GIF Lidar Workshop Agenda

Transcription

1 Workshop: Introduction to Lidar data and analysis April 5, GIF Lidar Workshop Agenda PART 1 Introduction to LiDAR LiDAR basics Types PART 2 LiDAR analysis DEM & Terrain products Vegetation products Applications and Effects on RS PART 3 Next steps Future of LiDAR? Resources 2 Lidar Workshop: Part 1 Introduction to Lidar Lidar basics Types of Lidar Lidar data Application examples 3 1

2 Introduction to Lidar Lidar = Light Detection and Ranging Lidar Data: 1. Range. The measurement of the speed which a pulse of light returns to a sensor is converted to elevation above sea level. R = ½(tc) R = range t = time c = speed of light 2. Intensity. The recorded maximum return from all the returns. Image modified from Lefsky et al with tree graphic from globalforestscience.org. 4 Lidar for Data Capture Advantages Collect large data sets quickly and economically 100 sq mi 1,000+ sq miles Sensor can be flown at night Acquisition in all seasons leaf on and leafoff Can be collected during cloudy conditions IF clouds are above aircraft Very accurate Disadvantages Cannot sense through rain, thick clouds, haze, wind (dust), or smoke Limited in thick forest or dense vegetation Surface materials may absorb laser Large data file sizes Small project areas not economical 5 Lidar Components Laser & receiver Records time to target Laser wavelengths can differ (e.g nm) 100 khz systems available today IMU Inertial Measurement Unit Gyroscopes and accelerometer Records roll, pitch, yaw of aircraft GPS Differentially corrected Provides cm accuracy of aircraft Allows cm accuracy of laser pulse On board computer Records data: Laser distance (intensity); IMU info; GPS info Converts into XYZ On board display Scanning mechanism 6 2

3 Lidar Differences Airborne or ground Type of scanning mechanism Single, multiple, or waveform returns Footprint size Posting density Application: atmospheric / terrestrial / bathymetric 7 Oscillating mirror Airborne Scanning Lidar The integration of a device such as an oscillating mirror allows for the development of scanning Lidar. Oscillating mirror Palmer scanner Rotating polygon Fiber optic array 8 Ground based Lidar Ground based lidar systems are hemispherical scanning laser range finders that fire millions of laser pulses and records detailed structural information at a range of up to 200 m. Data can be used to derive: canopy height basal area and stem density vertical foliage distribution leaf area index Sources: Omasa, et al

4 Lidar Returns Discrete return Single Multiple Waveform return Discrete return lidar devices major peaks that represent discrete objects in the path of the laser illumination. Waveform recording systems capture the entire signal trace for later processing. 10 Lidar Specifications Flying altitude ( m AGL) Laser pulse: Scan: Pulse rate frequency, measured in KHz (as high as 167,000 points per second) Laser is coherent light, but does spread Beam divergence, measured in milliradians 1. Scan angle, measured in degrees 2. Scan frequency, measured in Hz 11 Footprint Size & Density The laser footprint is approximately circular on the ground. Footprint size is a function of: Beam divergence Flying ht/speed Scan angle Posting density is a function of: Flying ht/speed Scan angle Scanning frequency Pulse repetition 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 1 m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 0.5m 12 4

5 Bathymetric lidar terrestrial lidar Terrestrial Lidar Lasers for terrestrial applications generally have wavelengths in the range of nanometers, where vegetation reflectance is high. One drawback of working in this range of wavelengths is absorption by clouds, which impedes the use of these devices during overcast conditions. 13 Bathymetric Lidar Bathymetric Lidar systems typically use a blue green laser centered on 532 nm and a raster scanning mechanism to acquire lidar data to measure bathymetry Source: NASA Experimental Advanced Airborne Research Lidar (EAARL) 14 Lidar Data Mass points: x, y, z, intensity 15 5

