) and comparisons for f A ≤0.4 are shown in Fig. 3. The bias is mostly invariant for clouds ≥7 km (determined by CALIPSO) for all f A . During daytime (nighttime) the differences are 1.5-2.5±2-3 km (2.5-3.5±2.5-3.5 km) with slightly higher variability for smaller f A . For clouds <7 km (determined by CALIPSO), both the bias and variability are a strong function of f A with Introduction EGU algorithm and the sensitivity of near-nadir viewing infrared radiances. Furthermore, the large negative bias for f A <0.02 demonstrates the presence of a significant number of spurious cloud retrievals near the tropopause (Kahn et al. 2007c). Figure 4 shows the sensitivity of inferred τ and D e using validated uncertainties in the 5 RTM inputs for three typical thin cirrus cases. Each probability density function (PDF) is calculated from 10 000 randomly perturbed RTM inputs that are then propagated into the retrieval of τ and D e as outlined in Sect. 2.1 and Yue et al. (2007). The perturbed inputs include T (z), q(z), T C , T S , ε and ρ, and are consistent with AIRS validation results with normally distributed 1σ errors of ±1 K, 10%, 12 K, 1 K, 0.01, and 0.01, respectively 10 (ε and ρ are mutually exclusive). Although T (z) and q(z) are uniformly scaled by 1 K and 10% errors, respectively, they are in practice a function of height, but more realistic error perturbations (e.g. Kahn et al., 2005) are not investigated here. However, the spatial and temporal dependences of cloud, atmospheric and surface uncertainties must be considered when quantifying the statistical significance of spatial and tempo-15 ral differences in cirrus τ and D e , including those that may arise from anthropogenic climate change and aerosol indirect effects (e.g. Chylek et al., 2006). This point is further highlighted in Sect. 3. Three typical thin cirrus cases are shown in Fig. 4a-c with corresponding values of τ=0.08, 0.33, and 0.71, and D e =30, 40, and 50 µm. Figure 4a-c demonstrate that the Gaussian retrieval "noise" of the RTM inputs leads to 20 non-Gaussian variability in τ and D e . A case with multiple retrieval clusters (or modes) is illustrated in Fig. 4a (τ=0.08 and D e =30 µm) with the three modes centered at D e =10, 30, and 120 µm. Although the 30 µm mode is expected because it is essentially the "noiseless" retrieval, the "spurious" modes at 10 and 120 µm arise from a combination of imposing bounds on the 25 range of simulated D e and RTM input noise. The PDF reveals that more scatter is observed in D e compared to τ, although most values associated with the 30 µm mode are within ±5-10 µm, in approximate agreement with Yue et al. (2007). In Fig. 4b 16193 Introduction

Potential biases in thin cirrus retrievals
Conclusions References Tables  Figures   ◭  ◮ ◭ ◮

Back Close
Full Screen / Esc

Printer-friendly Version
Interactive Discussion EGU (τ=0.33 and D e =40 µm), more scatter is observed in τ compared to D e , but most τ are within ±0.07-0.08. Nearly all D e are located within ±5 µm, with a modicum of scatter extending to 100 µm and a fairly sharp cut-off near 30 µm. In Fig. 4c (τ=0.71 and D e =50 µm), the scatter is observed to be more extensive in both τ and D e when compared to Fig. 4b. A spurious mode at τ=1.0 arises from imposing a limit on τ at 1.0 5 in the retrieval minimization. The characteristics of the thin cirrus retrieval noise may be dominated by particular atmospheric and surface quantities. To investigate this possibility, various combinations of T (z), q(z), T C , and T S are shown in Fig. 4d-i for the case detailed in Fig. 4b. Biases of q(z), T (z) and T C have similar impacts, namely, that positive biases in D e are 10 associated with compensating negative biases in τ and vice-versa (Fig. 4d-f;Huang et al., 2004). The physical cause is the spectral dependence of water vapor absorption in the atmospheric windows. For any increase (decrease) in the total column water vapor, the spectrum in Fig. 1 has a stronger (weaker) brightness temperature (T b ) slope and lower (higher) overall T b from 800-1200 cm −1 (e.g. Kahn et al., 2004;2005

Results
AIRS retrievals are presented for ∼2.5 million optically thin, single-layered cirrus clouds over the tropical oceans ±20 • latitude. A total of 29 days are used and are uniformly distributed over the four seasons (separated 48 days apart), starting 6 September 2002 and ending 4 June 2006. These days were selected to sample the entire seasonal cycle, to reduce computational limitations of a larger data set, and because of the constraints on the availability of AIRS V5 retrievals before public release. In Sect. 3.1, histograms of D e and τ are discussed and compared to retrievals from other surfaceand satellite-based platforms. Section 3.2 presents joint histograms of D e , τ, and RH i , and spatial and seasonal variations of RH i c and cloud-clear sky RH i anomalies. Lastly, Sect. 3.3 illustrates the dependence of RH i c on cloud geometrical thickness using coincident CALIPSO data, and the interpretation of the relationship of RH i c to cloud-clear sky RH i anomalies.

