Journal of Geophysical Research: Atmospheres

An improved radiance simulation for hyperspectral infrared remote sensing of Asian dust

Authors


Corresponding author: B.-J. Sohn, School of Earth and Environmental Sciences, Seoul National University, NS80, Seoul 151–747, South Korea. (sohn@snu.ac.kr)

Abstract

[1] The fast Radiative Transfer for Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (RTTOV) (Version 9.3) model was used for simulating the effect of East Asian dust on top of atmosphere radiances. The size distribution of Asian dust was retrieved from nine years of sky radiometer measurements at Dunhunag located in the east of Taklimakan desert of China. The default surface emissivity in RTTOV was replaced by the geographically and monthly varying data from University of Wisconsin (UW)/Cooperative Institute for Meteorological Satellite Studies (CIMSS) infrared surface spectral emissivities. For a given size distribution and surface emissivity, the effects of three refractive indices of Optical Properties of Aerosols and Clouds (OPAC) mineral aerosol, dust-like aerosol by Volz, and High Resolution Transmission (HITRAN) quartz were examined. Results indicate that the specification of surface emissivity using geographically and monthly varying UW/CIMSS data significantly improved the performance of the simulation of AIRS brightness temperature (TB) difference (BTD) between window channels, in comparison to the results from the use of default emissivity value of 0.98 in the RTTOV model, i.e., increase of the correlation coefficient from 0.1 to 0.83 for BTD between 8.9 μm and 11 μm, and from 0.31 to 0.61 for BTD between 3.8 μm and 11 μm. On the other hand, the use of Asian dust size distributions contributed to a general reduction of radiance biases over dust-sensitive window bands. A further improvement of the TB simulations has been made by considering the Volz refractive index, suggesting that hyperspectral infrared remote sensing of Asian dust can be improved using the proper optical properties of the dust and surface emissivity.

1. Introduction

[2] Dust originating in dry and arid regions, such as the Taklimakan and Gobi deserts, are frequently transported over vast regions in East Asia and on to North America, crossing the North Pacific, particularly during springtime [Duce et al., 1980; Chun et al., 2001; Husar et al., 2001; Kim, 2008]. This Asian dust can cause health hazards, interfere with transportation systems, and incur large costs related to cleaning city infrastructure and roads. Moreover, Asian dust particles, which can lead to both cooling by scattering sunlight back to space and warming by absorbing solar and infrared (IR) radiation, can significantly perturb the atmospheric radiation budget [Sohn et al., 2007; Costa et al., 2006; Kim et al., 2005; D.-H. Kim et al., 2004]. Unfortunately, in recent decades more frequent dust episodes have been reported in downwind regions, including Korea and Japan [Lee and Sohn, 2011], probably caused by decreased precipitation and reduced vegetation after the mid-1990s in eastern Mongolia [Park and Sohn, 2010]. Accordingly, it is natural to ask how well we monitor the extent and intensity of dust storms, knowledge of which can be effectively used for reducing social and economic losses caused by dust outbreaks.

[3] Since the advent of satellite technology, we have been able to routinely monitor the dust area and loading, mainly using ultraviolet and visible radiance measurements. On the other hand, previous studies have demonstrated a possible use of IR measurements for remote sensing of dust [Ackerman, 1997; Legrand et al., 2001]. As shown by the recent success of aerosol retrievals using IR measurements [DeSouza-Machado et al., 2010; Peyridieu et al., 2010; Z. Yao et al., Asian dust height and infrared optical depth retrievals from hyperspectral longwave infrared radiances, submitted to Journal of Geophysical Research, 2012], it may be beneficial to monitor dust-related features even during the nighttime [e.g., Lee and Sohn, 2012]. Considering that the retrieval of aerosol properties from satellite measurements requires an understanding of the underlying radiative transfer, a better description of dust optical properties and accurate radiative transfer modeling are a prerequisite for aerosol retrieval.

