SEARCH

SEARCH BY CITATION

Keywords:

  • AIRS;
  • humidity profile;
  • upper troposphere

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[1] In this paper we discuss a method to retrieve humidity profiles in the upper troposphere using cloud-affected radiances measured by the Atmospheric Infrared Sounder (AIRS). The magnitude of retrieval error for temperature and water vapor profiles caused by ignoring cloud presence depends strongly on the height of clouds as well as on the cloud fraction in the sensor's field of view. Error analysis based on radiative transfer calculations performed for 155 channels using a set of diverse atmospheric profiles shows that the relative humidity in the upper troposphere at pressure levels between 200 and 400 hPa could be estimated if the cloud height was lower than the height of the 800 hPa pressure level. Applying the channel ranking technique to evaluate the cloud influence in the observed spectra, we examined a high-resolution humidity map of the upper troposphere over land and sea.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[2] Water vapor in the atmosphere plays a significant role in Earth's weather and climate system. In addition to its importance for the direct absorption of radiation, water vapor is crucial for the formation of clouds. Relative humidity with respect to ice (RHi) in the upper troposphere is an important issue for ice-cloud microphysics as well as for radiative processes and forcing, although the amount in this region is very small relative to the total mass of column water vapor. Understanding the distribution of water vapor close to and beyond ice saturation in the upper troposphere will provide useful information for simulating ice clouds in numerical models of the climate system.

[3] The Atmospheric Infrared Sounder (AIRS), a hyperspectral infrared (IR) sensor in operation aboard the Earth Observing System (EOS)-Aqua polar-orbiting satellite [e.g., Aumann et al., 2003], provides measurements of thermal IR spectra. Calibrated radiance (L1B) data are converted to geophysical quantities (L2 product), such as vertical profiles of temperature and water vapor, using a set of retrieval algorithms [Susskind et al., 2003]. For accurate sounding, treatment for the effect of clouds on the observed radiances is crucial. The probability of a clear field of view (FOV) depends on the footprint size of the sensor instrument. In the case of AIRS, previous studies have shown that less than a few percent of the observed FOVs are cloud-free [e.g., Chahine et al., 2006].

[4] Since the great majority of satellite observations are cloud contaminated, some attempts have been made to use cloud-affected spectra for profile estimation. To produce AIRS L2 atmospheric profiles, a cloud-clearing algorithm is used in conjunction with Advanced Microwave Sounding Unit (AMSU) data [e.g., Susskind et al., 2003]. According to the results of validation studies that compared L2 data to radiosonde data, L2 profiles have an accuracy of ∼1 K root mean square (RMS) error in 1-km layers for temperature and ∼20% RMS error for the amount of water vapor in 2-km layers [e.g., Tobin et al., 2006]. From comparison with results from the Microwave Limb Sounder (MLS), AIRS water vapor was estimated to have a precision of ∼25 % in the upper troposphere [Fetzer et al., 2008]. A resent study of AIRS retrievals reported that these data have vertical resolution between 2.5 and 7.1 km for temperature and between 2.7 and 4.3 km for moisture [Maddy and Barnet, 2008].

[5] On the other hand, the horizontal resolution of the L2 profiles is ∼40 km at nadir although the footprint size of AIRS is ∼13.5 km. Because of the resultant large sampling volume, interpretation difficulty arises for the relationship between the relative humidity by L2 profiles and the detailed cloud microphysics [e.g., Gettelman et al., 2006; Kahn et al., 2008a]. The assumption of homogeneous cloud characteristics across the 3 × 3 arrays of AIRS footprints may cause a significant error in the retrieved profiles under some cloud conditions.

[6] Other than the synthetic approach used for the L2 product, the retrieval of profiles for cloudy atmospheres from infrared sounder data is possible in the vertical region over the clouds if we know some cloud parameters. A simulation study based on one-dimensional variational analysis (1D-Var) where the cloud height and cloud fraction in the FOV were dynamically taken into account, reported an improvement of the background RMS error for temperature and humidity profiles [Pavelin et al., 2008]. Although some cloud parameters such as cloud top height and cloud fraction can be estimated from AIRS radiances [e.g., Kahn et al., 2008b], retrieval of semi-transparent cirrus clouds and multi-layer clouds remains difficult.

