ATMOSPHERIC CORRECTION OVER COASTAL AND INLAND WATERS:

DOING BETTER WITH MERIS?

Richard Santer and Jérôme Vidot

MREN, Université du Littoral Côte d'Opale, 32 avenue Foch,62930 Wimereux, France,

Email:

Abstract

This work first illustrates the potential use of the oxygen absorption in improving the atmospheric correction over water. First, this oxygen absorption can be used to better flag cloud contamination, mainly with cirrus clouds. It can be used as well to better flag the sunglint complementary to the current radiometric classification. After flagging these anomalous circumstances, we propose simple methods to remove both the contribution of the cirrus clouds and of the sunglint. It appears more to be new developments rather than new operational methods. But the recommendations to conducts such studies are straightforward. First of all the simplified “6S” like formulation of the signal should be validated by reference to “exact” computation. Then, this formulation has to be applied on an operational basis at least to remove the Rayleigh contribution. The remaining will inform on the effective altitude of the scatters. Regular values will inform on the vertical distribution of the aerosols, higher values will flag cirrus clouds and lower values will reflect the sunglint contribution. Then a correct identification will offer possibilities to correct for cirrus clouds as well as for sunglint.

MERIS full resolution imagery offers the possibility to do water quality monitoring near to the coast line and or in inland waters. This opportunity raises first the problem of the adjacency effects. We simply illustrate this effect knowing than a theoretical formulation of the problem exits. For inland waters, both land and water algorithms exist to remote sense the aerosols. We propose simple comparison between the two as indicators of the quality of the atmospheric correction.

1INTRODUCTION

There are more and more demand to include Earth Observation in water quality monitoring. Traditional issues are: algae bloom detection, water eutrophication, sediment transport,…The European Community water directive impulses new monitoring activities close to the coastline or for inland waters. MERIS, in the full resolution mode, and thanks to it very good radiometric quality, is a potential complementary tool for monitoring. Level-2 MERIS products are quite suitable but there are room for improvements. The starting point to use the ocean colour products (as they are and/or through the development of new algorithms) is to achieve good atmospheric corrections. Several papers in this ENVISAT meeting are devoted to this task. Major problems are the uncertainty on the aerosol type through the necessary use of standard aerosol models [1]. Monitoring coastal waters North of Europe implies that the algorithms have to be improved to account for large solar zenith angles [2]. The contamination by the sunglint is also a major problem. Thanks to it multi spectral possibilities, the MERIS algorithm enables to decontaminate in the red-NIR (near infrared) spectral domain the atmospheric signal from the residual component of the turbid coastal waters. For inland water, this possibility improves the atmospheric correction scheme as well.

The MERIS bands at 753 nm and 761 nm are devoted, through a two band ratio technique, to derive the pressure at cloud top height [3] or, under clear sky conditions over land, to derive the surface pressure [4]. Over ocean, we know, thanks to meteorological sources, the pressure at sea level. But the 761/753 ratio can be use as well to detect cirrus clouds and variation of this ratio occur in the sunglint [5]. We want here to investigate this possibility in qualitative terms as well in quantitative values.

We then address the specific task of atmospheric correction over inland waters. A major problem is the so-called adjacency effect. Numerous illustrationsof this effect exist for MERIS [6,7]. Contrast between land and water is very high at 865 nm which implies a strong adjacency effect. The 865 nm band is used for aerosol remote sensing. Conversely, the water-land contrast is less pronounced in the visible part. Knowing correctly the aerosol model, the adjacency effects will therefore not impact much the atmospheric correction. Also, the aerosols over land are remote sensed over the dense dark vegetation (DDV) in a spectral domain where the land-water contrast is weak. Therefore, a compromise can be then: (i) use the land aerosol model around the lake and (ii) apply with this model the regular atmospheric correction over water [8]. This solution skips the traditional adjacency effect problem.

