Monitoring and Quantitatively Retrieving Dust Storm from Satellite Thermal Infrared Measurements Zhang P.1, Li J.2, Lu N. M.1, Hu X. Q.1, Schmit T. J.3 1 Key Laboratory of Radiometric Calibration and Validation for Environmental Satellites (LRCVES)/CMA 2 Cooperative Institute for Meteorological Satellite Studies/University of Wisconsin-Madison 3 Office of Research and Applications/NESDIS/NOAA No. 46, South street of Zhong-Guan-Cun, District Haidian, Beijing, 100081, China [email protected] Tel. 86-10-68409671, Fax. 86-10-62183844 Abstract In this paper, the possibility of monitoring and quantitatively retrieving dust storm from satellite thermal infrared measurements was investigated. It was found from satellite observations that for the dust clouds, the observed 11 μm minus 12 μm brightness temperature difference (BTD) is always negative, while the BTD of 8.5 μm minus 11 μm varies from positive to negative depending on the dust concentration. Based on the threshold classifications, a dust mask algorithm to identify the dust storm initiation and spatial extent was developed. The algorithm can be used to process data for both the daytime and nighttime. Under the Mie spherical scattering theory, the thermal radiation transfer through the single dust layer were performed by the widely used forward model DISTORT. Our calculations show that the dust-like aerosols can fairly well explain the observed BTD although both of the complex refractive index and particle size of aerosols will significantly influence the BTD. When the complex refractive index is fixed (dust-like aerosols in this paper), the dust optical thickness and effective radius of dust particles can be retrieved from the brightness temperature (BT) of the 11 μm channel and the BTD of 11 μm minus 12 μm channels, respectively. The integral dust columnar density can be also derived from retrieved dust optical thickness and effective radius. The algorithm developed here has been applied to the MODIS/Terra (polar satellite) and the SEVIRI/MSG (Geostationary satellite) data. Keywords: Dust storm, Thermal Infrared, Satellite, Brightness temperature INTRODUCTION Satellite monitoring is a powerful tool for studying the properties of large-scale dust storms. Since the 1970s, scientists have succeeded in identifying the outbreaks of dust storms from satellite images by use of the thermal infrared (TIR) window technique (Shenk and Curran, 1974; Ackerman, 1989). More recently, Wen and Rose (1994) developed an algorithm based on the TIR technique for the retrieval of particle size, optical thickness and total mass of volcanic cloud by using the AVHRR band 4 and 5. The concept of their algorithm is similar to that for the retrieval of cirrus using the BTD method (Wu, 1987; Giraud, et al., 1997). Dust aerosols have the very similar radiative properties with volcanic aerosols. Therefore, the principle is applicable to remote sense the dust storm. In this paper, we have developed an algorithm to identify and physically retrieve the dust storm with three satellite TIR window channels. Because only TIR channels are used, the algorithm is applicable to both daytime and nighttime conditions. The algorithm developed here has been applied successfully to the MODIS/EOS (polar satellite) and the SEVIRI/MSG (Geostationary satellite) data. DUST STORM MASK ALGORITHM Threshold method is used to identify the dust storm initiation and spatial extent. Because of the surface emissivity, different thresholds are selected over non-desert area and over desert, listed in Table 1 and Table 2. The dust storm mask results are given in Fig. 1 and Fig. 2 respectively. Fig. 1 shows the dust storm monitored by MODIS/Terra on 7 April, 2001 over most of northern China, which is the strongest among year 2001. Fig. 2 shows the dust storm monitored by SEVIRI/MSG on 3 March, 2004 over Sahara desert. As the comparison, the satellite RGB color composition images have been given too. Over non-desert area, the brightness temperature difference of 11 μm minus 12 μm (BTD11-12) alone can work well to identify the dust storm from the background. However, it