[7] In this paper, we investigate the influence of clouds on humidity profiles retrieved using AIRS L1B radiances. Using a simple cloud model, we estimated retrieval errors caused by the presence of clouds at fixed altitude in the upper troposphere. A channel-characterization method was used to determine the presence of clouds at the designated altitude. The humidity fields for a sample dataset were then compared to the L2 product, and the results are discussed briefly.

2. Methodology

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[8] To retrieve temperature and water vapor profiles, we applied a fast radiative transfer model (RTM) that uses the correlated k-distribution method [Mano and Ishimoto, 2004]. Simulations of L1B radiances were performed for 155 AIRS channels, which were selected by excluding blacklisted channels from the 324 channels used for the assimilation of AIRS radiances in numerical weather prediction (NWP) (e.g., A. D. Collard, Assimilation of AIRS and IASI at ECMWF, paper presented at Seminar on Recent Developments in the Use of Satellite Observations in NWP, 127–150, European Centre for Medium-Range Weather Forecasts, Reading, U. K., 2007). The modeled atmosphere was divided into 43 pressure layers in the range of 0.05 to 1013.25 hPa, with local thermodynamic equilibrium considered within each layer. The maximum a posteriori solutions for temperature and water vapor profiles, under the assumption of Bayes' theorem, were computed using the Gauss-Newton method [e.g., Rodgers, 2000].

[9] For a priori profiles of temperature and water vapor in the optimal calculations, profiles from Version 5 of AIRS L2 standard products at the nearest location to the L1B data point were applied. Here, we concentrate on relative humidity in the upper troposphere at pressures between 200 and 400 hPa. The selection of this range reflects our basic motivation to improve RHi at the upper part of the troposphere and from a limitation of water vapor retrieval using AIRS radiances. Note that the uppermost pressure level at which AIRS can detect water vapor is approximately 100 hPa and the retrieval of humidity is available at pressures approximately higher than 200 hPa depending on the magnitude of specific humidity and Jacobians for water vapor channels.

[10] Using the retrieved temperature and water vapor profiles, ice relative humidity was calculated from the equations by Murphy and Koop [2005] for atmospheric temperature less than 0°C. A background-error covariance matrix generated from NWP data by the Japan Meteorological Agency (JMA) [Collard, 2007] was substituted for the covariance matrix of the a priori profiles.

[11] The infrared cloud effect was estimated by adding a simple cloud model to the radiative transfer equations. Assuming a cloud with unit cloud-top emissivity and fractional cloud cover N, the upwelling radiance R at the top of the atmosphere can be written as [e.g., Matricardi and Saunders, 1999]

  • equation image

where Rclr and Rcld are clear-column and overcast radiances, respectively, B(T) is Planck's function at temperature T, ɛs is the surface emissivity, and t is the level-to-space transmittance (ts and tcld mean surface-to-space and cloud-to-space, respectively). The notation for dependence on zenith angle and frequency is omitted for simplicity.

[12] Retrieval errors caused by the existence of dense clouds were modeled and derived as follows. Using sample atmospheric profiles for temperature and water vapor, the associated synthetic L1B AIRS radiances are computed by the RTM for clear and overcast atmospheres, assuming a dense cloud with N = 1 at an arbitrary cloud height in equation (1). Then, temperature and water vapor profiles for the synthetic radiances are retrieved by optimal calculations with no cloud assumption. The cloud influence for the retrieved profiles is evaluated by comparison of the resultant profiles of the clear and overcast radiances. For simplicity, we ignore a random noise factor in the radiance calculations.

[13] If we apply the optimal estimation for overcast radiances with no cloud assumption, the resultant temperature and water vapor profiles largely differ from the “true” profiles (or the profiles estimated from clear-column radiances). Furthermore, cloud contamination can cause both positive and negative biases for the derived profiles depending on the cloud height. Figure 1 shows the estimated RMS errors for relative humidity derived by applying 117 European Centre for Medium-Range Weather Forecasts (ECMWF) profiles (the profile dataset was downloaded from Met Office web site) and modeled dense clouds with pressure levels from 400 to 1000 hPa. The satellite zenith angle of 50° over sea was assumed as a typical value for radiance calculations. Results of RMS errors for different satellite zenith angles show the same trend within several percentage differences.

image

Figure 1. RMS error of relative humidity for the modeled cloudy atmosphere with cloud levels from 400 to 1000 hPa. The 117 diverse ECMWF profiles were used for the statistical analysis.

