Image classification tutorial with ArcMap: 2.1 Atmospheric adjustments of Landsat 8 images

Previously we discussed how to adjust Landsat images for the reflectance TOA (top-of-atmosphere). Here is a diagram that will help us to understand what happens next:

Therefore the reflectance we have calculated is the% of light reflected to the totality of the incident visible light. But as we can see in the diagram,the satellite sensor measures two things at the same time: the light reflected by our targets on the earth’s surface, plus the light scattered by the particles suspended in the air.

We can push the atmospheric correction of the satellite images to remove the light due to scattering.

As we have already said, this is of no interest unless if you work on separate images in time. The diffusion percentage being the same for an image, therefore it is time lost to make this adjustment.



The instructions to perform this correction on Landsat images is available at http://www.gisagmaps.com/landsat-8-atco-guide/  

But if you are beginner, you will find difficult to understand the procedure. We’ll discuss the procedure by using the ArcMap tools.

DOS method (Dark Object Subtraction)

The method we will discuss is the most frequently used.
The subtraction of black objects is a simple empirical atmospheric adjustment method for satellite imaging, which assumes that the reflectance of dark objects includes an important component of atmospheric scattering.

The concept is simple: if you have a completely black object, and therefore does not reflect light at all, this object should have a value of reflectance equal to zero, will have a nonzero value at the satellite sensor. This value will be the reflectance due to the particles present in the atmosphere.

The subtraction of dark objects researches in each band the darkest pixel value. The broadcast is deleted by subtracting this value of each pixel of the band.

First thing: the adjustment we are going to perform does not apply only to the visible light and near infrared. Therefore we will work for Landsat 8images, on the bands 2 (Blue), 3 (green), 4 (Red) and 5 (near infra-red).

Second thing to know: the principle used is to calculate the scattering for band 4 (red) and to use a correspondence formula to determine the adjustment for the other three bands.

Third thing: we will work with bands 2 , 3, 4 and 5 in TOA reflectance but, also, in gross radiance (DN) for band 4 (red). In the user manual cited above this does not appear at first glance. Let’s see the procedure step by step:

1- the TOA reflectance for the four bands with is calculated with the following formula

TOA band = ( (DN * 0.00002)-0.1) / sin ( solar elevation)

2- by analysing the gross radiance (DN) values of the red band , we determine a radiance due to atmospheric diffusion ( DNred )

3- we transform this radiance to TOA reflectance

CORred = ( ( DNred* 0.00002) -0.1) / sin ( solar elevation  )

4- using an abacus we  enter the red band diffusion and obtain the diffusion for the bands 2,3 and 5 ( CORblue ,CORgreen , CORir )

5- we subtract the TOA reflectance values to obtain the reflectance values of the four bands adjusted for the atmospheric effect.

Red = TOA ( red) – CORred

Rblue = TOA ( blue) – CORblue

Rgreen= TOA ( green) – CORgreen

Rir = TOA ( ir ) – CORir

Let’s discuss in detail steps 2 and 4.

Calculation of the red band diffusion

There are several methods for determining the adjustment to be made for the red band. The recommended method is the so-called Frequency 50-0.008.You will find right here a detailed description of the method .

In practice, with ArcMap:

1- You must create the raster table of your red band:

  • Load the unadjusted red band (the gross values downloaded) in ArcMap.

 In the toolbox -> Data Management Tools -> Raster -> Rasterproperties -> Build Raster Attribute Table

Once the tool is executed , you can click on the layer in the legend window and select ”   Opentheattributetable «  

We are going to ignore completely the value 0 because those are the black pixels that surround our image. You are going to scroll down the recordings until finding a COUNT value equal to or greater than 50. In our example it corresponds to the 23rd recording where it is found a COUNT value of 54.

The radiance value to keep for the atmospheric adjustment of the red band will be the VALUE of the previous line (OID = 22) i.e. 5568.

We can proceed to the third step ,

3- we transform this radiance to TOA reflectance

CORred = ( ( DNred* 0.00002) -0.1) / sin ( solar elevation )

  CORred = ( (5568
* 0.00002) -0.1) /0.42631886   = 0.026647

This value corresponds to Frequency 50. As the method is Frequency 50 –
0.008, we remove 0.008 from this value:

0.026647 – 0.008 = 0.018647

It is this value we must subtract from the TOA value of the red previously calculated to obtain the adjusted values for atmospheric diffusion.This value corresponds to Frequency 50. As the method is Frequency 50 – 0.008,we remove 0.008 from this value: 0.018647

Adjustment of others bands

From the correction value of the red band, we obtain the values to be applied to each of the other bands :

Go to http://www.gisagmaps.com/l8-s2-relative-scatter-calc/You will find a window where you can enter the adjustment value for the red   

Re-enter your value ( here 0.018647 ), you will automatically obtain the values of the other corrections that will display

We obtain the following adjustments

  • 0.06309 for the blue band
  • 0.03442 for the green band
  • 0.00626 for the near infrared

To obtain the corrected bands we use the raster calculator by subtracting from each TOA adjusted band the correction constant we have found .

Leave a Reply

Your email address will not be published. Required fields are marked *