NASA ARSET: Mangrove Mapping with SAR Data | Tutorial

[00:00:00]hello this is erica potus and i will be your presenter today thank you for joining us on this third session of forest mapping and monitoring with sardata the focus of today's webinar will be on mangrove mapping this session has been assembled together with amber mccollum and juan ruiz torres perez from nasa ames and sean mccartney from nasa goddard for this training we have four two hour sessions and there is one left which is next thursday may 21st and we will be presenting the same content in two different live sessions one in english and one in spanish you can find all the course material on the website listed here and after each session we will have a question and answer portion feel free to type your questions in the chat box along the way and we will try to get at to as many questions as possible at the end we will also post the questions and answers on our website after the training so if we don't get to a question uh we but please take a look at the google doc that we will post on our website and also you can email us and we will get back to you as soon as we can we will have one homework which will be available on the course website and the homework will cover content from the lecture as well as from the exercises to receive credit for the homework you must submit all answers via google forms by the deadline and we will announce the deadline and the homework at the end of the final session on may 21st to receive a certificate of completion you must attend all live webinars and complete the homework now it does take some time to process these certificates so you can expect to receive them about two months after the completion of the course so the two prerequisites for this webinar series are the introduction to synthetic aperture radar and at the advanced webinar star for land cover applications you can access those through the links here and you also need to have a google earth engine account set up it's free but you do need to register all of the course materials are posted on the rset course page web page and you can access that through the last link on this page okay so this is where we are in this webinar series on mangrove mapping part three there's one more uh webinar and that will be on thursday on estimating forest stand height so by the end of this presentation you'll have an understanding of what are the challenges in mapping mangrove ecosystems and you'll understand these radar signal interaction with these ecosystems and we'll also do a demo where you'll be able to look at doing some time series change detection in mangrove ecosystems and do some general calculations of mangrove biomass so let's get started and just by providing a little bit of background on mangrove ecosystems and their importance mangroves are some of the most productive ecosystems in the world they're found in intertidal zones in the tropics and subtropics as shown on the figure on the left and these ecosystems are comprised of trees or shrubs they're found in shallow sandy or muddy areas that are adapted to esterine or saline environments and they are truly unique because they are the only trees capable of tolerating large amounts of salt in the water and this capability as well as surviving in low oxygen soils is a result of their root adaptations these ecosystems provide numerous services that sustain the livelihood of millions of people and some of these services in addition to carbon sequestration include the protection of coastline and infrastructure against severe storms nursery of fish and crustaceans and the production of lumber and charcoal in addition these ecosystems remove co2 from the atmosphere through burial of organic carbon in its sediments and they serve as huge carbon sinks though mangrove forests cover a small land area less than one percent they are estimated to be amongst the most carbon-rich ecosystems within the tropics so mangrove ecosystems have recently been included in the ipcc the inter governmental panel of climate change the climate mitigation strategy through a number of wetland supplements so their importance has really been elevated there are different types of mangroves primarily red black and white red mangroves are the most salt tolerant while white are the least mangroves can exceed 60 meters in height and are able to attain high values of above ground biomass in addition their structures can differ as you can see on the figure on the left some can have large curved roots while others don't these ecosystems are being threatened or lost the rate of loss throughout the 90s was twice that of terrestrial rainforest over the same period and it's estimated that about a third of mangrove forests have been lost during the last century some of the challenges in mapping mangrove ecosystems are the need for high resolution imagery since these ecosystems tend to cover narrow bands along the coastline also given that mangroves are found in tropical and subtropical areas and along the coast these are areas that are more prone to cloud cover and hence finding optical imagery that is cloud-free over these areas can sometimes be difficult and finally the conditions in these ecosystems is dynamic there are differences between high and low tide for example and these differences can impact the characteristics of the the the the satellite signal especially if you're using radar as well now let's discuss the sar signal characteristics and mangrove ecosystems so first let me start out by reviewing the back scattering mechanisms of the radar signal remember that the length of the wave will determine how it interacts with surface objects the wave will interact with features on the surface that are approximately the length of the wave the size of the components of a surface will determine its roughness so surface roughness refers to the average height variations in the surface cover from a plane surface and surface roughness is measured on the order of centimeters so whether a surface appears rough or smooth to a radar depends on the wavelength and the incidence angle a surface is considered smooth if the height variations are much smaller than the radar wavelength as a useful rule of thumb the higher the back scattered signal the rougher the surface will look like the surface being imaged will look like all right so let's review the [Music] uh backscattering mechanisms that we see here the first one is a smooth surface also known as a specular reflector and the smooth surface it acts like a mirror for the in incident radar poles such that much most of the incident radar energy is reflected away from the sensor and then that causes those areas to appear very dark open water surfaces tend to be specular reflectors the next scattering mechanism is a rough surface so when the surface height variations begin to approach the size of the wavelength then the surface will appear rough and the rough surface will scatter the energy approximately equally in all directions diffusely and a portion of the energy will be back scattered to the radar so surfaces that have some level of roughness will have some [Music] low back scatter but not as slow as low as a specular surface so then there is the opposite mechanism oh and the the rougher the surface the greater the energy uh scattered back to the satellite okay and then the other mechanism is volume scattering and volume scattering refers to the energy the radar energy being scattered within a volume or medium and it usually consists of multiple bounces and reflections within the medium okay whether it's a vegetation canopy whether it's a snowpack whether it's it's the soil and then the final mechanism is double bounce and that occurs when two smooth surfaces form a right angle facing the radar beam and so the beam bounce it bounces twice off the surfaces and most of the radar energy is reflected back to the radar sensor double bounce is commonly seen in urban areas and in areas where there's flooded vegetation and i'll quickly go through these examples just as a refresher on the intensity of the back scattered signal from these different surfaces okay so when you have a dark a a smooth surface a specular reflector like an open water body the signal will reflect away from the radar so in this case the radar is in the left and then the energy that's being emitted from the radar is being reflected towards the right okay when it hits a smooth surface so open water road and the pixel will appear very very dark and you can see that in this example the area that's circled with uh in yellow is part of the amazon river and as you can see it appears very dark [Music] and this next example shows rough bare surface so bare surface or a rough surface would be say an area that has been deforested a tilled agricultural fields water that has been roughened up by the wind so whenever you have a some level of roughness a small level of roughness you've got the signal being scattered diffusely and part of that signal then is returned to the radar and so for areas like these the back scatter intensity is low is very low but not as low as a specular reflector so not as low as the previous example this next example is volume scattering by vegetation and the area in the re in the yellow circle is a forest in the amazon basin and so the intensity of volume scattering depends on the physical properties of the volume such as the variations in moisture content and structure and also depend on the radar the wavelength polarization and incident angle but the point here is that volume scattering will have higher back scatter than surface scattering and and certainly specular scattering so this is an example of volume scattering in a forest and as you can see scattering may come from the leaf the leaf canopy at the tops of the trees the leaves and the branches further below the tree trunks and the soil at the ground level so the figure illustrates all of the possible scattering mechanisms within a forest there's direct scattering from the tree trunks there's ground crown scattering crown ground scattering ground trunk scattering trunk ground scattering crown volume scattering so there are potentially many different possible interactions of the signal with the vegetation or different parts of the vegetation and obviously the length of the wavelength will also determine uh which will be the primary or the dominant scattering mechanism in um in a in in a forest or wherever there's vegetation and then finally this example shows double bounce it's delineated here by the yellow circle and it's an inundated forest that means that the trees are on standing water and the signal is very very uh strong very high back scatter and the reason for this is because the signal bounces off the water which is a specular reflector off the water underneath the vegetation and it bounces onto the tree trunk or other components of the tree and then bounces back to the satellite so we can't measure the amount of standing water just that there is water above the surface but because these inundated vegetation is dominated by double bounds it's it appears very bright on radar images for mangrove ecosystems there are three types of scattering mechanisms there's direct or single bounce and yellow volume scattering in white and double bounce in red the dominant scattering mechanism in mangrove forest strongly depends on canopy structure trends and volume and double bound scattering vary much more than in other types of forests in particular in mangroves volume scattering decreases in closed canopies and double bounds increases in open canopies inundation at the time of data acquisition also can have an impact on the radar signal in open mangrove forests as you already know hv polarization is dominated by volume scattering and hh and vv polarizations on the other hand contain contributions from the ground because they penetrate deeper through the canopy than cross polarizations hhh is particularly sensitive to inundated vegetation and in such conditions will be impacted by the occurrence of double bound scattering now in mangrove forests however double bounds that that double bounce term that strongly impacts the hh channel may be reduced by the presence of roots those curved roots that are so distinctive in mangrove forests so what happens is that the the radar signal is scattered and attenuated by these roots l-band backscatter at both hh and hv polarizations is therefore several db lower for mangroves than for other types of forests and so here you can see that we have an l-band image from aylos pulsar and this is over gabon and the one on the right is an hh the one on the left is hv the dark areas along the coastline uh the all of those areas that are hugging the coastline within the red circle those are mangroves and you can see that they are darker than the surrounding forest this is an rgb of the previous image where hh is in the red and blue channels and hv is in the green channel you can see much better the mangroves appearing as a shade of dark green along the coastline now contrary contrary to inland forests were backscatter increases with veget with increasing vegetation due to volume scattering for mangroves backscatter from volume scattering is reduced in tall mangroves and is increased with shorter ones in some cases backscatter from volume scattering may become similar to that of inland forests however the texture of mangrove forests is tends to be smoother than just other non-mangrove forests and this is in part because of inland topography and to the overall homogeneity of the mangrove canopy structure if you have multiple polarizations you can use them to identify mangrove forests from other land cover classes especially at the longer wavelengths here we have a comparison of a mangrove ecosystem and the images on the top are l-band and the ones on the bottom are c-band the relative impact of the scattering mechanisms changes significantly with radar wavelength ground and double balance contributions increase with wavelength seaband backscatter mainly occurs at the top of the canopy and that strong canopy interaction provides for increased sensitivity to canopy structure and consequently a certain potential for distinction of mangroves and other vegetation so you can distinguish different types of mangroves with different canopy structures now the smoother canopies typically display a lower back scatter at both vv and bh polarizations than rougher canopies with a higher degree of structural variation at l-band there is attenuation of the signal due to those complex branches and aerial routes [Music] this table contains a very nice summary of the general penetration depth and dominant scattering mechanisms in mangrove forests for different bands for example at l band microwave penetration into the canopy is about as large as half the canopy height scattering is dominated by single direct bounce in tall forest and volume dominates in the shorter shrub mangroves double balance which means the back scatter will be very high increases significantly the lower the biomass and the more open the forest at a shorter wavelength such as c-band for example penetration depth goes about one third of the of the top of the forest height and the scattering mechanisms will consist of single direct bounce and volume scattering from the upper canopy with a small surface and double bounce components this double balance component will increase significantly in open forests and low biomass areas [Music] this table contains the backscatter range for mangroves for different bands and polarizations and these are general trends and may not be constant so for example backscatter may increase for low biomass shrub stands up to standing force where it then might decrease it's difficult to map the extent of macro force using radar alone in particular when the adjacent inland cover is forest or shruplands because of similar backscatter this can also be difficult with optical sensors because greenness may not be sufficient to distinguish mangroves from other vegetation types this example shows mangroves in cape york australia and on the left is an aylos pulsar image so that's hh in the red channel hb in the green channel and the ratio of hh to hb in the blue channel and then on the right are the same mangroves that can be discriminated from adjoining forests by using a landsat derived foliage cover product areas where mangroves have non or sparsely vegetated surfaces like sand or mud flats on their landward margins can generally be discriminated and mapped because of their higher back scatter at both l band hh and vv polarizations there several studies have recommended that the best approach to mapping mangroves is really a combination of data sets obtained from different sensors land cover classification can generally be performed with the radar backscatter as one of the layers along with data say from optical instruments such as landsat or sentinel 2 and there are global maps of mangrove extent derived from optical data that can be used to extract the area of mangrove force from radar data and then and these have recently been improved using elos data but it's generally more efficient to start with existing but reliable remote sensing products on where the mangroves are and then improve upon those with the radar data one solution is to confine the mapping of mangroves to just those areas where there's a higher likelihood of them occurring for example mangroves are unlikely to occur in sloping grounds or at elevations above sea level higher than 10 meters so that those baseline maps can initially establish where mangroves occur within uh some sort of reference year and then you can refine that baseline based on changes within areas uh they're most likely to occur and and then you can quantify those changes using radar data another approach is to use radar in combination with data sets from other sensors so a lan a land cover classification can be uh for example generated from radar data together with optical data from either landsat or sentinel 2 classification results can vary greatly though with this approach due to the availability of polarimetric layers and due to availability of radar data at different bands so you're going to get a different results whether you use l-band or c-band so there have been studies that have found that sentinel 1 did not provide significant improvement over landsat based land cover classification however other studies have shown increased accuracy over 10 using fully polarimetric radar sat2 data this image shows a classification of mangroves in queensland australia so the areas that are pale green have low biomass and the areas that have high green olive green are areas of high biomass forest without prop roots and the areas that are red are areas of high biomass bores with prop roots so while the literature is not definite on the backscatter signature of mangroves it's got quite a range the upper biomass level detectable with radar is similar to other forests so it's about 200 tons per hectare for key band 100 for l band 50 for c band and 25 for x band after that point the observed back scatter reduces due to absorption by the dense aerial root system found in mangroves so so it is different here the the backscatter response to what would be expected in a regular forest the regular forest as vegetation increases backscatter will increase until it saturates here as vegetation increases backscatter increases up to a certain point up to a certain biomass level and after that it actually decreases so you have higher biomass mangrove ecosystems that have low back scatter and we saw an example that we saw an image of that a couple slides ago now this holds specially for hv however with in observing scrub mangroves at hh and vv so these are low mangroves low biomass sometimes they these polarizations hh and vv also display high back scatter and that's due to the increased penetration within the canopy and the double bounds interaction with either the water surface or water saturated ground so this is a summary of the radar data available and note that there are legacy radar data sets current and upcoming the ones with the green box indicate that these data sets are freely available however you look at the legacy data sets jrs1 for example there is a global jrs-1 mosaic generated for 1996 that is freely available there is ers 1 and 2 data over select areas that you can download for free through the alaska satellite facility so csi jrs 1 and ers 1 and 2 you can download those data over select areas through the alaska satellite facility nbsat was a european satellite that operated at seabed from 2002 to 2012 and then halos one is a japanese sar that operated actually from 2006 to 2012 and then we have the current sar data sets the only one that's operational and the data are free are the sentinel one data these are from the european space agency again that is a c band sensor launched in 2014.

