22.214.171.124 Level 2 Algorithm Description
This chapter provides a description of the MERIS processing in terms of algorithms. It describes data processing to be applied to the MERIS pixels, in order to derive the MERIS Level 2 Reference Products, in Reduced Resolution as well as in Full Resolution. For a more detailed information, please refer to the “MERIS level 2 Detailed Processing Model”.
The Level 2 processing is in charge of processing TOA radiance measurements into geophysical parameters.
These parameters depend on the observed pixel (water, land, cloud) and provide information on:
· the surface properties:
normalised reflectance at surface, chlorophyll and other water constituent concentration (ocean); reflectance at surface, vegetation indices (land);
· the properties of the atmosphere above the surface:
aerosol type and optical thickness, water-vapour column content, cloud top height, optical thickness and albedo.
The general structure of Level 2 processing and products is presented in the following flow chart. The box numbers refer to the different step numbers of the MERIS Level 2 processing breakdown presented. Note that, in the diagram below, the paths labelled in Arial type indicate the main control flow, according to pixel type; those labelled in Times italic indicate the product flow. Please take note of the definitions on this page 2.3.3. . Click on the algorithm name in the diagram for a detailed description of the algorithm.
Figure 2.8 - General structure of Level 2 processing.
The detailed descriptions of the algorithms are contained in the subsections indicated (see index).
126.96.36.199.1 MERIS Pre-processing
188.8.131.52.1.1 Level 1b product check
If one or more band within the Level 1b product is not one of those in table 2-1, the product shall not be processed to Level 2.
If the Solar irradiance in the Level 1B product (GADS scaling factors) is 0 in any band, the product shall not be processed to Level 2.
184.108.40.206.1.2 Pre processing step
Pre-processing for geometry and meteorological parameters (step 2.1.0)
This step is done in order to derive geometry and meteorological parameters (pressure, wind) at each pixel, including invalid ones, from those provided at each tie point of the Level 1b annotation product.
Level 1b pixel classification screening (step 2.1.11)
The Level 2 pixel identification starts with the reading of the INVALID flag of the Level 1b product. If it is set to TRUE, then no further processing of the current pixel is done, the L2 product shall contain fixed values for all MDS (see section 10 below), with the "Invalid" flag set, and the next pixel is examined ; otherwise, the processing of the current pixel is pursued.
Pixel extraction and reflectance conversion (step 2.1.4)
If the Level 1b pixel is not flagged INVALID, the other L1B flags and the Top Of Atmosphere radiances at all bands are extracted from the L1B product. Radiances are converted to reflectances, using the Sun zenith angle cosine interpolated at the pixel and the Sun spectral flux read from the L1B product annotations.
220.127.116.11.2 MERIS Pressure Processing
18.104.22.168.2.1 Atmospheric pressure estimate (steps 2.1.5, 2.1.12)
The pressure is estimated for each pixel from the MERIS bands 10: 753.75 nm and 11: 760.625 nm, following two methods in parallel :
· "cloud top pressure" that uses a Neural Network algorithm, and
· "surface pressure" that uses a polynomial algorithm.
Spatial registration of the instrument has to be taken into account. A spectral shift index is computed from annotations of the L1B product and look-up-tables.
To retrieve the Cloud Top Pressure, Ptop,a neural net (NN) approach is used. The MERIS signals in channel 10, 11, the surface albedo asurf and the geometry (sun zenith angle, viewing zenith angle and azimuth angle) are used as input of the Neural Network. The net produces the cloud Top pressure Ptop. Depending on the surface albedo two different neural nets are used (one for surface albedo equal to zero, one for non-zero surface albedo). Neural Nets are selected according to spectral shift index.
The Neural Nets apply generic Neural Net functions to specific auxiliary parameters and inputs, to obtain the required outputs..
Each pressure estimate produces Product Confidence Data (PCD).
22.214.171.124.2.2 Atmospheric pressure confidence tests (steps 2.1.2)
A surface pressure test is performed (step 2.1.2). This test compares the difference between pressure estimates from MERIS measurements (step 2.1.12, 2.1.5 above) and pressure derived from ECMWF data (stored in Level1 data product annotations) to thresholds (function of qs, qv and surface type - LAND_F). Furthermore, the difference of the two estimates is tested against a confidence interval width.
126.96.36.199.3 MERIS Pixel Identification
188.8.131.52.3.1 Cloud screening (steps 2.1.2, 2.1.7, 2.1.8)
For pixels identified as LAND, tests are performed on the ratio of Rayleigh-corrected reflectance at several wavelengths (step 2.1.7). The coarse Rayleigh correction uses a algorithm to compute the reflectance due to Rayleigh scattering.
This algorithm enables setting the following Boolean parameters:
1. Bright pixel flag: set if the Rayleigh-corrected reflectance at a specified band (nominally 442.5nm) is higher than a threshold : BRIGHT_F;
2. Low pressure estimate from neural net : LOW_POL_F;
3. Low pressure estimate from polynomial :LOW_NN_F
4. Product confidence flag from neural net : PCD_NN_F;
5. Product confidence flag from polynomial : PCD_POL_F;
6. Consistency between the two pressure estimates : P_CONFIDENCE_F;
7. Rayleigh-corrected reflectance ratio 1: SLOPE_1_F;
8. Rayleigh-corrected reflectance ratio 2: SLOPE_2_F;
These parameters are used to index a decision table which provides the CLOUD_F flag (step 2.1.8).
