It is the cache of ${baseHref}. It is a snapshot of the page. The current page could have changed in the meantime.
Tip: To quickly find your search term on this page, press Ctrl+F or ⌘-F (Mac) and use the find bar.

Gayana (Concepción) - ATMOSPHERIC CORRECTION OF THE LANDSAT SATELLITE IMAGERY FOR TURBID WATERS

SciELO - Scientific Electronic Library Online

 
vol.68 número2  suppl.TIProcPREFACETHE CONTRIBUTION OF SATELLITE OBSERVATIONS ON FORECASTS OF SEA STATE OVER THE LAST DECADE, AND THE POTENTIAL IMPACT OF NEW CONFIGURATIONS OVER THE NEXT índice de autoresíndice de materiabúsqueda de artículos
Home Pagelista alfabética de revistas  

Gayana (Concepción)

versión On-line ISSN 0717-6538

Gayana (Concepc.) v.68 n.2 supl.TIProc Concepción  2004

http://dx.doi.org/10.4067/S0717-65382004000200002 

 

Gayana 68(2) supl. t.I. Proc. : 1-8, 2004 ISSN 0717-652X

ATMOSPHERIC CORRECTION OF THE LANDSAT SATELLITE IMAGERY FOR TURBID WATERS

 

Yu-Hwan Ahn, P. Shanmugam & Joo-Hyung Ryu

Satellite Ocean Research Laboratory Korea Ocean Research and Development Institute Ansan P.O. Box 29, Seoul, 425-600, Korea, Email: yhahn@kordi.re.kr


ABSTRACT

This paper describes methods for the correction of the atmospheric effects in the Landsat VIS/NIR imagery in relation to the retrieval of meaningful information about the ocean color, especially from Case-2 waters around Korean peninsula. Three atmospheric correction (AC) methods implemented and examined, using the TOA radiance or reflectance data, are 6S radiative transfer model, spectral shape matching (SSMM) and path-extraction methods. The results show that overall shape and magnitude of radiance or reflectance spectra of the atmospherically corrected Landsat VIS/NIR imagery by SSMM appears to have very good agreement with the in-situ spectra collected for clear and turbid waters, while path-extraction over turbid waters though often reproduces in-situ spectra, but yields significant errors for clear waters due to the invalid assumption of zero values for the black ocean pixels of the Landsat VIS/NIR bands. Because of the standard atmosphere with constant aerosols and models adopted in 6S model, a large error is possible between the retrieved and in-situ spectra. Validation suggests that accurate the retrieval of water-leaving radiance is not feasible with the invalid assumption of classical AC algorithms, but is feasible with SSMM.


 

INTRODUCTION

Retrieval of ocean color information from remotely sensed imagery is a two-step process: (1) Lw is retrieved as a product of atmospheric correction and (2) the constituents' concentrations are determined from Lw. With the intention of correcting the atmospheric effects in the ocean color imagery, several researchers have developed the atmospheric correction algorithms, which have increased in complexity from the simple single scattering algorithm designed for the Coastal Zone Color Scanner (CZCS) (Gordon, 1978) to the complex multiple scattering algorithms proposed for SeaWiFS and MERIS (Gordon and Wang, 1994; Antoine and Morel, 1999). The performance of these algorithms were examined and it was found that they perform better in case-1 waters, but they fail in Case-2 waters, which are rather highly influenced by inorganic mineral particles and additionally by other nonliving dissolved organic materials (DOM). The main reason for the failure of these algorithms over turbid waters is the invalid assumption of zero water-leaving radiance for the near-infrared (NIR) bands at 765 and 865nm, which is common for open ocean waters. Until recently, only few algorithms have been developed (Land and Haigh, 1996; Ruddick et al., 2000), which perform reasonably well in the correction of atmospheric effects, resulting from scattering and absorption of Raleigh and aerosol particles in the atmosphere and from direct reflection at the sea surface, from the LTOA measurements by high spectral resolution ocean sensor over Case-2 waters. However, not much emphasis has been given for the correction of atmospheric effects in the high resolution remotely sensed data, from Landsat-TM, SPOT-HRV and IRS-LISS-III sensors, acquired over the ocean because most of the algorithms are more applicable for deriving land surface/cover information (Popp, 1995). The useful algorithms and methods reported in the literature are as follows: 6S radiative transfer model (Second Simulation of the Satellite Signal in the Solar Spectrum) (Vermote et al., 1997), invariant object (Hall et al., 1991), histogram matching (Richter, 1996), contrast reduction (Tanre et al., 1998) and dark object (Brivio, 2001). With an aim to overcome the shortcomings such as assumption of zero water-leaving radiance at NIR bands and limitations in the availability of reasonable methods, the objectives of the present paper are framed as follows: (1) to describe two AC methods along with 6S model, namely, path-extraction and spectral shape matching, which are implemented and tested on the Landsat and SeaWiFS optical imagery; (2) to validate the results of our methods with the help of in-situ data and (3) to compare the efficiency of our methods with the classical algorithm such as 6S code.

