This project was part of a larger USGS study led by Mark Kozar to investigate water availability in the vicinity of the Leetown Science Center in the eastern panhandle of WV (Jefferson County) The specific objectives of the LiDAR mapping component were related to the need to map sinkholes as potential conduits to groundwater. Previous work mapping sinkholes had produced results that were insufficient to the needs of the larger project. We suggested trying LiDAR data as a mapping method. We had three objective for the LiDAR mapping; first - acquire high quality data, second – generate fine-scale surface models, and third – use a combination of land form analysis, photointerpretation, and statistical analysis to map and assess sinkholes in the Leetown area.
This picture is just to illustrate that LiDAR uses laser pulses from an airplane to determine spot heights on the ground.
This is a perspective view of the Leetown Science Center showing LiDAR height returns overlaid on an aerial photo. The point here is that the laser pulses interact with the vertical structure of the landscape. In a forested areas, the 1 st returns come from the upper tree canopy, but some laser energy penetrates tree canopies and returns from the ground surface (see inset). The challenge in using LiDAR data for surface investigations is to filter out upper canopy returns and rooftops to get to the true ground surface to create what’s often called a “bare earth” surface model.
We acquired the data for the project through a partnership with the USDA, Natural Resources Conservation Service. We were investigating contracting for our own flight, but we became aware that the NRCS was planning a flight for Jefferson County as a test of using LiDAR for updating the soil survey. We established a partnership where we agreed to do a QA/QC campaign of the data (acquired by a contractor) in return for immediate access to all the data. The data was delivered as 81 tiles (quarter-quarter quads) in first return, last return, and vendor processed “bare earth” representations. The data was acquired with a nominal LiDAR point spacing of 1 meter or less, although this spacing is reduced slightly after processing for bare earth.
Our QA/QC campaign consisted of using survey-grade GPS units to measure surface heights at 38 locations throughout the county in a variety of open land cover types, and then comparing our surface heights to LiDAR surface returns. We found that the surface heights from the LiDAR data acquisition were well within the contract specs of +- 0.15 meters. The bar chart at the lower right shows the mean difference between LiDAR derived height estimates and survey-grade GPS height estimates. The LiDAR data slightly underestimated heights on asphalt and gravel, and slightly overestimated heights on vegetated areas.
As mentioned earlier, one of the challenges in exploiting LiDAR data is to remove the returns from vegetation to get at the true ground surface. In the karst environment of the eastern panhandle, many sinkholes exist in woodlots, which were probably left unfarmed because of the difficulty of working the land. This slide shows an airphoto of a portion of the surface watershed of the Leetown Science Center to illustrate the LiDAR filtering task.
This image is of raw last return LiDAR data before filtering to illustrate that without processing, tree cover obscures surface features.
There are many filtering techniques for LiDAR data to remove returns from vegetation, and generally the data vendors have their own proprietary routines. Although the vendor in this case had filtered the data using their own technique, we found that there were still substantial artifacts from vegetation that obscured ground surface features. We partnered with Jeff Evans of the US Forest Service to implement a “progressive curvature” filtering technique that he came up with to remove vegetation returns from LiDAR data in western forests. His technique considers LiDAR returns at multiple scales to determine which returns are from the canopy and which are from the forest floor. The result is a technique that is very effective at removing vegetation.
This shows the result of implementation of the progressive curvature filter. The first view is of the raw, unfiltered LiDAR data that I showed previously. [HIT SPACE BAR]. The second view is of the same area after filtering for vegetation. Note the surface features that become apparent after removing vegetation.