[00:30:08]and then upcoming there's nysar we've talked about nysar that's a nasa and indian space agency satellite that will have an s-band and an l-band sensor and then there's a biomass which is a satellite that will have a p-band sensor and that's from the european space agency and that's um both of those are scheduled to launch at around the same time frame so that concludes the theoretical portion of this webinar and next we'll move on to the hands-on exercise okay so let me just start out by describing the exercise overview we will learn how to process and analyze radar images will actually be in radar images engine and we will be using then google earth engine as a platform for our analysis so we will be doing some land use change or biomass sorry mangrove change analysis and we will also estimate above ground biomass for mangrove forests in terms of the software and data that will be needed for this demo we will be using the sentinel toolbox and the analysis is going to be done on google earth engine so you do need to have an account active on google earth engine both of these are free there is no cost to set these up in terms of the data sets to be used we will be using global l-band pulsar 1 mosaics as well as pulsar 2 mosaics those are already on google earth engine we will be using jrs1 mosaics or mosaic so that is from the uh the the direct lead from the jaxa webpage so we will download those and then import those in or import the tiles into google earth engine for analysis and this is a 1996 radar mosaic it's an l-band mosaic and then we will be using a global mangrove watch vector and this delineates the mangroves it's uh globally and we will be using that to narrow our area of interest so let's get started with the l band sar mosaic tiles from jaxa so jaxa has generated these global mosaics from the data it's acquired from its satellites so back in the 90s the jaxa which is a japanese space agency had a satellite called jrs1 and it collected images at l-band there was an l-band radar sensor and so what the japanese have done is they've created a global mosaic from jrs1 for 1996 and it has also created global mosaics from its more recent l-band sensors which is halo's pulsar one that goes from 2006 it flew from 2006 to 2011 and then the pulsar 2 which was 2014 i believe and it's currently flying collecting data so these mosaics can be downloaded for free and you do need to have an account with jaxa in order to download these mosaics however the one that we are interested in is just the jrs one because the elos pulsar mosaics are on google earth engine okay so here is the web page the jaxa web page and in addition to the mosaics the backscatter mosaics they've also generated a product on forest non-forest so they've used these mosaics to generate these global forest non-forest maps these their yearly global force non-force maps and they've got a number of updates here um on the data sets that they've made available and they talk about the algorithm used to generate the forest non-forest maps and so these mosaics are at 25 meter resolution so in order to download the data okay you need to first register yourself and you go to down to number four and you click on this link and you then provide your information and you register yourself and they will send you a confirmation email with your password once you have your password then you go back and you select this link the data set download side is this one and so that link will take you to this page okay so this page contains the different mosaics that are available that you can download and we've got again it's 25 meters they're global for jrs one it's 1996 that's just hh polarization and then for other years uh there's there are mosaics available 2007 8 9 10 and then 15 16 17 and 18.

