RUS Webinar: Oil spill mapping with Sentinel-1 - OCEA03

welcome back again to another world webinar my name is Miguel Castro and today I will be guiding you through this webinar which topic is always spill mapping with Sentinel 1 data so as you can see we will be working in a different study area but I will give you more information about that so before starting let me tell you the objectives of this session you will learn two main things first how to do all spill mapping with Sentinel 1 data and second what is the route service and how it can help you in your projects with sentient data therefore for this exercise we will combine the route service and the capabilities of Sentinel 1 for oil spill mapping so just before starting be aware that this webinar is being recorded so you will be able to watch again the video and repeat the exercise so so I will give you more information on that later on but just for you to be aware so let's move and let's have a look very quickly to the outline of this webinar so we will first have a look to our study area for today some brief comments about that we will then cover very briefly the theory behind the use of satellite earth observation for this type of applications we will then describe the route service and what it is so that you can understand how it can benefit your projects and then we will use the route service to perform the exercise of today at the end we will we will have some time for questions and answers so just feel free if you have doubts during the session to send them to the specific tool we have if not if not you can just wait until the end so let's get started and let's start by our study area so this time we are traveling to provide in the middle east this country is located in a section of one of the driest and least host hospitable deserts on earth and the fourth by area the Arabian desert this tiny country was a British protectorate from 1899 until 1961 and all reserves in commercial quantities were discovered in the 30s from that moment the country underwent large-scale modernization by 95 this the country became the largest oil exporter in the Persian Gulf region so last August 10th 2017 so last summer and all spill was reported in the South of chabad near the al-karim area well the al-kafi offshore oil field is located so while the cause of the incident is not clear around 130,000 litres have been leaked based on conservative estimations made by external organizations so we have here this false cold composition of Cuba and our study area today is here in the south and of course we are looking at the oil spill over the ocean so we care about the ocean part of the image so let's have a look now to the remote sensing background and understand how satellites can help in this type of applications application so also Polish ocean pollution due to oil spills remains a major environmental hazard although oil tanker accidents are well-known they are not the main cause for this type of event illegal discharges from ships or offshore platforms drilling rigs pipeline accidents or natural lakes amongst others bring together most of source most of the sources for oil pollution in the ocean methods for the detection and tracking of oil spills and illegal all these charges are fundamental importance for improving the efficiency of my team surveillance systems the key advantages of spaceborne Earth Observation together with the characteristics of synthetic aperture radar sensors there is day and night and all weather sensing capabilities enables perform the basis of available and useful tool for the detection mapping and monitoring of oil spills the matterif data from the sentinel satellites enables a new approach for oil spill mapping and monitoring the combination of the temporal and spatial resolution together with relevant analysis can lead to improvement of the decision making process so for this exercise we will be using sea bonds our data provided by the Sentinel 1 a satellite the sending satellites are included in the space component of the Copernicus program of the European Commission and the European Space Agency and in case just couple of details about the sentinel one mission it is formed by a constellation of two twin satellites faced at 180 degrees to each other is an active an active sensor that works on the c-band providing data with a short a bit cycle and with different imaging modes so let's move now to the route service so first of all Cruise stands for research and user support for sentinel core products it is an initiative founded by the european commission and managed by either with the objective to promote the uptake of Copernicus sentient data and support R&D activities the service provide its free and open scalable platform in a powerful computing environment hosting a suit of open-source tool boxes pre-installed on vapour machines which allow you to handle and process the data derived from The Sentinel satellites so what does that mean in other words well with a large amount of data produced by The Sentinel satellites the challenge is no longer data availability but rather storage and processing capacity to solve that Roose offers virtual machines so that you have the appropriate computing environment to handle the data in addition to all that Roo's also provides specialized user helpdesk to support your load sensing activities with Sentinel data and a dedicated training program so if you are working with something on data or you think you will you can subscribe to us you can become a Rouge user you can get a good for machine for free plus the support for a team of experts that can help you in your processing and some training activities such as this webinar or even face to face it Thanks so you can find all the information of the route service in our Toulouse websites we have the routes - Copernicus top view and the loose training the first one you will find all the information about the project and it is in this website where you will be able to subscribe and in the second one we have all the information about the we do such as the webinar you are attending