2THE QUALITATIVE USE OF THE MERIS OXYGEN BAND

2.1Ocean Level-2 products in the presence of cirrus clouds

Clouds are identified trough a simple radiometric threshold. Cirrus clouds are therefore difficult to detect. Figs. 1(a) and 1(b) illustrate the possibility to detect thin cirrus clouds. But, when getting thinner in the northern part of the image (above Holland), they are no longer flagged. Fig. 1(c) is a MERIS cloud classification image with a bright pixel flag in yellow. Over land, the surface pressure is a MERIS Level-2 product derived from the oxygen absorption band. We also have the meteorological pressure first at sea level and second at surface level thanks to a digital elevation map. If the O2 surface pressure is lower by 40 hPa compared to the meteorological pressure, we raise a flag which indicates thin and high clouds, or in other words, cirrus clouds. This additional flag in black clearly illustrates the insufficiencies of the regular flag. They are no O2 surface pressure over ocean: the ocean is quite black around 760 nm and therefore the satellite signal does not provide any information on the surface. But instead, the two band ratio 761/753 may give potential information on the effective altitude of the atmospheric scatters.

(a) (b) (c)

Fig. 1:Cloud classification. (a) SeaWIFS RGB image over western Europe on 03 September 1999. (b) SeaWiFS classification, white for clouds; (c) MERIS level 2classification: yellow (standard) and black (using pressure product).

Figure 2 is another MERIS scene on which we clearly see cirrus clouds between Tunisia and Sicilia. The exploitation of MERIS Level-1B on two transects, Fig. 3, illustrates the deep increase of the atmospheric radiance when crossing the cirrus clouds. The 761/753 ratio increases because of the high altitude presence of scattering elements. The increase of the aerosol loading and/or the presence of a haze at sea level will not result in an increase of the 761/753 ratio.

The impact of the presence of cirrus clouds is illustrated in Fig. 4. The direct effect is to substantially increase the AOT and indirectly, the chlorophyll content. The correlation between the two products is very high.

2.2 Ocean Level-2 products in the presence of sunglint

MERIS sunglint flag was used to select on scene over the Mediterranean Sea (Fig. 5). On transects 1 and 2, the presence of sunglint corresponds to an increase of the TOA radiance (see red lines on Fig. 6 corresponding to the TOA radiance at 865 nm). At 761 nm, outside of the sunglint, the sea is black and therefore, the signal is purely atmospheric. Still, like in section 2.1, the 761/753 band ratio informs on the mean altitude of scatters. Conversely, the signal for high sunglint mostly originates from the surface and the oxygen absorption is higher. Schematically, the O2 absorption corresponds to the atmospheric scattering outside of the sunglint and gives the O2 transmittance at sea level when the sunglint is high.

Fig. 2. MERIS image with cirrus cloudsbetween Tunisia and Sicilia on the 04 February 2004. / Fig. 3. Radiance at 778 nm and 865 nm on the two transects and oxygen transmittance in percent.
Fig. 4. AOT at 865 nm and Chlorophyll-a on the two transects.
Fig.5. MERIS image of the 761/753 ratioon sunglintover LampedusaIsland on 16 July 2003. / Fig. 6. Radiance at 865 nm (red lines) and oxygen transmittance in percent (yellow lines)on the two transects of Fig. 5.

3 THE QUANTITATIVE USE OF THE MERIS OXYGEN BAND

3.1A 5S-likeformulation for the clear sky

In absence of O2 absorption and on the outside of the sun glint region, the reflectance  at TOA is derived from the 5S code[9] as:

(1)

where a and m represent respectively the aerosol and the molecular path reflectance. The surface reflectance is considered as null for the ocean. s and vare respectively the solar and the view zenith angle, and  the relative azimuth angle between the illumination and viewing directions. To simplify the notations in the equations hereafter we will omit the angular dependencies.

Similarly, we propose a simple formulation of the TOA signal which accounts for the oxygen absorption:

(2)

in which we introduce the O2 transmittances for the aerosols and for the molecules. These two transmittances will be computed using the primary scattering approximation. Actually this formulation, as reported in Eq. 2, is slightly more complicated than that in order to introduce the coupling terms between atmospheric scattering and Fresnel reflection. Nevertheless, this simplified assumption will not jeopardize what we will develop below.

The Rayleigh can be corrected and equations (1) and (2) become:

(3)

(4)

which gives by single ratio. On a theoretical basis, we can estimate simply having a relative vertical distribution of the aerosols. Then the theoretical O2 transmittance for the aerosols does not depend upon the aerosol model or the aerosol type, neither the aerosol loading.