Once a good “bare earth” elevation model was produced, we needed some way to quantify depressions. While sink identification and filling routines are built into ArcGIS, or can be added through extensions like ArcHydro, trials with these approaches showed that they vastly over-mapped sinks (identifying over 96,000 sinks in the Hopewell Run watershed). This is probably due to small artifacts in the interpolated DEM. Instead, I implemented a landform shape index first developed by Henry McNab of the US Forest Service for site productivity analysis. This approach quantifies local shape by comparing the elevation of each pixel against the mean elevation of surrounding pixels. If the focal pixel is lower than it’s neighbors, then it is sitting in a local depression. If it is higher, then it is on a local hill. By varying the size of the neighborhood, you can represent different scales of land form. I used a doughnut shaped neighborhood ranging from 2 meters to 10 meters (1-5 cells on a 2-meter DEM) in order to capture landform at a local scales relevant to mapping small circular depressions (potential sink holes).
This map shows the results of the terrain shape index with depressions in spades of red and hills in blue.
Note the small circular depressions that become apparent after implementing the landform shape index, highlighted here in yellow.
Field inspection showed that this was indeed a sinkhole 8-10 meters in diameter.
Here’s another view.
We mapped 94 small circular depressions in the Hopewell Run (Leetown Science Center) drainage using this method. We attempted to visit as many sites as possible that were flagged from the LiDAR surface model and landform shape analysis. We were only able to get to 55 sites due to private land access restrictions. Of those, we found an actual “throat” to a sinkhole in 16% of sites, but some form of circular depression in 85% of the sites. 15% of the sites were not easily explained as a natural depression feature. In some cases these were road cuts or breaks between rock outcrops. Overall, we felt pretty good about the detailed landform shape that we were able to quantify using the LiDAR data.
Here’s another view.
Once the data is processed for surface form, there are many other possible geologic applications including assessing sinkhole susceptibility by rock type, or using the bare earth models for fault line tracing, to identify structural anomalies, or to assess vegetation height and structure in relation to geologic features (geobotany).
Here an example map looking at the relative density of sinkhole formation in relation to geologic units. [Darker shades are higher density]
Structural geologists may be interested in a LiDAR surface model because it can help confirm and locate structural features. In this case, maps of fault lines were created based on field investigations of the Leetown area. [NEXT SLIDE]
Here are the field mapped faults overlaid on the LiDAR surface model. The LiDAR data in this case confirms the general location of the cross fault, but might also suggest some additional cross fault locations [CENTER OF IMAGE].
[ANY ADDITIONAL QUESTIONS, CONTACT JOHN YOUNG] Thank you… [ACKNOWLEDGEMENTS ON NEXT PAGE]
Using LiDAR to Map Sinkholes (EPAN09)
Using LiDAR to map sinkholes in Jefferson County, West Virginia John Young, USGS Leetown Science Center Kearneysville, WV
“ I read the news today, oh boy Four thousand holes in Blackburn, Lancashire And though the holes were rather small They had to count them all…” A Day in the Life The Beatles
Project Objectives <ul><li>* Part of a USGS study of water availability and threats to water supply of the Leetown Science Center (Kozar et al. 2007) </li></ul><ul><li>Acquire LiDAR data to allow for modeling of fine scale surface features </li></ul><ul><li>Generate fine-scale digital elevation models from LiDAR data </li></ul><ul><li>Use topographic analysis, aerial photography, and statistical analysis to map sink holes and potential groundwater recharge zones from fine-scale surface models </li></ul>