today or other events as face-to-face trainings that we organize as well so I really recommend you to visit those websites the webinar if you have some time or later on you can find a lot of useful information there in addition we have allotted three short videos in our YouTube channel where we summarize the main steps you need to perform to become our most user so those are how to how to register for rules how to request a bid for machines so they are very short couple of minutes and there you can find all you need in order to apply for the vector machines and start the users for your projects so without any further ado let's start with our exercise today and let's see how we can use Sentinel one for volts per mapping so for that I'm going to open my voice vector machines and here we are in the main news website so let's imagine I already registered for use I already apply from my willful machine so now I just have to login on the website and once I'm in I can just go to my dog sport and here I will have my here have microphone chain so I just need to wait until it's loaded and now I can access my data machine put my my credentials so you will receive all this information via email once you use this reference so here we are in the respectful machine as you can see it's a regular Linux environment and a couple of words about the VM first of all you have full administration rights that means that you can install and modify the VM as you want you can install software either if it's open source or commercial software but just remind that we do not provide commercial licenses for that so you need to have your own one but if you do you can install all the software you want you can also very easily upload and download files from the bit machine to your local computer and for that we just need to press the key combination ctrl alt shift if we do so we access this side menu and here we can just navigate through the folders of our were Beatle machine and for is if you want to download something we just need to navigate to the bath and for example double click on the 5 we're going to download we can also upload a file by following the same procedure we just need to click here in upload file and then we can select the file now a rock on computer to be sent to the via promising so for the user size today we will be using snap so snap is the is a software to process sectional data and as as you can as you may know when when getting the receive a promotion you get already a predefined list of software that is already installed and ready to use so one of the software that is available is snap so we can just double click here and open it so let's open the software and start the process on the image so in case this is the first time you are using sapling let me give you a very brief introduction to this interface so we have here the product product Explorer where all the files that we will be using will start to appear you will see that later here in the lower left corner we have some quick menus that we can use to manipulate our our image you will see later how we use this and here we have the main area well the where the image is displayed of course here on top we have all the tools we need to process our image for example of the raster tools the tools for optical sensors radar except Exeter so let's start by opening the image for that we can click in the open product icon or just in file open project and now we just need to navigate to the path where we have our image in this case this is the path and now here we have our image in the product expert so as you can see we have index 1 so what to say about this product well the first thing we can do is to expand the product and access some extra information for example we have the metadata folder where all the useful information about the image is stored so for example if we open abstract metadata we can see a lot of information of this product for example the mission the antenna pointing the acquisition mode if it was ascending or descending etc so you can check that later we also have the band's folder where actually we have the images that were captured by the sensor so if we open here we have amplitude and intensity so just remember intensity is the square of the amplitude so this is a virtual band that means that it is not physically saved on our on our hard disk it's calculated from the it's happened by the software directly so let's open up the DV for example so we just double click on on p.m. on the file and then the image will will be displayed okay there we go so this is what we see when opening a Centon one sorry image so if you come from the optical community you will see that it's very different but still we can see some useful things first of all we see those three let's call it rectangles or sub words so those are the consequence of the position mode that is used by Sentinel 1 in this case we are using interferometric wide acquisition mode and that means that the image is scanning through the through the track of the satellite and it's acquaint three different images that are later on merge I'm happier and and that's why we have those three let's say main areas we can use the pan mode to navigate over the image and we can also use the zoom tool to zoom in so if we go over here we can already start to see a dark region in the ocean okay we will talk about that later on I just want to point out that for example we see human areas in very bright so white pixels but also if we zoom into the ocean we see that there is a lot of random behavior in the backscatter so we have some dark pixels and then some very bright pixels so the design is coming back from the ocean is actually it's kind of random it's not the same when we look over land or its I mean not as random as over the ocean so it's it's obvious that we have to do some work to the image before we can create some useful results so this very dark region that we can see here it's actually the oil spill that took place that day the 10th of August 2017 and this is our study area if we look at the image at the complete scene we can see