184.108.40.206.3.2 Stratospheric Aerosol Correction (step 2.1.9)
When the switch to perform stratospheric aerosol corrections is set, all valid pixels are corrected for stratospheric aerosol transmission and scattering. Correction applies to all TOA reflectance bands, and to the TOA radiance bands used in further processing steps :
753.75nm; 761.25nm; 775 nm; 865 nm; 885 nm; 900 nm
The correction algorithm is similar to the one described in section ”Atmospheric Correction Over Land” below; the stratospheric aerosol parameters are read from auxiliary parameters sets. The algorithm runs on a 4x4 pixels window.
220.127.116.11.3.3 Gaseous absorption corrections (step 2.6.12)
Gaseous correction processing is organised in three steps, O3, O2 and H2O correction. Input is the TOA reflectance for the MERIS channels, corrected for stratospheric aerosols when applicable: r (b, j, f). Its output is the reflectance corrected for gaseous absorption (rng*). All three algorithms apply polynomial expressions using LUT technique. In step 18.104.22.168 the O3 transmittance is estimated over a 4*4 pixel window. In step 22.214.171.124 and 126.96.36.199 the O2 and H2O transmittances are estimated for each land pixel. Because the signals used are weak, an average of radiances is performed on water pixels to compute the transmittances.
188.8.131.52.3.4 Land Identification (step 2.6.26) and Smile Effect Correction (step 2.1.6)
The purpose of this classification is to identify using geo-physical data, land from water pixels in cases where the Level1b a priori classification leads to ambiguities which may occur from:
· geo-location error;
· land /ocean atlas error: uncharted land or water, etc.;
· transient emerged land: tidal flats, etc.
This cases are identified using the Surface Confidence Map, an atlas identifying zones of low-confidence in the a priori land/water classification map used in the level1b. When the Surface Confidence Map indicates high confidence classification, the Land Identification radiometric tests are by-passed and the a priori classification is kept.
First, a test on the reflectance corrected for gaseous absorption at 665 nm is performed to identify the darkest pixels. The TOA reflectance at 665 nm is compared to a threshold interpolated from a LUT.
For the pixels having a reflectance smaller than this threshold, a second test is made to compare the TOA reflectance at 665 nm with the TOA reflectance at 865 nm ; if the TOA reflectance at 665 nm is greater than the reflectance at 865 nm, the pixel is classified as water.
Land in water
The purpose of this test is to identify pixels of emerged land, flagged as "water" in the L1B product. It is the opposite of the Inland water test.
The purpose of the Smile Effect Correction is to correct TOA reflectance (already corrected for stratospheric aerosols and gaseous absorption) for small scale variations due to non-constant central wavelength of a given band across the field of view. Correction is made only for a subset of bands for which those variations can induce severe distortions after corrections based on fixed wavelength scheme (e.g. Rayleigh diffusion correction). This subset of bands, which is specific to each land and water surface type, should ensure smoothness of reflectance local variations with wavelength and allow a good estimation of the reflectance derivative using neighbour bands.
184.108.40.206.4 Total Water Vapour Retrieval
This algorithm is applied to land, water and cloud pixels. It is based on a differential absorption method using two spectral bands close to each other (one within the absorption band and the other outside the absorption band). The algorithm consists in a polynomial of the logarithm of the ratio of TOA radiance at band 15 (900 nm, within the water vapour absorption region) to TOA radiance at band 14 (885 nm, outside water vapour absorption).
Above land surfaces, TOA radiances are corrected for the spectral slope of the surface albedo prior to applying the algorithm. Above water surfaces, the algorithm takes into account the aerosol optical depth, except when Sun glint is significant.
Polynomial coefficients depend on illumination and viewing geometry, surface type, presence of glint (above water), cloud properties (above clouds), for any given pixel. That dependence is coded in look-up tables.
The spectral bands centred at 885 nm and 900 nm have proven to be the best suited for the retrieval of water vapour over all surfaces. The total water vapour is computed in two steps. In a first step, depending on the type of pixel: land or glint, water, cloud, polynomial coefficients are read from LUT. The ratio of TOA radiances (corrected for stratospheric aerosols if needed) at 885nm and 900nm may be corrected. The second step is a simple polynomial applicable to all surfaces.
220.127.116.11.4.1 Water vapour retrieval over land surfaces (step 2.3.1)
A correction is made for the surface albedo slope between 885 and 900nm. The polynomial coefficients take surface pressure into account.
18.104.22.168.4.2 Water vapour retrieval over water surfaces (steps 2.3.2, 2.3.5)
Outside the Sun glint region, the retrieval of the total water vapour over water surfaces is more difficult than over land surfaces because of the larger influence of aerosols. The polynomial coefficients take into account aerosol influence. Also in order to improve noise performance, 4x4 pixel averaging is performed as a pre-processing step.