6

7 Lidar Point Density Visualization of our high density Lidar product from NCALM: Mass points: x, y, z, intensity 19 Lidar Workshop: Part 2 PART 2 LiDAR analysis DEM & Terrain products Vegetation products 20 Lidar Data Products DEM Digital Elevation Model elevation points over a contiguous area DTM Digital Terrain Model elevation information about bare earth surface without the influence of vegetation or man made features DSM Digital Surface Model elevation information about all features in the landscape, including vegetation, buildings and other structures CHM Canopy Height Model Height information about vegetation features with elevation removed Canopy Height Model Digital Surface Model Digital Terrain Model Digital Elevation Model 21 7

8 Lidar Data (mass points) Field Data (from FFEH plots) Ground Above ground Slope DTM CHM DSM Aspect Trees Height profile Species Canopy cover Crown size DBH Tree height Height to Live Crown Lidar Data Analysis Flow 22 Digital Elevation Model DEM Available for the Sugar Pine area at 1m resolution. This can be sampled at coarser resolutions as needed. Aspect and slope products can be made from these layers. 23 Vegetation Analysis There are many ways to analyze Lidar data for vegetation information, in general, we can divide these methods into two types: 1.analyzing canopy height model (i.e. the last return data); 2.Analyzing the height profile through the full range of Lidar returns. Canopy size, tree height, spacing, canopy cover 24 8

9 Vegetation Analysis There are many ways to analyze Lidar data for vegetation information, in general, we can divide these methods into two types: 1.analyzing canopy height model (i.e. the last return data); 2.Analyzing the height profile through the full range of Lidar returns. 95 th 45th 10 th 25 Ground Interpolt LIDAR Data (mass points) filter Vegetation Individual Tree Level Analysis The canopy height model (CHM) can be used to derive many forest related variables. Watershed segmentation DTM CHM DSM Segmnt Trees Species Tree height 26 Crown size DBH Canopy cover Treetop Detection We also investigated a range of techniques to find tree tops: e.g. local maxima approach, shape matching 27 9

10 Individual Tree Detection Tree tops together with segmentation approaches allow us to automatically delineate trees across the landscape 28 Sugar Pine lidar products 29 Sugar Pine lidar products 30 10

11 Lidar Intensity Lidar light has very small bandwidth (2 5 nm); most multispectral imaging remote sensing bandwidths are often quite wide: nm. The intensity value is not directly comparable to the reflectance from the same object in optical remote sensing due to scattering and other factors. There are no guidelines on the interpretation of Lidar intensity images. But intensity still can be useful for analysis. 31 Lidar Applications From Lefsky et al. 2002: Only a few application areas have been rigorously evaluated, and many other applications are feasible. Developments in lidar remote sensing are occurring so rapidly that it is difficult to predict which applications will be dominant in 5 years. Currently ecological applications of lidar remote sensing: ground topography, 3D structure and function of vegetation canopies, forest stand structure attributes. 32 Lidar Applications: Vegetation Canopies From Vierling et al Lidar: shedding new light on habitat characterization and modeling. Frontiers in Ecology and Environment. Lidar can provide fine grained information about the 3 D structure of ecosystems across broad spatial extents. This data can be used to investigate: Animal habitat relationships; Fire behavior; Carbon and energy balance; Infiltration, canopy capture and sublimation. Canopy Percent High: 100 Low: 0 Lidar derived vertical distribution plots showing the percentage of laser pulse hits that occurred within a particular height classification. This kind of analysis helps us more fully understand the 3 D structure of vegetation

12 Lidar Applications: Forest Structure From Lucas et al Retrieving forest biomass through integration of CASI and Lidar data. IJRS Lucas et al used CASI and Lidar data to map biomass in a complex Australian forest. leaf branch trunk total 34 Lidar Applications: Powerline monitoring Power line route planning Detection of clearance violations Vegetation encroachment to minimize potential flash over and bushfire risks Tower location planning and catenary curves of lines Calculate overhead heights for through and high vehicle traffic analysis 35 Calculated using 30 m DEM Calculated using 3 m Lidar data Inundation with <= 1m slr certain uncertain Why National Lidar is Needed Beaufort County, North Carolina. Modeling Flood Inundation: if Sea Level Rises 1m 36 Source: ASPRS Special Session on National Lidar Moderator: Greg Snyder, USGS Geospatial Land Remote Innovation Sensing Program Facility 12