Thin cirrus properties
Several important physical properties of thin cirrus are illuminated in Fig. 5. A peak 15 frequency of occurrence is observed near 215-220 K for all values of τ, and a much smaller peak near 190-195 K for τ<0.25 (Fig. 5a). Although the vertical structure of thin cirrus resembles some active-based observations (e.g. Comstock et al., 2002), Fig. 5a shows that AIRS is limited in retrieving very thin cirrus within the TTL, consistent with the small lapse rates of T (z) typical of this region of the atmosphere. The small peak 20 near 190-195 K is dominated by τ≤0.1 and is largely within the range of the retrieval "noise" (Fig. 4). Furthermore, substantial uncertainties in T C for tenuous cirrus (τ≤0.1) are implied by a dispersed vertical profile relative to τ>0.1 (Fig. 5a). Kahn et al. (2007a) show that clouds with f A <0.05 (hence, small τ) may have uncertainties of 50-100 hPa or more that translate to T C uncertainties of 10 s of Kelvins, and a more dispersed

5
(analogous to small τ) shows a maximum of D e near 220-225 K with slight reductions in D e for both warmer and colder T C . For larger bins of IWP, D e increases gradually but consistently with increasing T C . With regard to AIRS, a positive correlation of D e and T C is observed for τ>0.25 and a broad maximum in D e is found for τ≤0.25 (Fig. 5b). However, the magnitude of D e is greater in Stubenrauch et al. (2004)  This is unsurprising given that the retrievals presented herein are limited to thin cirrus. Also, the differences in AIRS and TOVS derived D e likely arise in some part to differences in instrument and algorithm characteristics. In van Zadelhoff et al. (2004) larger values of D e are derived from surface-based lidar-radar measurements compared to AIRS, although much thicker cirrus clouds are included (τ<4) that 15 may preferentially contain larger D e . In Fig. 5c, histograms of D e are shown in order to quantify the scatter within each τ bin in Fig. 5b. Although significant overlap occurs between bins, the individual distributions are distinct and demonstrate a clear increase of D e with increasing τ, consistent with the cited studies above. Large frequencies of 10 µm retrievals for τ<0.1 20 and a significant reduction for the next highest τ bin are consistent with the retrieval noise discussed in Sect. 2.3 but a fraction of these may be valid. Filtering "spurious" from "valid" cirrus retrievals requires at a minimum validated FOV-dependent error estimates, but they are not sufficiently precise at present in operational AIRS products. Additionally, the small frequency peak near 120 µm is consistent with retrieval noise 25 (Sect. 2.3). However, after screening AIRS cloud layers for the likeliest candidates of single-layer thin cirrus, the operational retrieval may not detect all overlapping water and ice cloud FOVs. The "missed" lower layer of water cloud may correspond to a reduction in the magnitude (or reversal of sign) of the spectral radiance slope (Yang et al., 2003), forcing a solution of D e near the upper size bound of 120 µm. Figure 6 summarizes the annual mean spatial distributions of thin cirrus properties. Thin cirrus frequency resembles aspects of the distributions and magnitudes derived from other passive near nadir-viewing instruments (Fig. 6a) including the High-Resolution Infrared Radiation Sounder (HIRS) (Wylie et al., 1999), the Infrared Interfer-5 ometer Spectrometer (IRIS) (Prabhakara et al., 1988), and MODIS (Dessler and Yang, 2003). Areas with lower frequencies of thin cirrus, in particular those within the Inter-Tropical Convergence Zone (ITCZ) and western Pacific warm pool are dominated by thicker clouds (Fig. 6b). Mean fields of Z C reveal a tendency for somewhat higher cloud tops along the ITCZ and within the western Pacific Ocean (Fig. 6c). A small scan-angle 10 dependence in Z C is revealed, an average of about 0.5-1 km higher at ±49 • compared to nadir view. The dependence is negligible with regard to the impact on the interpretation of the joint distributions of T C with other quantities. Annual mean RH i c is 10-20% higher in the tropical Northern Hemisphere (NH) with significant longitudinal variation and a distinct maximum in the N. Indian Ocean basin (Fig. 6d). The calculation and 15 interpretation of RH i c will be discussed in Sects. 3.2 and 3.3.