In the Sun glint region, the same algorithm as above land is applied.
22.214.171.124.4.3 Water vapour retrieval over clouds (step 2.3.3)
Polynomial coefficients take into account the cloud optical thickness and the albedo of the underlying surface.
126.96.36.199.4.4 Range checks (steps 2.3.0, 2.3.6)
Range checks are performed on radiance at the algorithm input. Out of range radiance result in an exception, water vapour is not processed. When processed, the water vapour is also checked for range, the product is kept but a flag is raised when out of range.
188.8.131.52.4.5 Water vapour polynomial (function)
The algorithm consists in a second-degree polynomial equation using the logarithm of the ratio of TOA radiance at 885 and 900 nm and coefficients. All polynomial parameters are provided by steps 2.3.1, 2.3.2, 2.3.3 depending on the type of pixel.
184.108.40.206.5 Cloud Processing
Pixels output from the Pixel Identification module (step 2.1) and flagged as cloudy (CLOUD_F = true) are processed in order to retrieve the cloud albedo (step 2.4.1) and the cloud optical thickness (step 2.4.3). Cloud top pressure is already known from step 2.1. From the optical thickness and top pressure, the cloud type index is computed (step 2.4.8). It should be noted that cloud reflectance, written in the L2 product, is the TOA reflectance (corrected for stratospheric aerosol if needed) r (b, j, f), computed at step 2.1.
220.127.116.11.5.1 Cloud Albedo processing (step 2.4.1)
The cloud albedo processing relates the cloud albedo ac to the MERIS radiance in channel 10 (753.75 nm) using a polynomial regression technique. The polynomial coefficients are read from a Look Up Table as a function of geometry (sun zenith angle, viewing angle and azimuth angle) and estimated surface albedo.
18.104.22.168.5.2 Cloud Optical Thickness processing (step 2.4.3)
To retrieve the cloud optical thickness tc , the same technique is used as for retrieving the cloud albedo. A polynomial expression relates the cloud optical thickness tc as a function of the MERIS radiance in channel 10. The polynomial coefficients are read from a Look Up Table as a function of geometry and estimated surface albedo.
22.214.171.124.5.3 Cloud type processing (step 2.4.8)
The algorithm uses a simple classification table indexed by the cloud optical thickness and cloud top pressure, to provide a cloud type index.
126.96.36.199.6 Water Processing
The processing of water pixels is intended to provide the following Level 2 products:
· normalised water-leaving reflectance at bands 412.5, 442.5, 490, 510, 560, 620, 665, 681.25, 705, 753, 775, 865, 885 nm
· algal pigment index 1
· algal pigment index 2
· total suspended matter
· yellow substance absorption at 442.5nm
· photosynthetically available radiance (PAR)
· aerosol optical thickness at 865nm
· aerosol Angström exponent (775, 865)
· turbid case 2 water;
· yellow substance loaded case 2 water;
· water with excessive scattering;
· continental absorbing aerosol;
· desert dust absorbing aerosol;
as well as flags relevant to the quality of all products.
188.8.131.52.6.1 Water Confidence Checks (step 2.6.5)
Water confidence checks include processing of Sun glint, and flagging of low pressure and whitecaps. These steps use the wind and pressure provided in the L1B product annotations. All water confidence checks apply to water pixels within a 4x4 pixels window.
184.108.40.206.6.1.1 Glint processing (step 220.127.116.11)
Glint processing applies only to water pixels
· Glint estimation
The Sun glint reflectance rg is calculated (step 18.104.22.168.1) by interpolation in LUT produced using the Cox and Munk model (1954) as a function of geometry, wind speed modulus and direction. An estimate of glint reflectance is produced. Wind speed modulus and wind direction are computed from the W_u and W_v annotations.
Note: Sun glint reflectance is now an input to current step as its computation a been moved to section “Pixel identification” (step 2.1c).
· Glint classification (low, medium or high glint ?)
This glint reflectance is compared to a low glint threshold (step 22.214.171.124.3). If the glint reflectance is below this low glint threshold then no glint correction for this pixel is applied. If the pixel is not bright then it is further processed by step 126.96.36.199. If bright, it is flagged as ice or high aerosol load.
If the glint reflectance is above the low glint threshold then the glint reflectance is compared to a medium glint threshold (step 188.8.131.52.4).
If the glint reflectance is below the medium glint threshold then a medium glint flag is raised and the pixel is corrected for glint reflectance in step 184.108.40.206.7.
If the glint reflectance is above the medium glint threshold then no correction is applied and the pixel is flagged as uncorrected sun glint (step 220.127.116.11.5).
In addition, a specific glint threshold is used to derive a second “high glint” flag dedicated to Water Vapour Processing over water (step 18.104.22.168.1.9): As a matter of fact, when the Glint reflectance is very high, total water vapour over water surface is best determined using the Water Vapour Over Land algorithm, best suited for bright underlying surfaces.