[4] The calculation of radiative transfer with aerosols is generally based on the Lorentz-Mie scattering calculation with prescribed aerosol optical properties. However, unlike in the visible spectral range for which abundant measurements of optical properties are available [Patterson and Gillette, 1977; Levin et al., 1980; Hess et al., 1998; Dubovik et al., 2002; Myhre et al., 2003; Aoki et al., 2005], relatively few observations of IR dust optical properties are available in the literature [e.g., Volz, 1972, 1973; Hess et al., 1998]. Those observed refractive indices for dust have been routinely used for Asian dust studies [Gu et al., 2003; S.-W. Kim et al., 2004; Aoki et al., 2005; P. Zhang et al., 2006; D. F. Zhang et al., 2009]. Considering that the optical properties of dust depend on the mineralogical composition and thus on the dust source regions [Sokolik et al., 1998; Darmenov and Sokolik, 2005], it is of interest to examine how well the currently available optical properties can support our capability of radiative transfer modeling to simulate IR features associated with Asian dust. This type of research is now possible because finer IR spectral signatures of the dust are routinely measured on satellite platforms, such as the Atmospheric Infrared Sounder (AIRS) and the Infrared Atmospheric Sounding Interferometer (IASI); furthermore, the vertical structures of the dust layer can be probed by satellite-borne active sensors, such as the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO).

[5] In this study, we aim to understand the impact of various aerosol properties and surface emissivity on the IR spectral features. This effort consists of two steps: (1) improvement of IR radiative transfer modeling with optical and physical properties for Asian dust from sky radiometer measurements, various refractive indices, and the use of spatially and temporally varying surface spectral emissivity in a radiative transfer model, and (2) comparison of the simulations with hyperspectral measurements (here with AIRS radiances) for validating the results. Even though we only focus on AIRS in this paper because of the availability of collocated CALIPSO data used for the radiance simulation, the obtained results should also be valid for IASI.

2. Radiative Transfer Modeling

[6] In order to examine the performance of radiative transfer (RT) models for simulating the effects of Asian dust, we use aerosol size distributions obtained from ground-based sky radiometer measurements at Dunhunag located in the east of Taklimakan desert of China, in conjunction with three different IR refractive indices for the mineral dust. The influence of surface emissivity on satellite measurements is also tested with monthly land surface emissivity data. Details of those inputs used for the modeling are found in sections 2.2 and 2.3.

2.1. Radiative Transfer Model

[7] The fast Radiative Transfer for Television Infrared Observation Satellite (TIROS) Operational Vertical Sounder (RTTOV) (Version 9.3) model is used for simulating top-of-atmosphere (TOA) radiances. The RTTOV model was designed for fast calculation of radiances measured by space-borne IR and microwave sensors viewing the atmosphere and surface [Saunders et al., 1999, 2010]. For the simulation of TOA radiances including aerosols, it employs 10 aerosol components from the Optical Properties of Aerosols and Clouds (OPAC) [Hess et al., 1998] database plus a volcanic ash component (total 11 components) to allow mixing of the various aerosol components. Any specific aerosol is thus described through a mixture of the necessary species from these 11 components.

[8] The OPAC database describes four components of mineral aerosols, nucleus mode, accumulation mode, coarse mode, and mineral-transport mode. Those four mineral aerosol components are assumed to have the same refractive index, but different size distributions [Hess et al., 1998]. The refractive index of mineral aerosol is from various sources [i.e., Volz, 1973; Patterson and Gillette, 1977; Levin et al., 1980], and quartz absorption features [Schütz, 1979]. The size distribution parameters from d'Almeida [1989] and Schütz [1979] are used to describe aerosols over deserts [Köpke et al., 1997]. Those refractive indices and size distributions are mainly from observations over North African deserts, and partly from arid areas of the U.S. and the Middle East. Considering that mineralogical compositions and their optical properties exhibit a strong regional dependency [Claquin et al., 1999; Darmenov and Sokolik, 2005], it may be worthwhile to test if they are suitable (or which is more suitable) for better describing the radiative effect pertinent to Asian dust.

[9] TOA radiances except strong gas absorption bands should be sensitive to the surface emissivity because of the transmission of surface-emitted radiance to the TOA radiance through the transparent or semi-transparent atmosphere. Moreover Asian dust sources are located in Chinese and Mongolian arid regions. The dust is transported across large distances, over widely varying surface types with different emissivity properties. Because the default configuration of RTTOV assumes the emissivity over land to be a fixed value of 0.98, we used a land surface emissivity atlas (see section 2.3) to account for the spectral, spatial, and temporal variability of this parameter in our simulations. For sea surfaces, we used the RTTOV default routine, Infrared Surface Emissivity Model (ISEM)-6, which calculates the ocean surface emissivity analytically as a function of the wavelength.

