"Mapping total suspended matter from geostationary satellites: a feasibility study with SEVIRI in the Southern North Sea" a

Geostationary ocean colour sensors have not yet been launched into space, but are under consideration by a number of space agencies. This study provides a proof of concept for mapping of Total Suspended Matter (TSM) in turbid coastal waters from geostationary platforms with the existing SEVIRI (Spinning Enhanced Visible and InfraRed Imager) meteorological sensor on the METEOSAT Second Generation platform. Data are available in near real time every 15 minutes. SEVIRI lacks sufficient bands for chlorophyll remote sensing but its spectral resolution is sufficient for quantification of Total Suspended Matter (TSM) in turbid waters, using a single broad red band, combined with a suitable near infrared band. A test data set for mapping of TSM in the Southern North Sea was obtained covering 35 consecutive days from June 28 until July 31 2006. Atmospheric correction of SEVIRI images includes corrections for Rayleigh and aerosol scattering, absorption by atmospheric gases and atmospheric transmittances. The aerosol correction uses assumptions on the ratio of marine reflectances and aerosol reflectances in the red and near-infrared bands. A single band TSM retrieval algorithm, calibrated by non-linear regression of seaborne measurements of TSM and marine reflectance was applied. The effect of the above assumptions on the uncertainty of the marine reflectance and TSM products was analysed. Results show that (1) mapping of TSM in the Southern North Sea is feasible with SEVIRI for turbid waters, though with considerable uncertainties in clearer waters, (2) TSM maps are well correlated... Mapping total suspended matter from geostationary satellites: feasibility study with SEVIRI in the Southern North Abstract: Geostationary ocean colour sensors have not yet been launched into space, but are under consideration by a number of space agencies. This study provides a proof of concept for mapping of Total Suspended Matter (TSM) in turbid coastal waters from geostationary platforms with the existing SEVIRI (Spinning Enhanced Visible and InfraRed Imager) meteorological sensor on the METEOSAT Second Generation platform. Data are available in near real time every 15 minutes. SEVIRI lacks sufficient bands for chlorophyll remote sensing but its spectral resolution is sufficient for quantification of Total Suspended Matter (TSM) in turbid waters, using a single broad red band, combined with a suitable near infrared band. A test data set for mapping of TSM in the Southern North Sea was obtained covering 35 consecutive days from June 28 until July 31 2006. Atmospheric correction of SEVIRI images includes corrections for Rayleigh and aerosol scattering, absorption by atmospheric gases and atmospheric transmittances. The aerosol correction uses assumptions on the ratio of marine reflectances and aerosol reflectances in the red and near-infrared bands. A single band TSM retrieval algorithm, calibrated by non-linear regression


Introduction
Since the launch of SeaWiFS in 1997 and MODIS-AQUA and ENVISAT-MERIS in 2002, ocean colour data from sensors on polar-orbiting platforms have become an established source of information for monitoring of chlorophyll a and total suspended matter (TSM) in coastal waters ( [1][2][3]). These sensors provide data in mid-latitudes with typical space and time resolution of about 1km and 1 day in cloud-free periods. While these sensors give an enormous advantage in terms of spatial coverage when compared to in situ measurement techniques, cloudiness is a severe restriction in many regions. The daily revisit may also be a limitation in coastal waters with high frequency dynamics related to tide or wind-driven advection, resuspension or mixing/settling. For these two reasons the much higher temporal resolution that can be achieved from a geostationary platform is very attractive The technology required for building geostationary platforms for optical sensors is wellestablished thanks to the METEOSAT and GOES (Geostationary Operational Environmental Satellite) series of meteorological sensors which have been operational since the 1970s. However, the higher orbit required for geostationary platforms increases greatly the cost as compared to polar-orbiting platforms and gives a reduction in spatial resolution for the same optical system. Notwithstanding the cost issue, and possible problems related to viewing geometry, the potential for geostationary ocean colour is theoretically very high ([4]) and plans for launching such sensors are at various stages of development within national and international space agencies ([5,6]). This study provides a proof of concept for TSM mapping from geostationary satellites in turbid coastal waters.