· Glint correction in case of medium glint (step 22.214.171.124.7)
The glint reflectance at surface level is transferred to the Top Of Atmosphere by applying a direct atmospheric transmission term (including Rayleigh scattering on both sun-surface and surface sensor paths). The Top Of Atmosphere glint reflectance rg* is then subtracted from the TOA reflectance.
Note : step 126.96.36.199.8, because of its simplicity, is only shown in the corresponding equation section.
188.8.131.52.6.1.2 Low pressure water flagging (step 184.108.40.206)
Surface pressure (from ECMWF annotation) is compared with a threshold in order to flag low pressure, typically high altitude inland waters. Above such waters, the results of the atmosphere correction are disturbed.
220.127.116.11.6.1.3 Whitecaps Flagging (step 18.104.22.168)
The modulus of the wind speed is compared to a threshold. Above that threshold a whitecap flag is raised because white caps are likely to disturb the performance of the atmosphere correction.
22.214.171.124.6.1.4 Reflectance threshold on reflectance at 412 nm (step 126.96.36.199)
An additional “bright” pixels screening, specific to water pixels, is performed by comparison of the Glint corrected reflectance at 412 nm with a pre-computed threshold. It is intended to further sort out and flag those pixels affected by sea-ice, partial clouds or very high aerosol load.
188.8.131.52.6.2 Turbid water screening and corrections (steps 2.6.8, 2.6.10)
This section describes the algorithms used
1. in step 2.6.8, to detect Case 2 turbid waters based on radiometry (reflectance corrected for stratospheric aerosol, gaseous absorption, Sun glint); in the process, Rayleigh reflectance above water is computed;
2. in step 2.6.10, to compute the water-leaving reflectance for Case 2 turbid waters at 510, 705, 775 and 865 nm, needed before entering the atmospheric corrections processing over Case 1 waters (step 2.6.9) and provide an estimate of the total suspended matter used in turn to identified sediment dominated case 2 waters through a dedicated flag.
184.108.40.206.6.2.1 Rayleigh correction 1 (step 220.127.116.11)
This correction requires to estimate the Rayleigh reflectance in all useful bands, which will be re-used for further ocean pixels processing. The Rayleigh reflectance is interpolated in a LUT (RD 7, 3.6.1) as a function of pixel geometry, wind speed and wavelength.
The Rayleigh correction consists in subtracting the Rayleigh reflectance, corrected for atmosphere pressure in this step, from the total reflectance (already corrected for stratospheric aerosol, gaseous absorption and sun glint) in each band (RD 8, 2.7, 18.104.22.168.2).
22.214.171.124.6.2.2 Turbid water and White Scatterers identification (step 126.96.36.199)
Turbid water identification
The TOA marine reflectance is computed at 705 nm (channel 9) from Rayleigh corrected reflectances at 775 nm (channel 12) and 865 (channel 13) using the Angström exponent method. Then this TOA marine reflectance at 705 nm is compared to a threshold (interpolated in a LUT as a function of geometry). If it exceeds the threshold then the turbid water flag (Case2_S) is raised.
White scatterer identification
An estimate of the spectral slope of marine basckscatter is computed using Rayleigh corrected reflectance and pure water specific absorption. This estimated spectral slope is compared to a threshold below which the White Scatterer Flag is raised.
188.8.131.52.6.2.3 Turbid water correction (step 2.6.10)
When a water pixel has been detected as contaminated by a water signal in the infra-red by test 2.6.8, the algorithm called bright pixel procedure performs an estimate of the water-leaving reflectance at four bands used later by the atmosphere corrections above water (see 8.3 below). The algorithm is based on optical properties of the water and performs an iterative procedure with a combination of:
- single scattering aerosol reflectance;
- water-leaving reflectance;
- Suspended Particular Matter concentration.
The atmospheric attenuation of water-leaving reflectance is taken into account.
184.108.40.206.6.3 Clear water atmospheric corrections (step 2.6.9)
The objective of the clear water atmosphere correction is to identify and subtract from the TOA reflectances (corrected for stratospheric aerosols, gaseous absorption and Sun glint), the contribution of the atmosphere, which consists of molecular (Rayleigh) and particulate (aerosol) scattering and extinction. The correction is performed in order to provide normalised water-leaving reflectances.
A secondary objective is to estimate aerosol products: type and optical thickness.
The principle of the clear water atmosphere correction is to identify aerosol models which, together with a tabulated model of the molecular scattering and assumptions on the surface reflectance, fit the observed glint-corrected reflectance in the infra-red part of the spectrum (bands 705, 775, 865nm) and in a visible band (510nm).
The assumptions for Case 1 waters are that reflectance is null at all wavelengths beyond 700nm, and that reflectance at 510nm is nearly constant.
The output of the turbid water atmosphere correction (step 2.6.10, see section 220.127.116.11 above) provides as input estimates of the water reflectance at the bands used by the algorithm.
The algorithm provides one or two aerosol models and their properties in the visible and NIR wavelength domain, which allow to perform a correction of the atmosphere contribution and compute water-leaving reflectances.
The water-leaving reflectances output by the atmosphere corrections above water are normalised in order to remove dependency of the signal upon atmosphere conditions. The normalised water-leaving reflectance product r'w is defined as follows:
where Lw is the water-leaving radiance and Ed (0+) the down-welling irradiance.
