Basics of Airborne LiDAR

hi welcome to our video on lidar light detection and ranging so lidar is an active remote sensing technique and it works by pulsing a laser from an aircraft down to the surface and then measuring how long it takes that laser pulse to come back to a detector onboard the aircraft so that two-way travel time can be turned into a distance or a height of the aircraft which can then be turned into an elevation and we can use those elevations to construct very high-resolution digital elevation models or D EMS now a digital elevation model is basically just a raster or a grid of elevation values that can then be rendered by a computer to actually have a show topo graphic relief or show a landscape in this case this is a very high resolution lidar image of Paris so one of the huge advantages of lidar data is that the pixel size of the digital elevation model can be incredibly small down to even just a 1/2 meter or less which lets us capture extraordinary detail about changes in elevation and the shape of a surface whether it's a city or a natural landscape so here's an example of that here's a typical 10 meter resolution digital elevation model of a hill slope in California you can see you can roughly see where the peaks are you can roughly see where the river valleys are but you can't see any of the finer detail for example you can't you wouldn't even know there was a road snaking along this hill slope but once we get this 1 meter resolution lidar data we can suddenly start to see details like roads and gullies so it's hard to overstate what a revolution lidar data has been for earth scientists to suddenly be able to see the landscape in such incredibly beautiful detail so in this video we're going to do three things we're going to first look at how light our data is collected then we're going to look a little bit about the format that light our data comes in and then we're going to finish with a few slides on how you might create a digital elevation model from raw lidar data so here's what a lidar system looks like as I said it's an active technique that involves shooting a laser beam out of an airplane okay so you've got to have an airplane or in some cases a satellite on board that airplane is a GPS unit and so that allows us to keep track of the exact location of the airplane as its collecting this elevation data the airplane also has an inertial measurement unit this tells us exact information about the pitch or the roll of the airplane at any given time obviously an airplane is not perfectly stable so we need this an inertial measurement unit to tell us how exactly the airplane might have been tilted at any given time and therefore which direction the laser was pointing at any given time and then of course you have to let the laser itself commonly this would be an infrared laser a 1064 nanometer but also green lasers are used which are more effective at penetrating water and so the basic pattern of data collection is this laser pulses are being pulsed at an incredible rate thousands per second and the laser is basically sweeping back and forth as the plane is flying forward so you end up with a pattern of laser shots or elevations that is not at all grid like it's actually very much a zigzag pattern and we'll talk later about how we turn that in to a digital elevation model grid and then finally we need ground control points so things like this GPS we need to know certain locations on the earth surface very precisely so that we can then ultimately turn these elevations back into a map now one benefit of lidar is that the footprint is often very very small it's usually a circular footprint and for most terrestrial lidar the small footprint lidar it would be less than 30 centimeters so a pretty small little laser spot however there are certainly applications that use large footprint lasers with spots ranging well into the hundreds of meters one example of a large footprint laser is the Mars orbiting laser altimeter or mola this was an instrument that flew on board the Mars Global Surveyor satellite in an orbit of Mars from 1996 to 2001 basically collecting light our data the whole time and it had a spot size of about 168 meters this was a fairly coarse resolution lidar data but it produced this gorgeous map it literally mapped the detailed topography of the entire planet of Mars so it's hard to overstate how important this is much of what we know about the hydrology of Mars the volcanoes the lava flows so much of that information is wouldn't be possible without having this really detailed topographic data set now it's common that we collect two types of data with lidar one is the range distance and that would be the distance from the aircraft to the ground which is determined by this two-way travel times of the laser pulse and that's how we get elevation the second type of data would be intensity so this is really the fraction of photons that are returned back to the aircraft maybe you send out a billion and maybe you only get back half a billion so the number of returning photons determines the intensity of that radiation and that's actually controlled by the material properties of the surface some surfaces are really good at absorbing the laser wavelength other materials are really reflective they don't absorb it so for example a lot of lidar is done with an infrared laser and vegetation reflects very very brightly in the infrared so we tend to get high intensity returns off vegetation and so this intensity information can actually be used to be diagnostic of surface materials in some cases so let's look at a couple examples here we'll do range first so range is determined by the two-way travel time okay and these photons are of course traveling at the speed of light or C and the range then is given by the speed of light times the travel time that gives you distance and meters divided by two given that it was a two-way travel time and now let's look at intensity so here's an example of intensity this is a intensity map of an airport collected using a 1024 nanometer laser so this is a infrared laser and what you can see here is that the black asphalt of the runway is very good at absorbing this infrared wavelength so the intensity returned is very low and the asphalt therefore appears dark in contrast to grass around the runway this vegetation it's very reflective at 1024 nanometer and so it appears very bright so you can see that there's actually compositional information within the intensity of the lidar return now it's also important to note that reflections of lidar light on the ground can be very complex and it's quite common that each outgoing laser pulse would yield mult returns okay and so typically in a lidar data set each data point may actually not have a single range it may have multiple ranges that correspond to up to five different lidar returns from the surface so let's have a look here here comes the laser pulse it first hits this branch then this branch this branch and so on until finally it hits the ground each of these collisions bounces back a pulse back to the aircraft and the aircraft is recording a continuous signal and time so it sees each of these returns as a pulse of photons at a given time okay and typically what the processing has to do is establish some background level and then say okay anything above a background is going to be categorized as a discrete pulse okay or discrete return and the hope is that the last return is typically taken as the return off the ground it is now bear in mind the number of returns and the complexity of returns also provides information about the surface material so you can imagine if you shot a laser beam down onto a bedrock surface you might get only one return and that would be off the ground same if you shot it off the roof of a house you might just get one return off the roof in contrast if you shoot it into a forest of dense vegetation you're almost always going to get multiple returns and sometimes you can actually invert that information to learn about the structure of forest canopies