Download figure to PowerPoint

[14] As shown in Figure 1, the retrieval error for the relative humidity profile in the upper troposphere decreases as the cloud height decreases. Assuming that for pressure less than 400 hPa we can accept a RMS error of 10% due to the presence of clouds, then Figure 1 illustrates that this can be obtained only for clouds located at pressure levels higher than 800 hPa.

[15] Since the retrieval errors discussed here are those in cases of dense clouds covering the whole FOV, smaller retrieval errors are expected for semi-transparent clouds or for clouds with an N fraction less than unity with the same cloud heights. Thus, to achieve accuracy better that ∼10% for humidity retrieval between 200 and 400 hPa the condition of a cloud free atmosphere above 800 hPa must be met.

[16] The RMS error results over opaque cloud are basically the same even if we change the emissivity of the underlying surface. Since we applied a cloud model with unit emissivity, the resultant retrieval errors shown in Figure 1 have weak dependence on the surface under the cloud. This means that the retrieved humidity profile at pressures between 200 and 400 hPa is almost the same regardless of the surface difference, such as the difference of emissivity between land and sea. It should be noted that the above conclusion is based on crude cloud approximations in radiative transfer calculations. Multi-layer clouds and their scattering interactions should be taken into account for more exact estimation of the cloud effect.

[17] The cloud-free condition within P < 800 hPa can be inferred from the observed channel radiances (or brightness temperatures) even if no prior information for the clouds (such as L2 cloud products) is available. Here, profile retrieval was carried out after comparing observed channel TBs and those simulated using the a priori profile. As a test of cloud contamination, a “channel-ranking” approach [e.g., McNally and Watts, 2003] was applied. In this technique, the characteristic pressure level Pc by cloud contamination was estimated for each measurement channel. We determined the cloud-free condition over low clouds when the observed TB was the same as the simulated clear-sky TB for channels of Pc < 800 hPa within the threshold TB difference δ.

[18] Assuming a modeled cloud with N = 1 in equation (1) for an arbitrary pressure level, channel radiance R was calculated using the a priori temperature profile and the RTM. The characteristic pressure level for the channel is defined as Pc when the pressure level of the cloud is P = Pc and the calculated overcast radiance of the channel Rcld differs by more than the value α from the clear-sky radiance Rclr. Namely,

  • equation image

In this work, the value α = 0.01 was used [McNally and Watts, 2003], and the characteristic pressure level Pc was determined for 94 channels in the wavenumber range between 651.053 and 948.184 cm−1. The channels of this range were used for the temperature retrieval under the assumption of a constant CO2 volume mixing ratio. Here we assumed that variation in the CO2 ratio is small and the radiance Rclr depends mainly on the temperature profile. Comparing the calculated clear-sky TB and observed TB, the humidity retrieval was carried out when the TB difference was less than the threshold value δ for channels with characteristic pressure level Pc between 200 and 800 hPa. Figure 2 shows the TB (Figure 2, left) and calculated Pc (Figure 2, right) of 94 channels for a sample temperature profile.

image

Figure 2. Calculated brightness temperatures of 94 channels for the (left) sample a priori temperature profile without cloud and (right) characteristic pressure level Pc for each channel derived using equation (2). Humidity retrieval was carried out when the observed TB for the channels with 200 hPa ≤ Pc ≤ 800 hPa (denoted by arrow) were the same as those calculated for cloud-free atmosphere within the difference δ.

Download figure to PowerPoint

[19] The value of δ was estimated from a statistic for the difference between the observed clear-sky TB (clear flag in L1B data was used) and those calculated by our RTM for the a priori profiles (the nearest L2 profiles). Thus, δ was determined for each AIRS “granule dataset,” and the characteristic pressure levels Pc of the channels were estimated for every measurement point. Since a large value of δ leads to a tolerance for cloud contamination, a proper a priori temperature profile is desirable. We used the L2 temperature profile as the a priori profile for this reason. However, the same approach for detecting cloud-affected channels is basically available using other temperature profiles, such as numerical weather prediction data.