The water-leaving reflectance is used by other sub-steps and provided to the Product formatting (step 2.10). This step is applied to all pixels where ACFAIL_F is FALSE.
18.104.22.168.6.3.1 Path reflectance estimate (step 22.214.171.124)
When starting the atmospheric correction, we dispose of the (measured) total glint corrected reflectance ROGC, and of rR(l) for each wavelength. When turbid case 2 water has been detected by a previous step, we also have an estimate of the marine reflectance at TOA, trw_C2.
Atmospheric corrections need an estimate of the contribution of the sky to the total reflectance, or path reflectance, at four wavelengths.
In Case 1 waters, the Infra-Red (IR) contribution of water to the signal is neglected, so that we have, at 775 and 865nm:
rpath (l) = ROGC (l).
This is also performed at band 510nm, even though the water contribution is not negligible in the visible. That estimate will be useful in the MERIS aerosol model (see below).
In turbid Case 2 water, we subtract the water contribution so that
rpath (l) = ROGC (l)– trw_C2 (l)
126.96.36.199.6.3.2 MERIS aerosol model (step 188.8.131.52)
When starting the aerosol correction, we dispose on one hand of the path reflectance rpath at four wavelengths, and of the TOA reflectance ROGC(l) and Rayleigh reflectance rR(l) for each wavelength, and on the other hand of tabulated relationships linking the ratio rpath /rR to the aerosol optical thickness ta(l), for N aerosol models.
The central problem is the selection, among a set of aerosol models, of the two models that most closely bracket the actual aerosol. The principle is to rely on the look-up tables, which should allow :
- To calculate the values of ta(865) from the rpath(865)/rR(865) ratio, for several aerosol models,
- To extrapolate ta from 865 to 775 nm, for each aerosol model,
- To obtain the (rpath(775) / rR(775)) ratios from ta(775), for each aerosol model. These ratios computed from aerosol model, will be noted z(l) in the following.
- To select a couple of aerosol models, by comparing the actual (rpath(775) / rR(775)) ratio, and the various z(775) ratios as obtained at the previous step.
- To estimate the z(l) ratio in the visible bands from the knowledge of the spectral behaviour of this couple of aerosol models.
The successive steps of such a correction scheme are as follows. For a given pixel, and thus for a given geometry (qs, qv, Df):
(1) The ratio rpath(l) / rR(l) is computed at 865 and 775 nm, rR(l) being taken in tabulated values (at these wavelengths, and for oceanic Case 1 waters,
rpath = ROGC).
(2) A first set of N aerosol models is selected, which, in principle, is representative of clear oceanic atmospheres. For these N aerosol models, N ta(865) values are calculated from the (rpath(865) / rR(865)) ratio.
(3) N values of ta(775) are computed for the N aerosol models, from the knowledge of their spectral optical thicknesses (normalised by their values at 865 nm; tabulated values).
(4) N values of z (775) are computed from the N values of ta(775) for the N aerosol models, from the tabulated relationships between both quantities.
(5) The actual (rpath(775) / rR(775)) is then compared to the N individual values of z(775) obtained at step (4), and the 2 that most closely bracket the actual one indicate the two candidate aerosol models.
(6) 2 values of ta(l) are calculated for bands at 510 nm and 705 nm from the normalised spectral optical thicknesses of the 2 “bracketing” aerosol models. Step (2) is now inverted, to calculate two z(l) ratios from the two ta(l) at 510 nm and 705 nm.
(7) The following step lies on the assumption that the actual (rpath(l) / rR(l)) ratio falls between the two z(l) ratios calculated at step (6), proportionally, in the same manner as it does at 775 nm. rpath(l) is now estimated for bands at 510 nm and 705 nm.
(8) By making an assumption on the normalised water-leaving reflectance at 510 nm, the error in the atmospheric correction at 510 nm, Dr 510, can be assessed.
(9) A test is then made on this Dr510 value, if a number of conditions are met. If those conditions are not met, the correction is continued at step (10). Otherwise, depending on the test result, either the correction is continued at step (10), or it is carried out once more from step (2), by selecting however a different set of N’ aerosol models. In the latter situation, the correction is actually carried out for several aerosol databases, so that steps 2-8 are carried out several times; several couples of aerosol models are then selected (one at each time steps 2-8 are done), and the one which is retained at the end is the one that leads to the lowest Dr510.
(10) For every wavelength l of the visible domain, 2 values of ta are calculated from the knowledge of the spectral scattering coefficients of the 2 “bracketing” aerosol models.
(11) Step (2) is now inverted, to calculate two z(l) ratios from the two ta(l) for the visible bands, and then to obtain rpath(l) (see step 7).
184.108.40.206.6.3.3 Correction (step 220.127.116.11)
At then end of the MERIS model step, we now have an estimate of the path reflectance and aerosol optical parameters at all visible and NIR wavelengths where the atmospheric correction is required.