Corresponding patterns of D e (Fig. 6e) show that low spatial variability is observed except for higher values and variability found in (1) regions corresponding to low cirrus frequency in the southeastern Pacific and south Atlantic Oceans that also contain significant frequencies of overlapping cirrus and stratocumulus, and (2) the central Atlantic 20 and northwestern Indian Ocean basins that contain high loadings of aerosol τ (Fig. 6f). As discussed earlier overlapping clouds may cause high biases in D e ; similarly, thin cirrus overlying a lower layer of dust may cause high biases from changes in the radiance spectrum (Hong et al., 2006). Table 1 shows the differences of averaged D e in the NH and SH with a second set of D e values reported for the NH that are screened 25 for spatial regions with high averages of aerosol τ. Inter-hemispheric differences in D e are shown to be smaller after screening for these regions, and they further illustrate that the thinnest cirrus is most sensitive to underlying aerosol contamination. However, a residual inter-hemispheric difference remains after aerosol screening and may sug- EGU gest consistency with theories regarding impacts of heterogeneous ice nucleation on thin cirrus cloud frequency and D e (Kärcher, 2004). Before a robust conclusion can be drawn, cirrus retrievals must be extended both outside of the tropics and to optically thicker clouds. Retrieval limitations with regard to overlapping dust and cirrus could be misinterpreted as an aerosol indirect effect on cirrus and stresses the importance of 5 trustworthy error estimates and accounting for aerosol effects on cirrus retrievals.

Joint distributions of thin cirrus and RH i c
Coincident RH i c within and near the thin cirrus described in Sect. 3.1 are derived from T (z) and q(z) following Gettelman et al. (2006). Profiles of T (z) and q(z) are generally reliable for f A ≤0.7 (Tobin et al. 2006). However, the highest meaningful altitude of q is 10 closely tied to its magnitude since dry upper tropospheric regions induce radiances in corresponding water vapor sounding channels near radiance noise levels. Gettelman et al. (2004) showed with measurements from the Pre-AVE campaign that the sensitivity limit is around q=10-20 ppmv. In comparisons of spatially and temporally coincident RH i derived from AIRS and the Microwave Limb Sounder (MLS), values of q<30 ppmv 15 may be of questionable quality (E.J. Fetzer, personal communication). A further analysis of RH i between AIRS and MLS by Read et al. (2007) suggests q>20 ppmv is reliable in the tropics. Several important physical relationships between thin cirrus and RH i c are summarized in Fig. 7. Normalized frequency distributions of RH i c tend to be somewhat 20 broader for smaller τ, and peak frequencies are found between RH i c =60-80% with higher peak RH i c related to higher τ (Fig. 7a). The broad distributions of RH i c are suggestive that cirrus is geometrically thinner compared to the vertical resolution of T (z) and q(z) (to be discussed in Sect. 3.3). The RH i c histogram with τ<0.1 closely resembles all-sky distributions derived from TOVS   EGU to in situ measurements, including those from the Inter-hemispheric Differences in Cirrus Properties from Anthropogenic Emissions (INCA) campaign (Ovarlez et al., 2002;Ström et al., 2003;Gayet et al., 2004) that contain much fewer values of RH i c <80% than derived from AIRS. The segment of the histogram with RH i c ≥100 % is generally invariant with τ, but 5 subtle differences in the shapes exist for RH i c ≥120% (Fig. 7a). The observed distribution of supersaturation cannot be reproduced with validated biases and variability in AIRS T (z) and q(z) Gettelman et al., 2006;Tobin et al., 2006). Inspection of Fig. 7a reveals that the frequency of RH i c ≥120% is smaller relative