2.2. Dust Modeling and Improved Lorentz-Mie Calculation

[10] The scheme implemented in RTTOV to parameterize aerosol scattering is based on methodology of Chou et al. [1999]. This scheme (referred to as the scaling approximation) approximates the effect of scattering by scaling the optical depth by a factor derived by including the backward scattering in the emission of a layer and in the transmission between levels. This parameterization relies on the hypothesis that the diffuse radiance field is isotropic and can be approximated by the Planck function which allows the radiative transfer equation to be formulated in the same way as for clear sky conditions. In the scaling approximation the contribution of the thermal diffuse scattered radiation is simulated by replacing in the radiative transfer equation the absorption optical depth, τa, with an effective extinction optical depth, τe, defined as:

display math

where τs is the scattering optical depth and b is the integrated fraction of energy scattered backward for incident radiation from above or below. More details about the actual implementation in RTTOV are given in Matricardi [2005].

[11] It has been known that IR extinction process caused by dust aerosols is better explained by the inclusion of nonspherocity of dust particles in the Mie calculation [e.g., Hudson et al., 2008; Kleiber et al., 2009]. However, because its non-spherical effect is small over IR spectral range [Yang et al., 2007], we generate the necessary parameters (i.e., absorption and scattering coefficients, single-scattering albedo, and phase function) for the assumed spherical dust particles using the Lorentz-Mie theory. The size distribution for Asian dust is obtained from sky radiometer measurements (Figure 1). Three different refractive indices for mineral aerosol [i.e., mineral component from OPAC database, dust-like component based on the measurements of Volz [1972, 1973] [Shettle and Fenn, 1979], and quartz from the High Resolution Transmission (HITRAN) database [Rothman et al., 2009] (hereafter referred to as OPAC, Volz, and HITRAN, respectively)] are used for examining the impact of different refractive indices on TOA radiances. The backscattering parameter, b, in equation (1) is then calculated from the phase function obtained using a code provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) (M. Matricardi, ECMWF, personal communication, 2009). Optical parameters obtained from the above calculations are then included in RTTOV for the dust affected radiance simulations.

Figure 1.

(a) Size distributions of Asian dust retrieved from sky radiometer measurements at Dunhuang, China (thin solid lines), and the mean size distribution (white dashed line). (b) Mean size distribution from the sky radiometer measurements (thick dashed line), three modes for mineral dust components from OPAC (nucleus mode: Nuc, accumulation mode: Acc, and coarse mode: Coa), and the best fit for the mean size distribution (thick solid line).

2.2.1. Size Distribution of Asian Dust

[12] Sky radiometers (e.g., POM-01; Prede Co. Ltd.) measure direct and diffuse solar radiation at seven wavelengths of 315, 400, 500, 675, 870, 940, and 1020 nm, and then aerosol optical thickness (AOT), single-scattering albedo, Ångström exponent, and volume size distribution are retrieved from the measurements by applying an inversion technique developed by Nakajima et al. [1996]. Size distributions for Asian dust retrieved from sky radiometer measurements at Dunhuang, China, are used for the Lorentz-Mie calculation. The Dunhuang site (40.146°N, 94.799°E) is located at the eastern edge of the Taklimakan desert where Asian dust emission takes place all year, and thus the obtained optical properties at this site should represent the properties of Asian dust [e.g., D.-H. Kim et al., 2004]. Measured data are obtained from the sky radiometer network (SKYNET) data center in Chiba University of Japan (available at http://atmos.cr.chiba-u.ac.jp/). The size distribution representing Asian dust is obtained by taking an average of all available dust measurements over a nine-year period (October 1998–January 2007) selected by applying pre-determined dust indices (i.e., AOT at 550 nm ≥ 0.5, Ångström exponent ≤ 0.3 [Lee et al., 2002; Song et al., 2012]). Retrieved size distributions are given in Figure 1a in terms of number density versus particle radius.

[13] Since the sky-radiometer-derived size distribution for the Asian dust also includes contributions by other types of aerosol, we combine three modes for mineral aerosol components (i.e., nucleus, accumulation, and coarse modes) from the OPAC database [Hess et al., 1998] to best fit the mean size distribution from the sky radiometer measurements. The number mixing ratios for nucleus, accumulation, and coarse modes are 0.862, 0.136, and 0.217 × 10−2, respectively. The best fit result (thick solid line) is given in Figure 1b along with the three OPAC modes centered at a radius of 0.07, 0.39, and 1.90 μm, respectively. It is shown that the measured distribution is well fitted with three modes in the case of particles larger than 0.1 μm in radius, while particles smaller than 0.1 μm noted in the sky radiometer-retrieved distribution are likely erroneous. The large variations shown by particles smaller than 0.1 μm in Figure 1a strongly suggest that sky radiometer measurements are not suitable for retrieving optical properties for particles smaller than 0.1 μm (so-called nanoparticles). In this study, the size distribution depicted by the best fit curve is used for generating the optical properties of Asian dust.