Atmospheric correction algorithms and methods

Atmospheric correction with 6S model

The atmospheric correction of the Landsat optical imagery, acquired over the turbid waters of Jin-do and wan-do bays of Korea, has been carried out with the application of 6S radiative transfer model, released by Vermote et al. (1997). This was accomplished in the following steps: (1) for each band of the Landsat imagery, the recorded digital values are first converted to radiance at the level of TOA; (2) the was then converted to TOA reflectance, (unitless). is related to the above water reflectance (Rw, unitless) as where and (unitless) are the aerosol and Rayleigh reflectances, respectively, and (unitless) are the gaseous and diffuse transmittances, respectively; (3) the TOA measurements were then corrected for atmospheric effects using the 6S code. Finally, the AC reflectance, divided by p steradians, can be compared to the remote-sensing reflectance:

Path-extraction

Path-extraction is a conceptually simple and a more efficient method to extract path-radiance from the TOA signal. Significant advantage of this method is that it extracts the Lpath signal from the image itself without a detailed characterization of the atmosphere and modeling of its effects can be carried out in the atmospheric correction of the ocean color imagery. In fact, it is a first order technique where the lowest radiance value in certain band is subtracted from that band over the entire image. Thus, the principal assumption is that the TOA signal of deep-blue waters, that have the lowest radiance values in the VIS/NIR spectrum, is reduced to the path-radiance because deep-blue waters are often referred to as "black ocean" (Antoine and Morel, 1999). This radiance due to photons scattered by air molecules and aerosols, and also possibly reflected at the sea surface is assumed to be spatially homogeneous over the spatial range of, at least, one Landsat scene and the diffuse transmittance is taken to be 1 . Having done this simple procedure, one can easily retrieve the water-leaving radiance from the pixels of turbid waters, which can be comparable with the in-situ water-leaving radiance spectra.

Spectral Shape Matching Method (SSMM)

The idea behind this method is quite simple and easy to implement on any satellite imagery. The principal assumption of this method is that, there are some pixels in any given scene whose spectral form of radiances or reflectances is quite stable and known for typical turbid or clear waters, leading to the extraction of the path signal from LTOA by subtracting the known water-leaving radiance values obtained from the in-situ measurements as , where Linsitu-ref λ is the typical in-situ reference spectra for clear or turbid waters. The atmospheric path signal (Lpath (λ)) is the contribution of the photons, scattered and reflected between the sea surface and satellite sensor. To retrieve Lw values from the satellite VIS/NIR imagery, with the assumption of constant diffuse transmittance (td = 1) and homogeneous distribution of aerosols and other particles, the Lpath (λ) is then subtracted from the LTOA measured at the TOA as . In contrast to path-extraction, SSMM performs even better for Case-2 waters because Lw is not assumed to be zero The efficiency of this method is explored using Landsat- VIS/NIR imagery, and the results are compared with those of the path- extraction and 6S model.