[00:36:34]okay also 25 meters these are hh the and hv the radar backscatter mosaics are hh and hv and then there's also that product that i mentioned that they've generated from these radar backscatter mosaics for non-force maps okay they also have these mosaics that were assembled as part of what was called the global reinforced mapping project at the time over the tropics and these mosaics show you the data that's available right so it's the tropics but it's available for different years which is uh quite nice so you've got 1993 4 5 6 7 and 8.

[00:37:27]okay and then there are some low resolution products if you're not needing the 25 meter products you can download the same data sets at 100 meter resolution and at one kilometer resolution and at 0.25 degree as well the forest not forest maps okay so let's download the data set that we're interested in now our area of study will be or our focus area will be in myanmar and this is an area where there's been a lot of mangrove loss so we will select this grid here let's go back i might have done that too quickly it's this grid this one here and this is all in the powerpoint okay so once you select that grid you go to another one so we will select this area we're interested in in this coast right here okay and then we will sel we will download five tiles from here and these are these two these two and then this one so let me just show you in the powerpoint files the tiles that we're interested in so here you go we will be downloading these five tiles okay so you need to download each one one by one okay so you click on the first one and you select download and it will automatically download the file onto your computer okay so once it downloads you need to unzip that file and the format it's in it's an nv format so that means it's a raw file and then it's got a header file associated with it that has all the metadata relevant to that image so once that file is downloaded then what we do is we open the sentinel toolbox because what we want to do is convert that file from an nv file to a geotiff you don't necessarily have to do this with the sentinel toolbox you can do it with for example uh qgis you you guys recognizes mb files and you can just export it as a tiff so what we'll do is let's just go and open our file and the way to open that file is you go to import generic formats oops file import generic formats and select envy okay and then i've downloaded the files already and i've unzipped them and so what i'll do is i'll select the hh that's the one so each file it comes in a folder with the main file which is hh polarization and then it's got a file that has uh the date the date of acquisition of each pixel the incidence angle and it's got a a mask so the areas that were masked out when processing the data okay so we import we select the hdr import and it open it so you click on you open the the file name and there are three folders and then you click on back you open bands and double click on band one so that's the image and what we want to do is export to you go up to file export select geotiff and then select the name of the file would keep the same name and just add the tiff uh um extension at the end and then press export you'll do this with all of the five files because you want to have them as geotiffs okay so just make sure that when you load the next file and so let me just show you here very quickly let's load the next file just select this one number two when you really select the band okay it's easy to just kind of open it and then go up here and do the export and what happens is it's it's actually going to export this one so select the band and go through the same process export duotif okay so that ensures that you're exporting that band all right so once you have your files so we won't need sentinel the central toolbox anymore that's that's the only part where we need sentinel toolbox in this demo the next thing we do is we go to google earth engine okay so here we are on google earth engine and i've got my code editor here in the middle and what i do in order to import an image is i go into assets i click on new and then you can upload different types of files you can upload a shape file we'll do that you can upload a geotiff file so let's upload the geotiff files some image pops up and what you need to do is select your tiff image and you have to do this for each image right so so let's go and select that that we created and let's select your first one [Music] uploaded them code i'm just going to give it a name and i'll call it um jrs uh first okay first image and then i click on upload and if you go here on the right on the far right this window asks being uploaded and it tells you how much has been uploaded so it will give you an idea this is nice because then you know how long it's going to take to upload your image all right so once that finishes you'll see that there's there are no will spinning then just go go to that file which is the most recent one and click on that question mark okay and then this is the name and let's click on view assets so that when you click on view assets then this window opens and you can see these details about your your image so this is the image id this one up here and you can call your image through the uh using code and using this image id or it's much easier to just import that image so click here click on import and that's it okay so let's close e here and what happens is in your text editor there's something called imports up here and what you'll see is the image that you just imported it's called image jrs first image so i'd like to change the name rather than just having it being called image i always like to use a descriptive name so this is uh the jrs first okay and you do the same for all of the images so you import all of them and so i will i'm actually going to delete this because i already have the images imported but let me just show you what i've done here so under import i've imported the five images jrs images one two three four and 5 right so i've named them i've given them a descriptive name and then i can call them in the code the next thing we want to do is we want to upload another file and this one is on mangrove cover so man global mangrove distribution and this is a file that has been generated by global mangrove watch by it's a joint project but it's called the global mangrove watch product and you can access it through the link provided on the pdf so we go to this page and this page has the global mangrove watch files 1996 through 2016.

[00:48:59]and these are all vector files the one you're interested in downloading is 2010.