The water-leaving reflectance at the instrument level is then obtained as :
tu(l).td(l).r’w(l) = ROGC(l) - rpath*(l)
The following step consists in calculating the diffuse transmittance, downward td(l) and upward tu(l), in order to retrieve the normalised water-leaving reflectance at surface level r’w(l).
18.104.22.168.6.4 MERIS Ocean Colour Processing (step 2.9)
This chapter describes the processing to be applied to surface reflectances produced by the atmospheric corrections above water (see section 22.214.171.124.6.3 above) in order to derive ocean bio-optical parameters.
Different algorithms are used as shown in flow chart 8.5.2-1.
I. A band-ratio algorithm optimised for open ocean clear waters (so-called "Case 1") yields a geophysical quantity :
· Algal Pigment Index 1
II. Robust band-ratio algorithms valid for all water types, including yellow substance dominated (so-called Case 2 (y)) and waters with excessive back-scattering. These algorithms yield the following Product Confidence Data :
· flag for anomalous scattering waters
· flag for yellow substance-dominated waters
III. A robust algorithm based on the Inverse Modelling Technique (IMT), that can be applied to all water types, yields the following geophysical quantities :
· Algal Pigment Index 2 (mg.m-3)
· Yellow substance absorption (m-1)
· Sediment load (g.m-3)
IV. An algorithm to estimate the instantaneous value of the Photosynthetically Available Radiation (PAR)
Three different flags indicating the type of water sensed are used as they have an influence on processing quality:
· turbid waters (described in Section 7 of this document)
· yellow substance dominated waters
· anomalous scattering
Furthermore, range checks on input and output parameters are applied for quality control.
126.96.36.199.6.4.1 Case 2 (Yellow substance dominated) flag (step 2.9.4)
The presence of Case 2 water is flagged by Yellow substance (CASE2Y_F) flag (step 2.9.4). Input data are normalised water-leaving reflectance r'w(b, j, f). A LUT technique is applied for the retrieval. The procedure is described in RD8 (2.8).
188.8.131.52.6.4.2 Case 1 waters processing - Algal pigment index 1 (Chl1) retrieval (step 2.9.7)
Case 1 waters processing is based on a band ratio algorithm. Inputs are normalised water-leaving reflectance r'w(b, j, f) and ancillary data. The theory of data processing is described in RD8 (2.9). Processing is performed in two steps:
1. a band ratio estimate of Chl1 is selected among up to three possible ones, according to ratio value
2. an iterative procedure eliminates the influence of bi-directionality (parameter f_over_q1) on Chl1 estimate.
184.108.40.206.6.4.3 Case 2 anomalous scattering water flags (step 2.9.6)
The presence of Case 2 water is flagged by the Anomalous scattering (CASE2ANOM_F) flag (step 2.9.6). Input data are normalised water-leaving reflectance r'w(b, j, f) and algal pigment index 1. A LUT technique is applied for the retrieval. The procedure is described in RD8 (2.8).
220.127.116.11.6.4.4 Case 2 waters processing - Inverse modelling technique (IMT) (step 2.9.11)
IMT (RD 8, 2.12) uses an Inverse Radiative Transfer Model-Neural Network (IRTM-NN) to estimate the concentration of algal pigment index 2 (Chl2), yellow substance absorption (ODOC), and suspended particulate matter (SPM) concentration for Case 1 and Case 2 waters from normalised water-leaving reflectance at MERIS bands b442 to b665, b705, qs, qv and Dj. In this approach, Case 1 waters are treated as a special case of the Case 2 water algorithm. The IMT considers the complex nature of the water leaving reflectance and its parameterisation avoids any iterative procedures. The multiple non-linear regression method in this approach leads to high reduction in computing time and is therefore fast enough for operational mass production of Level 2 products, but it requires a careful and elaborate determination of the multiple coefficients (training phase). The IRTM-NN procedure is already programmed and requires only data input. The data output from this algorithm are Chl2 concentration, ODOC absorption, SPM concentration and a confidence flag.
The Neural Network applies generic Neural Network functions, as specified in AD 6, to specific auxiliary parameters and inputs, to obtain the required outputs. All specific aspects of the application are specified in section 18.104.22.168 below.
22.214.171.124.6.4.5 Photosynthetically Available Radiation (step 2.9.8)
Instantaneous Photosynthetically Available Radiation (PAR) is derived from the irradiance above each water pixels, under a tabulated relationship. This step is applied to all pixels where ACFAIL_F(j, f) is FALSE.
126.96.36.199.7 MERIS Land Pixels Processing
This chapter describes the algorithms to be applied to the MERIS Top Of Atmosphere reflectance in order to compute the MERIS land level 2 products:
· Top Of Atmosphere Vegetation Index (TOAVI);
· Surface reflectances in bands 412 to 885nm;
· Aerosol optical thickness and alpha above DDV
· Bottom Of Atmosphere Vegetation Index (BOAVI);
· Dense Dark Vegetation (DDV) flag;
as well as flags relevant to the quality of all products.
188.8.131.52.7.1 MERIS Top Of Atmosphere Vegetation Index (TOAVI) (step 2.2)
The TOA Vegetation Index computation algorithm takes as input the Top Of Atmosphere Reflectance output by step 2.1 .
