Aerosol impact and correction on temperature profile retrieval from MODIS

Transcription

1 GEOPHYSICAL RESEARCH LETTERS, VOL. 35, L13818, doi: /2008gl034419, 2008 Aerosol impact and correction on temperature profile retrieval from MODIS Jie Zhang 1,2 and Qiang Zhang 1,2 Received 24 April 2008; accepted 5 June 2008; published 11 July [1] Aerosols over desert areas can impact the temperature profile retrieval using infrared remote sensing techniques. Retrievals from Moderate Resolution Imaging Spectroradiometer (MODIS) Infrared (IR) radiances over the Gobi region in China show that atmospheric temperatures above the top of the boundary layer can be retrieved with a root mean square error (rmse) better than 2 K. In the boundary layer, the error in the temperature retrieval is largely due to limited spectral information and uncertainty of the radiative transfer model. The error in the retrieved atmospheric boundary layer temperature is positively correlated with aerosol optical depth (AOD) and estimated errors of skin temperature and surface IR emissivities. Taking into account the aerosol effect in the radiative transfer model, the temperature retrievals from MODIS are improved by approximately 0.38 K in the boundary layer depending on the aerosol optical depth (AOD). Citation: Zhang, J., and Q. Zhang (2008), Aerosol impact and correction on temperature profile retrieval from MODIS, Geophys. Res. Lett., 35, L13818, doi: / 2008GL Introduction [2] The atmospheric boundary layer is the most important part of atmosphere. Because it directly links soil, biosphere and atmosphere, and is sensitive to physical and chemical processes over land and sea surfaces, it is the most active layer in the exchange of matter and energy. The depth of the boundary layer over Gobi and the desert in Northwest China can be more than 4000 m [Zhang et al., 2004], which results in frequent regional disasters, such as hail storms and sandstorms. Therefore, understanding the thermodynamic structure of the boundary layer in this region is very important for predicting the weather disasters, investigating the transportation of sand particle and the aerosol effect on cloud formulation and precipitation. At present, the observation platform used for profiling the atmospheric boundary layer is the standard radiosonde. Only 8 radiosondes are located in the Gobi and desert region (10 10 ), and they cannot be used to observe the weather system; therefore, it is necessary to use remotely sensed atmospheric profiles that have good spatial and temporal coverage, the sensors include a spectrometer and radar used for infrared (IR) 1 College of Atmospheric Sciences, Lanzhou University, Key Laboratory of Arid Climatic Changing and Reducing Disaster of Gansu Province and China Meteorological Association, Lanzhou, China. 2 Institute of Arid Meteorology, China Meteorological Administration, Lanzhou, China. Copyright 2008 by the American Geophysical Union /08/2008GL remote sensing and microwave remote sensing [Andrea and Bosenberg, 2006]. The IR remote sensing method is widely used for atmospheric temperature and moisture soundings, such as TOVS (TIROS Operational Vertical Sounder) on NOAA satellites, MODIS (Moderate Resolution Imaging Spectroradiometer) from the Aqua platform, and so on. With a mature retrieval methodology [Seemann et al., 2003], a reasonable temperature vertical structure between 400 hpa and 800 hpa can be achieved from MODIS, with high spatial resolution, MODIS may reflect small scale atmospheric feature. However, since aerosol changes the thermodynamic structure of the boundary layer atmosphere [Alpert et al., 1998], and the IR longwave radiances also contain aerosol attenuation, and the aerosol component is not considered in the current operational MODIS profile algorithm, the temperature retrieval can be negatively affected if the aerosol effect is not taken into account properly in the radiative transfer model. In addition, surface IR emissivity uncertainty will also have impact on retrieval accuracy [Li, 1994]; a reliable IR emissivity database over Gobi and desert regions is necessary for temperature retrieval using IR radiance measurements. [3] The aerosol observations and MODIS data are used in this paper to analyze the impact of aerosol on temperature profile retrievals, and an aerosol correction technique is developed for improving the temperature profile retrieval from IR radiances. Results show that the boundary layer temperature can be improved by 0.38 K from MODIS when the aerosol effect is considered. With advancements in stateof-the-art IR sounding instruments such as the hyperspectral AIRS (Atmospheric Infrared Sounder) and IASI (Infrared Atmospheric Sounding Interferometer), and in the future CrIS (Cross-track scanning Infrared Sounder) and geostationary hyperspectral IR sounder, atmospheric profiles with high vertical resolution and accuracy can be achieved [Barnet and Blaisdell, 2003]; lessons learned from MODIS and the techniques tested on MODIS can be applied to both the current and future IR sounding instruments. With fewer spectral bands than hyperspectral IR sounders, it is easier to test the aerosol correction technique with MODIS. 