[00:49:07]so the 2010 file has a baseline of global mangrove distribution again these are vector files and if you want to learn about how these files were produced there's more information here and here but they've used the radar data from jrs and halos pulsar to generate these global distribution mangrove files okay so download this one and then let's upload it into google earth engine and the way to do that is go to google earth engine again select new under assets and this time select shape files and it'll be the same process you drag your shape file into the selection box to select files you give it a name and then you upload it again the same thing it will tell you how long it takes to upload this one will take a little longer because it's a larger file so here's what we do so going to data it needs at least the essay checks the shp and the [Music] the uh let's see the dbx the dbf files let me just double check on that so click on these three okay so i need to lease those three and then you can just rename it i named it mangrove mangrove vector just name it however you want and go through that process of upload and then same thing i've already uploaded it but same thing you'll have the image appear here under tasks with a cog wheel and it will tell you what percentage of the image has been uploaded and then once it's uploaded same thing you press on the question mark and you go to view assets and you import that file so now that file will appear in your imports list and you can see it here in my imports list i've named that file actually i've named it mangroves okay so it's a mangrove 2010 and i've named it mangroves to identify all right great so we've got those files uploaded and a lot of you were asking in the previous demos how you uploaded files so this is how you do it and the next step is in loading the pulsar mosaics so google earth engine has the pulsar global mosaics the yearly mos x it also has the forest non forest maps so just type pulsar up in the search window and click on pulsar 2 palsar yearly mosaic to get information about the the file okay there you go so the it tells you that they are hh and hv and there's the incidence angle information the date of acquisition and the processing information and there's a description again it's 25 meters the data are at 16 bit digital numbers and so you do need to apply a an equation to convert that dn value into decibels and we will go through that okay so the data have been processed corrected but they are in digital numbers so that's that's one of the things we need to do as well as apply a speckle filter okay and then one more thing is that these data are [Music] there are different years um so one thing you can do is import the collection and then filter it according to years or what i've done here is i've called the specific years so what we've done here is load the pulsar global yearly mosaics okay so in this case uh there's the we're loading the 2007 the 2010 and the 2017 yearly mosaics and um they're being called pauser 2007 2010 2017. okay and then we're also loading the srtn 30 meter dm again if you want to see more about the srtm digital elevation models on google earth engine this is the one that we selected and this is the name of that data set so we're just calling it srtm okay the next thing is the jrs1 images that we uploaded those were individual images so we want to create a mosaic right and in order to create a mosaic we use this command so we're calling the mosaic jrs 1996 and we're using the ee image collection so we're saying take all of these images the five images individual images and mosaic the next thing we need to do as you know with radar is that we need to filter the images so what we've done is applied the same smoothing filter that's been applied in other demos that we've done here we've defined the filter to be dirty it's slightly smaller you can certainly play around with that and what we're doing here is just saying okay so take the pulsar 2007 global mosaic hh and hv and apply the smoothing filter and then jrs it just has one band it's hh and apply the smoothing filter all right so once we do that then we convert the images into db and this is the equation to convert the images into sigma not in in db in decibels okay so it's the 10 log 10 of the dn number squared plus a calibration factor the calibration factor is -83 for pulsar for the pulsar mosaics and images and it's minus 84.66 for the jrs mosaic so what this is doing is it's applying the the equation with the specific calibration factor for jrs and then the same equation but with the specific calibration factor for the pausar mosaics okay and we're doing this for both uh hh and hv wherever there is hh and hv remember jrs has just hh all right the next thing we want to do is we want to define our area of interest and the area of interest that we want to define is off the code of myanmar okay so so what you do in order to define your region of interest is select the polygon select the draw shape polygon and this is what's going to come up the default is is is called geometry so you draw your area of interest and here we're drawing an area of interest along the coast here of myanmar and then you go in and you change the name of geometry you change that to by clicking on that cogwheel you change it to roi so that means region of interest okay i've done that already so this is just going to delete this i've done that already here this is my region of interest all right so once we select a region of interest then we want to clip the images to our region of interest our area of interest because remember these images are global right so we go ahead and we do a clipping according to that region of interest that we define so we're taking the db images that we created in the previous step here db for each of the images and we use this clip command according to the region of interest so we're clipping it according to the region of interest and we're calling those new images roi so the same name and then we're adding roi at the end all right and the other thing that we're doing is we're also clipping the srtm uh elevation image so that's a global product the srtm elevation image and so we are clipping it according to the roi and we're calling it srtm region of interest okay so once that's done then you need to add those images to layers in order to be able to visualize them so let's just go ahead and let's add them so we're adding all of them the srtm roi and so we're stretching the values here just to visualize it this is all trial and error i'm just this is you want to find the best stretch to to visualize your image so srt mroi and it's called srtm in the layers and then we're taking that vector that mangroves vector that is a global vector and we're adding it to layers as well and we say okay give it a red color so we're just calling it mangroves and then we're adding each of our images to the layers bar again we're stretching according to the the values in the image and this is this is trial and error you can play around with this until you find a stretch that that best allows you to visualize the image but in this case for uh hv we're using a stretch of minus 27 to minus 5 and for hh minus 15 to minus 3 and then for jrs minus 25 to 0.

[01:02:43]all right and then one thing that we've done here also at the end has been to create some rgb images so there are three rgb images um there is the there is the let's see the image the first one is looking at the difference between 2007 and 2017 only okay so we're using using the h band to look at 2007 which is in the red and blue bands and then we put 2017 in the green band the second rgb we're visualizing we're doing a our rgb between 1996 and 2017. so the 1996 image is in the red and blue bands and the 2017 is in the green band and then finally the last rgb is the three different dates the pulsar dates that's 2007 2010 and 2017.

[01:03:59]all right so let's just run that and let's take a look at our images and this is and you can just deselect that roi because it's kind of it's going to get in the way when you want to visualize the images so so the images we loaded are so let's start let's start with srtn okay that's our elevation file this is our mangroves vector file now this is a global file okay so that's the mangrove vector the global distribution this is our jrs file mosaic that we created 1996.

[01:05:11]and then this is pulsar 2007 pulsar 2010 and pulsar 2017.

[01:05:22]all right so really to visualize those let's upload the rgb so this is 2007 and 2017.

[01:05:38]so remember anything that is pink means that it was high in 2007 low in 2017 because pink is a combination of red and blue which is what we have in the 2007 channel and let's just zoom in so you can really see the amount of change that has occurred here in this area so this is between 2007 and 2017.

[01:06:06]all of the pink areas are areas where there's been loss mangrove loss and the green areas has been is is either there's been a regeneration of the mangrove so it was uh it had been cut in 2007 and it grew in 2017 either that or it could also be that there is the some sometimes these are cut and used for agriculture so there is some um some some agricultural crop in these fields it doesn't necessarily mean the green areas that there's been a return in the mangroves now let's visualize the the three dates the 2007 2010 and 2017.

[01:07:38]so remember that color wheel that we discussed back in the time series uh webinar on the combination of colors to understand the the change in different dates okay so anything that is yellow would mean that there is a mix of green and red and so whatever's in the green channel and whatever is in the red channel is high whatever's in the blue channel is low anything that is green means that only whatever's in the green channel is high um okay so now let's display the 1996 and 2017.

[01:08:49]deselect this so all of the pink areas are areas where there was loss between mangrove loss between 1996 and 2017.

[01:09:18]there is a slight offset when you zoom in you'll you'll see that there's a slight offset so the co-registration is not perfect but there are some interesting things that are quite obvious for example uh there has been an extension of the land right here in this snippet and down here as well in this area so this 1996 jrs mosaic is really a great mosaic if you want to go back and look at some changes historic changes you know especially being able to compare it with a very recent data the only thing about this mosaic is that it's it's just an hh polarization and one thing i'm not sure i mentioned but the jrs mosaics that i showed you how to upload as well as the mangrove distribution files are on the rsap webpage so you can grab them from there rather than going in and downloading them through the directly from the web pages i showed you all right so now that we've visualized the images let's let's look at some changes and what we're going to do here is do something similar to what we did in the time series analysis on the first webinar so we are going to create these ratio images and here we're creating ratio images for two different dates one is 2007 to 2017.

[01:11:40]and for that we're using the hb channel and the other ratio image is 1996 and 2017 we're using hh uh you will notice that there is a substrate here the reason for that is because remember these images are in their log so whenever you do a division with numbers that are log you subtract them okay so let's calculate the ratio images and then display them and run your code so you can see them here in the bar so you've got the 2007 2017 let's just look at that so the areas that are very bright are areas where there's been loss in mangroves and the areas that are very dark are areas where there's been some sort of return of vegetation so that is for 19 2007 2017 let's take a look at the 1996 2017.

