ESA Echoes in Space - Water: Oil Spill mapping with EnviSat ASAR

In this exercise we're going to use ENVISAT ASAR data to map oil spills And we're going to take as an example the Gulf of Mexico oil spill from 2010 So let's have a look at the data first of all Here, I'm opening two ENVISAT ASAR wide swath mode images So each one has a pixel spacing of 75 meters and here we can see the acquisition dates So if we go to the abstracted metadata Here we can find information about the product We can see that the image was acquired where the satellite was ascending We can see that the polarization is VV and the pixel spacing is 75 meters Let's have a look at the data we expand the bands folder we can look at the amplitude band And here we can see quite clearly the extent of the oil spill However we would see it more clearly in the dB in the decibel band so if we create a new band for decibel and then if we visualize that then we can see more clearly the extent of the oil spill and the other image we have is an archive image In which there is no oil spill And we'll see later on how we can use the archive image to better map the oil spill So what we goes through now is to process both the images in order to calibrate them geometrically correct them and mask out the land area and given that we're going to apply Various processing steps on both images. We're going to use the graph processing tool and the batch processing functions So we go to tools graph builder and here we will insert a list of functions so if we Right click with the mouse button here we can select the various functions we wish to apply So first we will apply a calibration And in the calibration if we select the calibration tab here we will select save in dB Then we will insert a multi looking and a multi look factor We will select as two we're going to reduce the dimensions of the image by a factor of 2 in order to Smooth out the speckle and also increase the speed of the processing The next step Will be to do the ellipsoid correction the reason why we do an ellipsoid creation and not a terrain correction is because We're not interested in using a DEM and taking into account the terrain because we're only interested in the sea area where there is very little terrain So here we go to ellipsoid correction Then finally we will apply a land sea mask so we go to raster Masks Land Sea Mask And here we will select mask out the land So now we will connect these various processes together So we will hover the mouse icon to the side of each process until the red triangular appears Then we will click and drag the red triangle to the next processing step So first was calibration then multi looking Then ellipsoid correction Land sea mask And then the final write output product We can make these a little bit tidier by arranging them in a certain way To make it clear Okay now we will save this processing chain We'll call it oil spill And click on save then we will close it and we'll close this window and Now first we will delete this Decibel band because we're going to batch process these images so we go to delete In any case we will have a decibel band in the final product Then we go to tools Batch processing and we select add opened And then here we will load our graph And here we have our processing chain which we will apply to these two images Then here we will select our output folder And then we will select run And here we have our final products open in the product Explorer window Okay, so we will close the input products in order to have only the output products Now let's have a look at them So here we can see that the land has been masked out The products have been geometrically corrected here we see the Gulf of Mexico with the land to the north And we can see that the products have been calibrated and converted to decibel we can see that by selecting pixel info And moving the mouse icon over the pixels and we can see the values in decibel Here we have our archived image processed in the same way Now we will try to Mask out the area covered by oil spill And you may wonder why we do not use one image and just apply a simple threshold If we look at the pixel values You will notice that there is a very clear difference between the area covered by sea and the area covered by oil spill and The difference is in this case if we move the mouse icon over the oil spill and over the sea You will see that over the oil spill We have decibels values of around -22, -23 decibels whereas over the sea we have - 10, -11 Decibels so we have a difference of around 10 dB So why not apply a simple threshold well let's try it we go to Raster band maths and here we select Threshold The new band, let's make the threshold minus 20 And then we create an expression here and we say mask out all pixels below minus 20 and we hope that the final product will be a mask of the oil spill We select ok then we select ok now notice that while on one side of the image The mask has worked well on the other side of the image it is completely ignored the area covered by oil spill That's because we have quite a strong near to far range variation in pixel values So if we again look at the pixel info We looked only at this area however if we look at this area the difference in DB is very different So while here we had values of around minus 20 dB Where we had oil spill here we have values of around minus 15, 16 dB Okay, and if we go to this site will again have a different value So It's not sufficient to apply a simple threshold on just this one image What we will try to do is to compare this image with an archive image where the near to far range variation is the same and that way we can apply a similar threshold, but take into account the near to far range variation So to do that we will merge the archive and the crisis image together In order to be able to perform operations on the bands so we will close these images in the viewer and here we will remove this Virtual band that we've just created, so we'll select delete And then we go to radar coregistration stack tools create stack And here we will select add opened And in the create stack tab we will select product geolocation as the initial offset method because we have not applied precise orbits to these products and the product geolocation is accurate enough for the Stacking that we would like to do And then here we select an output file name which We will rename keeping only the common parts of both file names and keeping the underscore stack suffix And we will save that to our output folder and select run Here we have our output stacked image, so let's check the registration between the two bands So to do that we will select one And then we will go to the layer manager and will overlay the other band selecting image of band and selecting the 2009 image and then we click on finish and here We can check