2. Retrieval Method [4] The retrieval methodology is a combination of the operational MODIS retrieval algorithm [Seemann et al., 2003] and a nonlinear physical iterative approach [Li et al., 2000]. The operational MODIS algorithm for atmospheric soundings is the statistical regression approach, and the regression equation is built upon a global training dataset that contains temperature, moisture and ozone profiles, surface skin temperatures and surface IR emissivities [Seemann et al., 2008]. For the sake of efficiency for L of5

2 operational processing of MODIS data, a physical retrieval was not employed in the profile retrieval. However, for the regional data processing, a physical retrieval can be used to improve the retrieval accuracy based on a fast radiative transfer model called Pressure-layer Fast Algorithm for Atmospheric Transmittances (PPAAST) [Hannon et al., 1996]. In this research, a statistical retrieval using the operational MODIS algorithm [Seemann et al., 2003] is used as the first guess for the physical retrieval, and a nonlinear iterative approach based on a one-dimensional variation (1DVAR) method and the discrepancy principal [Li and Huang, 1999] is adopted for the temperature profile retrieval. The profile at 101 pressure levels with the operational MODIS algorithm is used to generate the first guess for the physical retrieval. 3. Data Used 3.1. MODIS [5] The Moderate Resolution Imaging Spectroradiometer (MODIS) offers a new opportunity to improve global monitoring of temperature, moisture, and ozone distributions and changes therein. MODIS was launched onboard the National Aeronautics and Space Administration (NASA) Earth Observing system (EOS) Terra and Aqua platforms on 18 December 1999 and 4 May 2002, respectively. The instrument is a scanning spectroradiometer with 36 visible (VIS), near-infrared (NIR), and IR spectral bands between and mm [King et al., 1992]. The advantage of MODIS for retrieving the distribution of atmospheric temperature and moisture profiles is its combination of shortwave and longwave IR spectral bands ( mm) that are useful for sounding and its high spatial resolution that is suitable for imaging (1 km at nadir). The retrievals are performed using clear-sky radiances measured by MODIS within a 1 1 field of view (approximately 1-km resolution). MODIS has few available bands (0.66, 0.47, and 2.1 mm) that are used for retrieving aerosol optical depth (AOD), while the radiative transfer model used for the operational MODIS profile retrieval does not use the AOD information and the aerosol scattering is ignored in the sounding retrieval. This study illustrates that an aerosol correction in the MODIS profile algorithm is necessary over the Gobi and desert regions Radiosonde [6] In the Gobi and desert region in Northwest China (35 45 N, E), there are only 8 radiosonde stations that are located at Hami (42.48 N, E), Mazongshan (41.80 N, E), Dunhuang (40.02 N, E), Jiuquan (39.77 N, E), Zhangye (38.93 N, E), Ejinaqi (41.95 N, E), and Bayinmaodao (40.17 N, E). In order to study the boundary layer structure in the Gobi and desert region, the Institute of Arid Meteorology (IAM) made atmospheric profile observations every one or two hours at three stations (Dunhuang, Jiuquan and Zhangye), the observations lasted 3 months from June to July 2006 and from December 2006 to January The radiosonde observations at 14:00 (local time) are used for testing the retrievals of MODIS from Aqua (overpass time is from 13:20 to 14:40 local time). Therefore the time difference between the radiosondes and MODIS soundings is less than 1 hour, while the spatial distance between the radiosondes and MODIS soundings is limited to be less than 5 km in the comparisons. [7] The radiosonde is observed by the GFE(L) radar made in China. This instrument can observe many parameters including wind direction, wind speed, air temperature, pressure and moisture within 30 km vertical range, where the temperature error is 0.2 K, the pressure error is 2 hpa, the vertical resolution is 10 m, and the rising speed is 250 m per minute within the boundary layer and 400 m in FA (free atmosphere) Aerosol Optical Depth [8] The Sun Photometer is an effective apparatus for detecting atmospheric aerosol. NASA has successfully installed more than 180 sun tracking photometers CE318A made in France by the CIMEL company, which have formed a global AERONET [Holben