What is LiDAR? <ul><li>LiDAR = Li ght D etection A nd R anging (aka “airborne laser scanning”) </li></ul><ul><li>Laser pulses sent from aircraft in dense scanning pattern (0.5-2 meters apart) </li></ul><ul><li>Return of laser pulse recorded at aircraft </li></ul><ul><li>Time to return, speed of light, and altitude of plane used to compute surface height (z) of each pulse with high accuracy </li></ul><ul><li>Ground position of each laser pulse computed using differential GPS (x,y) </li></ul>
LiDAR returns: First (top of canopy, roofs), Last (ground surface) LiDAR returns (overlaid on aerial photograph)
Data Acquisition <ul><li>Acquired LiDAR data by partnering with USDA-NRCS </li></ul><ul><ul><li>Conducted accuracy assessment of LiDAR data acquisition in exchange for access to data </li></ul></ul><ul><li>LiDAR flight of Jefferson County WV, by Sanborn, Inc. in April 2005 (leaf off) </li></ul><ul><li>LiDAR data delivered as first return , last return , and “ bare earth ” (vegetation and buildings removed) </li></ul><ul><ul><li>81 tiles (1/16 th quad) </li></ul></ul><ul><ul><li>< 1.0 meter point spacing </li></ul></ul>
<ul><li>USGS QA/QC Campaign: </li></ul><ul><li>38 stations established throughout county in variety of land surface types </li></ul><ul><li>Survey-grade GPS used to collect surface height data </li></ul><ul><li>GPS surface heights compared to mean LiDAR return height within 2 meters of GPS points </li></ul><ul><li>All but 1 checkpoint were within ± 0.15 meter RMSE (< FEMA spec) </li></ul>
Problem: Find method to locate surface sinks, even under forest canopy Color Aerial Photo, 0.6 meter pixel resolution, 2003
Data: LiDAR, acquired Spring 2005, delivered Fall 2005 Raw (last return) data gridded to 2m surface
Progressive Curvature Filtering <ul><li>Evans and Hudak (2007) proposed a method for processing LiDAR data to find ground surface in forests of the interior western U.S. </li></ul><ul><li>Method uses an adaptive, iterative filter that considers scale of variation </li></ul><ul><ul><li>Fits “thin-plate splines with tension” at multiple scales to examine local curvature and determine which points to filter out </li></ul></ul><ul><li>Effective at removing vegetation </li></ul>
Data processing: Progressive Curvature Filter (Evans and Hudak, 2006) PCF filtered data gridded to 2m surface Raw (unflitered) last return data, gridded to 2m
Data processing: a modification of McNab’s (1989) “Terrain Shape Index” TSI = DEMgrid – focalmean(DEMgrid, annulus, 1, 5) Focal cell higher than surrounding cells = convex Focal cell lower than surrounding cells = concave * Graphic after F. Biasi (TNC)
Data processing: Landform analysis Landform shape in 10 m window, red = concave, blue = convex
Data processing: Landform assessment Find compact bowl features, possible sinkholes
Field validation results 94 sites mapped, 55 visited on ground Sink (throat) found: 16.4% Probable sink (no throat): 43.6% Depression: 25.5% Not a sink: 14.5%
Other applications of LiDAR <ul><li>Site analysis </li></ul><ul><li>Geologic structure </li></ul><ul><li>Fault line tracing </li></ul><ul><li>Hydrology / floodplain assessments </li></ul><ul><li>Vegetation height/structure </li></ul><ul><li>Slope analysis </li></ul><ul><li>?? </li></ul>
Conclusions <ul><li>PCF filtered LiDAR data provided a high resolution “bare earth” elevation surface </li></ul><ul><li>Converting elevation values into a landform shape index was effective at highlighting depression features </li></ul><ul><li>Caveat: </li></ul><ul><ul><li>Depression ≠ sinkhole </li></ul></ul><ul><ul><li>(but it’s a good place to start looking!) </li></ul></ul>For additional info, contact John Young (email@example.com)
Acknowledgements <ul><li>Jared Beard , USDA-NRCS soil scientist, Moorefield, WV (collaborator) </li></ul><ul><li>John Jones , USGS Geography division, Reston, VA (collaborator) </li></ul><ul><li>Bob Glover , USGS Geography division, Reston, VA (GPS survey) </li></ul><ul><li>Jeffrey Evans , USFS, Moscow, Idaho (filtering algorithms) </li></ul><ul><li>Kenny Legleiter , (formerly of) USDA-NRCS, Ft. Worth, TX (LiDAR flight contracting) </li></ul><ul><li>Data acquired by Sanborn, Inc. under contract to USACE/USDA-NRCS </li></ul>