Results and discussion

Atmospheric correction of Landsat-7 ETM+ Imagery

With the assumption of uniformly mixed aerosols within this layer and computation of TOA total radiances for molecular atmosphere using 6S radiative transfer code, the rpath signal was assessed to give rise to the extraction of Rw above the sea surface (Fig. 1a). The normalized value of path reflectance Rpath(N) for maritime aerosols is significantly low, when compared to that for urban and continental aerosols. This implies that, high reflectance can be expected from oceanic boundary layer aerosols. Using the 6S code, the path reflectance Rpath of maritime aerosols and compound atmospheres, assessed from RTOA of the ETM+ VIS/NIR bands, was then compared to that of the path-extraction and SSMM methods. Fig. 1b compares three atmospheric path reflectances of 6S code, path-extraction and SSMM methods. The normalized value of Rpath(N) (using ETM+ band 4) decreases with increasing wavelength. One should note that 6S model yields high Rpath(N) values throughout the visible channels, indicating low-level contribution of the atmospheric path signal to the Rw. On the contrary, the Rpath(N) values assessed using path-extraction and SSMM methods are comparatively low, which in turn yields high reflectance in all visible bands of ETM+. Consequently, the subtraction of path reflectance from RTOA of the ETM+ VIS/NIR bands gives rise to the atmospherically corrected (Rac) water reflectance spectra, which can be compared with the in-situ spectra. Figs. 2a-d illustrate the total, path and path-corrected reflectance spectra for clear and turbid waters, using 6S radiative transfer model and SSMM. The result of path-extracted was not included in these figures, as it did not have a significant difference from the SSMM method. It is obvious that the total and path reflectances decrease with increasing wavelength. It should be noted that though the spectral form of clear waters is nearly accurate with the 6S model, it often produces unrealistic estimates of Rw values for such waters (Fig.2a). Conversely, the SSMM method is found to be more efficient in the retrieval of accurate values of Rw and thus the spectral form for both clear and turbid waters (Figs. 2b and d). Appraisal of the results of 6S model, path-extraction and SSMM suggests that, there is a need to compare the derived spectra with the in-situ spectra. To do so, the atmospherically corrected reflectance was divided by π value to obtain the remote-sensing reflectance Rrs(l). During the Landsat-7 overpass on 16 April 2001, simultaneous ground radiometric measurements of surface reflectance (Rrs) spectra over clear and turbid water regions were conducted using the ASD (Analytical Spectral Devices) spectroradiometer (covering the spectrum of 0.35-1.1mm). In order to compare the ETM+ band Rrs spectra with the in-situ spectra, the ASD measurements were aggregated using the ETM+ sensor spectral response functions for each band as follows

where S(l) is the ETM spectral responsivity and Rrs(l) is the remote sensing reflectance at 2nm intervals for a given waveband and n is the number of spectral responsivity or Rrs for a given ETM waveband. Fig.3 compares remote sensing reflectance spectra of the ETM+ VIS/NIR bands and in-situ ASD measurements. It is evident that the 6S radiative transfer model leads to the essential overestimation of the Rrs values throughout the VIS/NIR spectrum and most importantly, such high values were never observed during our field campaigns in these waters. The main reasons for such unrealistic estimation of Rrs values by 6S model are as follows: (1) the selection of generalized atmospheric conditions and a standard aerosol model and concentrations, (2) the introduction of white cape reflectance and (3) the basic assumption itself. However, the 6S radiative transfer model offers users to define their own real time and ground-observed variables, which can improve the retrieval of water-leaving reflectance in an accurate manner. On the contrary, the SSMM and path-extraction produce spectra, whose form and magnitude are consistent with the in-situ spectra, though slight discrepancies may be associated with these methods, which are attributable to the surface adjacency effects or sub-pixel mixture problems.

Figures. 1a and b: Comparison of the results of three aerosol models (a) and comparison of the atmospheric path reflectance estimated from the 6S code, path-extraction and SSMM methods (b). Note that the path reflectance Rpath is normalized at ETM+ band 4.


Figs. 2a-d: Schematic representation of the total (at the TOA), path and path-corrected reflectance spectra obtained from the 6S model and SSMM method, for clear and turbid waters.


Figure 3: Comparison of the atmospherically corrected remote sensing reflectance (Rrs) spectrum, estimated using 6S model, path-extraction and SSMM, with the in-situ Rrs spectrum.

Atmospheric correction of Landsat-5 TM Imagery

The efficiency of SSMM was also explored using the two Landsat-5 TM imageries, acquired on 5 May 1999 and 7 May 2000, over the highly turbid waters of Jin-do and Wan-do bays of Korea. Note that the 2nd image was acquired almost exactly after a period of one-year of obtaining the 1st image. The results of 6S model are not included due to lack of ground information. Figs. 4a and b display the color composite images, generated from the Landsat-5 TM bands and show dynamic characteristics of resuspended sediments having similar patterns in both the images. A number of field campaigns were conducted for simultaneous radiometric measurements (such as Lu, Lw, Ed-, Ed+, Lsky) with collection of water samples for the determination of SS, Chl, and DOM concentrations. In order to compare the spectral form of in-situ and TM measurements over typical turbid and clear waters, the in-situ Lw (mWcm-2µm-1sr-1 ) values were converted to equivalent TM radiances as described in the previous section. The derived reference radiance spectra of clear and turbid waters are illustrated in the top panels of Figs. 4a and b. As the two Landsat imageries were acquired during the same season and month, with similar characteristics of water masses, it is therefore assumed that, the mean values of Lw of typical turbid or clear water pixels that can be retrieved from the TM bands remains nearly constant under the clear atmospheric conditions. This assumption allows us to extract the path signal from both the imageries using SSMM. Figs. 5a and b illustrates spectral variation of the atmospheric path signal (Lpath) (mWcm-2µm-1sr-1 ) over clear (a) and turbid (b) waters, during the overpass of Landsat-5 TM on 5 May 1999 and 7 May 2000. This is to compare variations in the atmospheric path signal between the two imageries. It is observed that the atmospheric path signal over clear and turbid water regions decreases monotonically, with increasing wavelength for both the Landsat-TM imageries. This implies that absorption by gaseous molecules dominates the longer wavelength part of the EM spectrum, while scattering by aerosol particles is more prominent toward the shorter wavelength part of the EM spectrum. A notable difference in the two atmospheric path signals over clear and turbid waters results from the high Lpath signal of the hazy atmosphere (especially in the southern part of the scene) (please see Fig. 4a) and low Lpath signal of the clear atmosphere (Fig. 4b).

