An overview of forest remote sensing technologies

just lift it off gently and then set it down okay hello today we're going to be talking about remote sensing and my goal today is to give you a general overview of what all right stop stop stop cut it down hello today we're going to be talking about remote sensing in the context of forest ecosystems and i'm going to try to give you an overall sampler of basically everything that's out there bearing in mind of course they teach entire courses on all of this so i want to start by categorizing remote sensing technologies into five different categories the first is imaging and what i firm refer to as imaging it's basically two-dimensional imaging of the earth that can be with the drone that could be very high resolution or it could be from a satellite image reading doesn't necessarily have to be in the colors that humans see we can image with multi-spectral colors we can take infrared and shortwave infrared and get a lot of information about the forest that way the second category is photogrammetry and photogrammetry is the three-dimensional reconstruction of two-dimensional imagery and you can do this with very high resolution data and the way this works is the same way that your eyes actually see 3d if we're taking images from different angles we can actually reconstruct three-dimensional environments using uh basically trigonometry next we have radar and the way that works is that usually a satellite will send out a radio wave it goes down hits the earth and then comes back and the brightness of what's coming back and tell us a lot about what's down there now the thing about radar is uh what you get back depends a lot on the frequency of the radio waves that you're shooting out uh so really short radio waves will bounce off different objects and really long wave radio waves the fourth category of remote sensing technologies i would call hyperspectral hyperspectral is similar to imaging except what makes hyperspectral imagery different than regular imaging is that where regular imagery is maybe only capturing three wavelengths of light uh red green and blue and from that you're able to infer all the different colors that you can see with the human eye hyperspectral imagery is able to capture up to a thousand wavelengths of light so whereas the human eye is only capable of looking at three different colors hyperspectral imagery is able to capture a thousand or more different colors with very thin bands of light but again not all hyperspectral imagery is equal and then finally the absolute king of remote sensing for forest context is lidar and lidar works just like radar it shoots a light pulse a laser beam out of an airplane usually that laser goes and hits the ground and it comes back and the amount of time that it takes for the laser to return basically tells us how far away that object is and so from this we're able to reconstruct these beautiful three-dimensional renderings of the forest again though uh light are it can be categorized into a couple different categories mostly lighter are these three-dimensional point clouds that i'm talking about uh however there are some larger footprint light art that basically just tells you about the vertical structure of a large area all right so those are our five categories but we have what we have to also consider is that data looks dramatically different at different resolutions so if we split those five categories up into three different resolutions low resolution or maybe we're talking about imaging the earth where every pixel is 200 meters wide medium resolution usually referring to pixels that are between 10 and 30 meters so between 30 and 90 feet and then finally high resolution and high resolution imagery usually captured from a drone or from an airplane what's special about this is that the high resolution images every pixel is doesn't actually refer to a clump of trees it actually refers to an object so when we're when we go from different levels of resolution we have to completely change the way that we and analyze this data it's actually much more difficult to analyze high resolution imagery because each tree each object in the scene make is made up of maybe a hundred different pixels whereas the other way around if one pixel has maybe a dozen different trees in it you can infer a lot about the color of that one particular pixel you don't have to worry about shadowing so you can see why this is quite a difficult video because there's a lot of different remote sensing technologies and then there's also a lot of different resolutions that we can image at some other things to keep in mind is the difficulty of dealing with data so some of the data sets that i mentioned are very challenging to work with hyperspectral data when you're capturing a thousand bands for every pixel on an image just imagine how difficult it is to actually pour through that data and get something useful out of it likewise super high resolution imagery can be very challenging to work with because just the data sets are so large repeat time is another thing that we have to consider low resolution imagery may be kind of lame but you can also take an image of the earth every single day and you can process that image the same day so we're able to create these maps of the earth on a daily basis which we could never do if we're dealing with pixels that were only one meter wide and then finally availability this is the biggest one a lot of these data sets are tremendously expensive hyperspectral is probably the most expensive data set lidar is right behind that most people do not want to pay for that and so we'll see a lot of remote sensing folks trying to do more with cheaper data and it doesn't always work out so let's talk