et al., 1998]. CE318A has 8 channels in visible and near-infrared wavelengths, with wavelength centers at: 1020 nm for 870p1, 670 nm and 440 nm for 870p2, and 870 nm and 936 nm for 870p3, respectively. The widths of these wavelengths are all 10 nm. The view angle is 1.2 while the precision of tracking the sun is 0.1. The observed time is 30s and the data include 3 components. [9] In wavelengths 440 nm, 870 nm and 1020 nm without water vapor absorption, only aerosol attenuation and Rayleigh scattering are contained. Therefore, Rayleigh scattering optical depth s rl and aerosol optical depth (AOD) s al can be estimated by and s l ¼ s rl þ s al ; s rl ¼ p s p 0 0:0088l 4:05 : where l is wavelength, p s is the pressure of the surface, and p 0 is standard pressure, equal to hpa. 4. Aerosol Correction Technique 4.1. Aerosol and Atmospheric Transmittance Correction [10] Observed result shows that sand dust size is usually um though it can be larger than 11 um during a sand storm; under clear skies, the aerosol particle size is um. According to the ratio of the particle diameter with the frequency of the spectrum wavelength, the particle is Mie scattering at the wavelength less than 11 um and Rayleigh scattering for the other MODIS wavelengths, and should be considered in the radiative transfer model. Tomasi et al. [1983] proposed to combine absorption and scattering effects in one formula, that is, s al ¼ e bl a m a : where m a is the relative air mass for aerosols, b is the Angstrom coefficient, and a is the wavelength exponent. There is a negative exponent relation between optical depth and transmittance. ð1þ ð2þ ð3þ 2of5

3 Figure 1. Effect of aerosol optical depth (AOD) on the weighting of the skin temperature. [11] In the physical retrieval processing, the radiative transfer is simplified by assuming that scattering by the atmosphere is neglected. Then, atmospheric transmittance t l can be expressed as below t l ¼ t ms t g t o : [12] When particle scattering is considered, the transmittance [Weiss and Norman, 1985] and aerosol transmittance t alj at level j can be described by ð4þ t l ¼ t ms t g t o ðt r t al Þ ð5þ t alj ¼ p j p t p 0 p t t al ; where t ms, t g, t o, t r, t al is the spectral transmittance due to water vapor continuum attenuation, mixed gases absorption, the ozone layer absorption, molecular scattering and aerosol scattering and absorption, respectively; t alj is estimated according to analytical results from sun photometers observed t al ; and p t is the pressure on the top of the boundary layer Sensitivity Experiment [13] Figure 1 shows the relation of AOD to the weighting of skin temperature. It is assumed that the temperature is ð6þ 30 C, and AOD is 0.2. When AOD increases from 0.2 with increment from 0 to 0.5, weighting of water vapor absorption bands 27, 28, and 29 are decreased; the weighting of ozone absorption band 30, temperature sensitive bands 25, 33, 34, 35, 36, and IR window bands 31, 32 are also decreased. If AOD is underestimated, atmospheric transmittance and weighting of skin and air temperature will be overestimated, and resulting in an underestimation of the retrieval temperature profile, ultimately leading to an increase in the absolute error. Therefore, AOD correction can improve atmosphere transmittance, weighting of skin and air temperature, and finally the temperature profile retrieval. 5. Retrieval Results 5.1. Retrieval Results Without Aerosol Correction [14] Three retrieved profiles close to the three radiosonde stations are selected and compared with radiosonde observations at 14:00 local time. The retrieval bias and the root mean square error (RMSE) are calculated. Figure 2 shows the RMSE of the physical retrieval from 32 cloud-free cases (Figure 2a), and the retrieval error of one case on 29 June 2006 (Figure 2b) and retrieved profile of this case (Figure 2c). The RMSE is less than 1 K at levels from 600 to 340 hpa, the error is increased at the levels above 340 hpa, and there is a peak value 2.3 K at about 200 hpa. Below 600 hpa, the retrieval errors are increased with altitude near the surface, the largest error is 2.7 K, but the Figure 2. (a) Retrieval error of temperature profiles from 32 cases, (b) error of one case at 15:00 local time on 29 June 2006, (c) statistical and physical retrieval results of the case (dashed line indicates statistical retrieval, thin line is the physical retrieval, thick line is the observed result). 3of5