Figuras 4a and b. Color composite images generated from the Landsat-5 TM bands (band 4=red, band 3=green, band 2=blue), acquired on 5 May 1999 (a) and 7 May 2000 (b), showing dynamic characteristics of turbid waters around the southwest coastal areas of Korea. Top panels indicate reference spectra for clear and turbid waters.


Figures 5a and b. Spectral variation of the atmospheric path-signal (mWcm-2µm-1sr-1 ) over clear (a) and turbid waters (b), during the overpass of Landsat-5 TM on 5 May 1999 and 7 May 2000.

To assess the importance of the SSMM method, the Landsat VIS/NIR imagery (7 May 2000) was converted to radiance (mWcm-2µm-1sr-1 ) at the level of TOA (not shown here). It was observed that even at low SS concentration it never reached the zero value at the near infrared spectral domain. Fig. 6a shows spectral variation of the Lpath signal estimated, using SSMM for the spatially homogeneous atmospheric conditions, over turbid and clear waters. It is apparent that the variation of the Lpath signal over turbid and clear waters is conspicuously less within the coverage of one Landsat scene. Therefore, the mean value of these signals was taken and compared with that of the path-extraction method (Fig. 6b). It should be noted that the high values of Lpath signal associated with path-extraction might introduce a significant error in the retrieval of water-leaving radiance for clear waters, as a consequence of the assumption of black ocean (i.e. LTOA = Lpath + td Lw;td = 1, Lw = 0). Despite of a lack of number of in-situ observations, validation of the SSMM was performed with the help of in-situ match-up data of upwelling radiance measured during our field campaign in May 2004. Consistent with what is seen with the example from the Jin-do and Wan-do bays, there is a very good correlation between in-situ and retrieved Lw spectra from clear waters. As seen in Fig. 7a, under the clear atmospheric conditions deep 6 blue waters scatter light strongly, particularly at the blue wavelength, exhibiting a very steep slope in spectra towards the shorter wavebands of the TM imagery. The steepness lessens for clear waters (Fig. 7b) and a noticeable decline in Lw spectra is evident at this waveband, with a peak at 560nm, when turbidity increases (Fig. 7c). This peak is however no longer stable, and essentially shifts from 560nm towards 660nm, when the level of turbidity remarkably increases due to the occurrence of strong tidal currents and bottom circulations (Fig. 7d). It is thought that a significant discrepancy between the in-situ and retrieved Lw spectra, particularly for turbid waters, is the result of sub-pixel variability of SS concentration and possibly of the surface adjacency effects, which are caused by the interference of multiply scattered signal from neighboring pixels (Liang, 2001).

Figures 6a and b. Spectral variation of the atmospheric Lpath signal (mWcm-2µm-1sr-1 ) over clear and turbid water atmospheres, during the Landsat-5 TM overpass on 7 May 2000.


Figures 7 a-d. Comparison of in-situ ASD radiometric measurements of water-leaving radiance (mWcm-2µm-1sr-1 ), collected during May 2004, with the spectra retrieved from Landsat-5 TM imagery using the SSMM method.