[01:13:26]you see that there was there's been quite a bit of change there in that time period okay so the next thing we do is calculate the threshold that we're going to apply and this time we'll do it a little differently so last time what we did was we looked at the histogram and then based on the standard deviation we multiplied the standard deviation by 1.5 and we calculated the threshold based on the mean minus 1.5 okay so what we're going to do here is draw polygons over the areas of change to determine our thresholds so let's just bring up the hv 2007-2017 okay so what you do is you go to the polygon shape here and you select the draw shape let's deselect roi and select new layer and then zoom in to [Music] those bright areas so what we want to do is we want to select some polygons and then determine the statistics of the values within those polygons to then set our threshold so let's just select this area here so you get the point right so now what we'll do is go into geometry and just give that a name so what i've done as has been to give this a name and call it loss so these are areas where there was mangrove loss 2007 2017 okay and press ok now repeat that so since i've done this i'm not going to save it i've got it right here now repeat that for the areas where there's been a gain so select those polygons over those very dark areas and then also select over areas where there's no change there's no mangrove change and those are the areas where it's not either dark or bright and another way to do that is to overlay that mangrove vector file so let's just overlay the mangrove vector file on this ratio image okay so you can overlay it using the the tool here next to the the file name so there we're overlaying just to make sure that wherever you pick an area that you want to characterize as no change that it actually falls within the mangrove vector file and this is also a good way to look at the areas that are defined as mangrove by the vector file so so you can select an area here again you select the polygon say new layer and then zoom in select that area oops apologize i selected line we actually want a polygon so let's make sure we select polygon okay and then so you can see as you overlay this mangrove vector file that this this is from 2010 right and there have been changes in the mangrove cover in this area right from 2010 those bright areas are areas where there's been mangrove loss okay and then name your polygon appropriately so what i've done is i've created six different categories right so one is called or column classes so one is called lost 2007 2017 gained 2007 2017 and then no change 2007 2017. that's the last one i selected now you want to go through that same process just upload the 1996 19 1996 2017 racial image and do the same thing select areas where it's bright there are actually not very many dark areas and i will deselect so these are the areas i selected so the areas that are purple are areas that i that are bright that means there was a mangrove loss the areas that are yellow means that there was some sort of gain and then no change are areas that are cyan and when you overlay that mangrove vector layer you can see how much change there's been since 2000 uh since 1996.

[01:22:45]okay so we've got our polygon selected now we want to extract the statistics of those polygons and the first thing we do is we generate something a variable called reducers and that reduces it combines the statistics that we want to calculate for the polygons so it combines mean and standard deviation and then what we do is we compute the mean and standard deviation for each of our polygon classes so for lost 2007 2017 i think we selected six polygons so basically it's computing the statistics for the pixels in all of those six polygons that have been labeled as lost 2007-2017 okay and we've done that for for each one so we've got six of these statistics that we're computing and then we are printing the results to our console window here and so what we're saying in this print statement is print out the mean and standard deviation for loss 2007 2017 and so here's the statistics for that and we've done that for each one each of the six one six uh polygon classes okay so once you run it you go to the console and you can see here then that the for 2007 2017 the pixels in the polygons that were labeled as loss the mean is 15 and the standard deviation is 2.28 okay then for gain and for no change we're not going to use these but this is just a reference of what the values are for for no change and then the same thing the statistics down here are for 1996.

[01:25:15]all right so what we do is we take those statistics to set the limits for the thresholds that we're going to apply and uh so we've got um two thresholds we've got a loss and then we've got a gain threshold for the 2007 2017 and 1996 and 2017.

[01:25:41]so how do we set these thresholds we basically took the mean so in this case for loss we took the mean looked at the mean it's 15 looked at the standard deviations 2.2 so we multiplied that standard deviation times 2 and then subtracted that from the mean so that that came up to and we rounded some numbers here but that came up to 10.4 okay and then we did the same thing for the for gain and the same thing then for the 1996 gain and loss there is something different that we did here for the the gain in 1996.

[01:26:28]the it was quite large the standard deviation and so instead of setting that standard deviation multiplying it times 2 and this was all done by trial and error we multiplied it by around one and a half to set the threshold for gain all right so then we've created these masks that we display and so you you you save and you run and then we've got the results here what we're displaying let's just go back so we're displaying loss 2007 2017 labeled as vegetation loss vegetation gain and then loss 96 2017 and gain so these are our images so let's look at loss 2000 and we're displaying it as the color red so these are just the pixels where there's loss what we can do is just overlay the image so let's overlay the in this case the pausar hv 2017 image to show the areas that are being identified as lost in fact let's better yet even let's just overlay the the different that ratio image so these are the areas identified as areas where there's been loss okay we are missing out on some of the the bright uh pixels when you overlay these on the difference image and uh what you can do is you can just decrease that threshold so instead of using 10.4 maybe try 10.2 okay so let's just run this again and try 10.0 and see how that looks okay so we're selecting more of those brighter pixels but there might also be more noise you see over here there's some some some noise and and so um setting these thresholds really does require some level of trial and error all right so let's take a look at the 1996 the vegetation loss from 1996.