to all-sky conditions (Gettelman et al., 2006). This is consistent with in situ 10 observations (Ovarlez et al., 2002;Gayet et al., 2004) and modeling studies (e.g. Haag and Kärcher, 2003) that demonstrate the existence of upper bounds on RH i c consistent with heterogeneous or homogeneous nucleation or some mixture thereof. Although a well-defined "cut-off" is not observed in Fig. 7a like in Haag and Kärcher (2003), this is unsurprising given the global and tropical nature of the histograms, that the strongest 15 anthropogenic "signal" is expected at higher latitudes (e.g. Ovarlez et al., 2002), and the inherent noise of T (z) and q(z) act to broaden the RH i distribution (Gettelman et al., 2006). The sensitivity of the slope of the RH i c histograms to assumed thresholds of q is quantified in Fig. 7b that shows the percentage of in-cloud q between 15-30 ppmv. 20 No values with q<15 ppmv are included for consistency with the sensitivity limitations suggested by Gettelman et al. (2004). For all values of τ, about 5-15% of in-cloud observations are between 15-30 ppmv. The exception is for RH i c >80% and τ<0.25, where 20-60% of observations are between 15-30 ppmv, the precise value depending on the magnitude of RH i (Fig. 7b). Higher numbers of occurrences of q that approach 25 the sensitivity limits of AIRS are observed for thin and cold cirrus near the base of the TTL. On one hand, this implies that the shape of the distribution of supersaturation within the optically thinnest clouds, hence the ability to discriminate between cirrus nucleation mechanisms, may be dependent on the estimate of AIRS water vapor sen- EGU sitivity. On the other hand, this appears not to be important for cirrus with τ> 0.25. In Fig. 7c, T C decreases with increases in RH i c , consistent with observations (Gettelman et al., 2006). The relationship of D e and RH i c is shown in Fig. 7d. While little dependence of D e on RH i c is observed for the smallest τ bins, it is observed for larger τ, in particular the largest values of D e correspond to RH i c <50%. These 5 results are qualitatively consistent with the precipitation of large ice particles within sub-saturated air layers that have the potential to survive for several hours or more (Kay et al., 2006). In contrast, INCA observations show modest increases in D e that correspond to RH i c >100%   Fig. 6d). There is a very weak increase in D e with decreasing RH i near ∼80% in the INCA observations, qualitatively consistent with Fig. 7d, although this dependence may not be statistically significant. Despite the marginal significance according to the uncertainty estimates in , mutual increases in D e and RH i c are expected because of higher amounts of available water vapor for uptake to ice crystals. However, the INCA observations were sampled in SH and NH mid-latitudes over a more confined dynamic range of T C 15 compared to this work. Furthermore, the observations of Gayet et al. (2004) are not necessarily samples of cirrus with low IWP or τ as viewed by a near nadir-viewing satellite. Given the aforementioned differences in the sampling and observing points of view, it is unsurprising that the characteristics of D e -RH i c relationships differ between AIRS and those of Gayet et al. (2004). Understanding these discrepancies requires 20 AIRS retrievals of thicker cirrus outside of the tropics, and is a topic of future research.
In summary, joint histograms of cirrus quantities and humidity obtained from a single sensor like AIRS allow for studies that are not constrained by temporal and spatial sampling differences inherent in aircraft in situ, surface, and satellite multi-platform analyses. This new capability facilitates a more robust investigation of the physical 25 mechanisms that control the behavior of cloud and humidity distributions.
