3.5 Six-channel SST retrieval

The difference maps you created in the previous section form the basis of the atmospheric correction algorithms used to derive an estimate of the SSTskin temperature. In the final part of this lesson, you are going to use a linear combination of Tb data in both views to derive a six-channel SSTskin retrieval from AATSR Tb data.

Constraints on SST retrieval

The operational SSTskin algorithms used by AATSR depend on available Tb data. For example:

• Often the forward view data may be corrupted by clouds due to the larger FoV and cannot be used for SSTskin retrieval. In this case only three-channel SSTskin retrieval is possible.
• During the day the 3.7µm channel is unavailable because of the strong contribution of reflected sunlight at this wavelength, and a four-channel (nadir 11µm and 12µm nadir and forward view data) must be used.
• Daytime data with cloud-corruption of the forward view means the only option is a two-channel (11µm and 12µm nadir channels) algorithm This is similar to that used in other infra-red satellite sensors.

In our case, we have all 6 Tb channels available for the majority of the image and so we will compute a 6 channel SSTskin retrieval. Ideally the AATSR SSTskin retrieval algorithm should be implemented on a pixel-by-pixel basis. In Bilko the easiest way to implement the algorithm is to calculate the SSTskin retrieval algorithm coefficients in advance, and format each of these as an image with identical dimensions to the AATSR GBTR data to be processed. We also have to ensure that the algorithm is applied only to cloud-free pixels.

Retrieval coefficients

We are going to implement the 6 channel SSTskin algorithm, which uses seven coefficients derived from radiative transfer modelling of the atmosphere. This approach is different from that adopted for most other sensors where satellite data are co-located in space and time an then used in a multiple-linear regression to derive suitable SSTskin algorithm coefficients. In the case of the 6 channel algorithm a single set of coefficients is applicable in all atmospheric regions due to the extra information provided by the forward view Tb data. We will first investigate the SST algorithm coefficients before applying these to AATSR Tb data and generate an SSTskin map of the Balearic and Algerian basin on the 27th July 2002.

Activity: Open the Bilko image set AATSR_6ch_coefficients.set. This is a set of seven 32-bit floating point images, each corresponding to one of the seven retrieval coefficients. When the Redisplay dialog opens, change the stretch to Autolinear and click All to accept this stretch for all the images.

The stack that opens has the image sst_coeff_d0.dat on top. This is an image of the Offset value used in the 6 channel SSTskin retrieval equation. Vertical stripes in the image correspond to the across track sections of the AATSR swath. Note how these bands become narrower towards the edges of the image.

It is easier to visualise what this image represents by drawing a transect across the entire image making sure that the transect is horizontal:

1. Select the line tool from the top tool-bar and draw a line across the image.
2. Use the Go-To dialog to make sure the transect line is horizontal (DY=1)
3. Open a new transect document ([CTRL+N] > TRANSECT )
4. Right-click on the transect document, select Scale and set the Minimum to 0.65 and the maximum to 0.75.

The transect shows how the offset (red line) has a minimum value of 0.6977 at nadir and increases towards the edge of the swath with a maximum of 0.7215 at the edge (band 39). The coefficient is symmetrical about nadir, and each band is visible as small plateau of equal values, the width of which decreases away from nadir.

Right-click on the transect and change the Scale back to Auto by checking the Minimum and Maximum check-boxes in the Scale dialog, and examine the across-track changes in the other coefficients. Note how some of these are negative, with an absolute value that increases towards the edges.

Question 1
Considering the viewing geometry, can you explain the across-track changes in the absolute values of the seven coefficients?

In order to account for differences in viewing geometry, the AATSR swath is split up into sections across-track; 1 central sector at nadir and 38 sections each to the left and right. For each of these sections, a separate set of SSTskin algorithm coefficients is applied. For 2 and 4 channel SSTskin retrieval, the coefficients also have to account for the difference in atmospheric state associated with tropical, mid latitude and high latitude atmospheres. Using mean differences for these three zones, the coefficients are interpolated to the exact latitude of a given pixel.

Note: The method described here is a refined version of the SSTskin retrieval algorithm originally used for ATSR-1 and ATSR-2. However, ATSR-1 and ATSR-2 data have now been re-processed using this method in order to obtain a consistent time series with AATSR.

Computing SSTskin

You are now going to connect the AATSR Tb data together with the AATSR SSTskin algorithm coefficient maps and compute the SSTskin over our study area.

1. Open the formula document AATSR-dual-6ch-SST-Algorithm.frm and familiarise yourself with the content of this file.

2. Open one of the ATS~.N1 files, select the bands folder and, in the right frame, select both nadir and forward view 11µm, 12µm and 3.7µm brightness temperature data sets by clicking on them once.

3. Right-click on the selection, and chose Open Connected to open all the bands as a connected stack.

4. In the Redisplay dialog, set the Null value to 27315 (273.15°K or 0°C), the Minimum value to 28615 (286.15°K or 13°C) and click All to apply this to all images in the stack.
   Note: Opening multi-channel images in this way is more efficient than opening each band individually.