2.2.2. Refractive Indices of Dust

[14] Refractive index is an important parameter that controls the optical properties of dust particles and, thus, the radiative signature of dust. We employ and test three different refractive indices (i.e.: OPAC, Volz, and HITRAN). The refractive indices of OPAC and Volz have been commonly used to simulate dust impacts or retrieve dust properties from measured IR spectra [Sokolik, 2002; DeSouza-Machado et al., 2010; Peyridieu et al., 2010]. The refractive index of quartz from HITRAN is also tested because quartz is one of the major components of Asian dust [Jeong, 2008]. The real and imaginary parts of the three refractive indices are given in Figure 2. Generally, larger absorption can be expected from OPAC in comparison to Volz over most of the IR spectrum. A more fluctuating feature is obvious for HITRAN, suggesting very large absorption near 9 μm and a smaller absorption near 12.5 μm. OPAC depicts absorption peaks near 9 and 12.5 μm, similar to HITRAN but with a flat distribution.

Figure 2.

(a) Real and (b) imaginary parts of the refractive indices for OPAC mineral aerosol component (solid line), Volz dust-like aerosol (dashed-dot line), and HITRAN quartz (dashed line). The vertical dashed lines depict the spectral locations of four channels (ch 770; 11 μm, ch 565; 12 μm, ch 1247; 8.9 μm, ch 2330; 3.8 μm) used for the BTD calculation.

2.2.3. Extinction Coefficients

[15] The IR extinction coefficients are calculated with the three different refractive indices for the Asian dust size distribution given in Figure 1b. In Figure 3, the extinction coefficient of OPAC shows a minimum value of 1.7 × 10−4 km−1 near 8 μm, and a maximum of 4.9 × 10−4 km−1 at around 9.5 μm. HITRAN appear to be quite similar to OPAC, except for spectral regions around 7–9 μm and near 12.5 μm, where HITRAN shows distinct peaks. The extinction coefficient of Volz seems to follow a similar pattern to OPAC, although the magnitudes are generally smaller than the others, showing values between 1.2 × 10−4 and 3.5 × 10−4 km−1 over the IR spectral region. Volz shows a relatively smoother distribution of the extinction efficient. It is expected that the simulated spectrum using HITRAN may show significantly different features from OPAC or Volz because of the strong extinction coefficient near 8.5 μm.

Figure 3.

Extinction coefficients for the mineral aerosol component of OPAC (solid line), dust-like aerosol of Volz (dashed-dot line), and HITRAN quartz (dashed line) over the IR region. The coefficients are calculated with the given Asian dust size distribution of Figure 1.

2.3. Surface Emissivity

[16] In this study we specify the surface emissivity using the global IR surface emissivity data produced by UW/CIMSS [Borbas et al., 2007; Seemann et al., 2008] (available at http://cimss.ssec.wisc.edu/iremis/), which were derived from the baseline fit method and principal component analysis regression. A conceptual model for a baseline fit is developed on the basis of high spectral resolution laboratory measurements (i.e., the Moderate Resolution Imaging Spectroradiometer (MODIS)/University of California, Santa Barbara (UCSB) (http://www.icess.ucsb.edu/modis/EMIS/html/em.html) and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) emissivity libraries [Salisbury and D'Aria, 1994; Baldridge et al., 2009]) with 10 hinge-point wavelengths (3.6, 4.3, 5.0, 5.8, 7.6, 8.3, 9.3, 10.8, 12.1, and 14.3 μm) to capture the spectral shape between 3.6 and 15 μm [Seemann et al., 2008]. We use the land surface emissivity produced each month with a 0.05° spatial resolution at 416 wave numbers over the IR spectral range, allowing for climatological seasonal variations of the surface emissivity [Borbas et al., 2007]. The latest version of RTTOV (Version 10.1) includes this global IR emissivity atlas as part of the package. Over the ocean, we use the emissivity computed from the ISEM-6 [Sherlock, 1999] which is part of RTTOV.