3.2 A like-5S formulation for the clear sky contaminated by cirrus cloud

First we can useas a cirrus cloud flag. If then we raise the cirrus cloud flag. is computed for an aerosol vertical profile in the lower troposphere which minimized it. This flag has to be consolidated on empirical basis. There are circumstances for which we know that this flag will not work such as transportation of Saharan dust in the upper troposphere.

In absence of O2 absorption, the cirrus cloud contribution is added to the signal:

(5)

where c is the cirrus cloud reflectance. In Eq. 5, the aerosol and molecular signals are not attenuated by the cirrus cloud. The cirrus cloud transmittance is assumed to be 1 which accounts for the predominance of the forward scattering. In other word, the direct solar beam and the atmospheric radiance are attenuated during their respective path through the cirrus clouds, but what is loss in direct extinction is retrieve in the forward peak.

In the oxygen absorption, Eq. 2 becomes:

(6)

in which we introduce the O2 transmittance for the cirrus cloud .can be computed for a standard altitude of the cirrus cloud and does not depend upon the cirrus cloud model nor the type, neither the optical thickness.

The Rayleigh can be again corrected and Eqs. 5-6 become:

(7)

(8)

In Eqs. 7-8, the aerosol and cirrus reflectances are supposed to be constant between 753 nm and 761 nm. If we can reasonably estimate and , Eqs. 7-8 are a simple system from which a andc can be simply derived. Error on the estimate of and will results in random errors on the estimate of a andc while the presence of cirrus clouds is a bias.

Fig. 7. Estimate of the reflectance of the cirrus cloud. / Fig. 8. Mean altitude of the scatters.

The feasibility of such approach has to be demonstrated. Fig.7 gives the idea of the reflectance of the cirrus clouds. Reflectances of 3 points on each transect have been calculated from points represented on Fig. 2. To obtain the reflectance, we use Eq. 7 and we get ρaover the clear sky pixels (from points marked Ref. 1 and Ref. 2 on Fig. 2). Cirrus clouds are white in the near infrared; still white if there are thin enough. For thick clouds, we certainly need to introduce a cirrus cloud transmittance. Aerosol signal was removed at 753 nm and 761 nm above cirrus clouds to get . The oxygen transmittance respectively over the clear skyand over the cirrus clouds was converted in pressure and reported in figure8.in altitude This conversion used the relation ship (mP2 versus ) introduced in the land algorithm. If for the clear atmosphere the pressure is of physical significance, the mean pressure of the cirrus clouds does not agrees with expected values of cirrus cloud altitude of around 10 km. More as to be done and was reason is that the (mP2 versus ) is out of it applicable range. One possible reason is that the clouds identified in Fig. 2 do not correspond to cirrus clouds by to medium altitude clouds. Synergy with a thermal sensor like AATSR may help to cancel ambiguities.

3.3 A like-5S formulation for the clear sky contaminated by sunglint

Let us do first the Rayleigh correction both outside and inside the O2 band.In absence of O2 absorption, the sunglint contribution is added to the signal as:

(9)

where g is the direct to direct sunglint contribution with no atmosphere. This contribution is white. In Eq. 9, the aerosol and molecules, through the total optical thickness , attenuated g.

In the oxygen absorption, Eq. 2 becomes:

(10)

in which we introduce the O2 transmittance at sea level . can be computed for a standard surface pressure or for the estimated surface pressure as provided by the MERIS auxiliary data.

First we can use, as ratio of the terms of Eqs. 9-10, as a sunglint flag. If then we raise the sunglint flag. is computed for an aerosol vertical profile in the lower troposphere which maximized it. This flag has to be consolidated on empirical basis.

If we can reasonably estimate , Eqs. 9-10 are a simple system from which a and g can be simply derived. Error on the estimate of will results in random errors on the estimate of a and g exp(-m) while the presence of sunglint is a bias.

The final objective of this sunglint estimate is to remove the direct sunglint from the signal. What we have is g exp(-m) at 760 nm and what we want is to remove g exp(-m) at 778 nm and 865 nm which are the two key bands to remote sense the aerosols. From g exp(-m), we can extract after correction of the Rayleigh is g exp(-m). At first order, we can neglect the spectral dependence of .