actually that there is a trend of from left to right we could say that here it's more bright than that here on the right side and this is the consequence of the right side looking direction of this our sensor so the the Czar's the satellite it's orbiting on the other end speaking to the right and it's sensing in that duration and that's why in the near range we have a brighter areas than in the full range of the image so let's talk a little bit more about how the backscatter over the ocean is created so the back scatter of the SAR signal over the ocean is mainly a result of sea roughness and it's determined determined by small surface waves that are usually called gravity capillary waves those are also linked to the superficial ocean currents and wind speed wind speed and direction so all the films decrease the sea surface roughness and hence the backscatter this cause spills to appear darker insert images and spill free areas however the contrast between polluted and non polluted areas depends on different parameters such as the wave height the wind speed the type of oil or even the sensor characteristics such as the wave length authorization and I want to just stop a little bit in polarization if we look at this image this is the V vaporization so vertical vertical we are transmitting to sign out in the vertical position and we are only receiving in vertical but if we open them a VH the amplitude VH we will see that the same area okay starting the image okay we will sit at the same area we cannot actually see the oil spill there is a very nice feature in stamp we can combine both views if we go to windows tire horizontally and here we see the difference in VH well we don't see at all the oil spill while in VV it's very obvious so as you can see even the sensor characteristics can determine your ability to detect an oil spill a lot so let's start our processing the first thing we are going to do is to subset dementia as you can see the area the image the product covers a very big area and we just want to focus on this specific region so for that we subset the image so we select the product and we go to raster subset and here we can create a subset in different ways we can for example draw our region in this thumbnail here we can also define some pixel coordinates or apply subsets by using lock lock on coordinates so in my case today I'm using pixel coordinates so let me just define the area I'm just going to input the parameters here okay so here we have our subset area I will just click okay so now the output of this operation will appear here in the product Explorer as you can see it has index 2 and we have this keyboard subset which is linked to the operation that we have done so when you create a subset don't don't forget to save the product because if not it will not be safe I would just save it here and there it goes so now we can keep moving so the next step is going to be a speck on filter so as you can see the image oh sorry let me just show you the subset area we can expand the product and go to bands and open again completed video of course nothing have has changed it's just the same image but a smaller region so the next step we want to do is to reduce this random behavior we have over the ocean so as as you see if i zoom in a lot you will see that we have of a funny behavior of the sign-on and we want to reduce this a little bit to improve the contrast bit in the area that is affected by the oil spill and the area is free of pollution and for that we apply this spec of filter so talk sorry to apply the spec of filter we select again the product and we go to radar speckle filtering single product spectrum filter we are using one image today so no need to do multi temporal aspect of it so here we have the common interface of snap of the tools of snap we have a first tab which is input output parameters you know here we specify the input product and the output directory and then we have the processing parameters tab where we set the parameters of the tool so in our case we just need to make sure that we are saving the the product in the appropriate folder and for the processing parameters we will leave everything as default and now we can just press right so the process is let's say very fast it takes almost 20 seconds so I will just run it life so you can see it okay so almost done and now we will have a look to the output of this tool and we will see the difference of fab force our image that has this speckle filter applied and and sorry much that that doesn't have is so let's close here and let's so we have now this product number three and if we go at the end we see that we have this suffix s pay K which is linked to the speckle filter we have performed so let's open the Bands folder and again open the band amplitude V V and again let's link both useful we can compare directly so as you can see the image looks a little bit darker that's that's okay no problem but if we zoom in we can see that this random behavior or this variation has been reduced still you can see some patterns but it's not as random as in the as in the image without the spectral feature so that's good we have already we'd use a little bit of this noise in the image and if we go to the dark area where the oil spill is we can see that it's let's say that the darkness is much more homogeneous than in the image without the specular so this will actually help us a lot in order to create this extent from the image so let's close the image number two and now let's keep moving with our processing so just to give you a theoretical note on what the spec of filter is speaking noise like feature is a common phenomenon in source systems it confers two star images a granular aspect and random spatial variation and may decrease the utility of sorry imagery the source of this noise is attributed to random interference between the coherent returns and the principle of Stegall filtering is to use the variance of the complex spec of scattering and improve the estimation of the unspeak of scattering coefficient so this is actually what we have done so next what we