ESA Echoes in Space - Water: Water Body Mapping with Sentinel-1

In this exercise we're going to map water bodies using SAR data exploiting the difference in backscatter between water and land surfaces For the exercise we're going to use a Sentinel 1b image a subset of a Sentinel 1b image And this image was acquired on the 15th of october 2016 We can see that from the file name, or if we go to the metadata. We can see the image acquisition date here as well If we look at the world view tab we can see where the image was acquired so over an area in northern Germany and If we expand the bands folder we can see the image contains a VH vertical transmit and horizontal receive Band and also a vertical transmit and vertical receive so two different polarizations And if we double click on each we can open them in a viewer And if we select window tile evenly we can see them side-by-side What we will do is we will Calibrate the image to convert it to Sigma naught backscatter Then we will do a speckle filtering then we will try to create a mask of the water bodies And we will create that mask by exploiting the difference in backscatter between water and land surfaces So first we go to radar Radiometric calibrate and here we will calibrate the image to Sigma naught So we will select Ok we select run And to reduce some of the image speckle we will apply a speckle filter So we go to speckle filtering single product speckle filter Here we will select the Lee speckle filter, and we'll apply a window size of 5 by 5 pixels And then we select run Let's have a look at the speckle filtered product we select bands then we can first close the Existing windows which are opened and in fact It may be easier to visualize them in decibel if they're in a logarithmic scale so we convert We convert each band To dB so here we select linear to from dB and here also We select linear to from dB And we'll do the same thing for the image prior to speckle filtering Let's open an RGB composite of each image Here we can select for example VV as red VH as green and VV as blue And then we can do the same for the speckle filtered image VV as red VH as green and VV as blue this is one way of combining the two polarizations in an RGB composite and then we can compare them side by side by selecting tile horizontally So notice how the speckle filtered product looks a lot cleaner Less covered by speckle we see lots of areas where magenta dominates so this corresponds to VV backscatter and we see other areas where green dominates corresponding to VH backscatter These areas are likely to be areas of vegetation causing volume scattering such as forests Over the water bodies on the other hand we see a very dark response This is because we have specular reflection in both the VV and the VH What we will now do is we will do a geometric correction Taking into account also the terrain so we go to radar Geometric terrain correction Range Doppler Terrain Correction and here we select the speckle filtered product and we can leave everything as default and We select run And here we have our terrain corrected image Let's convert the bands again to decibel so again we right-click and select linear to from dB And again for the VV channel Let's now save these decibel bands to file so we right-click and we select convert band And we do the same for the VV, and then we save the product And now the product is saved What we will now do is we will look at the image histogram for one of the bands, let's take VV in decibel To do that we go to analysis, histogram and then we select in the product Explorer window the VV in decibel and We click on the refresh icon And here we have the histogram Notice how we have two peaks here we have a very large peak and a much smaller one with lower values and a much larger peak corresponds to the pixels over land as there are many more pixels over the land Whereas a smaller peak includes the much fewer pixels which are over water surfaces We're going to exploit the difference between these two peaks to find a suitable threshold In order to mask out the water bodies Notice how around might is minus 20 and minus 19 dB we have the boundary between the two peaks So what we can do to mask out the water bodies is to create a new band So we'll go to raster band maths where you'll create a new mask which we will call water bodies and here we will create an expression which will just be a simple threshold so here we're going to select the VV band in decibels and we're going to Select only The pixels which are below minus let's say 20 decibels And then we select ok and now ok again and here we see only the pixels which are below 20 decibels in the VV channel And these are all pixels, which lie in the lower peak of the histogram And what we see is a mask that corresponds more or less to the water bodies Although there are some problems here because we see lots of pixels lots of dark pixels Within the water body So what we can do is we can go back and create a new mask And adjust the threshold so here we'll call it water bodies Let's make this threshold minus 19 decibels, so we'll write 19 in the name, so we know which mask this is then we select again VV band in decibels and here we select minus 19 And we select ok and ok again And now we have a new mask with less Dark pixels in the water bodies it's not perfect, but at least we have some idea of where the water bodies are We can now compare the two masks by selecting window tile evenly We can close this window here and view only the two masks Side-by-side we can do we can select our horizontally So this was just a quick exercise on water body mapping using SAR data exploiting the difference in backscatter between water and land surfaces and applying a simple thresholding technique I hope it was interesting and thank you for watching

Channel: EO College

Views: 17,447

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

Id: ToGnUMgevhE

Length: 11min 12sec (672 seconds)

Published: Mon Oct 30 2017