3. Humidity Plots for the Upper Troposphere

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[20] Figure 3 shows retrieval results for RHi at pressure levels of 254 hPa (Figure 3a), 322 hPa (Figure 3b), and 397 hPa (Figure 3c). Plots at the same pressure levels of relative humidity interpolated from the L2 product are also presented for comparison (Figures 3d, 3e, and 3f). As a reference for the high-cloud region, Figure 3g shows the observed TB at wavenumber 800 cm−1.

image

Figure 3. A sample of humidity retrieval for AIRS granules taken on 27 Apr 2007 at 04:41 UTC. Retrieved humidity on pressure levels of (a) 254, (b) 322, and (c) 397 hPa. (d)–(f) Interpolations from L2 profiles (data of “highest” and “good” quality for temperature and water vapor are plotted). (g) Observed brightness temperatures at wavenumber 800 cm−1; the bright area indicates dense clouds with high altitudes. The point size is approximately relative to the resolution of the dataset (13.5 km for L1B and 40 km for L2).

Download figure to PowerPoint

[21] The standard deviation of the difference between observed TB and calculated TB for the a priori profiles was approximately 0.5 K for clear-sky L1B data, and three times the standard deviation, δ = 1.5 K, was adapted as the threshold TB difference to allow for random noise. On this sample day, cirrus clouds had widely covered the Japanese islands after passage of a cold front from west to east.

[22] In this result, relative humidity in the upper troposphere between 200 and 400 hPa could be obtained at more than 8000 L1B data points, whereas only 204 points were identified as clear scenes in the L1B spectral analysis. This result suggests that, in this case, a significant portion of the observed data is that with low cloud contamination. The existence of low clouds is also corroborated by MODIS cloud products.

[23] Although our humidity distribution is roughly consistent with that of L2 products, our analysis results show a more detailed humidity structure. Furthermore, in Figure 3b, areas of ice supersaturation are detected in the high-humidity region around 39°N, 132°E, while the plots of the L2 profiles do not show this humidity saturation. The retrieved field is expected to have the same accuracy as the RMS errors shown in Figure 1; however, it is uncertain whether the obtained supersaturation is genuine or not. Validation of the retrieved RHi using independent data (such as humidity data derived by a NWP model) is required.

[24] Even though the vertical range of the retrieval is limited and additional errors due to low clouds may be contained in the results, it is expected that new information on the RHi distribution in the upper troposphere will be provided by the re-analysis of the cloud-affected radiances.

4. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[25] The influence of cloud contamination on relative humidity retrieval was investigated by simulating AIRS L1B using a fast RTM for a set of 117 diverse profiles. Temperature and water vapor profiles were retrieved using AIRS L1B radiances generated for 155 channels. Simulations were carried out for two test cases: in clear-sky conditions and by introducing dense cloud in the FOV. Retrievals were then performed assuming clear-sky radiances. The retrieval error is defined as the difference between the retrieved profiles for cloudy and cloud-free radiances. Results of our model calculations indicate that the retrieval error depends strongly on cloud heights. If dense cloud is located in the upper part of the atmosphere, the error in the humidity profile becomes very large. On the other hand, the retrieval error is relatively small in the upper troposphere if the cloud height is low. A statistical investigation of the model calculations revealed that the relative humidity in the upper troposphere at pressure levels between 200 and 400 hPa can be obtained within ∼10% RMS error if the cloud height is lower than that at the pressure level of 800 hPa. The channel-ranking technique was applied to estimate such a low-cloud condition, and the retrievals of relative humidity were examined for sample granule data. In a comparison to the plots of L2 products, our approach resulted in higher-resolution humidity maps in the upper troposphere over land and sea.

[26] The analysis described here has the potential to detect additional areas of ice supersaturation in the upper troposphere. Such humidity information obtained by an infrared sounder can contribute not only to validation and parameterization of some atmospheric models but also to knowledge of ice-cloud formation. In particular, analysis of the data as matched with other A-Train sensors, MODIS, CloudSat, and CALIPSO data [i.e., Kahn et al., 2008b] will provide new information on ice-cloud formation and its relationship to ice supersaturation.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References

[27] This work was supported by the Ministry of Education, Culture, Sports, Science, and Technology through a grant-in-aid for scientific research (19340132).

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Methodology
  5. 3. Humidity Plots for the Upper Troposphere
  6. 4. Conclusions
  7. Acknowledgments
  8. References