4 Figure 3. The physical retrievals with (dashed line) and without AOD correction (solid thin line) along with the observed value (thick solid line) of (a) one case on 29 June 2006 and (b) another case with AOD of (c) RMSE after AOD correction (dashed line) and the RMSE improvement (thick solid line) due to AOD correction from all 32 cases. error at surface is about 1 K. Research shows that the atmospheric boundary layer (ABL) is more than 4 km at noon over Gobi and desert in Northwest China [Zhang et al., 2004]; the top of the boundary layer is near 600 hpa. Therefore, the physical retrieval is good for above the top of the boundary layer, but poor for within the boundary layer. [15] Figure 2b shows the physical retrieval error of a typical case with a large retrieval error due to the large aerosol effects (AOD is 0.38 according to the observation). The error is less than 2.5 K at levels about 600 hpa. Below 600 hpa, the retrieval error increases with altitude near the surface where the error is between 2.5 K and 5 K. Figure 2c shows the statistical retrieval (dashed line), the physical retrieval (thin line) along with the observed profile of this case (thick line). The statistical result is far less than the observed values. The error can be as large as 10 K, the physical retrieval is able to reduce the error at the levels above 600 hpa, below 600 hpa, retrieved values are are not consistent with observed values at levels near the surface. Figure 2 shows that the presence of high concentrations of aerosols introduces large errors in temperature retrievals below 600 hpa. Therefore, modification of the retrieval algorithm using an aerosol correction on the radiative transfer model may lead to improvement in retrieval accuracy Retrieval Results After Aerosol Correction [16] Based on the equations (4) to (6), many cases are corrected by taking into account the effect of AOD, herein skin temperature is retrieved from MODIS, and emissivities are from the regression. Figure 3a provides a contrast of the physical retrievals of the case (the same as were shown in Figure 2 below 600 hpa with and without AOD correction, respectively. The results without AOD correction are the same as that shown in Figure 2 (thin line). When AOD correction is performed, the retrieved values increased (dashed line). Another case shown in Figure 3b has an AOD of 0.23; the physical retrieval results are less than the observed values (thin line). When AOD correction is performed, the retrieved values are larger than the observed values (dashed line). Figure 3c is the vertical distribution of boundary layer temperature RMSE for all 32 cases after AOD correction and the RMSE improvement due to the AOD correction. The largest RMSE of 32 cases is 2.34 K, the average RMSE is 2.02 K. RMSE improvement is from 0.15 to 0.58 K while the average RMSE improvement is 0.38 K in the boundary layer. This result shows that the errors can be reduced when the aerosol absorption and scattering is considered in the retrieval. However, the retrieval accuracy after AOD correction can still not reach the 2 K range because: a) MODIS has low spectral resolution; b) the study region is very asymmetrical, which results in large emissivity variability; analytical results show statistical emissivities in our study are very different from the MODIS emissivity product ( ) [Seemann et al., 2008], and better handling emissivity in retrieval should be studied; and c) the radiosonde error should also be considered. [17] At present, global aerosol stations and an aerosol network have been formed, and hourly data are released at web Moreover, the Cloud- Aerosol Lidar and Infrared Pathfinder Satellite Observations(CALIPSO) was launched on April 2006, which will provide the possibility for understanding AOD distribution, and should be useful for estimating atmospheric transmittance and retrieval of atmospheric profiles, especially for hyperspectral IR sounder profile retrievals. 6. Conclusion and Discussion [18] On the basis of the first guess value from the statistical retrieval, a physical retrieval algorithm is used for atmospheric profile retrieval from MODIS IR radiances, and the retrieval error is analyzed. It shows that the AOD in Northwest China is an important factor affecting temperature profile precision from IR remote sensing. The conclusions include: [19] 1. Physical retrieval results have good consistency with radiosondes above 600 hpa. However, in the range of the boundary layer, retrieval results are poor. The temperature error is around 2 K the largest error is 2.7 K if aerosol correction is not taken into account, and the error is increases relative to the decrease in the pressure level. [20] 2. The global aerosol distribution is very different, and aerosol also leads to notable radiant, climate and optical effects. This is one of the important factors affecting physical retrieval of atmospheric temperature. The retrieval error is positively correlated with the aerosol optical depth. If aerosol effects at infrared spectral bands are considered, the atmospheric transmittance can be better estimated. 4of5