are going to do is to convert the band to the decibel scale so this will allow us to have a baritone trust between the the affected area and the rest of the image and for that it's very easy we just need to go to the the image to the bank that we want to convert and we'll right click and select linear - from decimal so we are changing the the scale that is used to assign the colors to the pixels of the image it's just a logarithm a logarithmic scale the decibel scale so it's well known so we just press yes and now we see that the this band has been added to the product and it has this video remember this means that the band's doesn't exist it has been created by the software but if we close the product right now it's not saved so we want to save it after that we just right-click and select convert band so now the band has been saved and it doesn't have the the fee anymore so let's open it and let's compare both views again that's a little bit okay so so as you can see on the left side we have the image the the regular image and on the right side we have the decibel sour image so as you can see it's I remain a little bit brighter and we can see the contrast contrast much better now we are almost ready to do our oil spill mapping and what we need to do first is to characterize a little bit more to the the back scatter that is coming back from the polluted area so I'm going to close this view now and what we are going to do is to create a profile plot we can use this pool here in snap which is a drawing line tool so these two crates draws a line over the image so let's draw a line for example over here like this so we are going all over through the polluted area then we are in this area where there is no pollution and I so on so now we want to see the evolution of the pixel values of the back scatter all over this line and so that's the profile plot and to see this profile plot we go to analysis profile plots already we can see some trends but let's improve this first of all let's zoom in a little bit and we just select this area to zoom in okay and now we can select a bigger box size to smooth a little bit the trend that we are saying let me just change this parameter okay and let me just zoom in again okay so and I will get bigger as well okay so what we see here is the evolution of the pixel values in the decibel scale all over our or image and actually if I click on the graph I can see the value in this specific position where if you look in my if you look in the yellow line you will see that while I move all over the graph I can see exactly where in my line I am I can very easily recognize the pixel values to the specific location so the line starts in an area that is not dark so I never that's not affected by the whole spiel so we have a back scatter in the decibel scale of 19 let's say nineteen nineteen point five more or less as soon as we reached a dark area it goes down to sixteen then we have here some mixture of pixels are polluted and open it but if we go a little bit more here in this fear dot sorry in this dark area we see that the value is very stable in the range of 16 now when we reach this non dark area again the values go up and when we reach again the polluted area it goes down and so on and so on so this gives you already an idea of how the presence of an oil spill effects the back scatter and this actually helped us a lot to delineate and map the extent of an oil spill and this is actually what we are going to do in the next step so just remember we are going from a value of 19 let's say let's say 19 to a value of 16 in the dark area so if there if we do not have oil spill we have 19 if we do have an oil spill we have 16 in this specific case so let's move on our next step will be then the old school mapping so to identify oil spills in the ocean who will use this Sentinel one data and a dedicated tool that snap offers for this purpose however it has to be highlighted that only possible oil spills are detected since some specific conditions of the ocean can generate similar patterns to the ones of an oil spill what do I mean by that well here we have a dark area and this is clearly an oil spill because we know but it can happen that due to the conditions of the ocean due to specific wind spins wind speed and directions we can see dark areas over the image for example here that can be detected by the algorithm as an oil spill but actually they are not is just an addition of circumstances that are creating this dark area so this is something that does that we cannot change something that happens and that's why when doing oil spill mapping with with SAR we always talk about possible detections we cannot be a hundred percent sure unless we have blunt truth data or some type of validation data of course the the extent the form sorry the shape of the of the dark region the extent if we also tragedies with other data for example if it's coming from a ship we can know the location so specific characteristics of the of the adult area can tell us already if it's really an oil spill or not but actually it can happen that that you are detecting something as a known spill while it is not so in this specific case validation data is very important as in many other applications applications of applications or with remote sensing data so we all spill detection tool that we are going to use includes two pre-processing steps mask out the inland areas and volumetric calibration so that pixel values truly represent the radar back scatter the reflecting surface after those pre-processing steps dark spots are detected using an adaptive threshold algorithm where the local min backscatter level is estimated using pixels in a larger window after that a threshold is set to specific amounts of decimals below the local min calculated before and pixels within the window with values lower than the threshold are detected as dark dark spots finally the detected pixels are clustered into a single cluster and those with sizes smaller than a predefined area is selected by the user are eliminated so let's do that we go to radar SAR applications ocean