Although dedicated geostationary ocean colour sensors have not yet been launched into space, it is possible to test the application of TSM mapping in turbid waters with the existing SEVIRI sensor on the METEOSAT Second Generation platform (MSG). SEVIRI, with its few and broad spectral bands in the visible and near-infrared, is not capable of detecting chlorophyll. However, it has been established previously for the Advanced Very High Resolution Radiometer (AVHRR) sensor ([7]) that the use of a single broad red band, combined with a suitable near infrared band for aerosol correction, is sufficient for quantification of TSM in turbid waters. In fact, TSM concentration can be considered as quite an "easy" parameter to retrieve because of the strong signal: turbid regions can be identified from top of atmosphere radiance ([8]) or even in photographs taken from space ([9]). Mapping of TSM is of interest in many turbid regions because of its link with sediment transport problems (dredging, dumping, geomorphology) ([10]) and because of its impact on the availability of light for primary production ([11]). Remote sensing algorithms for TSM retrieval are now quite mature and satellite data is becoming more and more used in coastal monitoring and as support for modelling ([12]). TSM has therefore been adopted as the output parameter for this feasibility study.
The specific objectives of this study are to test the feasibility of mapping TSM in the Southern North Sea using the SEVIRI sensor and to determine whether high frequency  The normalized spectral response function, ω(λ), for the 4 visible/near infrared bands of SEVIRI is shown in Fig. 2

Study area: SEVIRI subscene
The northern boundary of the study area corresponds to a maximum SEVIRI satellite viewing zenith angle of 64° and is shown in Fig. 3. For a location at the top of this subscene, total airmass (defined as θ is the viewing zenith angle and 0 θ is the sun zenith angle) ranges between 3.4 and 5.4 over the timeframe 06:00-18:00UTC on June 30 th 2006 (see Fig. 4). Rayleigh reflectances start to dominate the water-leaving signal for larger airmasses, rendering data unusable for ocean applications. The area of the Southern North Sea and Western Channel chosen for this study is a relatively shallow sea region, with water depth mostly <50m, and is subject to strong semidiurnal tidal currents e.g. with typical amplitude of 1 m/s. Suspended particulate matter originates from a variety of sources ([15]) including river discharges, inflow from the Atlantic Ocean ([16]), coastal and sea bottom erosion, atmospheric dust, primary production and dredging and mining operations. In the shallower near shore regions, tide and wind resuspension of bottom sediments is particularly important giving high TSM concentrations, even exceeding 100 mg/l, e.g. in some Belgian nearshore waters. The deeper offshore waters have generally much lower concentrations, down to 2 mg/l or less. Most of the region is vertically well-mixed except for some haline stratification along the Dutch coast and some thermal stratification in summer at the Northern limit of the domain ([17]). For most of this region TSM is not highly correlated with chlorophyll a -see Fig. 2

Atmospheric correction of SEVIRI imagery
Solar radiation reflected by the ocean-atmosphere system to the SEVIRI sensor does not come uniquely from the sea. Depending on wavelength, geometry and water turbidity, between 35% and 95% of the measured light flux can have an atmospheric origin due to light scattering by air molecules and aerosols. These atmospheric influences need to be eliminated from the total signal at the top of the atmosphere (TOA) to obtain the marine reflectance ( w ρ ), which is defined by multiplying the above water upwelling radiance 0 w L + by π and normalizing by the above-water downwelling irradiance 0 Total reflectance at TOA ( TOA tot ρ ) is obtained through calibration of SEVIRI level 1.5 data.
Count data ( K ) are transformed into total radiance at TOA, TOA tot L [in Wm -2 sr -1 µm -1 ] by applying the calibration coefficients ( f c ) and offset values ( 0 r ), provided with the SEVIRI level 1.5 native format file headers (see Table 1), as follows ([14]): where 0 λ is the band central wavelength.