13 Effects of LiDAR on RS: DEM/DSM extraction Complexity of the problem has been essentially eliminated Accuracy significantly improved 37 Lidar Workshop: Part 3 PART 3 Upcoming research in LiDAR LiDAR resources 38 Why National Lidar is Needed Key national applications: Updating USGS NED Floodplain mapping and risk assessment Emergency response, public safety, critical infrastructure Urban infrastructure planning and development Ecosystem structure, function and health Habitat assessment and management Water management Source: ASPRS Special Session on National Lidar Moderator: Greg Snyder, USGS Land Remote Sensing Program 39 13

14 Coordinated mission addressing earth surface deformation, ecosystem structure, and the dynamics of ice 8 day revisit frequency Combines two sensors including, 1. an L band Interferometric Synthetic Aperture Radar (InSAR) system with 35 m spatial resolution and multiple polarization 2. a multiple beam Lidar operating in the infrared with 25 m spatial resolution and 1 m vertical accuracy NASA DESDynI Mission 40 DESDynI Ecosystem measurements allow: estimation of forest height with meters accuracy estimation of threedimensional forest structure direct measurements of live above ground woody biomass (carbon stock) and structural attributes such as volume and basal area. More info here: 41 Status of USGS Lidar Advisory Committee Began October, 2008 Mission: Increase availability and use of consistent, accurate, timely and contiguous national Lidar data USGS disciplines coordinating to advance: Data availability, standards and applications The Center for Lidar Information Coordination and Knowledge (CLICK) A national Lidar program concept Stakeholder interaction (NSGIC, AASG, MAPPS, ASPRS, FGDC, IFTN, NDEP)

15 LAStools converting, viewing, and compressing LIDAR data in LAS format 43 University based research Simple tools for LiDAR data management No analyses executables, API, and source code: lastools.zip txt2las.exe las2txt.exe lasinfo.exe las2shp.exe shp2las.exe las2tin.exe las2dem.exe las2iso.exe lasmerge.exe las2las.exe lasthin.exe lasview.exe laszip.exe LAStools: example 44 Terrasolid converting, viewing, and compressing LIDAR data in LAS format Software developed in Finland Based on Bentley s Microstation Very good management / analyses Expensive

16 liblas converting, viewing, and compressing LIDAR data in LAS format liblas is a C/C++ library for reading and writing ASPRS LAS versions 1.0, 1.1 and 1.2 data Great information on software and resources 46 LiDAR news: General updates on LiDAR in science and otherwise Forum based Lists upcoming conferences, events 47 Studio Clouds from Alice Labs Based on either 3DMax or Maya Very promising Learning curve? 48 16

17 Topographic Laser Ranging and Scanning by J. Shan & C. Toth (editors) Introduction to theories and principles Airborne and terrestrial laser scanners Waveform Analysis Management of data Filtering Forest inventory topics Feature extraction Much, much more 49 Appendix 50 Lidar Applications: Topography Photogrammetric techniques have long been used to collect topographic information from stereo imagery. From Dietrich and Perron, The search for a topographic signature of life. Nature: The influence of life on topography is a topic that has remained largely unexplored. Erosion laws that explicitly include biotic effects are needed to explore how intrinsically small scale processes can influence the form of entire landscapes, and to determine whether these processes create a distinctive topography. Dietrich et al. are using Lidar and Radar to compare topographic landscapes on earth and mars