3.4Discussion

Both for cirrus cloud and for sunglint, through the 2 linear system of degree 2 (Eqs. 7-8 for the cirrus clouds and Eqs. 9-10 for the sunglint), we propose a simple approach to correct for the two undesirable contributions.andhave to be known as accurate as possible. On theoretical basis, the first requirement is to accurately measure the oxygen transmittance with severe constrains on the spectral characterization of MERIS and of the so called smile effect.Then, you compute and versus: (i) the geometrical conditions (the total air mass), (ii) standard vertical profiles for the aerosols and (iii) standard mean altitude for the cirrus clouds. Points 2 and 3 require accounting for geographical and seasonal variabilities. Another approach is empirical in which the 761/753 ratio is determine from MERIS versus the above parameters: geometry, time and location. This approach has to be considered because the characterization of the instrument does not to be perfect as well as the theoretical approach.

Fig.9.Right: MERIS spectral response and. The dashed lines correspond to the 10 narrow channels within oxygen absorption bands used for this in-flight experiment. The full lines correspond to the broad channels of the nominal MERIS band setting.

Left: monochromatic atmospheric transmission for the oxygen absorption bands region

Also, this degree 2 system is only well posed if substantially differs from or . Making more sensitive to the three scenarios(clear sky, presence of cirrus cloud, presence of sunglint) can be address in a sensitivity study. The outputs of it should be to propose the best spectral definition for band 11 (at 761 nm). On an experimental basis, MERIS acquired for specific orbits, a specific spectral arrangement, illustrated in Fig. 9, for in flight spectral calibration in the oxygen band. These data can be used to optimize the approach. The use of 2 oxygen spectral bands can be investigated as well.

We report here two examples for which the presence of the oxygen MERIS band may help. This approach can be generalized to other cases in which two sources (Rayleigh scattering excluded) are; located at two different altitudes. The case of transportation of aerosols in the high troposphere is reported in [5]. A significant contribution of the stratosphere by a volcanic eruption is another one.In the MERIS ATBDs, there is a description of a routine which indicates how to correct for the stratospheric aerosols. This routine, implemented in the ground segment, is not activated.

4MERIS OVER INLAND WATERS

4.1Illustration of the adjacency effects

The so-called adjacency effect corresponds to photon reflected and scattered toward the sensor where a substantial contrast exits between the target and it surrounding. One salient condition is observations over waters near by the land, both coastal and inland water. The effect is easier to illustrate in coastal water because it decreases from the coastline to off shore. Also, the better spatial resolution you have the better adjacency effect you see. Unfortunately here, we do not have to present MERIS full resolution image. Nevertheless, even in reduced resolution, we can comment on the adjacency effects. The MERIS image, Fig. 10, generally suggest than the AOT increased in a fringe along the coast line. More explicitly, on a transect perpendicular to the English coast; we can interpret the adjacency effect. The x scale report the distance in pixel (1.2 km by pixel).

Fig. 10. Illustration of the adjacency effect: Left: MERIS AOT at 865 nm (05 august 2003) above South East England. Upper right: AOT at 865 nm on the indicated transect. Lower right: MERIS level 1 reflectance versus wavelength; above land (full line), above water at coast line (small dashed line) and off shore above water (dashed line)

Over land, AOT is about 0.3 and slightly less over the ocean. Two different algorithms operate over land and ocean and the agreement is good. The maximum AOT is just at coastline, and then the AOT sharply decreases within few kilometres. Typical spectral signatures of the MERIS TOA radiances are also reported in Fig. 10. Land becomes brighter in the NIR. Conversely over the ocean, the atmospheric signal increased towards the visible. Just at the coast over water the TOA radiance is white. Addition scattered light is provided by the close land and more and more in the NIR. That explains this spectral behaviour. On the aerosol product, the adjacency effects increases the AOT at 865 nm. For the Angström coefficient the impact is not so high because the land surface reflectance does not substantially differ between 778 nm and 865 nm. The photons reflected by the land are simply an additional white source to the direct solar beam.