Total radiance at TOA, TOA tot L is then converted to total reflectance at TOA ([14]): where d is the Sun-Earth distance in astronomical units ( E is the extraterrestrial solar irradiance at TOA at 1AU (in Wm -2 µm -1 ). Calibration correction factors for the SEVIRI solar channels, A 0 , provided in [24], are applied to TOA tot ρ (see Table 1).
where TOA r ρ is the contribution to the reflectance from scattering by air molecules, TOA a ρ is the reflectance resulting from scattering by aerosols and from multiple scattering by aerosols and molecules. Both terms include photon interactions with the air-sea interface and are influenced by gaseous absorption. TOA wc ρ is the reflectance from foam and white caps and TOA g ρ is the contribution from sun glint for which the only scattering event is specular reflection of direct sunlight at the air-water interface. Sun glint reaches the sensor only for viewing zenith and 75° and v θ ranges between 55° and 64° over the entire study area and in the timeframe 08:00-16:00UTC. Measurements are clearly made outside the directions contaminated by sun glint and therefore TOA g ρ can be neglected. TOA wc ρ is small for wind speeds lower than 10 m/s and largely corrected for in the aerosol correction for maritime aerosols ([26]). The difference between the spectral variation of white caps ([27]) and aerosols gives an uncertainty on the marine reflectance of maximum 0.0009 for wind speeds of 10m/s. 0 T and v T are the sun-sea and sea-sensor atmospheric transmittances, respectively. It is assumed that the effects of atmospheric gases (ozone, water vapour, methane, carbon dioxide), aerosols and air molecules, can be treated separately by decomposition of 0 T and v T according to: are the sun-sea (sea-sensor) atmospheric transmittance factors for aerosol and Rayleigh effects and 0 for absorption by atmospheric gases. Thus correcting Eq. (4) for the two-way gas transmittances and Rayleigh transmittances (assuming that marine aerosols are generally low in the atmosphere) gives: For notational simplicity, the prime symbols and TOA notation are dropped hereafter. The gaseous absorption correction is performed using the Msixs software package, based on 6S code by [28]. This allows the computation of the transmittances of ozone, and water vapour in the VIS0.6 and VIS0.8 bands and for CO 2 and CH 4 in the NIR1.6 band.
The correction for absorption by water vapour, which is strong in the VIS0.8 band (see Fig. 2), is done using the precipitable water content (PWC) obtained from the National Weather Service's National Centers for Environmental Prediction (NCEP) meteorological data. PWC can be quite variable during the day, therefore, PWC data is obtained at four times during the day at 0:00, 6:00, 12:00 and 18:00 UTC and linearly interpolated in between. Ozone column content is obtained from AIRS (1.0° x 1.0° daily Level-3 product) through NASA's Giovanni application, and daily averaged over the study area. CO 2 and CH 4 concentrations are obtained from climatological values for these gas vertical profiles (Mid latitude summer atmospheric model, [29]). Then, the 6S software is used to simulate the transmittances of these two gases for different airmasses. Finally a second order polynomial interpolation of the transmittances as a function of airmass is applied to derive the CO 2 and CH 4 transmittances within the field of view of SEVIRI.
The Rayleigh scattering component, r ρ , and two-way Rayleigh transmittances, 0 r r v t t are calculated from geometry (sun and sensor zenith and azimuth angles), atmospheric pressure and wind speed using lookup tables constructed from the Successive Order of Scattering code ([30]). Atmospheric pressure and wind speed are spatially averaged over the study area from 6-hourly NCEP atmospheric pressure data. A daily mean pressure is computed, while a temporal nearest neighbour is taken for wind speeds. The Rayleigh corrected reflectance is defined as: To obtain w ρ , the aerosol reflectance a ρ and the two-way aerosol transmittances, 0, remain to be computed. a ρ can be computed wherever the water reflectance is negligible, i.e.
typically in the clearest case 1 waters. However, for other waters further knowledge or assumptions are required regarding the spectral behaviour of a ρ , which varies with aerosol size and refractive index. To obtain w ρ in the VIS0.6 band, ) 6 . 0 ( w ρ , two approaches are considered using either the VIS0.8 or NIR1.6 band to correct for aerosol scattering. This is discussed in Appendix A and shows that the (VIS0.6, VIS0.8) band pair minimizes the uncertainty on the desired ) 6 . 0 ( w ρ . This band pair is therefore selected for the processing described hereafter. The relations and unknown quantities are, after Eq. (6): ) with 6 unknowns. To close the system, further assumptions regarding the spectral shape of aerosol and marine reflectances are needed. In this study the following assumptions are made: 1. The ratio of marine reflectances in bands VIS0.6 and VIS0.8 is assumed to be a constant value in space and time: A similar assumption was made previously for the SeaWiFS 765nm and 865nm bands ([32]). For the SEVIRI bands this assumption is less valid for very turbid waters. The consequences for the TSM products are discussed later.