the registration between the two products What we should look at here is the coastline Because that should be the same it should be identical for both images And we can see that the registration is very good. There's no noticeable difference Now we will try to Mask out the area covered by oil spill taking into account also the archive image And what we will do is we will look at the difference in decibel between the two images To do that we will open the two images In two separate viewers in order to have them both open then we will go to Pixel info So we will go to view tool windows pixel info And then as we move the mouse icon over Each image we will see the values in decibel for both images so let's go to the area covered by the image in which we see oil spill and Here if we move the mouse icon over the image We will see that over the sea There is very little difference in decibel Both images are more or less the same whereas over the oil spill we have a much lower value for the crisis image in which oil spill is present and a much higher value over the archived image in which there is no oil spill because the oil spill dampens The water surface and become smoother and we have a higher specular reflection so we have a lower value in dB We will exploit this difference to mask out these oil spill areas and notice how the difference in dB Is this more or less the same From near to far range so also here We have a difference of around 10 dB and here we have a difference of around 10 dB The difference in the pixel values are more or less the same So what we will do is we will go to raster band maths and here we will create an expression. We'll call it difference Threshold And here we will create our expression however we will not choose 10 as our threshold because that is In most areas the maximum difference, so we'll choose a slightly lower value but not too low to be too close to the to be too close to zero So first we will subtract the crisis image. Sorry the archived image from the from the crisis image And we will set a threshold on the result of that, so let's say minus 6 dB Which is not quite -10 dB, but it's also not too close to zero and then we select ok and ok and here we have The areas of oil spill masked out however notice how The result is quite noisy. You have many pixels here classified as oil spill when in fact they're more likely to be speckle or due to the difference in backscatter caused by the waves and if we look at the archived image And if we zoom in we can see a quite a high variation of high to low backscatter Probably due to the waves and due to speckle so we will try to remove that and one way of doing that is to Apply a smoothing filter on this archive Image so to do that we right-click on the band In a product Explorer window and here we select filtered band And what we'll do here is to create our own filtered band And we will Specify this filter here Okay, we will create a kernel size of let's say 23 by 23 pixels so quite a large kernel and we will remove all of the Filled pixels by left clicking and dragging the mouse icon over the filter pixels and here in the filter properties we will select the type of filter we wish to apply And we will apply a mean filter so here we're going to apply a very heavy smoothing using a kernel size of 23 by 23 pixels and here we apply a mean filter, so we will mean all of the pixels in a 23 by 23 moving window across the image And we will rename that mean 23 pixels Mean 23 pixels Okay, and then we close that And then we select okay and here we have our smoothed archived image We will then apply that difference threshold again, but using this smoothed archived image So we go to raster We're going to band maths and then here we will create another difference threshold band But we will call it smooth Okay and here we edit expression and we apply the same expression so the 2010 image minus the smoothed 2009 image Less than -6 dB and we will select ok and ok And here we have a much clearer mask of the oil spill or the areas covered by oil spill So now we no longer have some of these Wrongly classified areas ok we still have some points, which may be due to waves or Not due to oil spill, but in general we see quite clearly the area covered by oil spill And we can compare this image with the crisis image If we close all of the viewers and keep only the difference threshold And the oil spill image we can see the difference So what we will now try to do is to quantify the area covered by oil spill So there are several ways we can do that. We could just go to the histogram and look at the number of pixels Covered by oil spill so the number of pixels with the value of one in the mask image So here having selected the Refresh icon here We can see we have calculated the histogram of this mask image and we can see that there are pixels with a value of zero and With a value of one the pixels with a value of one correspond to the oil spill so if we zoom in to this part of the histogram We can see the number of pixels with a value of one and Knowing the dimensions of the pixels Which in this case are 150 meters by 150 meters the original pixel spacing was 75 meters But then we did a multi looking so we ended up with a pixel dimension of 150 by 150 meters Knowing the pixel dimensions and knowing the number of pixels covered by oil spill we can then estimate the area covered by oil spill Our pixel counting is quite a crude method of calculating an area affected by a certain phenomenon But at least it gives us some some idea, and it's a good quick estimate Another way of counting the pixels is By applying a mask so here we go to the mask manager. We select the function icon and here we apply an expression, so we take a smoothed image and we select the pixels Which are above zero So these are all the pixels with a value of one and then if we go to analysis statistics We select the mask image we select The master we've just created, and then we hit refresh and Here we see the number of pixels with a value of one So here we can do a quick calculation So if we know that the pixel dimensions are 150 meters and so in each pixel we have 150 square meters When we multiply that by the number of pixels with a value of 1 which in this case which is given here so here we put in 1358710 so we put in this number And here we get the extent of the area in square meters Okay, so this is around 30,000 square kilometers So here we have created a very quick estimates of the area covered by oil slick at this particular point in time Which we have calculated to be around 30500 square kilometers I hope it was interesting. Thank you for watching

Channel: EO College

Views: 5,967

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, oil, deepwater horizon, spill, snap

Id: 8Q2QnNbBT7A

Length: 24min 8sec (1448 seconds)

Published: Mon Oct 30 2017