CONCLUSION

New and more efficient methods, to perform atmospheric correction of high spatial resolution Landsat VIS/NIR imagery and high spectral resolution SeaWiFS ocean color imagery for turbid waters, have been described and tested. Inadequate ground information and standard atmospheric models by the 6S model, misled the retrieved water-leaving radiance to be severely overestimated in the visible bands, as the consequence of essentially underestimation of the path reflectance. The remote reflectance spectra by this algorithm are not comparable with the in-situ spectra. The very high signal from turbid waters undergoing interference by adjacency signals and complex multiple interactions with the aerosol particles in the atmospheric system also result in the introduction of significant errors in the atmospheric correction of the ocean color imagery over turbid waters. Perhaps the classical AC algorithms developed for Case-1 waters need to be treated separately by taking into account the optical properties of turbid waters. In contrast, the path extraction relying on the image itself performs reasonably well for turbid waters, but yields significant errors for clear waters due to invalid assumption of black ocean radiance (as the consequences of typical excessive aerosol path radiance removal). On the other hand, the SSMM seems to be a practical approach because it relies on match-up in-situ data and the image itself. A preliminary validation of the SSMM made by the comparison of TM or SeaWiFS-derived water-leaving radiance spectra with in-situ measurements for highly turbid waters of the Jin-do and Wan-do bays suggests that, a good reproduction of the water-leaving radiance spectra by this method is possible, with less errors for both clear and turbid waters. Unlike other AC algorithms or methods, the SSMM considers typical clear and turbid water spectra to derive a constant value of the path signal, which is to be extrapolated over the entire sub-scene of interest towards the retrieval of desired water-leaving radiance. From this study, it was found that these two methods described in the context of the correcting atmospheric effects in the Landsat VIS/NIR and SeaWiFS imagery are also highly applicable to other satellite-based (for example, MERIS, MODIS and POLDER) and probably also, airborne ocean color sensors because they do not require number of bands used in the atmospheric correction and modeling of water-leaving radiance at the two NIR wavebands and modeling of DOM absorption at the near ultraviolet band. In the context of chlorophyll retrieval in turbid waters, an atmospheric correction, based on SSMM, is highly complementary to an ocean color model based on red bands and exploiting the chlorophyll absorption peak near 670nm, the chlorophyll fluorescence at 685, or ratio of both (Gower and Borstad, 1990).

REFERENCES

Antoine, D. & A. Morel, 1999. A multiple scattering algorithm for atmospheric correction of remotely sensed ocean color (MERIS instrument): principle and implementation for atmospheres carrying various aerosols including absorbing ones, International Journal of Remote Sensing, 20, 1875-1916. [         [ Links ] 1]

Brivio, P.A., C. Giardiano & E. Zilioli, 2001. Determination of concentration changes in Lake Garda using image-based radiative transfer code for Landsat TM images, International Journal of Remote Sensing, 22, 487-502. [         [ Links ] 2]

Gordon, H.R., 1978. Removal of atmospheric effects from satellite imagery of the oceans, Applied Optics, 17, 1631-1636. [         [ Links ] 3]

Gordon, H.R. & M. Wang, 1994. Retrieval of water-leaving radiance and aerosol optical thickness over the oceans with SeaWiFS: a preliminary algorithm, Applied Optics, 33, 443-452. [         [ Links ] 4]

Gower, J. F. R. & G.A. Borstad, 1990. Mapping of phytoplankton by solar solar-stimulated fluorescence using an imaging spectrometer, International Journal of Remote Sensing, 11, 313-320. [         [ Links ] 5]

Hall, F.G., D.E. Strebel., J.E. Nickeson & S.J. Goetz, 1991. Radiometric rectification: Toward a common radiometric response among multidate, multisensor images, Remote Sensing of Environment, 35, 11-27. [         [ Links ] 6]

Liang, S., H. Fang and M. Chen, 2001. Atmospheric correction of Landsat ETM+ land surface imagery-Part 1: Methods, IEEE Transaction on Geoscience and Remote Sensing, 39, 2490-2498. [         [ Links ] 7]

Richter R., 1996. A spatially adaptive fast atmospheric correction algorithm, International Journal of Remote Sensing, 17, 1201-1214. [         [ Links ] 8]

Ruddick, K.G., F. Ovidio & M. Rijkeboer, 2000. Atmospheric correction of SeaWiFS imagery for turbid and inland waters, Applied Optics, 39, 897-913. [         [ Links ] 9]

Tanre, D., P.Y. Deschamps., C. Devaux & M. Herman, 1998. Estimation of Saharan aerosol optical thickness from blurring effects in Thematic Mapper data, Journal of Geophysical Research, 93, 15955-15964. [         [ Links ] 10]

Vermote, E.F., D. Tanre., J.L. Deuze., M. Herman & J.J. Morcrette, 1997. Second simulation of the satellite signal in the solar spectrum, AN overview, IEEE Transactions on Geoscoence and Remote Sensing, 35, 675-686. [         [ Links ] 1]