2. The ratio of aerosol reflectance in bands VIS0.6 and VIS0.8 is assumed to be spatially homogeneous over the subscene of interest:  The ratio of aerosol reflectance (6,8) ε is obtained on an image-by-image basis from the ratio of Rayleigh and gas corrected reflectances for SEVIRI bands VIS0.6 and VIS0.8 over clear waters pixels (shown in the red polygons in Fig. 3). The distribution of these reflectance ratios is approximately normal (see example in Fig. 6). For every SEVIRI scene, (6,8) ε is estimated automatically from the mean and the uncertainty on (6,8) ε is quantified from two times the standard deviation (see Appendix A). Fig. 6 shows some discretisation effects coming from digitisation of the top of atmosphere signal which is of the order 0.0013 for the VIS0.6 channel and 0.0015 for the VIS0.8 channel. For every SEVIRI scene, the aerosol  where the VIS0.6:VIS0.8 ratio of two-way aerosol transmittances (γ).is given by: Using Eq. (13) and Eq. (10), the system (Eq. (7)-(8)) can now be rewritten:  Fig. 7. Radiative transfer simulation with Hydrolight 4.2 gives a reflectance at 0.635µm of 0.00028 for pure sea water with no suspended particles, implying a theoretical offset of B=-0.07 mg/l. The non-zero offset found above is specific to the in situ measurements used for calibration. Since any offset for the SEVIRI instrument and processing is likely to be very different from the in situ measurement bias, the algorithm is applied here without the offset B.
For some pixels in clear waters, (0.6) w ρ was found to be negative (up to -0.01). This is thought to be typical of the uncertainty of the method, including all aspects of atmospheric correction and sensor calibration. For these pixels, TSM concentration was set to 0 mg/l.

Estimation of uncertainties on the derived products
As derived in Appendix A, the uncertainty on the marine reflectance in the VIS0.6 band associated with the uncertainties on the key assumptions of spatial homogeneity of VIS0.6:VIS0.8 band ratios of aerosol and marine reflectance (Eq. (9) and Eq. (10)) is given by Eq. (A4): The important contributions to this uncertainty can be seen more clearly by simplifying with  The uncertainty on TSM, due to uncertainties in the aerosol correction, as derived in Appendix A (Eq. (A11)) is given by:

Processing steps: from SEVIRI Level 1.5 data to TSM concentration
The atmospheric correction is schematically depicted in Fig. 8 and can be summarized as follows: 1. Calibrate SEVIRI level 1.5 data to obtain TOA tot ρ in VIS0.6 and VIS0.8 SEVIRI channels from Eq. (2) and Eq. (3) 2. Correct for gaseous absorption, Rayleigh transmittance (Eq. (5)) and Rayleigh scattering (Eq. (6)) 3. Compute the calibration parameter (6,8) ε from the ratio of Rayleigh and gas corrected reflectances for SEVIRI bands VIS0.6 and VIS0.8 over clear waters pixels in every scene (Eq. (10)), estimate   Fig. 9 shows the uncertainty on the derived TSM concentration associated with uncertainties in the atmospheric correction (obtained from Eq. (24)). TSM concentration ranges between 0 and 34mg/l, with 14 out of 8704 pixels exceeding concentrations of 23mg/l. For these pixels the assumption on the ratio of marine reflectances in bands VIS0.6 and VIS0.8 is less valid (see Eq. (9) and Fig. 5). Absolute uncertainties in TSM concentration (see bottom panel in Fig. 9) range between 0.03 and 9.6mg/l. Uncertainties are highest near clouds and near the coast, due to strong atmospheric turbidity (high aerosol scattering). For turbid waters with TSM concentrations above 10mg/l, relative uncertainties range between 4 and 51% (108 pixels, 2 outliers removed, with a mean uncertainty of 20±10%). Moderately turbid waters, with TSM concentrations between 5 and 10mg/l (167 pixels, 2 outliers removed), are mapped with an average accuracy of 32±18%, ranging between 2 and 86%. Zero TSM concentrations were found for 3274 pixels, with absolute uncertainties between 0.4 and 5.2mg/l. For the clearer waters, with TSM concentrations between 0 and 1mg/l (3705 pixels), relative uncertainties generally exceed 100% and such concentrations must be considered as below the detection limit of this method.  