about imaging first and let's start at high resolution imagery high resolution imagery as i define it is basically sub meter imagery although you could really even define it as you know like 25 centimeter imagery uh it's very expensive you either need to play a drone to collect it drones if you're flying large areas you know their batteries run out you can't fly 50 000 hectares of forest with the drone you can purchase this from a couple of commercial companies but still very expensive the thing about high resolution imagery there's a lot of potential out there but you have to get into these object detection techniques you don't get very many bands with high resolution imagery mostly there's a trade-off between getting more spectral information and getting more spatial information and so the promise of high resolution imagery is that maybe you can determine something about the forest based on the structure of the trees or the shapes of the trees as opposed to looking at the ratios of different colors medium resolution imagery is perhaps the most common type of remote sensing technique out there the most common sensors out there that people will use are sentinel 2 and landsat and these are basically 10 to 30 meter resolution satellites they're free they capture images of the earth multiple times a month and they're multi-spectral so they're not just measuring red green blue they're also measuring a couple of infrared bands and then shortwave infrared bands and shortwave infrared bands are especially good at telling us a little bit more about the ages of forest whether or not there's been a disturbance and even a little bit about how much biomass there is the thing about medium resolution imagery is that it saturates very quickly and so there's a lot of people out there trying to use it to make maps of forest carbon uh but in reality it's it's just not able to actually measure everything you need to measure about the force in order to make an estimate and so people are able to make these maps but in high biomass forests like the one behind me they really don't do a very good job of predicting how much carbon there is so you have to watch out finally with imaging we get to low resolution images modis is the most common one out there right now these satellites are taking pictures of the earth at 250 or 500 meter resolution it's very grainy but it's free and you get it every day and you get a lot of different bands with it so you can do a lot of different analyses based on the the ratio of this reflectance versus that reflectance uh the next one on this list was called was photogrammetry and photogrammetry is basically just taking 2d images and stitching them together it's usually drone based it can be done with airplane data of course too you can produce maps of forest carbon with photogrametry but the biggest problem with photogrammetry is that once you create these digital surface models these 3d renderings of the forest surface you can't actually see the ground unless there are gaps in the trees and so why does this matter well if we want to know how high the tree is off the ground and correlate that to how much carbon it's storing we actually have to know how high the ground is to begin with and so if we're overestimating tree height by say three meters then we're going to be way off on our carbon estimate so photogrammetry can kind of be used to estimate carbon in the right circumstances but nobody's really demonstrated it at scale quite yet the other thing to look out for is that there are a couple of companies out there like maxar and planet who are capturing very high resolution imagery from multiple angles we're not quite at the point where we can reconstruct three-dimensional data very well from them but i think you know in the coming decade this is going to be a really big area because eventually i think we could get the resolution needed and the algorithms needed to actually measure the tree's height without having to necessarily shoot a laser beam out of a lidar satellite the next one on our list is radar and radar has many different colors to it and what i mean by that is that different types of radar shoot different different radio waves different wavelengths of light and those radio waves will bounce off the forest differently so the shorter radio waves the x-band and the c-band basically just bounce off the surface of the forest they're not able to penetrate down into the forest and tell you basically how thick and how deep the forest is and so expanded seed band radar not very good for carbon but it is good for disturbance detection because you can basically say whether or not the surface has been removed l-band radar is fairly good for carbon but it does saturate in the absolute highest biomass situations so there are two l-band satellites out there there's a-loss pulsar and then there's uh saocom by the argentine space agency and those are those are fairly good for making maps across a lot of the world of forest carbon the most useful radar for biomass is is p band radar and the trouble with that is that a it's it's very difficult data to actually collect you need a massive satellite uh and b uh it's not really good for any other applications other than measuring the forest so nobody's really putting those satellites up there but there is a satellite in development by the european space agency that should be in orbit around 2024 that may get as forest carbon estimates that are just unprecedented now finally the last thing about radar is that like all the other sensor types there's different resolutions that you can work with most people are working with medium resolution radar and so they're looking at a single pixel and looking at the uh the brightness of that pixel there is high