Before computing TOAVI, a spectral test is done on every Land pixels in order to flag any pixels that are not vegetated. Then, on the vegetated pixels, TOAVI or MERIS Global Vegetation Index (MGVI) is estimated in two steps. First, the information contained in the blue band at 442 nm is combined with that in the bands at 681 and 865 nm traditionally used to monitor vegetation, in order to generate "rectified channels" at these latter two wavelengths. The "rectification" is done in such a way as to minimise the difference between those rectified channels and the spectral reflectances that would be measured at the top of the canopy under a standard geometry of illumination and observation. The proposed algorithm assumes that ratios of polynomials are appropriate to generate both the "rectified channels" and the final spectral index, MGVI.
The MGVI has been optimised to assess the presence on the ground of healthy live green vegetation. The optimisation procedure has been constrained to provide an estimate of the Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) in the plant canopy, although the index is expected to be used in a wide range of applications.
184.108.40.206.7.2 Atmospheric correction over land (step 2.6.23)
This scheme is detailed in RD 8, sections 2.15, 2.17. The input of the algorithm are TOA reflectances, corrected for gaseous absorption and stratospheric aerosols, above all valid land pixels. The outputs of the algorithm are:
- Rayleigh corrected reflectances for all pixels;
· Dense Dark Vegetation (DDV) flag for all pixels;
· aerosol model index and Angström exponent for pixels flagged as DDV;
Processing of the atmospheric correction over land is performed in 13 MERIS bands (basic set of 15 bands described in section 2, minus the O2 absorption band at 761.25 nm (band 11) and H2O absorption band at 900 nm (band 15)).
220.127.116.11.7.2.1 Rayleigh Correction Processing (step 2.6.15)
Rayleigh correction processing is organised in several steps. First the Rayleigh reflectance is computed for every 4x4 pixels window. Then, Rayleigh transmittance and Rayleigh spherical albedo are computed for every 4x4 pixels window. Then, the TOA apparent reflectance corrected for gaseous absorption r*ng is corrected for Rayleigh contributions for each pixel in order to derive the top of aerosol reflectance rtop. The 4x4 pixels windows do not overlap.
18.104.22.168.7.2.2 Dense Dark Vegetation (DDV) Screening (step 2.6.13)
DDV screening consists in flagging pixels identified as DDV by comparing a spectral index, the Atmosphere Robust Vegetation Index (ARVI), to a tabulated threshold which depends on Earth location and on the date of data acquisition. The ARVI is built using MERIS bands 3 (442 nm), 7 (665 nm) and 13 (865 nm).
22.214.171.124.7.2.3 Aerosol above DDV (step 2.6.17)
Aerosol type and optical thickness are estimated for DDV pixels. The same set of aerosol models is used for all pixels in a 32x64 pixels window (both in RR and FR); these windows do not overlap.
Using the aerosol parameters, we can estimate the aerosol reflectance ra at MERIS bands 12 (775nm) and 13 (865nm), and the corresponding optical thicknesses. The alpha product expresses the spectral dependence of the optical thickness: .
126.96.36.199.7.3 MERIS Bottom Of Atmosphere Vegetation Index (BOAVI) (step 2.8)
The processing of the BOA Vegetation Index computation is only applied to land pixels. The algorithm is the MERIS Terrestrial Chlorophyll Index (MTCI).
The products delivered by the atmospheric corrections processing are used as input to the BOAVI algorithm.
188.8.131.52.8 MERIS Level 2 Product Formatting Algorithm
MERIS processed data samples corresponding annotations and flags are collected from previous steps and formatted according to Level 2 product description in R-8.
Each MERIS Level 2 geo-physical product is derived from a MERIS Level 1B product (herein after called "parent L1B product") and auxiliary parameter files specific of the MERIS Level 2 processing.
The MERIS Level 2 product is composed of : the Main Product Header (MPH), the Specific Product Header (SPH), one Summary Quality Annotation Data Sets (SQ ADS) ), one Global Annotation Data Sets (GADS), one Annotation Data Sets and twenty Measurement Data Sets.
The MPH allows to identify the product and some of its main characteristics.
The SPH contains references to external data files and Data Sets descriptors, as well as general information applicable to the product such as sensor characteristics, PCD and metrics summary. A large amount of SPH contents can be directly derived from the parent L1B product SPH.
The first ADS (SQ ADS) contains information on the quality of the product.
The GADS contains all the data scaling factors.
The second ADS contains information on geo-location, measurement viewing and illumination geometry and auxiliary environment parameters for a subset of the product pixels: the tie-points. One ADSR includes the set of tie points corresponding to a given satellite location. It is the same as in the parent Level 1B product.
The Measurement Data Sets (MDS) contain geo-physical parameters derived by the L2 processing. The products are distributed in order to obtain
· maximum homogeneity of the information: the "reflectance" bands, for instance, contain reflectance whatever the underlying surface is;
· maximum storage efficiency: the bytes allocated for a given pixel will be used to store different parameters, relevant to the surface observed.
The Flags MDS (20) contains all information needed to decode and check for quality the distributed pixel information.