[01:31:24]so there's a lot that being that's been selected here okay so the next step here is to let's clip those radar images and elevation to the mangrove vector file and the way you do that is you select what we're going to do in this case is we're going to select the original unclipped file and there's a reason for this that i want to show you in the next step so the original unclip file so this is the global mosaic of 2007 hh hb or and and 2017 and 1996. i'm gonna clip it here according to the areas where there's a mangrove coverage okay and then this file is being called mangroves 2007 hh so all of the files that are called mangroves are the clipped files and then we'll add those to the layers bar so if you hit run let's deselect this so you've got these mangrove files where we're just we're extracting the the back scatter only over the areas where the vector file is indicating that they're mangroves okay so since this is global it might take a little while but here you have the backscatter for the different years and polarizations of the areas where the mangrove vector file indicates that there's at least as of 2010 that there was uh there were mangroves there so we can just pan around here a little bit a huge mangrove region here obviously bangladesh and so the next thing i'm going to do is show you how to calculate above ground biomass yeah so there might also be a little more noise when you low lower that threshold so the next and final step i want to share with you is calculating above-ground biomass of mangrove ecosystems and there is a relationship between canopy height and mangrove biomass and there's been work done by a number of people uh mark zimmerd richard lucas sansaci on establishing these relationships they've done a lot of collection measurements in situ to establish these relationships between biomass and canopy height and develop allometric equations so in terms of using what we'll do is we'll use the srtm dm because that provides an assessment of canopy height and you're thinking how does it do it well srtm the the wavelength did not that dm is using a wavelength that did not completely penetrate through the vegetation so what we're seeing actually in the srtm dm is a is is is vegetation height in some places where the signal is not penetrating all the way through and so that is the case in uh in in mangrove ecosystems that signal is not reaching the ground it is penetrating somewhat through the canopy and it's been established that the uh the that that level of penetration in on a general basis right so the maximum height is generally 1.6 times the um the sr srtm elevation so what srtm really is measuring is the basal area weighted height that's also called lori's height so there's an assessment of the the actual elevation of mangroves with srtm and since we know that these areas are flat wherever you find mangroves because they are lying along the coast you don't have the problem of like finding them in in mountains or hills or additional topography then you can assume that srt srtm dm was what that's telling you in areas along the coast is actually the height of the mangroves so there's an equation here in this web page in this powerpoint that shows the relationship between above ground biomass and canopy height and and so so we're using the basal area weighted height which is 1.08 times the srtm elevation and we apply this formula so let's go back to the google earth engine and so here we have the backscatter dm so here we're looking at the clipped the the clipped back scatter for global mangrove ecosystems so what we did is we took the db image and we clipped it according to that mangroves vector file so now we've got the the backscatter only for mangrove ecosystems globally okay and you'll notice that in the code i also i have two parts one i left uh comment commented out but you can change this so i have either the clipped if you want to see just backscatter for the mangrove areas or it's unclipped so if you want to see the mosaic for the entire globe and then you can overlay the mangrove vector file so let me just show you right now the way it is you're we're clipping all of the images to the mangrove file okay so if you want to see the the file for the entire globe okay so then it's it's mislabeled but if you just want to see you know the entire mosaic and then you want to overlay the mangrove vector file you do this and it might take a little bit to to load so so there if you want to do that instead you can but in this case i've clipped it and let's just go back to what i showed the clipped one okay so to calculate biomass you use this equation the one you just saw in the powerpoint and this is how you write that equation so it's taken that elevation and it's um it's multiplying it by 1.08 raising it to power of 1.53 times 3.25 okay and so let's just run this and we can see some of the results okay so now what we're running is the clipped so first of all let's just load elevation and there's some interesting mandrel of ecosystems around the world obviously that there's the sunderbans here but if you go to gabon there's some very high mangroves in gabon let me just zoom out so here near libraville you've got some very high mangrove ecosystems and we load the elevation so you can see so it's right here so this is the elevation file for those areas and if you want to see the value of elevation just click on go to inspector and click on an area and you'll see that you'll see the elevation value so you'll see the value for that file let's see i must be clicking on the wrong area okay so here that pixel i clicked on it says it's three meters and then the biomass is is the it's 19 tons per hectare okay if we click on another area where the vegetation is higher then you have an elevation of 49 meters biomass is much higher all right so you can generate this and this will give you just a general assessment of biomass um for mangrove ecosystems now there is a caveat with this because in some places the elevation might be zero it might be showing zero or it might be showing like minus one so there you have to take the absolute value because uh it you're not going to get a a number a biomass above ground biomass value if your elevation is negative so you have to convert it into absolute values okay but that gives you a a nice general assessment of of biomass all right so with that uh i this demo uh concludes and now we will open it up to your questions and i see a lot of questions have been coming in already through the chat okay first question would the methodology be the same for wetlands um yes you can use the same methodology for wetlands now the difference with mangrove ecosystems is that it's not as straightforward because mangrove ecosystems remember they're a mix of like inundated vegetation it depends with the tide is high or low and non-inundated vegetation the structure is is very different than a regular forest so the thing about mangrove ecosystems is that you usually have like in force you usually have an increase in backscatter as the vegetation biomass increases okay until there's a point where that signal just saturates it doesn't increase anymore it just it doesn't um it just reaches a certain level and and so the higher the back scatter you kind of assume that the higher the vegetation biomass for micro ecosystems it's it's different the back scatter will increase up to a certain point as as the biomass of the mangroves increases up to a certain point and from that point on it actually decreases because the signal just gets attenuated it has to do with that that um the structure of these echo systems so that signal gets attenuated and you might have like high density high biomass mangrove ecosystems that have low back scatter but then again in mangrove ecosystems you might have sparse or low density mangroves where the signal is penetrating and you're getting that double bonds component and you're getting just a very high back scatter so so uh so mangrove ecosystems tend to be a little tricky with wetland ecosystems you can apply the same methodology in terms of doing a threshold or even running a classification as we did in the second session so it's much more straightforward actually in wetland ecosystems if you see a high bright backscatter you know that that wetland is inundated this is these are inland wetlands is it suitable to use sentinel 1 to assess flooding areas and or tree heights within the urban environment yeah so the thing about using first of all sentinel 1 is not gonna penetrate as much through vegetation so it's a little more challenging to look at flooded areas where there's dense vegetation you're just not gonna pick it up because that signal is not penetrating through the vegetation canopy in flooded areas though i think i'm sure so there are two types of flooding right so there's from the radar perspective there's the flooded vegetation and then there's just open water just potted water without any vegetation so you could pick up puddled water in urban areas a flooded vegetation you could pick it up too it's just a little more challenging because in urban areas that same backscatter mechanism for flooded vegetation is also the same one for urban areas right so you've got double bounds dominating in urban areas you've got double bounce dominating in flooded vegetation is it suitable to use sentiment to assess flooding areas and or tree height so tree height again probably not in urban areas just because it's so sparse you might be able to use it in areas where there's vegetation and then again if you're estimating biomass with sentinel one you really it will probably it just work in areas with low or low vegetation not in forests or it won't work as well and for us how paul insar and tomos are this is question number three work in tree height estimation okay that's a great question i think someone is getting ahead of themselves here because uh if you wait until thursday you will learn about this there is the session on thursday it's focused on estimating tree height okay so don't miss that one on the slide 20 why are there two hh in the three band composite image is one of the hh l band and the other p band let me quickly take a look at slide 20.