applications always spill detection so this is applets already included in snap then we just need to set the parameters so the first one is the input image we are going to use so number 3 the one with the spec of filter then we will apply a landsem mask so we want to remove all those areas that are not in the ocean because doesn't make sense to run this application over the LAN so we can just remove that so we leave all the parameters here as default to perform the Lancie mask then in the calibration tab we leave everything as default here we will convert the pixel values to 2 to calibrate the pixel values where we can be sure and that the backscattered that is coming back the sensor truly represents or has a physical meaning and then in those three consecutive tabs we will perform the ozpin mapping so we can select the input bands and we will just use the Sigma V V remember at the beginning the the VH position doesn't help a lot in this case so no need to use it and then we need to set the background window size and the threshold shift so for this specific application we are using a broad background window size of 1400 in terms of window size and then we need to apply a thresher v so here we are creating this window and we are saying and we are telling the software or the algorithm that we are expecting shift in the values of 3.5 by 3.5 well if you remember the plot I was showing before we were seeing that the pixels sorry that the the decibel values were dropping from 19 to 16 more or less so that's 3 3.5 more or less so that's why I am selecting this parameter Y 1400 well this is something that you have to try one running this type of algorithm there is no predefined value that can be used of course you can do some quick tests but actually the idea is to analyze the back scatter from the oil spill and then try different window sizes to see how well it's performing and depending on the shape and extent of your oil spill you might want to change those parameters so those works for this specific case in Kuwait but if you are repeating this exercise later on you might want to change them or even to try several times then the next step will be the clustering so here we will set the value to the default one so no need to change anything and finally in the right tab we just need to specify the name of the output product and the folder where we are saving our output so for the name I'm just going to add s be K for speckle and I'm also adding the parameters that I've used so hundred and 3.5 the way if you repeat the same analysis with different parameters you can keep track of your methodology and your results so after that we can press the Run button and have a look at the result so I'm not going to run the process because it takes a little bit so I will show you directly the result so here we have the output and now if we expand it we can go to the band's folder and open Sigma DV I would close this one we can also actually open the walking so we see here Sigma V V this is the output of the oil spill detection but we are not still checking the mask so you can also open the the Sigma V V in decibels so to improve a little bit the visualization okay so here we can see that actually the the contra is much better so now let's have a look to the mask that the algorithm has created this mask is here and it's named Sigma Y V V oil-spill big mask so we open it we can see the mask now appearing here there it goes by the by default the the mask would be white but you could change the color I've done that before if you go to the core manipulation top table you can change for example to two green if you want I mean it's up to you so in this case we are using red but now in order to really visualize how well the tool has performed let's combine both layers in the same view so for that we go to lay your manager we sorry we go here we go to linear manager we select + and image of band and now we select the Sigma 0 V V oil-spill mask so this is the band we want to overlay with the actual view we have so we click finish and there we go so here we have the image and the oil spill mask that has been created by the ability as we can see it works pretty well I mean if we zoom in a little bit we see that for example in the edges I mean it's very it's it's delineating the dark area very well of course in some areas not working in very well for example here and this is something that that depends on your window sighs mainly I would say not that much in your in your decibel thresher or not you are specifying if you change the threshold in my meter job that you will include some of those pixels but said that the window size that you are moving to analyze the image is more important so for example here it works pretty well because all this area is dark and here we have a clear spot and it's not detected or for example here but of course in some other areas for example here these are there's a I mean there's areas that are not detected for example also here so it's it doesn't work completely but it gives us good delineation of the area that is affected and for the purpose of oil spill mapping or of monitoring the progress of an oil spill it is good enough something that I want to highlight it is those false positives so here we have for example some detection of the oil spill also here and those are actually look-alikes are classified as oil spill why do I know that this is a lookalike well just because seeing the extent of this main area it doesn't make sense to have for this specific all spill doesn't make sense to have an oil spill here since there is no connection between this area and this area so this is the consequence of the localized phenomena that can happen insert images and that can be identified as as all spills actually so we are almost done with the methodology the only thing now that we need to do is to project the image properly because right now we are still working on this our geometry but we have to let say turned image properly sedates projected so we are going to do is its then to reproject our image in a specific coordinate reference system