resolution radar which is very new 25 centimeter resolution from space and using that you can actually see the individual shadows of every tree crown so i think there's a lot of potential there to model carbon uh but nobody has yet it's very new technology next we have hyperspectral data and like i said hyperspectral data is the most expensive and rarest of the bunch it can basically only be collected with very expensive airplanes or with very expensive drones hyperspectral data is the only data that can be used to actually get tree species and biodiversity estimates now every single tree species out there they all may look green to us but they all have very slightly different shades of uh of color that can only be detected with a hyperspectral camera in addition to species you can get massive amounts of data about the ecosystem itself using hyperspectral so you can get information about nutrient content like nitrogen and phosphorus and zinc and calcium and all these crazy little trace minerals that tell you everything you could want to know about the forest ecosystem hyperspectral remote sensors are just the coolest people ever they have they have ecology figured out already we just have to catch up to them but like i said hyperspectral data is extremely rare and expensive it has no other applications outside of agriculture and forestry so uh you know the us government isn't flying it for for any particular reason it's very difficult to work with just a small area of forest will be a terabyte of data or more and it's also it's kind of the wild west in terms of data extraction for every single pixel and if we're talking about one meter pixels there could be tens of millions of them or more for a small area we've got thousands of numbers attached to them and so actually extracting the relevant data from that is is quite the challenge and it's it's a very interesting machine learning problem that i think it still really needs to be worked on more the last important thing about hyperspectral data is that there's a difference between full versus partial hyperspectral and nobody will ever tell you the difference but partial hyperspectral doesn't include include shortwave infrared bands and those are the bands that tell you all about the useful stuff like species and nutrient contents and stuff so full hyperspectral is much more rare you can purchase partial hyperspectral data from from satellite companies in orbit it's not that expensive but you're also not getting the really everything that you need to know about uh in order to make these predictions of the forest finally we come to what is absolutely the most essential on this list which is lidar now like i said you can kind of categorize this into two categories discrete lidar in which we're looking at individual points in which we create this three-dimensional point cloud and then large footprint lidar where we're basically looking at an area that's maybe 20 meters wide and we're getting height information about that area so with discrete lidar you can tell a lot about the forest basically everything that you'd want to know with a normal field inventory short of species so carbon tree number leaf area index basal area you name it you can tell it using lidar and the reason for this is that we're able to see not only the complete structure of the canopy we're also able to see the ground and so we can use this to develop relationships between what we're predicting such as carbon uh another cool thing you can start to do with lidar is you can actually start to count trees in certain ecosystems now this only works in the simplest cases but in places like california where everything's are nice conifers you can actually start counting uh you know individual trees and and labeling them and seeing how big they are and keeping track of them so lidar is amazing it's tremendously expensive it tends to be about as expensive as a field inventory itself and so it's a hard sell every major forest timber company out there already has a lidar program so they're already sold on it the carbon world has not sold on it yet and you know i think there's a couple reasons for that and one of them is just the carbon world moves kind of slow one thing that's nice about lidar is that it's free everywhere in the us so the federal government has basically collected the entire country already it may be a couple of years old for the forest you're looking at but it's a really good starting place if you want to know how much carbon is on that forest the area based lidar you know it's a little yet less useful we basically we get it from space there's a an iss module called jedi yes they're shooting lasers from space so they're calling themselves jedi and so it's going around and shooting laser beams and kind of ran randomized samples and basically measuring forest height in those areas jedi was launched a little over a year ago i haven't seen great science come out of it yet i think there's still a lot of calibration issues but hopefully we can get to the point where we can make carbon maps using that that are you know pretty good now when we're talking about remote sensing one thing that looms over all remote sensing is is saturation and what i mean by that is that when measuring a forest especially as a forest gets larger and larger and older pretty much all sensor types will get to the point where they're not giving you useful information anymore for predicting carbon and one of the reasons for that is that you know as trees are growing their canopies are changing and the trees are getting taller and taller but once they get to a state where they're fairly old their canopies remain fairly static and they're basically just accumulating carbon on their trunks and they're getting