is insufficient over desert area because the desert surface and dust layer have the very similar behaviors at 11 μm and 12 μm. As an important remediation, 8.5 μm can provide the additional information to distinguish the suspended dust from the desert surface effectively. Moreover, the brightness temperature difference of 8.5 μm minus 11 μm (BTD8-11) is a pretty good index to indicate the strength of dust storm. By combination the thresholds of BTD11-12 and BTD8-11, the mask algorithm can automatically identify the outbreak and the relative strength of dust storm from these three TIR observations. Because of the spectral variation of surface emissivity, the surface usually contributes about -1.0 to 0.5 K in BTD11-12. Therefore, the threshold of BTD11-12 was set as -0.5 K over non-desert area and -1.0 K over desert separately. There is a substantial advantage to use BTD as the criteria to identify dust storms because BTD is less dependent on the surface temperature than BT. a b Fig. 1 Dust storm on 7 April, 2001 over most of northern China a) RGB composition image from MODIS channel 1 (645nm), 4 (555nm) and 3 (469nm); b) Dust storm mask results by threshold method listed in Table 1. b a Fig. 2 Dust storm monitored on 3 March, 2004 over Sahara desert a) RGB composition image from SEVIRI/MSG 11 μm, 11-12 μm, 8-11 μm; b) Dust storm mask results by threshold method listed in Table 2. Threshold BTD11-12<-0.5 and BTD8-11>0 BTD11-12<-0.5 and BTD8-11<0 BTD11-12>0 and BTD8-11>0 BTD11-12>0 and BTD8-11<0 0>BTD(11-12)>-0.5 Mask Flag 1 Description relative strong dust region 2 relative weak dust region 3 ice cloud 4 low cloud or surface 5 uncertain region Table 1 Thresholds for the identification of dust storm over non-desert area Threshold BTD11-12<-1 and BTD8-11>-5 others Mask Flag 0 Description Dust 1 No Dust Table 2 Thresholds for the identification of dust storm over desert area FORWARD MODEL AND SIMULATION A two-layer TIR band forward model is developed to simulate thermal infrared radiation transfer through dust layer and to simulate the satellite measurements (Fig. 3). The forward model consists of the DISORT model (Stamnes et al., 1988) and a Mie scattering code (Zhang and Shi, 1997). With the selected dust aerosol parameters (i.e., complex refractive index, particle size distribution and particle shape), the single scattering albedo, ω0, asymmetry factor, g, and extinction coefficient ratio, k, of the dust layer can be calculated from the Mie code. These three optical parameters of dust aerosol are necessary for the radiation transfer calculation. By inputting them together with given optical thickness of dust layer into DISORT, the satellite observation can be simulated for arbitrary wavelengths. Fig. 3 Conception of two-layer IR band forward model In the two-layer forward model, satellite observation in the TIR band is composed of two parts, the radiance emitted from the dust layer and that transmitted from the underlying surface. The observed radiance by the satellite through such semitransparent dust layer can be expressed as I i = ε c B(Tc ) + t atm B (Tg ) (1) where Ii is the observed radiance in an given channel, εc, and tatm are emissivity and transmissivity of the dust layer in the central wavelength of given channel, Tg and Tc are the underlying surface temperature and equivalent top temperature of the dust layer. The emissivity of surface is assumed as blackbody in (1). Several assumptions are made to simplify the calculation. These are: 1) the underlying surface is homogeneous; 2) the dust cloud is a single layer parallel to the surface; 3) the atmosphere above the dust layer and between the surface and the dust layer are clear windows and 4) the dust layer completely fills the field of observation. Other assumptions in the simulation also include spherical particle shape, logarithm normal particle size distribution, and complex refractive index of dust-like aerosol for aerosol model. Fig. 4 shows the simulation results from the forward model. Green, blue, black lines represent the results calculated from Tg = 290 K and Tc = 270 K, Tg = 290 K and Tc = 250 K, Tg = 290 K and Tc = 230 K, respectively. Lines with †, ◊ and □ represent the results obtained by assuming dust effective radius to be 13, 8 and 2 μm, respectively. In each line, 20 optical thicknesses from 10 to 0.1 are considered in the direction of abscissa. Fig. 4 Simulation results calculated by the forward model. Abscissa is BT11 and ordinate is BTD11-12. Green, blue, black lines represent the results calculated from Tg = 290 K and Tc = 270 K, Tg = 290 K and Tc = 250 K, Tg = 290 K and Tc = 230 K, respectively. Lines with †, ◊ and □ represent the results obtained by assuming dust effective radius to be 13, 8 and 2 μm, respectively. In each line, 20 optical thicknesses from 10 to 0.1 are considered in the direction of abscissa. From Fig.4, it can be seen that BT11 decreases with increasing optical thickness while BTD11-12 decreases with the increase of particle effective radius. They indicate that it is possible to retrieve the optical thickness and particle size of dust from the combination of BT11 and BTD11-12 information. Fig. 4 also shows that BTD11-12 decreases with the decrease of the temperature difference between Tg and Ts. The larger the temperature difference of Tg minus Tc is, the larger the BTD11-12 caused by dust particle size. They imply that both Tg and Tc will strongly affect the calculation of BT11 and BTD11-12. Therefore, the accuracy of Tg and Tc will determine the reliability of the retrieved results. DUST STORM RETRIEVAL ALGORITHM An algorithm has been developed to retrieve the optical thickness and particle effective radii of the dust layer (Zhang, et al., 2005). The retrieval processing relies on the BT11 and BTD11-12 pattern. Forward simulations have showed that BT11 and BTD11-12 are sensitive to the optical thickness and particle size of the dust layer. Whilst, the underlying surface temperature, Tg, effective top temperature of dust layer, Tc and the chemical composition of dust aerosol, i.e., the complex refractive index of dust aerosol will influence the calculated BT11 and BTD11-12 too. However, if Ts, Tc and refractive index can be derived from the pre-knowledge, BT11 presents an almost linear relationship with dust optical thickness and BTD11-12 presents an almost linear relationship with dust particle size. By fitting the satellite observations to the pre-calculated BT11 and BTD11-12 look-up table, the optical thickness and particle effective radii of the dust layer can be retrieved. When dust optical thickness and particle effective radius have been retrieved, dust column density can be obtained by using of following formula uniquely n(r ) = ⎡ (ln r − ln r0 )2 ⎤ 1 exp ⎢− ⎥ 2 2(ln σ ) ⎥⎦ 2π r ln σ ⎢⎣ N0 (2) m = ρ sandsV0 = ρ sands ∫ v(r )dr (3) 4 v(r ) = πr 3 n(r ) 3 (4) r2 r1 N0 = τ ∫ r2 r1 Qeπr 2 n(r )dr (5) () where in (2), the lognormal distribution (LND) is selected for dust aerosols, n r is the particle number size distribution, N0 is the number of the particles per cross section of atmospheric column (m-2), r is the particle radius, r0 is the modal radius of LND, ln σ is the standard deviation of ln r. From (3) to (5), m is the column aerosol loading in mass, ρ sands = 2.5 g / cm 3 () ρ sands is the mass density of sands and we take in this paper. v r and V0 are the volume size distribution and the volume of the particles per cross section of atmospheric column, Qe is extinction coefficient of particle decided by Mie scattering. It is clear that the dust column density, N0, is related to the optical thickness, τ, by (5). In addition, the inversion of the optical thickness to the dust column density will be influenced by both the complex refractive index and the particle size distribution of aerosols. The methodology of the fitting and retrieval process is illustrated by Fig. 5. The kernel of the process is a pre-calculated look-up table (LUT). The LUT is calculated with 9 Ts (from 310 K to 270 K in 5 K increment), 9 Tc (from 270 K to 230 K in 5 K increment), 20 optical thickness (from 10 to 0.1 in equal logarithm increment), 29 particle effective radius (from 0.01 μm to 10 μm in equal