18 NEXTMap Affordable high resolution digital geometries and imagery data Interferometric Synthetic Aperture Radar (IFSAR) technology produces: digital elevation models 5m to 25m, 1m orthorectified radar images, and other value added products (contours, TINs) NEXTMap California program, based on IFSAR mapping of California in NEXTMap Terrain On Demand

19 You will be working with the following datasets that supplied. flightline.las Sample of raw data from the sensor Clip_LiDAR.las Small sample of multiple flight lines combined into a single file (to increase the density of points) Clip_LiDAR_above.las contains only above-ground points from Clip_LiDAR.las Clip_LiDAR_grd.las contains only ground points from Clip_LiDAR.las Naip_2005.tif Imagery collected by the National Agriculture Imaging Program over the same area as the Clip_LiDAR.las Because LiDAR analysis is still in its infancy, there is not a single tool out there that can do it all. Various software packages complement each other in different areas of functionality; for example, PointVue is excellent for visualization but does not offer much analysis options. In the following exercises we will be using the following software programs: PointVue LE, LiDAR Tools plugin for ENVI, and ArcGIS. Visualizing data We will now look at some characteristics of the raw data using PointVue 1. Open PointVue from the Start menu. 2. From the main menu, choose File / Open, navigate to your working directory and open flightline.las. This is the raw data out of the sensor. The program opens up with the rotating tool selected, so you can immediately drag the screen with your mouse to look at the data from different perspectives (hint: use the wheel for zooming). It s easy to get carried away with all the spinning and lose your point of reference. Don t worry; you can always go back to the original view through the menu: View / Top. Look around the data. Does it contain any obvious artifacts? Use the spinning controls and your mouse to look underneath the point cloud. You will notice outlier points far away from the majority of the data these are erroneous data called low points. It may help to change the displayed point size to make these points more visible: from the main menu, choose View / Settings / General tab / Point Size. Try 2 or 3. Let s change the depth exaggeration to really see the 3-D nature of the point-cloud data: From the main menu, choose View / Settings / General tab / Depth Exaggeration: change to 5 and click Apply. 19

20 Recall that there are a number of different LiDAR hardware systems. Can you figure out what type of sensor collected this data? (e.g. Palmer scanner, oscillating mirror, fiber scanner, rotating polygon). Oscillating mirror Palmer scanner Rotating polygon Fiber optic array To figure out the type of sensor used to collect this data, use PointVue to zoom in to the edges of the flighline dataset to the level of individual points. Basic.las file information PointVue also allows us to easily obtain some basic information about LiDAR files. This information including sensor type, projection, etc. is inherently kept within.las files but may be more or less complete depending on the supplier of the data. To view this information, choose Tools / File Information from the main menu. Note that we can use this information to find out the dimensions of the data (X, Y, Z) as well as how many points comprise the point cloud. Let s use the above information to calculate the average density of the working file. First, obtain the X and Y dimensions of the file by calculating the difference between the respective maximum and minimum values. Note that the units in FileInfo screen are reported in the projection of the file (UTM / NAD83, Zone 10 N, i.e. meters). Divide the total number of points by the area of the point cloud to find number of points per meter. This is one of the most basic and common ways to describe LiDAR data. Now open Clip_LiDAR.las in PointVue. This is a small clip of the final product delivered by the supplier. Right away you should notice that the point cloud appears more solid than the previous file. This is because Clip_LiDAR.las is a combination of a few overlapping flight lines to increase density of the data. Recalculate the density of this dataset using the steps above. LiDAR intensity In addition to the X, Y, and Z position of points, LiDAR sensors are capable of collecting the intensity of the data. We can look at the intensity directly or in combination with the height (rainbow-color) information. Let s explore some of the advantages of having LiDAR data intensity. With Clip_LiDAR.las still open, make sure that PointVue s viewing perspective is from Top. By quick visual inspection, are there any roads in this dataset? 20