Global and seasonal distributions of RH i c are presented in Fig. 8. Although the limited number of days leads to some small-scale spatial variability, in general higher RH i c corresponds to greater thin cirrus frequency (Fig. 6a) as in Sandor et al. (2000)  EGU other studies. Furthermore, significant regional, hemispheric, and seasonal asymmetries are observed. The largest seasonal variations occur over the N. Indian and N. E. Pacific Ocean basins, with the largest RH i c observed during JJA (Fig. 8c) Fig. 9. The average RH i c is shown to be almost always greater than the all-sky (clear sky + thin cirrus) RH i average with the difference dependent on the location ( Fig. 9a and b). In-cloud/clear sky RH i anomalies are largest in the fringes of the tropics transitioning to the subtropics, and near the coast of S. America, while the smallest values are observed near the ITCZ, the S. Pacific convergence zone, and throughout the tropical W. Pacific basin. The in-cloud/clear sky RH i differences tend to be largest in dry regions dominated by subsidence in the fringes of the tropics and sub-tropics, but they are smallest in moist regions of frequent convection and upward motion in the deep tropics. 15 Some illustrations of observed regional and inter-annual variability in RH i c are shown in Fig. 10. In general, very small differences in the characteristics of RH i c distributions are observed between the NH, SH, and S. Indian Ocean basins compared to the global average. The exception is that RH i c is on average ∼10% greater over the N. Indian Ocean (Fig. 10c). The inter-annual variations in RH i c for the same region are shown in 20 Fig. 10a and they vary by 5-10% in the 2002-2006 period for RH i c <150%. In contrast, the corresponding yearly variations in RH i c over the S. Indian Ocean are shown to shift around 2-5% for RH i c <150% (Fig. 10b). Although the components contributing to the large variations of RH i c within the N. Indian Ocean basin are uncertain, these regional variations can only be identified from the satellite point of view. Regional differences in 25 cloud/clear sky RH i anomalies (Fig. 9c) are probably due in part to actual differences of RH i c and instrument and/or retrieval sensitivity as a function of geophysical scene type. To partly address this question, the relationship of geometrical cloud thickness (∆Z ci ) to RH i c is addressed in Sect. 3.3. 7,2007 Thin cirrus and relative humidity from AIRS In situ observations of cirrus (e.g. Ovarlez et al., 2002;Ström et al., 2003;Gayet et al., 2004) consistently show that RH i c peaks in frequency around 100%. In Sect. 3.2, AIRS retrievals were shown to vary between 60-80% with higher RH i c correlated to higher τ (Fig. 7a). However, some INCA observations indicate the existence of RH i c <70% 5 within some very tenuous clouds . In that study, the frequency of those observations is much less than comparable ones made with AIRS. Many of the clouds in sub-saturated conditions described in Spichtinger et al. (2004) are very tenuous and often occur at the edge of thicker clouds, and are probably below the AIRS sensitivity limits of detection and characterization. We investigate the hypothesis 10 that ∆Z ci can provide an explanation for the discrepancy in the frequency of low RH i c . The nominal vertical resolution of AIRS-derived RH i is about 2-3 km but a majority of cirrus clouds have smaller values of ∆Z ci (Comstock et al., 2002, their Fig. 8). Thus, for a typical and idealized single-layered cirrus cloud, any given observation of "incloud" RH i is some undetermined blend of clear air (likely to be dry) that is vertically 15 adjacent to the cirrus layer (likely to be moist) resulting in a low bias of inferred RH i c . However, more complicated vertical and horizontal humidity structures most certainly exist, for instance those in highly supersaturated clear sky adjacent to cirrus (Jensen et al., 2005). To quantify the relationship between ∆Z ci and RH i c , observations of cloud vertical structure from the Cloud-Aerosol LIdar with Orthogonal Polarization (CALIOP) 20 (Winker et al., 2007), located on the CALIPSO platform, are collocated to AIRS FOVs to identify a set of single-layered cirrus that extend over the entire AIRS FOV. Results are presented for both the global (±70 • ) and tropical (±20 • ) ocean basins (Fig. 11). Over the global oceans, the mean RH i c increases from 60% to 85% for ∆Z ci increases of 0.2 to 3 km while at higher values RH i c slightly decreases. Over the 25 tropical oceans, the mean RH i c increases from 60% to 75% for ∆Z ci increases of 0.5 to 3 km while at higher values RH i c is generally invariant. The larger overall RH i c over the global oceans is consistent with a greater frequency of ice supersaturation in the ACPD 7, 2007 Thin cirrus and relative humidity from AIRS EGU mid-latitudes compared to the tropics as observed by nadir-viewing satellites Gettelman et al., 2006). Therefore, the mutual increases of RH i c and ∆Z ci demonstrate that low biases of RH i c occur for a majority of thin cirrus. Upon first inspection of Fig. 11, the feasibility of a "correction factor" to simultaneous observations of RH i c and CALIOP-derived ∆Z ci is suggested. Although this may be possible on the regional scale using large sets of observations, the large scatter in RH i c implies this is not the case for individual AIRS retrievals. Further reconciliation of cloud vertical structure with AIRS-derived humidity, for instance quantifying the spatial correlations of RH i c and ∆Z ci to help isolate the existence of other mechanisms that cause regional differences in Figs. 8-10, is a subject of ongoing research.