3. AIRS Spectral Radiance Simulations

[17] In order to test the performance of the RT model, we simulate AIRS TOA radiances using RTTOV with a specification of the dust vertical profile using CALIPSO aerosol products. CALIPSO carries a two-wavelength polarization lidar, the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), and it provides high resolution vertical profiles of clouds and aerosols [Winker et al., 2007]. The level 2 aerosol layer product used in this study contains the top and base altitude of the dust layer, AOT, and feature classification flag of each detected aerosol layer. Dust pixels and associated properties are obtained only if the CALIPSO pixel is determined to be dust-loading and CALIPSO AOT at 532 nm (hereinafter AOT532) is greater than 0.1. This is because AOT532 < 0.1 is considered to be dominant in background aerosol. Dust profiles are discarded if CALIPSO level 2 data indicate cloud contamination.

[18] Atmospheric temperature and moisture profiles are also used as inputs to the radiative transfer model. Six-hourly (e.g., 0000, 0600, 1200, and 1800 UTC) temperature, moisture, and ozone profiles are obtained from the ECMWF Reanalysis (ERA)-Interim data, which are given at fixed 37 pressure levels (1000, 975, 950, 925, 900, 875, 850, 825, 800, 775, 750, 700, 650, 600, 550, 500, 450, 400, 350, 300, 250, 225, 200, 175, 150, 125, 100, 70, 50, 30, 20, 10, 7, 5, 3, 2 and 1 hPa) in a 1.5° × 1.5° grid format [Dee et al., 2011]. Surface variables (skin temperature, surface temperature, surface dew point temperature, and surface pressure) are also extracted from ERA-Interim data.

[19] For the simulation, AIRS observations are collocated with the vertical profile of dust from CALIPSO measurements, and with surface and atmospheric conditions of ERA-Interim data. In doing so, ERA-Interim data are interpolated temporally into the AIRS observation timeline using the two closest analysis times, and spatially to the pixel location in interest using surrounding data points. ECMWF surface pressure is used for defining the vertical extent of the atmospheric column at a given geographical location.

[20] For the simulation, AIRS observations are collocated with the vertical profile of dust from CALIPSO measurements, and with surface and atmospheric conditions of ERA-Interim data. In doing so, ERA-Interim data are interpolated temporally into the AIRS observation timeline using the two closest analysis times, and spatially to the pixel location of interest using surrounding data points. Since CALIPSO and Aqua satellites show about a two minute time difference in the overpass, both measurements at the nadir view are considered to be collocated. However, because the spatial resolutions of AIRS and CALIPSO are 13.5 km and 5 km at nadir, respectively, all CALIPSO aerosol measurements within each AIRS pixel are averaged to construct collocated CALIPSO and AIRS measurements. Total 16,387 AIRS - CALIPSO collocated pairs were constructed over the two-year period (January 2008–December 2009) in the region 15°N–55°N, 70°E–150°E (i.e., over East Asia). CALIPSO dust information and collocated ERA-Interim temperature, moisture, and ozone profiles, and surface variables are used as inputs to the RTTOV model. Only nighttime simulations are performed, because diurnal variations of surface variables, such as the skin temperature, are much larger during the daytime and thus it is more difficult to resolve the expected diurnal variation during the daytime by simple interpolation.

4. Comparisons With Observations

[21] We first simulated AIRS radiances using the RTTOV v9.3 model without any modifications, employing the OPAC optical parameters of dust described in section 2.1. In this step, the mineral coarse mode is selected and a fixed land surface emissivity (0.98) is used. The results are taken as a reference for comparison with results from the improved simulations developed in this study. The AIRS radiances affected by dust are simulated using ERA-Interim surface and atmospheric variables and CALIPSO-derived AOT532 with dust top and bottom heights as inputs to RTTOV. The simulations are performed for the total 16,387 AIRS field-of-views, which are collated with CALIPSO measurements.

[22] Comparisons are made between simulated and observed brightness temperatures (TBs) in the scattergrams given in Figure 4 with associated statistics (correlation coefficient, mean bias, and root mean square error). AIRS has over two thousand channels [Chahine et al., 2006]; however, because the window channels are much more sensitive to dust than other gaseous absorption channels, we chose only four window channels (channel numbers 770, 565, 1247, and 2330 corresponding to 11, 12, 8.9, and 3.8 μm spectral bands, respectively) for the scattergram comparison. The TBs at those channels are referred to as TB11, TB12, TB8.9, and TB3.8, respectively. The dashed line in each diagram represents a best fit regression line.

Figure 4.