21 Click the Apply Intensity button (black-red gradient) to add the intensity in the view. Notice that the road is now clearly visible. Explore the dataset and notice how the view changes when you apply or deactivate the intensity overlay. Analyzing the data In this portion of the lab we will try to extract trees from clip_lidar.las, one of the basic utilization of LiDAR. First, let s create a digital surface model (DSM), a digital earth model (DEM), and crown height model (CHM). We can think of the DSM as an interpolated surface that would result if we spread a malleable sheet over a forest or a city. It would include the outer shapes of trees, buildings, cars, and anything else above the ground. The DEM, conversely, simulates how the earth surface would appear if we removed all above-ground objects, including trees, buildings, etc. A CHM is the difference between the DSM and the DEM; it s like the DSM without the effect of the topography. We will use a LiDAR plug-in for ENVI to generate these surfaces. ENVI is a comprehensive software package developed for analyses of remotely sensed imagery. In particular, it has extensive tools for analyses and visualization of hyperspectral data. As most (if not all) other major remote sensing / GIS software suites, it was not designed specifically for analyses of LiDAR and for this reason we will use a separate plug-in. The plug-in was preinstalled on the GIF machines but you can freely download it from the ITT website. Generating DSM First, we will create a DSM. Open ENVI + IDL from the Start menu. From the main ENVI menu, click LiDAR / Rasterize LiDAR Data, and choose clip_lidar_above.las. clip_lidar_above.las is a subset of points from the original clip_lidar.las file classified as above ground. Notice that the working directory also contains clip_lidar_grd.las which is a file with only points that were classified as ground points. The classification is typically performed by the data provider or software such as TerraScan, both beyond the scope of this lab. The data can then be separated into multiple files based on classes, pulse return numbers, etc using free set of command-line tools known as lastools. Using lastools is slightly more technical so please ask in lab if you re interested in further details. Make sure the following options are selected: Select return number: All Returns Enter raster spacing [m]: 0.25 Projection: UTM, NAD83, Zone 11 N Select products: [1] Max, [2] min, and [3] mean elevations, [4] slope, [5] aspect, [8] intensity, and [9] Bare Earth Model Create a results folder and choose a sensible name for the output (e.g. DSM_025m.img ) When the algorithm finishes successfully, an Available Bands List window will appear. This window contains all files that currently open and can be viewed within ENVI. Notice that image you just generated appears in the list of files and has a subset of layers or bands 21

22 (e.g. Maximum Elevation, Minimum Elevation, etc.). Select the Maximum Elevation band and click the Load Band button below; the interpolated surface will appear in a new window. Double-click inside the Image window or select Tools / Cursor Location to view the value under the cursor. Note that the units are meters. Let s make the visualization more interpretable. 1. Select Tools / Color Mapping / ENVI Color Tables, scroll down and select Rainbow + white, 2. Stretch the histogram to the inner 98% of the data by clicking Enhance / [Scroll] Linear 2%. Take a look at the other bands, such as Minimum Elevation and Intensity. Notice how they differ. Maximum Elevation looks as though it may be a good approximation for canopy height. However, note that we can see inherent artifacts from LiDAR data within the canopies. We can fix that by applying a spatial median filter 1. Choose Filter / Convolution and Morphology 2. In the Convolutions Tool, choose Convolutions / Median 3. Apply kernel size of 5 4. Click Apply to File and select DSM_025m.img. Name the output DSM_025m_median5.img When ENVI finishes filtering the data, the output will appear in Available Bands List window and a new Display Window will be automatically opened. Let s compare the cross-sections of the two filtered and unfiltered rasters. 1. In the window with the original LiDAR raster, choose Tools / Profiles / X Profile. 2. Do the same for the image window that opened after the convolution operation 3. From one of the image windows, choose Tools / Link / Link Displays This is a good way to make sure that your data is in the ballpark and that it does not contain any astronomical outliers. Note that the tree profiles now look much less noisy. Also note that as you move the cursor, the Cursor Location / Value window now shows heights for the two datasets and the two Display windows show identical locations. How do the values from the filtered and unfiltered images compare over tree tops? Use Enhance / [Zoom] Linear to easily find the tree tops. Generating DEM 1. Go to LiDAR / Rasterize LiDAR data and choose clip_lidar_grd.las. 2. Choose the correct raster spacing, projection (same parameters as above) 3. Select [2] Minimum Elevation and [9] Digital Surface Model outputs 4. Click OK. How does the DEM compare to the DSM from above? Compare the X profiles of the two rasters as described in the section above. Generating CHM We will now create the CHM. As mentioned above, the CHM is the difference between the DSM and the DEM (i.e. CHM = DSM DEM). Matrix algebra is made really simple in ENVI so this difference can be calculated in just a few steps: From the main ENVI menu, go to Basic Tools / Band Math Type the following expression: b1 b2 (this stands for band 1 minus band 2 ) 22