wider and wider and so that's hard for a remote sensing technology to actually distinguish the other reason for that is that a lot of these technologies just can't penetrate deep forest for example so radar the radio waves may just not go deep enough into the forest for you to actually tell how deep it is so this is the problem with pretty much all remote sensing technologies that we're constantly kind of battling lidar is the only technology that really doesn't saturate and oftentimes the way this manifests itself for people who are making maps or making models is that in high biomass forests they'll basically just not be able to estimate they'll basically consistently underestimate and they'll label uh basically anywhere that's forced as the same value and they'll basically they'll over generalize so they may label anywhere that has force as basically the average the same value on my left you'll see a predicted versus observed figure and what this is basically showing you is the actual biomass versus what the model is is saying and saturated models of all types tend to basically curve away from the one-to-one line and what this means is that they're basically no longer predicting biomass very well in these high biomass forests so let's quickly summarize this by describing what these technologies are good for if we want to estimate carbon lidar is king sometimes you can use photogrammetry from drones uh but then you have to be able to see the ground in order to do that some radar can do it l-band radar can sometimes do it and p-band radar should pretty much always be able to do it but there isn't the p-band radar satellite up there right now for species and biodiversity hyperspectral is the only way to do it for history so for saying uh what may have happened on this forest over the last 30 years for that you want medium-resolution imagery because the landsat satellite has actually been in orbit since 1984. uh you could even go earlier than that to 1972 and it's fantastically useful for telling us what happened on the landscape during those periods and so it can be really good for a carbon project to demonstrate whether or not that carbon project is justified for disturbance detection to figuring out whether or not there's been a recent clear cut of the forest medium resolution imagery is great and then c and x-band radar is great these are free technologies there's a bunch of people doing that already and they're basically able to use these technologies to like on a weekly basis say whether or not there's been forest disturbance now degradation is a subtler form of disturbance and what often happens in these high-risk areas is that bad guys will go in and take just the largest trees on the landscape and because the largest trees can can store a massive amount of carbon they're actually removing a lot of carbon from the forest by only removing a couple of trees per hectare so nobody's really good at detecting degradation yet but high resolution imagery is probably going to be the way to do that and it's probably going to come from like a provider like maxar or planet finally if you want to actually get into inventory planning and managing how we want to actually grow the forest out in the future and whatnot lidar is key so lidar can get you pretty much every inventory measurement that you would normally use in order to make these estimates with the exception potentially of species now having said all that there are lots of people who have basically mixed and matched sensors in ways that are probably not great people will often use different sensors because they don't have access to expensive lidar data for example and the result of this is that maybe their results are only good at scale so for example there's a lot of people using a landsat imaging device to estimate how much biomass there is in the forest because of the saturation issue they're not going to get good results that sensor just does not measure in the forest to tell you how much carbon there is in high biomass areas but because there's this pressure to publish only positive stuff they'll often take these results and and publish them as success stories and so you really when reading remote sensing papers there's a lot of positive papers out there but a lot of bad maps associated with that and you really have to read between the lines and figure out what their use case was for these maps a lot of the people making these maps maybe they're only doing it to map the entire country maybe they don't really care if their biomass number is is dramatically wrong for an area the size of a carbon project and so it takes a little bit of expertise to read these papers read between the lines read the discussion section and find a couple of lines that say the map isn't any good and the problem here is that you know for those uninitiated for like new master students for you know engineers who don't have a background in this stuff they read these papers and they think all right we can do this we can use landsat to model biomass or we can uh you know or we can use landsat to estimate tree species and and in reality you can't that information just isn't there i guess what i'm saying is that you have to be careful you have to pick and choose which sensor you're going to use with some diligence and and because there's this pressure to publish out there you really have to make sure that you're you really have to dig in and be skeptical of a lot of the papers that you see in this field that said as we discussed remote sensing can be used to basically measure everything that you'd want first you just gotta pay a little moolah you

Channel: Elias Ayrey

Views: 4,260

Keywords:

Id: 9j95H_6hZ5U

Length: 21min 46sec (1306 seconds)

Published: Tue Jan 18 2022