Elevation Models from LiDAR

by Willy Li (CCGT Intern) and Blake Lytle (CCGT UAV and LiDAR Specialist)

LiDAR point clouds are a rich source of geospatial information and can be converted to high-resolution digital elevation models (DEM), such as digital terrain models (DTM) which represent the bare-ground surface and digital surface models (DSM) which include vegetation and building data. One of the challenges we face, however, is the sheer amount of data to process. Individual LiDAR files, such as an ASPRS LAS, may contain hundreds of millions of points and be several hundred megabytes to a few gigabytes in size. Large areas of interest, such as a county, may be represented by billions of data points which equates to many gigabytes or even hundreds of gigabytes of data.

Processing this much data could take days to run on a single machine, and that’s a long time to wait for your results. In many cases, a standard workstation computer will actually run out of memory and crash when trying to process an entire dataset by itself. Therefore, we need to use a type of divide-and-conquer system. In the methods section we will further explore the approach we used to accomplish processing a single county in a little over an hour with our GalaxyGIS Cluster.

Aerial LiDAR data has been collected across much of South Carolina by the SC Department of Natural Resources (SCDNR) and can be easily downloaded from many sources, such as the the National Oceanic and Atmospheric Administration (NOAA) Data Access Viewer. We obtained point cloud data for several counties in South Carolina and used the RapidLasso LASTools as part of a processing workflow to efficiently generate a DTM and a DSM for an entire county at a time using GalaxyGIS.

Advantages of Distributed Computing

We have applied our workflow to over a dozen counties in South Carolina comprising 13,545 individual LiDAR files -- 415 GB of data in compressed form -- and counting! This would have taken weeks, if not months, to accomplish on a single computer.

To evaluate time savings, consider the Dorchester County, SC dataset. It contains 253 individual LiDAR files and is 10.5 GB when compressed. Each file took an average of 9 minutes to process on GalaxyGIS, so the full dataset would take 38 hours on a single computer. We processed all the data on GalaxyGIS in 62 minutes, a 97% reduction in computation time!

Explore the elevation datasets created using GalaxyGIS in the interactive map to the right.

Methods

Our processing algorithm is 3-step process consisting of:

1. Splitting the point cloud data with its neighboring files for buffered input.

2. Distribute the data to our GalaxyGIS cluster for calculating DEMs.

3. Mosaicing our data back together to obtain our final DEMs.

Step 1 - Splitting

Running all of the point cloud data sequentially to calculate the DEMs will require intensive amount of computational power and time. Calculating DEMs from a large set of point cloud data on a desktop computer would likely take days to run, and it could crash your computer. For this reason, we decided to split our data and process them in parallel.

Each LAS file needs its extent buffered slightly to prevent gaps in the data. We generate a neighbor table to identify adjacent tiles and then zip the relevant files together as a single package for a single job on the GalaxyGIS cluster. Our workflow works for point clouds without classification codes, such as raw LiDAR or from photogrammetric processing, or can be modified slightly if the point clouds are classified.

Step 3 - Merging

Once all of the pieces return to the submit machine, we merge the individual rasters into a Mosaic Dataset in ArcGIS and, if needed, export to a single file.

Step 2 - Processing

Next, we submit the zipped data along with the processing script and a submit file to our scheduler, HTCondor. HTCondor distributes the jobs among the nodes of our GalaxyGIS cluster and each node processes its individual job as shown:

The processing node is sent the package of adjacent LAS tiles.
A buffer is applied to the central tile, then the adjacent tiles are clipped to the buffered extent.
The points representing bare earth are classified as ground points using the lasground tool.
Buildings and vegetation points are determined using lasclassify.
Point clouds are converted to 1) a bare earth terrain model and 2) a surface model using las2dem. This tool creates a temporary TIN and then interpolates to a raster grid.

Upon completion, the output rasters are sent back to the submitting machine.

Research Use Cases

The parallel LiDAR-processing resources developed by the Geospatial Center have been applied to several research projects here at Clemson which span multiple disciplines. We hope to see their use applied more and more in the future, so please contact us if you need assistance.