Scatterplots of simulated and measured AIRS TBs for (a) 11 μm (ch 770), (b) 12 μm (ch 565), (c) 8.9 μm (ch 1247), and (d) 3.8 μm (ch 2330) with associated statistics. The TBs are simulated with OPAC mineral aerosol refractive index and with OPAC mineral coarse mode size distribution. Dashed lines represent best fit regression lines.

[23] In Figure 4, all simulations of the four window channels demonstrate good agreement with observations, as shown by high correlation coefficients around 0.97. Regression equations suggest a similar pattern for all 4 channel simulations. However, mean biases for TB8.9 and TB3.8 appear to be larger than for TB11 and TB12; mean biases for the TB11, TB12, TB8.9, and TB3.8 channels are 0.54, 0.05, 1.5, and 1.9 K, respectively. The root mean square errors (RMSEs) of the simulations are in the order of 3–4 K, although TB8.9 and TB3.8 show slightly larger RMSEs.

[24] Although there is a good agreement of the simulations with observations as shown in Figure 4, it is still not clear whether the influence of the dust is properly simulated. The TB difference (BTD) using the 11, 12, 8.5, and 3.8 μm channel TBs is commonly used for detecting the presence of mineral dust in the atmosphere [Ackerman, 1989, 1997; Darmenov and Sokolik, 2005]. In order to examine whether the modeling approach adopted in this study is capable of simulating the dust effect accurately, we compare the simulated BTDs with AIRS measured values, i.e.: TB11–TB12 (BTD11–12), TB8.9–TB11 (BTD8.9–11), and TB3.8–TB11 (BTD3.8–11).

[25] Scatterplots of simulated versus measured BTDs are given in Figure 5. The BTD11–12 shows a weak linear relationship with correlation coefficient, mean bias, and RMSE of 0.67, 0.49 K, and 0.86 K, respectively (Figure 5a). The simulated BTD8.9–11 (Figure 5b) displays a much lower sensitivity with a wider spread extending to about −10 K of the measured BTD8.9–11. The comparison between observed TB11 and TB8.9 indicates that observed TB8.9 is always lower than observed TB11 and the low bias of TB8.9 seems to be larger for higher TB11 (not shown). On the other hand, a well-matched one-to-one correspondence is noted in the simulated results (not shown). Thus, the large discrepancy shown in BTD8.9–11 (Figure 5b) is likely attributable to the use of a fixed surface emissivity in the simulation.

Figure 5.

Scatterplots for simulated and measured (a) BTD11–12, (b) BTD8.9–11, and (c) BTD3.8–11. The TBs are simulated with OPAC mineral aerosol refractive index and with OPAC mineral coarse mode size distribution. The default land surface emissivity of 0.98 is used.

[26] The simulated BTD3.8–11 are also compared with observations (Figure 5c). In general, simulated BTDs show a relatively large positive bias of 1.36 K in comparison to the corresponding observed values. The correlation coefficient of 0.31 is low and the RMSE of 2.02 K is relatively large, suggesting that the RTTOV model may not reproduce the dust features observed in the combination of the 11 and 3.8 μm channels.

[27] Overall, as noted in the comparison of BTDs, the default dust optical properties and fixed land surface emissivity used in RTTOV should not be adequate for simulating Asian dust signals in IR measurements, especially for brightness temperature differences. Poor agreement of the simulated BTDs for the 8.9 and 3.8 μm channels suggest that the dust optical properties or land surface emissivity near 8.9 and 3.8 μm used in the RTTOV as default are likely to be unrepresentative of real situations, and hence, the default values should be updated with more realistic values.

[28] To examine the impact of the land emissivity on the simulation, the default value of 0.98 in RTTOV model is replaced by temporally and spatially varying UW/CIMSS surface emissivity data. The simulated BTDs are plotted against measurements in Figure 6. In the scatterplots, better agreement of the simulations with AIRS observations is clear, especially for BTD8.9–11 and BTD3.8–11. For example, the almost random pattern of simulated BTD8.9–11 shown in Figure 5b now shows a linear relationship, with a correlation coefficient of 0.83. Better agreement is also visible in BTD3.8–11 as shown by the increased correlation coefficient from 0.31 to 0.61. The improvement of BTD11–12 is less as shown by only slightly improved statistics.

Figure 6.

Same as Figure 5 but for the use of UW/CIMSS surface emissivity data.