[01:50:20]okay yeah so the reason there are um 2hh is because we're looking at an image from one date right and so we're create we have two bands hh and hv this is alos pulsar and you i'm creating an rgb right so i need to put something in the third band and you can in this case i've put the same polarization in two bands in a band and the first are the the red and the blue but you can put say in the blue you can also do a ratio of the bands hh over hv and put that in the in the blue band i just uh just use this combination hhh vh uh just so that you can just for visualization and so you can understand really the unique information content in each of the polarizations but these are images from the same date okay next question how do you come up with the table in slide 23 so let's yeah so this is a table uh this is coming from a great resource online it's called the surveyor handbook and this handbook contains chapters on on different applications of tsar and one chapter is focused on mangroves this was led by mark zumar and he has this table in here and he assembled this table from his publications and measurements in the field as well as that from others that have been doing using radar to measure mangroves for a very long time okay six if we look at the concept of antenna pattern then is it right that in a given swath of a homogeneous land cover the mid-range of the illuminated swath have lightly higher back scatter values than the near range and far range will have the least when compared to the midden area yes actually but this should be corrected when you apply the the the processing the correction so you do have an antenna pattern and it looks like an umbrella and so what you do to remove that through an algorithm is you take a homogeneous area say a forest that goes yeah uh homogeneous across the image horizontally and so you take the backscatter and you see how that varies as i said it's like a curve right and so then you apply the inverse of that curve to correct for that antenna pattern but that should be corrected and usually um the the antenna the variations in antenna pattern it depends on the data set um but usually they're small but if you want to really see if there's any anything related to antenna pattern you can just do what i just said just just look at the pixel variation across the image over a homogeneous area okay so is it possible to access google earth engine google earth images as well as as well in google earth's engine for validation purposes while performing your classification if yes what code should we use i yeah i i'm not sure i have never tried that so this would be a matter of looking to see what's been done and if so how it's been done but i agree that having this sort of capability would really facilitate things in terms of just loading and an image rather than having to put it in your code and and visualizing just just bringing it up like with a google earth interface question 8 jax says officials said in late 2019 that they will make the halos 2 data free is there an update on that great point not that i know of as of today not that i know of but yes they have said that they will make the halos 2 data free and that is going to be huge because it is l band data it is global data and i i think it would be just an amazing data set to have in addition to sentinel one question nine can tsar predict mangrove as healthy or damaged um i'm so i'm trying to think what healthy means say say you have a change in colors in in the mangroves so i know sometimes there are mangrove die-offs and they just change they just turn brown you're not not going to detect that because remember radar is sensitive to structure and and moisture right so with optical you can see those differences in color but you might be able to detect if there's been a degradation in the mangrove because of say a huge amount of leaf loss right or in terms of damage that there's been a degradation of the mangrove through logging you might be able to see those changes with the radar to perform a land cover classification using sentinel 1 and landsat 8 for example is it 20 samples of approximately thousand square meters of ground truth too little oh so 20 samples so let me see no i think that that would be fine and what you can do really if you're not certain if you have enough ground truth is cut that out you know cut a third out and run your your uh ground truth and and see how your results end up and then add those other at at the part that you didn't cut out and see what the results are and if they change significantly then you you don't have enough points question 11. i'm finding that snap creates a bunch of large size temp files that quickly clog up the hard drive is there a way to automatically flush these temp files after project i don't know i mean i i feel your pain because i've run into this before and there might be a way in the preferences the best way to figure this out is there is a forum on uh sentinel toolbox where people post their questions and uh usually there's an answer there already for for what you have in mind so it it's good to uh to visit that forum and see if someone has has had this issue already and if the question has been answered question 12 is it possible to map the lost force between 2010 and 2020 is it necessary to resample the sentinel one from 10 meters to 25 meters to be compatible with pulsar one yeah okay so the thing about using two different bands is that they are they have different wavelengths right so you are going to get slightly different information if you're looking at deforestation you're probably you're going to see that you know if an area has been cleared if there was mangrove there and there is no mangrove now um but you don't i think the best approach here when you're using radar images with two different bands is um it's just keep in mind that the the information content is is different it's slightly different because of that penetration capability question 13 are there any constraints with respect to storage size restrictions for uploading files i i think there are yes i'll ha i'd have to go back and look and see what those constraints are so there are a lot of forms online for google earth engine i mean there are a lot of people that are using google earth engine for analysis and uh there are there's people that post questions or the forums where people post questions and and they get answered so you might want to do these sort of searches and and see if your your queries are resolved why not use question 14 why not not use halo's world 3d 30 dm instead of the srtm since we're using the alos pulsar or is it not available in google earth engine um so the halos world 3d the um you can try that too now let me see one thing is we calculated biomass at the end right so that equation has been derived based on c band actually so so srtm used um c-band and so it's base it's derived based on the penetration depth of the srtm signal into the vegetation if you use a dm that was derived using a wavelength that's longer so i i'd have to look at the uh okay so it looks like okay so the als dsm global tier 3 30 meter dm is derived using a stereo pair of optical images okay yeah so you can certainly use this you just need to figure out what is the that allometric equation needs to be modified because that signal if it's using optical images that signal is not penetrating through the vegetation up to a certain point it's just seeing the tops and so that equation needs to be modified accordingly right next question can we use this tool pulsar mosaic to analyze other vegetation types not only mangrove or other forest plants uh yeah yeah absolutely you can use it it is a great resource the only thing about the pulsar mosaic is that if you're looking at things that are dynamic you're looking at like at wetland inundation dynamics or other things that might be dynamic if you want to look at like the scar of a fire a recent fire then you're going to lose that sort of information in a global mosaic and it'll be important to read in that date file that's associated with the mosaic just to know the date that your that the the date for each pixel question 16 why do we use gamma naught here but we use sentinel sigma naught previously is one better more appropriate than the other how do you make this decision finally could you explain a bit more about beta naught gamma naught and sigma naught um so we're not using gamma naught here let me double double check so these images are i'll have to go back and double check i don't think this is gamma naught these are sigma naught and they are expressed as just digital numbers which we are converting into into db but i need to double check that all right if you send me an email i i'd be happy to uh to clarify this why question 17 why did you first convert the images to db and only after that clipped them would it be faster to clip them first and convert later yeah absolutely the reason i did that was because at the end i wanted to show you how you can look at mangroves globally and how you can use that vector file to clip the entire global mosaic and and and and just look at backscatter over the mangrove areas but yeah if you don't want to look at any other place just clip it from the beginning question 18 how do i know which range is best for starch image why min 5 max 40 okay i'm not sure what is meant here by starch image but i'm going back to the code and looking at 5 and 40. so that's the srtm that's the srtm elevation and the reason i just stretched it this is just for visualization this has nothing to do with analysis i just stretched it so i could see well that variation in the coast which is it's very um subtle right so if you're looking at it it's just a couple of meters along the coast so i just stretched it and to that range question 19 with regard to exporting the final map the map has to have the basic element of a map such as north arrow grid scale bar and legend so how do we export our final map with the mentioned basic elements so there is a code to do this and if you send me an email i'd be happy to share the code with you i don't have it just accessible with me at the moment i do need to kind of put it together but there are ways to export the image so in the previous in the last webinar there is a command there at the end that exports the image you can export it as a geotiff but i know there are ways to add additional things like a scale and all of that color bar okay question 20 is it possible to repeat all these steps using only the snap toolbox that's a great question i think it would because the snap toolbox can also deal with vector files and the only thing about using the snap toolbox is that you have to download these images to your desktop you need to download the mosaics you need to cut them so it's just a little more cumbersome but i believe you can do all of this you can do band math on on the snap tool box do we have question 21 do we have to be concerned about the coordinate system or the projection or shape before we load it to glue earth and is there any adjustment we're supposed to do for the shape file for a particular region um not that i'm aware of i believe google earth engine takes this into account to match the position of the different pixels even though they might be in different projections so as far as i know there isn't any adjustments now you saw that the jrs image had a slight offset from the sentinel the from the pulsar image and that has to do just with the the uh the geolocation of of the jrs image and uh you could correct for that i i just i don't know how it would be done on google earth engine but there are ways to correct for it say with a software like envy by using control points and then stretching the image question 22 i think the back scattering of mangroves is similar to other vegetations with similar canopy if so how do we map a particular vegetation separately are there any criteria used to distinguish the different vegetation types okay so it's not clear to me whether this person is referring to different vegeta mangrove vegetations or different types of just inland vegetation so you could um you can use radar to separate different types of vegetation polarimetry really helps that's when the more polarizations you have the the better you can do this okay because you're basically classifying things according to structure so assuming that different vegetation types have different structure then you can you can separate them question 23 you mentioned that the core registration of the images is not perfect okay we there were there ways to correct this or is this officer so small and systematic that it is not necessary to correct yeah well it it is uh it is very small and when you run when you apply threshold or you run a classification uh you know even if you run a classification with the jrs and halos pause our images together you're going to see this offset or if you have classification results and you're comparing things you're going to see this all set there might be ways to deal with it so maybe like in in coastal areas you apply a mask to to to to just mask out those coastal areas so it's not a super big deal like you can still see the change for example but uh there will be issues when when you're doing some analysis with them that you need to deal with you can correct it if i'm not sure how you would correct it say in the sentinel toolbox but i do know with envy what you do to correct these type of things is you choose these control points so you you select the same points in in each in two in two images you select the