Summary and discussion
Observations of upper tropospheric microphysical, optical, and bulk cloud properties on the global scale, along with in-cloud and clear sky relative humidity are important for monitoring aspects of Earth's climate, and for evaluating and improving simulations of future climate change. In this work, distributions of thin cirrus optical depth (τ), 15 effective diameter (D e ), and relative humidity with respect to ice (RH i ) within cirrus (RH i c ) over the tropical oceans are simultaneously derived from the Atmospheric Infrared Sounder (AIRS). A rapid radiative transfer model (RTM) that is coupled to a thin cirrus parameterization is used to simultaneously retrieve thin cirrus τ and D e using a least-squares minimization method applied to simulated and observed AIRS spectral 20 radiances between 8-12 µm (Yue et al., 2007). The retrieval approach is applied to ∼2.5 million observations of single-layer thin cirrus over the tropical oceans. The RTM uses available AIRS Version 5 L2 Standard and Support products to represent the atmospheric, surface, and cloudy states needed for the inference of τ and D e . Results of recent AIRS Version 4 and 5 validation studies relevant to the RTM inputs and retrieval 25 approach are summarized. Cloud top temperature (T C ) frequency statistics for thin cirrus reveal a dominant mode EGU that peaks between 215-220 K and is associated with convective detrainment, while a much less dominant mode near 190-195 K is observed near the base of the tropical transition layer (TTL). The vertical cloud structure of thin cirrus derived from AIRS highlights strengths and limitations in using passive infrared radiances to infer thin cirrus quantities near to and within the TTL. Joint increases in D e and T C are shown for cir-5 rus with τ>0.25, consistent with other results based on surface, in situ and satellite retrievals. It is shown that numerous thin cirrus with τ<0.1-0.2 are spurious because of limitations in the RTM inputs. This limitation is quantified by propagating the validated uncertainties in the atmospheric, surface, and cloudy inputs through the RTM and cirrus retrieval. 10 The methodology of Gettelman et al. (2006) is used to derive RH i within (and outside of) cirrus. In-cloud supersaturation is observed for 8-12% of thin cirrus in the tropics and is several factors greater than reported in all-sky conditions (Gettelman et al., 2006). Even higher frequencies of supersaturation occur for the coldest cirrus. The shape of RH i c frequency distributions is shown to depend on τ with a peak frequency 15 located between 60-80%, and higher values of RH i c are associated with larger τ. Comparisons of in situ and AIRS-derived RH i c illustrate AIRS has a dry bias within cirrus, much more so for τ<0.25. Although distributions of in situ and AIRS RH i c have similar characteristics, a consistently broader distribution is observed for AIRS.