One MDSR includes the parameters for all pixels corresponding to a given time sample of MERIS. The term "product line" will be used hereafter to name the MDSRs of the different MDS for the same time sample, i.e. with the same MDSR index.
Information coming either from parent Level 1B product, from external data sources, or generated by any processing step are gathered, organised, scaled and coded according to the “Envisat-1 Products Specifications - Volume 11 – MERIS Products Specifications” document ( R-8) to build the Level 2 product file.
184.108.40.206.8.1 Main Product Header
Main product header is formatted as described in the “Envisat-1 Products Specifications - Volume 5 – Product structures” document ( R-7). The Error Message MPH field summarises the errors encountered in processing.
220.127.116.11.8.2 Specific Product Header
Specific product header is formatted as described in the “Envisat-1 Products Specifications - Volume 11 – MERIS Products Specifications” document ( R-8).
18.104.22.168.8.3 Annotation Data Set "Summary Product Quality"
The annotation data set is composed of one Annotation Data Set Records (ADSR) for every 8 tie frames, i.e. every 128 (Reduced Resolution) or 512 (Full Resolution) product lines.
Each ADSR, following “Envisat-1 Products Specifications - Volume 11 – MERIS Products Specifications” document ( R-8) is composed of :
· Start time of the measurement or MJD, modified Julian Day of time sample
· Attachment Flag
· % of water pixels having absorbing aerosols (wrt water pixels)
· % of water, % of DDV land, % of land, % of cloud pixels (wrt valid pixels);
· % of pixels w/ low polynomial pressure (wrt valid pixels);
· % of pixels w/ low Neural Network pressure (wrt valid pixels);
· % of pixels w/ out of range inputs for water vapour processing (wrt valid pixels);
· % of pixels w/ out of range outputs for water vapour processing (wrt valid pixels);
· % of pixels w/ out of range inputs for Cloud processing (wrt cloud pixels);;
· % of pixels w/ out of range outputs for Cloud processing (wrt cloud pixels);
· % of pixels w/ out of range inputs for Land processing (wrt land pixels);
· % of pixels w/ out of range outputs for Land processing (wrt land pixels);
· % of pixels w/ out of range inputs for Water processing (wrt water pixels);
· % of pixels w/ out of range outputs for Water processing (wrt water pixels);
· % of pixels w/ out of range inputs for Case 1 processing (wrt water pixels);
· % of pixels w/ out of range outputs for Case 1 processing (wrt water pixels);
· % of pixels w/. out of range inputs for Case 2 processing (wrt water pixels);
· % of pixels w/. out of range outputs for Case 2 processing (wrt water pixels);
The counters are accumulated according to every pixel in the time interval between a Q-ADSR (included) and the following one (excluded) and dumped to the Q-ADSR. The last Q-ADSR of the product may relate to a smaller number of product lines than the others.
22.214.171.124.8.4 Global Annotation Data Set - Scaling Factors
Global Annotation Data Set is formatted as described in R-8. Scaling factors and offsets are read from an auxiliary data product.
126.96.36.199.8.5 Annotation Data Set "Tie Points Location and corresponding Auxiliary Data"
Annotation Data Set "Tie Points Location and corresponding Auxiliary Data" is the same as found in the parent L1B product.
188.8.131.52.8.6 Measurement Data Sets
There are 20 MDS:
· MDS 1 to 13 for the Normalised Reflectance for any valid pixel, at those MERIS bands not dedicated to gaseous absorption measurements: b412, b442, b490, b510, b560, b620, b665, b681, b705, b753, b775, b865, b885;
· MDS-14 for total water vapour for any valid pixel;
· MDS-15 for Algal Pigment Index I (water pixels) or TOAVI (land pixels) or Cloud Top Pressure (cloud pixels);
· MDS-16 for Yellow Substance and Total Suspended Matter (water pixels);
· MDS-17 for Algal Index II (water pixels) or BOAVI (land pixels);
· MDS-18 for PAR (water pixels) or Cloud Albedo (cloud pixels) or surface pressure (land and bright pixels);
· MDS-19 for Aerosols Angström exponent and optical thickness (water, land pixels) or cloud type and Optical Thickness (cloud pixels);
· MDS-20 for the associated flags for any pixel;
with the same record structure : an MDS is composed of one Measurement Data Set Record (MDSR) by product time sample. The structures are specified in R-8.
The normalised surface reflectance MDSR contains, according to R-8 :
· start time of sample in MJD2000 format;
· quality indicator (0 if nominal, -1 if no data are available; in such a case the data field of the MDSR is filled with zeroes);
· one (scaled) normalised surface reflectance value per pixel (1121 in RR, 2241 in FR, 1153 in FR imagette, 4481 in FR FullSwath).
Geo-physical parameters are expressed in counts using the scaling factor and offset stored in the GADS. Each value is stored in one or two bytes.
The flag MDSR contains :
· start time of sample in MJD2000 format;
· quality indicator
· one flag set (three bytes) per pixel (1121 in RR, 2241 in FR, 1153 in FR imagette, 4481 in FR FullSwath).
The coding of flags is specified in R-8.