[29] The improvement of BTD8.9–11 and BTD3.8–11, demonstrates that specification of more accurate surface emissivity is crucial for improving simulations of dust-affected IR radiances, especially near the 8.9 and 3.8 μm spectral bands. This improvement is particularly important because arid regions have relatively low and variable surface emissivity in the 3.5–4.5 and 8–10 μm bands [Salisbury and D'Aria, 1992, 1994], and these are the bands we often use for monitoring dust storms. In the previous studies, dust has been monitored by using the BTD between two split window channels [Li et al., 2007] or using three window channels [Zhao et al., 2010]. Efforts have been made to relate the BTD to a quantitative measure of dust intensity [e.g., Zhang et al., 2006]. However, results obtained in this study strongly suggest that a robust simple BDT criterion may not be adequate because of strong influences of geographically and temporally varying emissivity on the TOA TB.

[30] To test the impact of the assumed dust refractive index on the AIRS TB simulations the size distributions for Asian dust retrieved from the sky radiometer measurements (section 2.2.1) are used for the simulations, together with OPAC and Volz refractive indices for mineral dust and the HITRAN refractive index for quartz given in Figure 2. For comparison, coarse mode and mineral-transported distributions are also included. While the coarse mode has a mode radius of 1.9 μm, the size distribution for the transported component has a mode radius of 0.5 μm [Hess et al., 1998]. The three simulations are based on a combination of the OPAC refractive index for mineral dust with three different size distributions (i.e., Asian dust, transported mode, and coarse mode). In the simulations, moderate to strong dust cases showing collocated CALIPSO AOT532 > 0.5 are chosen in order to reduce the uncertainties arising from surface variables. Of total 16,387 AIRS-CALIPSO collocated data, 2,350 cases satisfying the criteria remain. Mean biases from the observations are presented over the two spectral regions in Figure 7 after taking the average of 2350 simulations, and associated RMSEs for the 4 window channels (11, 12, 8.9, and 3.8 μm) are given in Table 1. Also plotted in Figure 7 are results from the use of different refractive indices for the given Asian dust size distribution.

Figure 7.

Mean bias (simulated TB minus measured TB) for dust with CALIPSO AOT532 > 0.5 over spectral ranges (a) 750–1150 cm−1 and (b) 2400–2700 cm−1. The TBs are simulated with three refractive indices (OPAC: OPAC mineral aerosol component; Volz: Volz dust-like; HITRAN: quartz from HITRAN database) and with different dust size distributions (AD: Asian dust size distribution; TR: mineral-transported in OPAC; CO: mineral coarse mode in OPAC). Surface emissivities over the land are specified by UW/CIMSS surface emissivity data. The vertical red lines indicate the four channels (ch 770; 11 μm, ch 565; 12 μm, ch 1247; 8.9 μm, ch 2330; 3.8 μm) used for the BTD comparison.

Table 1. RMSEs of the Simulated Versus Measured TBs for AIRS-CALIPOS Dust Pixels if CALIPSO AOT532 > 0.5a
 11 μm (ch 770)12 μm (ch 565)8.9 μm (ch 1247)3.8 μm (ch 2330)
  • a

    Total of 2350 cases. TBs are simulated with three refractive indices (OPAC: OPAC mineral aerosol component; Volz: Volz dust-like; HITRAN: quartz from HITRAN database) and with different dust size distributions (AD: Asian dust size distribution; TR: mineral-transported in OPAC; CO: mineral coarse mode in OPAC).

HITRAN-AD2.29 K2.50 K3.21 K2.48 K
Volz-AD2.40 K2.31 K2.36 K2.47 K
OPAC-AD2.67 K2.68 K2.72 K2.62 K
OPAC-TR2.75 K2.81 K2.89 K2.70 K
OPAC-CO3.63 K3.52 K3.49 K2.96 K

[31] It is shown that the size distribution from the coarse mode (OPAC-CO in Figure 7) only results in the largest underestimate, of up to −1.5 K over most of the window region in the thermal IR band. Use of the OPAC transported mode and Asian dust distributions (i.e., OPAC-TR and OPAC-AD) give rise to significantly reduced TB differences by up to −1 K. It is also clear that a noticeable improvement has been made by use of the Asian dust model throughout the thermal IR window region of interest (Figure 7a).