5. Connect the btemp images with the coefficients in a new stack:

   	From the menu bar select Image > Connect
   	select all the Tb channels and all the coefficients,
   	add 1 blank image, check Stacked, and click OK.

   The coefficients are now in the new stack, so you can safely close the original coefficient stack, AATSR_6ch_coefficients.set

6. Use the Selector to check that the order of the images is as described in the connect configuration for the formula AATSR-dual-6ch-SST-Algorithm.frm.
   Note that in this case the names of the coefficient files have been chosen so that the alphanumerical default order of the images matches that in the formula document specification.

7. Apply the formula document to the connected tile:

   	Click on the formula document to activate it,
   	select Edit > Copy from the menu bar ([CTRL+C]),
   	Activate the stack, and
   	use Edit > Paste ([CTRL+V]) to apply the formula.

   A new image will be generated containing the AATSR 6 channel SSTskin map. To see this, use the selector or [TAB] to @14.

8. Save the new image as AATSR_6ch_SST_1.dat (file type BILKO.dat), and close the two image stacks, leaving the formula and coefficient stacks open.

9. Repeat 2-8 above to calculate SSTskin for the second of the AATSR scences, saving this as AATSR_6ch_SST_2.dat

10. Close all stacks, and the formula document, then open the two SSTskin images you just created, and open the btemp_nadir_1100 band from each of the two AATSR scenes. Place the images on the computer screen so that you can compare the retrieved SSTskin data with the original Tb100 data.

Question 2
How does this SSTskin temperature map differ from the 11µm Tb data?

The operational SSTskin algorithm used by the AATSR attempts to minimise the noise within an SST image by smoothing the atmospheric signal using a 3x3 kernel mean smoothing filter. The smoothed data are then added back onto the 11µm Tb data set to generate the final SST-skin image. We will implement this final step of the AATSR SSTskin retrieval process using Bilko tools. Finally, we will mask out cloud and land pixels to complete the SSTskin map.

Question 3
How would you apply a Bilko formula to compute the difference between the SSTskin map (AATSR_6ch_SST_*.dat) and the corresponding btemp_nadir_1100 data?

1. Apply the formula you just created to the two sets of images to create two difference maps, one for the Algerian coast and one for the Gulf of Lions.

2. Apply a 3x3 mean smoothing filter to the difference maps:

   	Select the whole image (keystroke [CTRL+A])
   	Select Image > Filter > Mean from the menu bar
   	In the Filter Size dialog set the filter kernel to 3 rows and 3 cols.

3. Save the filtered difference maps as AATSR_SST_Tb_diff_1.dat and AATSR_SST_Tb_diff_2.dat, so that these names correspond to the numbering used for the SSTskin images you saved earlier.

The images you have just created represent a smoothed measure of the atmospheric attenuation inherent in the AATSR TOA Tb11 data. You may need to adjust the display to view the image content - using a Linear Redisplay stretch with Min = -800 and Max = 800 works well.

Question 4

a)	How would you explain the difference between land and sea areas in these difference maps?
b)	How do current eddies and river plumes appear in the difference maps?
c)	How would you explain the difference between these structures and areas with more uniform Tb? What effect might this have on the final atmospheric correction for these areas?

Activity:

1. Open the Bilko for Windows formula document Final_AATSR-dual-6ch-SST-Algorithm_processing.frm and familiarise yourself with the contents of this file.

2. Connect the required images together as described in the formula document. (To do this you will also need to open the bands containing the cloud flags).

3. Check that the order of the images in the stack matches the connect figuration in the formula. If necessary, move one of the images by choosing it in the selector, and pressing the number key on the keyboard which corresponds to the new position for this image (e.g type [3] if you want the selected image to be @3).

4. Apply the formula to create a masked map of AATSR SSTskin measurement, and save this as ATS_SST_1.dat.

5. Repeat 1-4 for the second image, and save it as ATS_SST_2.dat.

Question 5
Some of the fine scale features have been erroneously masked as cloud in the new images. Use the [TAB] key to look at these regions ins the unmasked SST image (@4). How do these areas compare with other fine scale features that have not been given a cloud mask? Bearing in mind your answer to question 4, how would you explain this? (Taking transects across masked and unmasked filaments in the stacks may help your analysis).

The final step in this lesson is to merge the two scenes into a masked AATSR SSTskin measurement map on a Lat/Lon projection.

1. Close all windows, and re-open the two masked SST scences - ATS_SST_1.dat, and ATS_SST_2.dat.

2. Resample the two scenes as described in section 3.2 and use the formula merge_aatsr_scenes.frm to merge the two rectified images into a single scene.

3. Finally save the merged image as ATS_SST_20020727.dat, close all other windows, and Redisplay the merged and masked scene using the linear stretch from earlier (Null <= 27315, Min = 28615).

This image is an example of the mixing of water masses in the Mediterranean Sea driven by the wind and the dominant currents of the region. During the day, thermal stratification builds up near the surface, particularly in areas with low wind speeds, such as the region NW of Mallorca. This is known as diurnal stratification, because it builds up due to solar heating during the day, and breaks down overnight.