logarithm increment) and fixed refractive index in this paper. The dust retrieval algorithm has been applied to both MODIS/Terra and SEVIRI/MSG data. Two examples are showed in Fig. 6 and Fig. 7, corresponding to the retrieval results from China case on 7 April, 2001 by MODIS/Terra and from Sahara case on 3 March, 2004 by SEVIRI/MSG respectively. Fig. 5: Flow chart of fitting and retrieval process Fig. 6 The retrieval results on 7 April, 2001 over most of northern China Fig. 6a, b, c show the retrieved dust optical thickness (dimensionless), effective radius of dust particle (in µm) and integral dust columnar density (in 106 µg m-2) respectively. Fig. 7 The retrieval results on 3 March, 2004 over Sahara desert Fig. 7a, b, c show the retrieved dust optical thickness (dimensionless), effective radius of dust particle (in µm) and integral dust columnar density (in 106 µg m-2) respectively. CONCLUSION A dust storm mask and retrieval algorithm has been developed. Because the algorithm is based on the satellite TIR measurements, it can be used in both daytime and nighttime conditions for automatic processing. The algorithm has been applied to both MODIS/Terra and SEVIRI/MSG data. The dust mask and retrieval results from the case study have showed that: 1) BTD11-12 alone is not sufficient to identify the outbreak and the spatial extent of the dust storm. 8.5 um channel provide the additional information. Over desert, this channel is especially helpful to distinguish the suspended dust from the desert surface effectively. While over non-desert area, the brightness temperature difference of 8.5 μm minus 11 μm (BTD8-11) is a pretty good index to indicate the strength of the dust storm. 2) Optical thickness, particle effective radii and integral dust column density can be retrieved from the satellite measured BT11 and BTD11-12. While, the effective surface temperature of the underlying (including the surface emissivity effect), the effective top temperature of the dust layer, and the complex refractive index of the dust aerosol will considerably influence on the reliability of the retrieval results. The uncertainties of these factors will decrease the accuracy of the retrieval results. The detailed error analysis is required. The algorithm will be improved in the future on 1) the more accurate forward model to include atmospheric profile and surface emissivity effect, 2) better determination of the effective surface temperature of the underlying and the effective top temperature of the dust layer, 3) Validation from ground-based measurements and other satellite results, such as TOMS results. REFERENCES Ackerman, S., (1989) Using the radiative temperature difference at 3.7 and 11 μm to track dust outbreaks. Remote Sens. Environ. 27(2), 129-133. Giraud, V., Buriez, J.C., Fouquart, Y., Parol, F., (1997) Large-scale analysis of cirrus clouds from AVHRR data: assessment of both a microphysical index and the cloud-top temperature. J. Appl. Meteorol. 36 (6), 664–675. Stamnes, K., Tsay, S.C., Wiscombe, W., Jayaweera, K., (1988) Numerically stable algorithm for discrete-ordinate-method radiative transfer in multiple scattering and emitting layered media. Appl. Opt. 27(12), 2502-2509. Shenk, W.E., Curran, R.J., (1974) The detection of dust storms over land and water with satellite visible and infrared measurements. Mon. Weather Rev. 102(12), 830-837. Wen, S., Rose, W.I., (1994) Retrieval of sizes and total masses of particles in volcanic clouds using AVHRR bands 4 and 5. J. Geophys. Res. 99 (D3), 5421-5431. Wu, M.L., (1987) A method for remote sensing the emissivity, fractional cloud cover and cloud top temperature of high-level, thin clouds. J. Clim. Appl. Meteorol. 26(2), 225-233. Zhang, P., Shi, G.Y., (1997) Optical characteristic Mie scattering calculation using an aerosol distribution observed by balloon. Proceedings of the 6th Chinese Aerosol Conference, pp.57-61. Zhang, P., N. Lu, X. Hu, C. H. Dong., (2006) Identification and physical retrieval of dust storm using three MODIS thermal IR channels. Global and Planetary Chance, 52, 197-206.
© Copyright 2026 Paperzz