same point and you select points around the image and then basically you do you stretch one image to the other so that it it fits um over one over the other and yeah it can be a little cumbersome though it can take some time to do this question 24 how can you obtain metrics from the ratio step for example how can you quantitatively determine the percentage gain lost in biomass from 2007 to 2017. okay this is a good point i didn't show this here but in the previous um in the in the first webinar that we had for this series i showed how to determine statistics so basically what you're do what you do is you create a polygon over the area where you want to determine statistics or say just apply that mangrove the the mangrove vector file however you want to do it and then you can calculate the statistics for the area within that vector file or within that polygon that you drew remember to check when you select an area it has to be a polygon not the line otherwise if you select a line uh you calculate the statistics of pixels that touch the line you need to select a polygon to calculate the statistics of the pixels falling inside that polynomial okay question 25 can i apply the same techniques to riparian vegetation mapping uh yes you would be able to apply a similar technique to riparian vegetation mapping yes now that's a little different because if you're here we're looking at change right so so we're looking at that large change in the signal and and that's why we're setting a threshold now if you want to do mapping of vegetation you want to run a classifier and that would be more relevant to the second webinar that we did where we used optical and radar so you can just try running a classification and just using the radar data but make sure that you have good ground truth to classify your classifier to select your training areas question 26 which sensor would you recommend for mapping coral reefs deep water coral excluded yeah that is a little bit outside of the scope of my expertise i don't look at coral reefs obviously you you want some sort of optical sensor and i know that there have been studies uh doing this type of thing so you need to [Music] it's i think it this requires a little bit of research to see what's been done out there but um jpl the the nasa center where i worked they had a very large project in using an airborne sensor to map coral reefs benthic cover question 27 can i use earth engine to download the data in net cdf format i believe you do have the option of downloading the data as netcdf but yeah it's something that i i'd have to double check wouldn't it be better to use the pulse rdm that has a higher resolution than the srtm dm yeah possibly i the the pulsar dm i don't know what the characteristics are of that dm and whether it's global um and then when you're applying that biomass equation um it should have the same characteristics as srtm otherwise that equation needs to be modified how do we define the formula to find the threshold loss and gain so what we're doing here there is a little bit of uh trial and error in sending and setting these thresholds right so you kind of start out with just something general that you applied in this case we applied we said times two okay so take the standard deviation times two and then subtract that from the mean and you take a look at your results and you might have to change that slightly the the threshold slightly depending on your results question 30 it seems like it could be problematic to manually grab the values and calculate your thresholds can we access the mean standard deviation objects within the stats filters to calculate thresholds using a function yeah absolutely this is a great point i just wanted to go through the steps and and have people um just understand what the values are and how to calculate these but absolutely you can just automate this and take that mean and standard deviation that were calculated and put them into a function where you're setting these thresholds directly question 31 do you expect the area of loss of mangroves identified or due to human activity or are these areas lower elevations and have been affected by a modest right rise in sea level yeah that's a great point you know in some areas uh yeah i think probably 99 of these areas have been lost due to human activity i mean they're they're inland you would it not their inland but they're kind of um a little bit in you go in a little bit right so if you want to take a look at um loss due to sea level rise um you would have to let me see what's the best way to do that is first start with an area that you know there's been loss to sea level rise and see how that looks one of the things i did here is i brought up google earth and i looked at this area and it was obvious that these are all areas that have been converted to farmlands or aquaculture i mean the the human imprint is clear they're very geometrical structures so you want to take a look at uh what's the impact of sea level rise uh maybe off the gulf coast you'll see some not only effects of sea level rise but also the subsidence of the land 32 can the same method be applied in grasslands and shoplans if yes what do we need to change apart from the study area in terms of of theory okay so if you're looking at grasslands and shrublands first of all you'll want to use sentinel 1 because that is sea bend and that will be more sensitive to grasses and and shrubs okay and let's see you want to understand what are these signal characteristics for each so if you want to classify grasses and shrubs you don't use a threshold you want to use a classifier you want to train classes okay but i think the main thing here is you want to use sentinel one question 33 rather than apply thresholds might an unsupervised classification be better with say 10 classes or so okay so uh i'm not sure an unsupervised classification will be better you might want to try it i mean all of these things part of this is exploring what works better right so so the main point here is we're looking at change right so you want to be able to whatever method you're using is that you're capturing that dramatic change between um two different images from different dates can these biomass calculation methods be for mangroves be used for terrestrial forest ecosystems there is something similar for terrestrial forest ecosystems and in fact this one it was based on just the canopy height there are equations there's been a lot of work by saatchi for example on forest biomass estimation if you go to that severe handbook and you might just want to google it do a google search for a severe soar handbook uh there is a whole chapter on forest biomass estimation and there is a tutorial associated with that so you can follow that or replicate that tutorial to understand how forest biomass is is calculated but there is a relationship between as i mentioned radar backscatter and biomass and so that's what those methods are primarily based on and you need a lot of ground data and develop some allometric equations in order to to do those estimates how accurate are these data to use in commercial projects yeah that's a good question i mean i think the accuracy is going to depend on on the area obviously we're using first of all a mangrove and i'm assuming you're talking about the above ground biomass so we're using a mangrove vector file in this case from 2010 you want to make sure that um whatever you're assessing is is accurate right so you want to update whatever that vector file is indicating in terms of the presence of mangroves with what's current there might have been mangroves that were cut since 2010 and then make sure that the biomass is relates to whether there was a mangrove there or not right so srtm remember that was back from 2000 uh it's a 2000 as well so how accurate is it to actual biomass um that is yeah i'm i'm [Music] there are papers that have been published on the accuracy of these products and i think the best thing is to refer to them smart at all have published a lot on this question 36 hi can we use sentinel c band for above ground biomass estimation in a dense canopy forest uh now that that wouldn't be a good band because you you don't have much penetration in a dense canopy forest you want to use l band question 37 what is the difference when using ascending or descending orbits in star time series analysis so the your geometry is going to look a little different um if it's an ascending orbit or descending so when you're doing time series analysis you want to as with any like classification you want to you just use or ascending or descending in your analysis and you can do two separate analysis and just compare the results but you want to just keep to one or the other is there an automated way to correct the offset of the images from different years i don't know of an automated way there might be but i don't know question turning out how can you decide scale in the calculation of the statistics of a polygon so the scale would represent the resolution of the image so basically you're taking all of the pixels within all of the polygons labeled as loss for example at the original resolution of the image question 40 how do we know we need to multiply the standard deviation by 2 or by 1.5 to define the threshold is there a rule no there is not a rule as mentioned you this is uh you have to test this and see what works best in terms of what you want to identify is it possible to use srtn to estimate vegetation height in places with higher elevation it seems that using srtm for estimating mangrove works because we know that the ground at zero meter is okay yes um yeah actually i think there have been some efforts along these lines and using people have been using for example um digital terrain models so uh if you have a digital terrain model where you know like what is the elevation at the at the surface at the terrain level and then you have something like srtm which is actually giving you the height of the of the vegetation then you you sub you use you subtract them is it possible to use halo's dm to calculate mangrove biomass again as mentioned you need to account for that in that equation the fact that the halos dm used optical data and so that um you don't have a signal penetration through the canopy and so you need to account for that basically the halos dm is really is is just an assessment of the top and you need to modify the equation how do we arrive on the calibration factor that's a really good point so that is provided by jaxa they gave us the calibration factor and you kind of have to look around and in the document documentation for these mosaics to find what that calibration factor is so um that's that's how i came up with that i didn't invent it it is a standard calibration factor provided by jaxa what is the formula you use for biomass calculation is it a global allometric equation for mangroves could you elaborate so i can if you send me an email i can point you to some publications so this is just a general yeah these are allometric equations for mangroves and i can yes i can provide you more details through publications do you have any recommendation for the choice of the threshold in the case of amazonian forest again if you want to do a classification of like forest type for example um you you want to you want to do a classification not a threshold you want to use the threshold method when you're looking at change so for example if you're looking at or if you're looking at areas that are vastly distinctive okay then you can apply a threshold so if you're looking at say areas that have been deforested there's been some sort of vegetation loss between different years you can use a threshold if you have one image and you want to separate water from everything else you could probably use a threshold because water is such is so is so much lower usually in radar images than everything else sometimes not if there's one but usually or if you have an image where you have flooded force and you want to separate that from everything else usually flooded for us inundated vegetation has such a higher backscatter than everything else that it would work just applying a threshold okay question 46 can you explain why you have to use absolute values when the elevation is zero or negative because uh yeah if you have a negative value in that equation that i provide so you're raising something to the power of uh you'll get a complex number is it appropriate to use sentinel 1 or sentinel 2 in urban area development also can this tool be used to analyze the quality of water in a wetland or stream yeah so this is sort of a little this question is a little different um yeah probably sentinel 2 would be your best bet for looking at urban area development and if you're looking at water quality you definitely want to use something optically based sentinel one any radar is not going to give you an assessment of water quality all right so unfortunately we are out of time but we will get to the rest of your questions is there any any more questions we will get to these and we will post this google doc online remember there is one more session with this webinar series and that is forest height estimation we will have a guest speaker for that professor paul cicada from university of massachusetts amherst that would be this thursday same time stay tuned the homework will be announced at that during that session as well as the due date so see you all on thursday for the last part of this webinar series thank you for tuning in thank you for all the questions and see you on thursday bye bye