The TSM map given in Fig. 9 shows considerable spatial noise in the low concentration regions. This was traced to corresponding digitisation noise in the top of atmosphere reflectance data, which was of the order of 0.0013 for the VIS0.6 channel. This can be translated approximately to noise in the TSM map by multiplying by the factor A/C, giving 0.305mg/l. Fig. 10 shows a spatial comparison between SEVIRI TSM and MODIS TSM maps, acquired simultaneously on July 18 th 2006. These products show comparable spatial patterns and comparable TSM concentration ranges but the SEVIRI image is clearly affected by spatial noise and lacks the spatial resolution necessary to pick up the finer spatial structures seen in the MODIS image. A cross-validation analysis was performed between corresponding TSM products and the resulting regression is shown in Fig. 11. The error bars for MODIS TSM show the standard deviation from the spatial mean obtained from downsampling the MODIS pixels to the SEVIRI grid. The error bars for SEVIRI TSM show the estimated uncertainty on the TSM value as obtained from Eq. (24). A good correlation was found with a correlation coefficient of 0.91 and a coefficient of determination, R², of 0.83. Similar correlation coefficients were found for the other 3 match-ups on June 29 th at 12:45 UTC, July 15 th at 12:45 UTC and July 17 th at 12:35 UTC (see Table 2). It is noted that the slope of the regression line is close to one and that the offsets are close to zero. Discrepancies between the two datasets could be caused by differences in atmospheric correction, differences between SEVIRI and MODIS AQUA bandwidth and central wavelengths and differences in spatial resolution.

High frequency TSM dynamics
The high frequency variability of TSM concentration is studied on a cloudfree day on June 29th 2006 at five different pixels/stations (shown in the top panel of Fig. 9)  For P1, TSM concentrations are close to zero. The absolute uncertainty on these low concentrations ranged between 1.0 and 3.5 mg/l, with a mean of 1.5±0.6mg/l. Uncertainties on TSM concentrations for P2 ranged between 12 and 29%, while TSM concentration varies by a factor 2 over an 8h period. Between 10:00 and 14:15 UTC, relative uncertainties for P3 ranged between 16 and 34%, but increased above 45% outside that timeframe due to atmospheric turbidity. For P4, relative uncertainties of 19 to 96% were found. Relative uncertainties ranged between 39 and over 100% at P5. The variability of TSM concentration in the coastal pixels P2 and P3 suggests a periodicity related to the tidal cycle, such as could be expected from resuspension/settling and/or advection. At the other stations, possible diurnal variability cannot be discerned due to the high noise level. The period and the spatial variability of the amplitude and phase of this high frequency variability in turbid waters will be analysed with a more complete dataset and in conjunction with hydrodynamical model results in a future study.

Conclusion and future prospects
Though the geostationary meteorological SEVIRI radiometer, with its few and broad spectral bands, high radiometric noise level and medium ground resolution, is not designed for ocean colour remote sensing, this study demonstrates the feasibility of this sensor for TSM concentration mapping in the Southern North Sea. It is shown that TSM products from SEVIRI are highly correlated with similar data from MODIS. Also, SEVIRI TSM uncertainties in turbid waters are comparable to MODIS TSM uncertainties, though, in clear waters SEVIRI TSM uncertainties are very high due to radiometric noise and/or digitization effects.
The striking advantage of SEVIRI is the superior temporal resolution: every 15 minutes compared to once per day for MODIS. This study provides the basis for future mapping of TSM from geostationary sensors at very high temporal resolution. Important applications of the availability of very high temporal resolution TSM maps include: 1. Mitigating the problems of cloudiness encountered with the current generation of polar-orbiters (MODIS, MERIS, SeaWiFS), through daily averaging of the SEVIRI images in areas of low TSM variability.