5 Moreover, the weighting of skin temperature and atmospheric temperature can be realistically estimated, and the retrieval precision of atmospheric temperature can be improved to approximately 2 K; however, the retrieval error is still large due to the low spectral resolution of MODIS. If the hyperspectral IR sounding instruments such as AIRS, IASI and the future CrIS and geostationary advanced IR sounder are used, temperature profiles with high vertical resolution and good accuracy are achievable [Barnet and Blaisdell, 2003]. It is our next plan to use AIRS and IASI for atmospheric profiling over the Gobi and desert regions. [21] 3. In nonlinear physical retrieval processing, emissivities are estimated by statistics, and they are directly used for the retrieval of the atmospheric profile. However, the difference between statistical emissivities and the MODIS emissivity product is quite large (0.08), and emissivities change with complex surface and seasonal alternation; better handling of the surface IR emissivity in the sounding retrieval needs to be studied. The emissivity database in uniform regions should be tested in atmospheric profiling in the future. [22] 4. This research gives a statistical algorithm for estimating vertical distribution of AOD from CE318 sun tracking photometers, but it should be further tested if it is used in other regions next. At present, ground and satellite Laser radar(calipso) have been used for aerosol spatial distributing [Berthier et al., 2006], and the data will be used for aerosol vertical distribution and particle size, which is better for understanding aerosol scattering effects and atmospheric transmittance distribution. [23] Acknowledgments. The authors would like to thank Doctor Li J. for providing the operational MODIS retrieval algorithm, the 101-level MODIS radiative transfer model, and giving good ideas and suggestions for improving the paper. References Alpert, P., Y. J. Kaufman, Y. Shay-El, D. Tanre, A. da Silva, S. Schubert, and J. H. Joseph (1998), Quantification of dust-forced heating of the lower troposphere, Nature, 395, , doi: / Andrea, A., and J. Bosenberg (2006), Determination of the convective boundary-layer height with laser remote sensing, Boundary Layer Meteorol., 119, Barnet, C. D., and J. Blaisdell (2003), Retrieval of atmospheric and surface parameters from AIRS/AMSU/HSB data in the presence of clouds, IEEE Trans. Geosci. Remote Sens., 41, Berthier, S., P. Chazette, P. Couvert, J. Pelon, F. Dulac, F. Thieuleux, C. Moulin, and T. Pain (2006), Desert dust aerosol columnar properties over ocean and continental Africa from Lidar in-space Technology Experiment (LITE) and Meteosat synergy, J. Geophys. Res., 111, D21202, doi: /2005jd Hannon, S. E., L. L. Strow, and W. W. McMillan (1996), Atmospheric infrared fast transmittance models: A comparison of two approaches, Proc. SPIE Int. Soc. Opt. Eng., 2830, , doi: / Holben, B. N., T. F. Eck, and I. Slutsker (1998), AERONET: A federated instrument network and data archive for aerosol characterization, Remote Sens. Environ., 66, King, M. D., Y. J. Kaufman, W. P. Menzel, and D. Tanre (1992), Remote sensing of cloud, aerosol, and water vapor properties from the Moderate Resolution Imaging Spectrometer (MODIS), IEEE Trans. Geosci. Remote Sens., 30, Li, J. (1994), Temperature and water vapor weighting functions from radiative transfer equation with surface emissivity and solar reflectivity, Adv. Atmos. Sci., 11, Li, J., and H.-L. Huang (1999), Retrieval of atmospheric profiles from satellite sounder measurements by use of the discrepancy principle, Appl. Opt., 38(6), Li, J., W. Wolf, W. P. Menzel, W. Zhang, H. L. Huang, and T. H. Achtor (2000), Global soundings of the atmosphere from ATOVS measurements: The algorithm and validation, J. Appl. Meteorol., 39, Seemann, S. W., J. Li, W. P. Menzel, and L. E. Gumley (2003), Operational retrieval of atmospheric temperature moisture and ozone from MODIS infrared radiances, J. Appl. Meteorol., 42, Seemann, S., E. Borbas, R. Knuteson, H. L. Huang, and G. Stephenson (2008), Global infrared emissivity for clear sky atmosphereic regression retrievals, J. Appl. Meteorol. Climatol., 47, Tomasi, C., E. Caroli, and V. Vitale (1983), Study of the relationship between Angstrom s wavelength exponent and Junge particle size distribution exponent, J. Clim. Appl. Meteorol., 22, Weiss, A., and J. M. Norman (1985), Partitioning solar radiation into direct and diffuse, visible and near-infrared components, Agric. For. Meteorol., 34, Zhang, Q., G. A. Wei, and P. Hou (2004), Observation studies of atmospheric boundary layer characteristic over Dunhuang Gobi in early summer, Plateau Meteorol., 23, J. Zhang and Q. Zhang, Institute of Arid Meteorology, China Meteorological Administration, Donggang East Road 2070, Lanzhou, Gansu , China. (gs-zhangjie@163.com) 5of5