ESA Echoes in Space - Land: Forest mapping with ALOS PalSAR

In this video tutorial we're going to look at two ALOS PALSAR acquisitions over a part of the Amazon rainforest in order to extract deforested areas each image includes two polarizations horizontal transmitted horizontal receive and horizontal transmitted and vertical receive First we will compare the two polarizations To determine the deforested areas and then we're going to compare the two images to extract the areas which were deforested in between the acquisition dates So let's begin first we will open the images in the product Explorer window Notice how we have the two images here and for each image we have two different polarizations so in fact we have four separate image files Each image was acquired over latitude 6 degrees south and longitude 56 degrees west and one image was acquired in 2007 and the other was acquired in 2010 and each image has two polarizations And here we see the polarizations HV for horizontal transmitted vertical received and HH for horizontal transmitted and horizontal received Let's have a look at these images And if we compare the two polarizations from the same date, let's see how they look We can compare these side-by-side by selecting window tile evenly and in the navigation tab if we if we ensure that these icons are selected to link the viewers and link the zoom and the mouse icon Then we can scroll and zoom Simultaneously just by inspecting images that does not seem to be a great difference between the HH and the HV if we want to Perform some analysis on these images it helps to have the two polarizations in the same file that way we can perform ratios calculate the difference between the images and Do what do further analysis? to combine the two polarizations into the same image file select raster geometric operations collocation and ere we can select the master and the slave image we wish to combine together So let's begin with the 2007 image so we select as the master the HV channel and as the slave the HH channel and Here we can call the output image HV and HH We also should rename The final image bands of this combined product to reflect the the master and the slave images so The master is HV so here we type in HV and The slave is HH so here we type in HH These are the bands of the output product So these are the input and these are the output so we want the output to be named according to the input And then we select run And here we have the combined product So if we look at the bands of this image we now see the two polarizations of the same image acquisition date 2007 in the same image file Now that these are in the same image file we can perform calculations on the two bands For example if we go to the band maths we can calculate the ratio We can create a new band. We can call it HH over HV avoid using the forward slash symbol here because the forward slash is one of the characters you cannot use in the naming of new bands Then if we deselect virtual then we will write the band to a file and If we click on edit expression Here we can type out our expression so HH divided by hv and okay And if we click on OK again Then here we have our ratio And from the ratio we can see more clearly the difference between the HH and the HV We can go further and create an RGB composite of the HH HV and the ratio between the two so if we close all the existing image windows and click on Select the name of the file in their product explorer window then go to window open RGB image window then here we can select HH HV and the ratio as blue So notice here that all of the deforested errors appear in blue Meaning that the ratio between HH and HV is very high over These deforested areas So the HH is much stronger than the HV component in these areas whereas over the forested areas The difference is not so great We can exploit this characteristic to extract the deforested areas But first we will do a speckle filtering in order to remove some of the image speckle because if we zoom in We can notice that there is some image speckle If we select radar Speckle filtering Single product speckle filter here we can select a speckle filter to apply we will choose the Lee filter and we will select a window size of 5 by 5 and we will select as the input bands Only the single polarizations and the ratio and here we run assured that the output file and directory is as we want it And we select run We can now compare the speckle filtered product and the non speckle filtered product, so if we open the same RGB combination of the speckle filtered product We can compare them So here we select HH is red HV is green and the ratio is blue the reason why we select HV is green is because over vegetation you have a lot of volume scattering and Where you have volume scattering you Tend to have a depolarization of the signal and a depolarization of the signal corresponds to a high return in the cross-pol and given that vegetation tends to be green is intuitive to assign the HV channel to green Now we select ok Here we have the speckle filtered RGB composite We can view these side-by-side by selecting window tile horizontally Here we can see the difference between the speckle filtered and non speckle filtered product Notice how much of the speckle has been removed in the Lee filtered image we will now try to mask out the areas that are deforested and to do that we will apply a threshold on the HV channel first of all we need to see what the pixel values are like over the deforested areas for the HV channel so if we go to the HV image we double click on it and If we select pixel info well first, let's close these other windows then we select pixel info here and here we can see the Backscatter of the HV channel Okay, actually these values are in decibel this image has already been calibrated Converted to decibel Prior to the exercise Here we can see that over the over the deforested areas we have values of less than 20 db and over the forested areas we have values of around minus 10 11 12 13 db So perhaps a suitable threshold in between the deforested areas and the forested areas is around minus 15 db So what we will now do We'll go back to the product Explorer window select our input image, and we will go to the band maths and create a mask including only values which are below minus 15 db And we'll call this deforestation mask We will write this to the file and here we select edit expression we select the conditional operator and here we select the HV channel and we specify a condition that the HV channel has to be below minus 15 db and if it is below minus 15 db Then the mask will have a value of 0 otherwise the mask will have a value of 1 And we select ok Again we select ok and here we have our mask So deforested areas have a value of 0 and the areas covered by forests Have a value of 1 we will now perform exactly the same steps on the 2010 image Which I have already done prior to this exercise So I'll open that here And we will compare the image from 2010 with the image from 2007 to see the extent of deforestation To do this first we need to combine the two images so we go to again to raster geometric operations collocation and here we input the 2007 image as the master and the 2010 image as a slave We will write as suitable output file name so we'll call this 2007 and 2010 and here we will rename the output master and slave to reflect the input so here we will call this 2007 And we will call the slave 2010 This is just the suffix at the end of the original file name And we will select run And here we have our combined image We will now try to extract the areas that were deforested in between the two acquisition dates in between 2007 and 2010 by comparing the two masks And we'll do that by simply subtracting one mask from the other So again we go to raster band maths Here we will we will write the name of the new band, which will be 2007 minus 2010 Again we'll deselect the virtual in order to write it to a file Then we click on edit expression and here we type out the difference between the two masks so we select 2007 minus 2010 And we select okay the reason why we subtract the 2010 from the 2007 image is because we expect there to be more deforested areas in the 2010 image We then select ok And then we select ok again And here we have the mask including the deforested areas in between 2007 and 2010 Having a value of one and all other areas having a value of zero Let's compare this with the two masks So we have The mask from 2007 and the mask from 2010 and we compare these So notice that in each mask we have deforested areas, but you also have areas that do not correspond to forests such as the river By subtracting one from the other only the differences are highlighted here in the difference mask However, we still have some residual pixels from Areas that do not correspond to deforestation such as the river Nonetheless it gives us a good general overview of the areas that may have been deforested This was just a very simple exercise on creating a mask of deforested areas in between to SAR image acquisitions I hope you found it interesting. Thank you for watching

Channel: EO College

Views: 9,317

Keywords: MOOC, Radar, EO College, remote sensing, science, e-learning, ESA, space, data, Sentinel, Earth, environment, course, microwave, uni jena, Copernicus, TerraSAR, ERS, ALOS, Satellite, SRTM, SNAP

Id: nC-lHl5HRU8

Length: 16min 16sec (976 seconds)

Published: Sun Oct 22 2017