We hypothesize that the dry bias is directly a consequence of the geometrical thick-20 ness (∆Z ci ) for a majority of cirrus being less than the vertical resolution of AIRS temperature T (z) and humidity q(z) profiles (∼2-3 km). This is confirmed with coincident cloud vertical structure observed by the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO). Both RH i c and ∆Z C correspondingly increase until the nominal vertical resolution of AIRS is reached. Thus, the narrow extent of ∆Z ci 25 compared to the broad vertical resolution of T (z) and q(z) complicates the interpretation of RH i c . RH i c is systematically biased low because of vertical smoothing over clear (likely dry) and cloudy (likely moist) layers, although real geophysical variability most certainly includes much more complicated vertical and horizontal humidity struc- EGU tures. Furthermore, the frequency of supersaturation in cirrus is mostly insensitive to assumptions about the minimum sensitivity of AIRS water vapor when τ>0.25, thus the shape of RH i c generally appears to be robust. Global distributions of RH i c have significant seasonal, latitudinal, and regional variations that correspond to large-scale dynamical variability, cirrus frequency, and clear 5 sky/in-cloud RH i anomalies. The smallest (largest) anomalies tend to occur in moist (dry) regions associated with regions of vertical ascent (descent). The temporal and spatial differences in correlations between τ, D e , and RH i c may indicate fundamental differences in these joint relationships. In particular, slightly larger values of D e in the NH compared to the SH could suggest consistency with theories regarding impacts of heterogeneous ice nucleation on thin cirrus cloud frequency and D e (Kärcher, 2004). However, retrievals must be extended outside of the tropics and to thicker cirrus, and a thorough error analysis including aerosol effects must be calculated before a robust conclusion can be made. Furthermore, the N. Indian Ocean basin has a ∼10% high bias in RH i c compared to the global average and substantial inter-annual variations of 15 5-10% in average RH i c not seen elsewhere.
The cloud and humidity distributions presented herein demonstrate that AIRS provides a unique perspective on temporal and spatial variations of clouds and water vapor not available from any other aircraft in situ, surface, or contemporary satellite observing platform. Investigations of the physical mechanisms that control the behavior of joint 20 cloud and humidity distributions using a single sensor are not constrained by temporal and spatial sampling differences inherent in multi-platform, surface-based, and in situ aircraft analyses. Although in situ measurements have established fundamental relationships of D e and RH i c , AIRS now provides a global view within a vast variety of geophysical conditions. The joint distributions will be useful towards investigating ACPD 7, 2007 Thin cirrus and relative humidity from AIRS

EGU
Future work must also take into account estimates of uncertainty of joint cirrus and humidity properties at the AIRS FOV-scale. A few general approaches are possible, including (1) the propagation of scene-dependent error estimates through single FOV retrievals, or (2) the replacement of atmospheric, surface, and cloud quantities with other sources of satellite observation. Method (1) will require further advances in AIRS 5 error estimation for practical application. However, method (2) is feasible with the current A-train configuration and operational and research-mode generation of retrieval products. For instance, τ, D e , and T C derived from CALIPSO and MODIS can be used in synergy with AIRS to form a range of uncertainty in the joint distributions to investigate whether they are fundamentally similar or dissimilar regardless of the instrument 10 used. For T (z) and q(z), global NWP models or instruments like MLS can be used with AIRS, although the vertical structure of q(z) from model forecasts are problematic despite reasonable total column water vapor, and MLS only retrieves T (z) and q(z) in the upper troposphere and hiher in altitude. For T S , microwave-derived SSTs may be useful over oceanic regions (e.g. Wentz et al., 2000). Other observing platforms may 15 be used as well.
The estimates of uncertainty will provide more robust conclusions of interhemispheric, regional, and temporal variations in joint cirrus and humidity distributions, and a greater certainty of the physical mechanisms that control their structure. Future work includes extending this analysis to regions outside of the tropics, to thicker cirrus 20 clouds using a radiative transfer model with multiple scattering, and to clouds with more complicated vertical configurations. Swath instruments like AIRS and MODIS provide the ability to use Lagrangian parcel trajectories to track cirrus cloud origin and evolution (e.g. Massie et al., 2002;Mace et al., 2006 Rev., 114, 1167-1199, 1986 Lett., 29, 1813, doi:10.1029/2001GL014440, 2002 (Remer et al., 2005) between 0.1-0.5 are removed.   Fig. 11. Relationship of RH i c to geometrical cirrus cloud thickness derived from collocated AIRS-CALIPSO observations (Kahn et al., 2007c). Only AIRS FOVs with oceanic, homogeneous, single-layered clouds for ±70 • lat (black) and ±20 • lat (gray) are used. Horizontal bars indicate 1σ variability for each 0.5 km cloud geometrical thickness bin.