for example now importantly it's very common to take the first return and the last return and turn them into very different products and so here a digital terrain model is actually shown here in B and this is what you might call a they're earth digital model this is made from the last return okay which is supposed to be from the ground so in theory this should show you information about the surface texture of the ground okay so that would be a digital terrain model this one a is a digital surface model this might be a digital elevation model that's made from the first return which is typically going to be off of that that tree canopy right so take a look at these these are from the exact same area okay the digital surface model shows the height of trees then it shows areas that have been clear-cut where there's fewer trees and then it shows an alpine meadow here where there's no trees at all so this gives us this amazing rendering of the forest canopy right the digital surface model whereas the digital terrain model gives us system sees through the forest and gives us this amazingly detailed rendering of the actual land surface okay so you can start to imagine how powerful these different things are the geologist might care a lot about the digital terrain model whereas the biologists might care more about the digital surface model okay so now that you have an understanding of some of the basics of lidar let's talk a little bit about the format of the data and you can start to get a sense that there's a lot of information that needs to actually be stored and light our data is typically stored in a point cloud format okay and that usually has a dot la s extension a last file and we can think of the last file as you know basically a table that has all these different entries in the table so for example it has the X and the y coordinate of each shot location it has the intensity and in this case it has the return number so this would be perhaps you know the first return the second return the third return how many total returns and importantly here it would also have the classification so it's quite common that light our data is classified in terms of whether it was being water whether it was being forests or other categories so for example here's a little clip from our own Addison County lidar data set and things are broken down into all the different classes here so it has a total of one and a half billion lidar shots so these are going to be big file 424 million of those were deemed ground ok one hundred and ninety-three thousand were deemed water ok so you'll find that different lidar data sets have different classification categories but almost always the data will show up having been classified during the processing stage ok so now let's say you want to take that light our data and make it into a digital elevation model or a de M remember the fundamental problem we're facing here which was that the light our shot points are not spaced on a regular grid they were collected as the laser raster Daken force zigzagging so they tend to have a zig zag pattern and of course we would like to convert that zig zag pattern into a grid like a raster and the way we're going to do it is we're going to need to resample the lidar data so we know the coordinate of each of these laser shots and we're going to resample those into a gridded model so just to remind you we're going from a point cloud data into a grid or a raster format here so what's the basic way this works basically what you do is you start with an empty grid you just you know you decide okay I want a grid of one meter by one meter pixels and each of those pixels then get to coordinate and so then you can base it we overlay your grid pixels with the actual shot points okay and then you do a search you basically say okay I'm going to start at this pixel and I'm going to search for lidar shots in some radius R of my pixel okay and then I'm going to take maybe I find three or four light our points I'm going to compute an inverse distance weighted average I'm basically going to compute a weighted mean elevation from those light our shot points that were in my vicinity and in the case where I don't find any light our shot points I would first end up with a no data or a null pixel which I would then interpolate all right so let's look at a little example of this here this isn't perfect but it's pretty close this is an example of an how you might get a mean elevation by taking an inverse distance weighted mean all right so here's P this will be the pixel value that we're trying to fill okay and each of these red dots or red pluses represents a light our shot point so let's say we did a search within maybe two meters and we came up with three shot points okay one of those was 23 units away one was 58 centimeters in one with 35 centimeters okay and we don't know the elevation here yet this is what we're trying to figure out and we're going to do a weighted average of these three lidar shot points okay - 33 20 and 580 so those might be the elevations perhaps also in centimeters okay so how do we do this how do we do an inverse distance weighted here's the formula the elevation of the shot point is given by the sum of all the elevations times one over their distance from the actual point P so what does that mean it means that if you were further away if D was bigger then that means you're going to be multiplying Z by a smaller number and you're going to get weighted less and then we're going to divide all of those by the sum of 1 over D ok so in this example we'd have 230 over 23 plus 320 over 58 580 over 35 we sum those on the top and then we divide them by the sum of 1 over each distance and so the key thing is that this point right here 230 is going to get the highest waiting because it is closest right and this is going to get the lowest waiting because it is farthest away and so in this example this was an inverse distance waiting or weighted mean and in this case we use the weights to be 1 over R it might be more common to use a weighting of 1 over R squared now in this case if you think about it this is actually penalizing for distance even more so this is this is waiting faraway points even less basically and so just to finish this up what controls the resolution of a de M that you could actually make from like our data points and obviously what's control what controls this is basically we'd love to have the smallest pixel size that we possibly can however if we have pixel sizes that are too small we may end up with a lot of empty or null pixels right and so this experiment shows up lidar data that had an initial shot density of about 3.6 shots per meter squared okay so each meter squared at 3.6 laser shots on average okay and here's four different attempts to make DMS here's a quarter meter DM a half meter a one meter and a two meter when we attempt the 2 meter de M or 2 meter pixels we're only left with 471 null pixels so basically almost every pixel had at least one shot point in it we get down to one meter pixels we're only left with 666 so still most pixels had a shot point if we try to go to 1/2 meter we end up with 60,000 null pixels so here we've gotten to a point where a large number of the pixels actually didn't have a shot in them or didn't have a shot within the search radius and so obviously the optimal size here is going to be someplace between 1/2 meter in a 1 meter pixel resolution just to reiterate that obviously people want smaller pixels because that may record more surface detail however the the minimum pixel size you can get is limited by the shot density which of course is ultimately controlled by how the data was collected during the data collection awesome so that wraps up our video about lidar we talked a little bit about how the data is collected what the format of the data is and then how we can create a digital elevation model from that data thanks for listening

Channel: Middlebury Remote Sensing

Views: 14,971

Keywords:

Id: GbJ9xPf0gwg

Length: 21min 37sec (1297 seconds)

Published: Fri Apr 07 2017