[32] The impact of the three refractive indices for the Asian dust size distribution on the AIRS TB simulations shows that use of the Volz refractive index for mineral dust provides the best agreement with observations over the thermal window region. Note that the refractive index from the HITRAN quartz does not reduce the bias over the 750–870 cm−1 region, probably because of the peak in the extinction coefficient of the HITRAN quartz over the same wavelength region of Figure 3. RMSEs calculated from 2,350 cases at 4 channels show that seasonal and geographical fluctuations of the bias from the AIRS measurements are in the order of 2–3 K (Table 1). Consistent with the mean bias result, the combination of Volz refractive index with AD distribution (Volz-AD) yielded smallest RMSEs (2.3–2.5 K) in general.

[33] In contrast to the significant radiance differences in the longwave IR when particle size distribution and refractive index are varied, the shortwave IR window region (2400–2700 cm−1) in Figure 7b shows less sensitivity to both size distribution and refractive index. Different from smaller biases shown in the shortwave IR region the relative errors are much larger (2–4% of radiance) because of the smaller emitted radiant intensity over those bands (not shown). Larger relative errors suggest the difficulty of using shortwave IR channel measurements to monitor the dust features.

[34] One might concern about influences of gas absorption due to uncertain input data (particularly due to uncertainties in water vapor profile). Sensitivity tests were conducted by allowing ±10% of uncertainty in water vapor profiles. Results indicate that the uncertainly range caused by ±10% error in the water vapor profile is within at most 0.1 K except some strong water vapor absorption bands (not shown). Since BTDs are between brightness temperatures at weak water vapor absorption bands, and CO2 and O3 absorptions should be much smaller there, obtained results from Figure 7 and others should be valid even if uncertainties in gas absorption exist.

[35] By using the Asian dust size distribution, the Volz refractive index for the mineral dust, and the UW/CIMSS surface emissivity, three BTD distributions are calculated to show the improvement in simulations when compared to Figures 5 and 6. The results are given in Figure 8. All three BTDs show improved accuracy compared to those shown in Figures 5 and 6. It is concluded that among the available refractive indices for the dust, the use of the Volz refractive index gives the best performance when combined with the Asian dust size distribution and UW/CIMSS surface emissivity.

Figure 8.

Same as Figure 5 but for the use of the Volz dust-like refractive index, the given Asian dust size distribution, and the UW/CIMSS surface emissivity data.

5. Summary and Conclusions

[36] In this study, we mainly focused on improving the accuracy of simulated AIRS radiances for Asian dust conditions. To facilitate this, the size distributions for Asian dust were estimated from multiyear sky radiometer measurements at a site (Dunhuang) in the east of Chinese Taklimakan desert. Spectral surface emissivities were specified with the UW/CIMSS emissivity atlas in order to better describe the surface contribution to the TOA radiance. For a given size distribution and specified surface emissivity over the East Asian region, the impact of three different IR refractive indices for the dust (OPAC mineral aerosol, dust-like aerosol observed by Volz, and quartz from HITRAN) on the simulated AIRS radiances has been studied for the dust conditions prescribed by CALIPSO retrievals. The simulations were compared to measured AIRS TBs in the IR window regions.

[37] The results indicate that the specification of surface emissivity using geographically and monthly varying UW/CIMSS data significantly improved the accuracy of simulated IR TBs. This is important as BTDs are frequently used for monitoring dust outbreaks. Hence an accurate description of land surface emissivity is a prerequisite for remote sensing of dust from IR spectral measurements. It was also found that Asian dust can be more accurately simulated by using the Volz refractive index combined with a retrieved Asian dust size distribution with the UW/CIMSS surface emissivity data. It is suggested that radiative transfer modeling with the best estimate of the optical properties of the dust and surface emissivity is necessary for the optimal retrieval of dust properties using hyperspectral IR measurements from space.

[38] While UW/CIMSS data were used to improve the simulations, it is of much interest to examine how well results from other emissivity data compare with findings in this study. The surface emissivity database produced by Zhou et al. [2011], for example, should be comparable to the UW/CIMSS data and thus also applicable to the hyperspectral sensor. Of course, the comparison results will improve our understanding of the use of radiative transfer modeling capability for monitoring the dust from the hyperspectral sounder measurements.

Acknowledgments

[39] The authors would like to thank three anonymous reviewers for their constructive and valuable comments, which led to an improved paper. The RTTOV model development is part of the EUMETSAT funded NWP satellite application facility activities. Part of this work was done while the first author (H.J.H.) resided at UW/CIMSS as a visiting scientist. This work was funded by the Korea Meteorological Administration Research and Development Program under grant CATER 2012-2060.