and to perform this step who will use the ellipsoid correction so in sum you have different options to perform geometric Corrections for our imagery in this case we are using the ellipsoid correction and not the terrain correction or range Doppler terrain correction since our study area is in the ocean we do not need to correct for geometric distortions of this our back scatter because those distortions do not happen in the ocean so let's let's do that and I'm skip on radar geometric terrain correction sorry ellipsoid correction your location grid so what we are going to do here is to use the geolocation of the image to project it in our projection setting so we have again the same input-output layout so just make sure you are using the appropriate input and that you are saving the output image in the appropriate path again and this is the name we will give it and we will add the e C suffix to the products so that we can know that we have performed this type of operation in the processing Prommy to stop you just need to change the map projection so we want to project the image in a UTM zone and there is a very convenient way to do dialing in snap we can just click here and in this drop down menu we can go down and select UTM automatic so the software will locate the image and we'll apply the parameters that belongs to this specific UTM zone so in this case is zone 39 so that's great now we can just press run and and have a look to the results so again I'm not running this analysis to save some time so I will show you directly the result so let me open the product there it goes and here we have so I will close this view here and now let's open the output the Sigma V V in decibels as before and now we will add on top of that the mask that is also projected there we go this okay okay let me just read it again maybe I was too fast on oh there it goes okay so if you are aware of the position of collide in the wall you can see now that they must make sense we have the ocean to the right and the land to the left before it was the other way around because the image was acquired by the sentinel one satellite while it was in a descending orbit and it was looking to the right so that why that's why I did much was turned over so now we have this output and we can then export for example the output from snap let's imagine we want to put this in a GIS system where we want to perform some post processing analysis such as some buffer around the detected area or for example we want to overlay this with with some base map etc so it's very easy to do to do that we can for example open the oil spill mask and if we want to export it we can just right click and say eh export view as image so here we can slip what we can first of all navigate to the path where we want to save the image say select the format for example jyothi's and then we can just give it a name for example mask and just press safe so the okay well in this case let's just okay so it seems it was saving it as a thief but I've done already the process as a thief you can also explore the image by going to file export yep if it's the same as right clicking on the image so let's have a look to this output product no no no solver for that I'm closing now I'm minimizing snap and let's open QGIS so it's very convenient because in that way you can do all the analysis in snap and you can for example export the final product as geo teeth you can then send that to your computer or still keep working on your virtual machine so let's open the diff file the j5 sorry and here it comes so we are within GI yes we can run any of the post-processing analysis that you might think are relevant in this case for this webinar I'm just putting this information with a base map so for example we can go to web if you have the upper layers plug-in we can overlay this with some Google suite base map and for example we can see here the extent of the oil spill still the look-alikes that were detected the road network etc so you can think about many many applications and many ways of combining this information with with external data that can improve even more the ability to use the output from central imagery to this type of application and the decision-making process that is required to - type of situation so before finishing this webinar I just want to highlight couple of take-home messages that I think are relevant so let me just go back to my presentation so the the key message that we have to remember is that with the new Sentinel satellites the challenge in satellite remote sensing is no longer data availability but rather have to store and process all the information in addition to that it is necessary to explain how the data can be used support users in their application the route service is here to solve those problems by providing dreadful machines to store and process data and by offering a dedicated helpdesk supported by a team of remote sensing experts to help you in your projects with sending later so before moving to the Q&A session let me tell you again how you can repeat this exercise by your own so for that you have to go to the loose Copernicus website and register as a root user and while you apply for the root speed for machine and remember we have three short videos in our YouTube channel where we explain all those steps you can just write here this code which is OSHA zero three if you do so we will know that you want to repeat this webinar and we will provide you with all the training kit that you need that is all the input images and all the outside data and step-by-step guide to perform the same analysis I've shown today so that talk from my side I hope you have learned something new today so again thank you very much for attending this webinar I hope you have learned something new today I hope you find it you found it interesting so thank you also to the rest of the team that has been involved in the in the webinar and I'll see you in the next one ciao

Channel: RUS Copernicus Training

Views: 10,634

Keywords:

Id: Jm1kIZBhJK0

Length: 43min 36sec (2616 seconds)

Published: Wed May 16 2018