Mapping the Rice Kingdom

The Antebellum era gave rise to rice culture in North America, which in turn provided such rich benefits that the slave-driven economy increased ever more rapidly. Growing rice was brutal work involving clearing of ancient forests by hand, re-channeling of surface and tidal water over huge areas and moving mountains of earth in building dikes. Africans with special knowledge of rice culture from their home mangrove estuaries were prized and it is argued that without them, the rice kingdom of the Carolinas may never have evolved. Historian Edda Fields Black of Carnegie Mellon University - and Clemson collaborator - studies rice culture in Africa and its historical connections to South Carolina. Her book, "Deep Roots" illuminates the profound legacy of West African rice farmers and their thousand years of technology on the history, economy, and culture of the American South - and world. Now, geospatial scientists are uncovering the extent of their impact on the land itself.

Rice agriculture was abandoned in South Carolina after the Civil War; while it has enjoyed slight resurgence recently and at points in history it never has returned to its apogee and consequently the landscape today is littered with abandoned rice fields. Many of these function as valuable freshwater wetlands. Instead of rice, these wetlands today are home to thousands of migratory waterfowl. Ducks are a valuable commodity for hunting as well as birdwatching revenues and landowners are very interested in restoring and maintaining historical rice fields - for cultural and wildlife reasons. Other wildlife use the wetlands as well, including myriads of reptiles and amphibians, birds, fish, and invertebrates. We suspect the wetlands may provide a range of other ecosystem services including flood and nutrient retention, protecting downstream communities and ecosystems. To better understand the magnitude of these impacts, Clemson University is partnered with Low Country stakeholders to use remote sensing and mapping technologies to discover the true extent of rice fields. To date the partnership has mapped over 200,000 acres of inland and coastal rice fields. The maps will be field-validated and completed by August of 2018.

-Dr. Robert Baldwin

The remnant drainage features such as dykes and canals are difficult to identify in imagery data (left). With the help of CCGT and the GalaxyGIS cluster, the researchers created high-resolution hillshaded DEM's from LiDAR. The elevation data have abundant drainage features (right) and were used to map the extent of rice fields boundaries with unprecedented detail .

This project has the following major collaborators from Clemson University: Dr. Rob Baldwin, Dr. Daniel Hanks, Mr. Richard Coen, Mr. Michael Gouin, and the Clemson Center for Geospatial Technologies.

Stakeholders and Funding Agencies: The Nemours Wildlife Foundation, The Nature Conservancy, ACE Basin Project, Folk Land Management, Inc., Margaret H. Lloyd-SmartState Endowment; James C. Kennedy Waterfowl and Wetlands Conservation Center

Tree Cover Analysis using LiDAR-based High Resolution DEM's

The advent of the availability of LiDAR data covering large land areas in high levels of detail provides opportunities for new analysis of both topography and land cover. Because LiDAR returns can be classified by the surface from which they are reflected (ground, trees, water, buildings, etc.) models of ground surface as well as anything above ground surface can be made. Using this data, opportunities to measure tree heights and map forest canopy structure abound.

Our research seeks to link existing stream quality data with land cover and stream conditions for the watersheds that contribute to the sampling points from which this data was taken. Past research suggests that one of the important metrics that may be associated with the water quality at these points is the percent of the watershed which is covered in trees. Above ground heights greater than 1.5 m were identified to classify tree cover. This height cutoff was chosen based on literature suggestions as well as to include the large number of stands of young trees in the study areas. Values for percent forest cover will be adjusted for building presence based on detailed imager-based land classification. Further measurements for this study regarding tree cover may include the percent of the floodplain in tree cover, the percent of 30m stream buffer in tree cover, as well average tree height within these buffers and/or watersheds.

- Dan Callaghan and Dr. Chris Post

The LiDAR processing resources at CCGT were used to create high-resolution canopy height maps from terrain and surface models derived from LiDAR to characterize water quality in streams.

Project Members

If you have any question about how this project was conducted or the results, please feel free to contact one of the project members below:

Blake Lytle balytle@clemson.edu
Willy Li xiang3@clemson.edu
Patrick Claflin pat@clemson.edu
Patricia Carbajales-Dale pcarbaj@clemson.edu
Connor Kinzie ckinzie@clemson.ed

References

LAStools, "Efficient LiDAR Processing Software" (version 170608, academic), obtained from http://rapidlasso.com/LAStools