23 Click Add to List Select the equation from the list and click OK. While B1 [undefined] is selected, choose the Maximum Elevation band from the DSM_025m_median5.img image. Then select B2 and choose the Minimum Elevation from the DEM we just generated. Name the output CHM_025m and click OK. The CHM is calculated and will appear in the Available Bands List. Take a look at the output. Change the Color Table to differentiate the elevations. Compare the CHM to the DSM and/or DEM by linking all of the Display windows and looking at the X-profiles. Notice the actual data values underneath the cursor. Extraction of trees We will be using an IDL program written by the Popescu research group from Texas. This is an experimental program written for research purposes and is described in a few journal articles. Because the program has not been developed into commercial-grade software, there are still a number of caveats we must be aware of for it to function as intended. For example, the program expects ENVI-type rasters but does not recognize files with.img extension. Instead, the images must be set of two files: the image file with no extension, and a corresponding header file (e.g. CHM_025m and CHM_025m.hdr ). In our case, I suggest the following steps: 1. In the results directory, copy and paste the CHM_025m.img file within the same folder 2. Rename Copy of CHM_025m.img to CHM_025m This results will now be able to Go to TreeVaW directory and double-click init_treevaw.sav to start the program. 1. Choose CHM_025.img for the Image File 2. The output will be a comma-delimited ASCII file so it would make sense to name it CHM25_TVoutput.txt 3. Go to Settings / Set Parameters 4. You can experiment with the parameters to obtain optimal results. For now, let s choose Min. Exp. Crown Width of 5. The parameters are explained in the Help file (Help / About TreeVaw). 5. Hit Save and then Execute. Let s analyze the results in ArcGIS Analysis in ArcGIS From the Start menu, open ArcGIS In ArcGIS, add DEM_025m.img Right-click the added layer and choose Properties Go to the Symbology tag and set the following properties: Stretched Band: Maximum Elevation Stretch type: Minimum-Maximum 23

24 Edit High/Low Values: HIGH: 1700, LOW: 1450 Color Ramp: Precipitation (right-click and uncheck Graphic ) Invert: Yes Now add CHM_025m.img. We can apply the same Symbology by importing it from the previous layer. In Symbology properties, click Import and select the DSM_025m.img layer. change the Min/Max values (0-60m works well). Finallly, we can add the tree detecftion results. Remember that they were outputted in ASCII format, so we will need to add their coordinates semi-manually. Add XY Data Import CHM25_TVoutput.txt file from TreeVaW Project it to UTM NAD83 Zone 11 N. Export the imported layer as a shape file. Let s compare the res How do the model trees compare to the CHM? Q12. What kind of errors do you notice in the results? Q13. Use the measuring tool to compare the actual tree crown widths to the attribute table from modeled results? Q14. Any interesting relationship between the height and crown radius in the results? Add NAIP imagery (supplied in your data directory) and compare to the CHM. Q15. How do the two datasets compare / align? Save your work (the next step might kill Arc). Add the clip_lidar.las file. Zoom in and analyze the point density (you can probably do by visual inspection). Q16. How does the point density compare in bare vs vegetated areas? What causes the difference? 24