Nate Green and Jacob Hartle Forest and Natural Resource Management 3262 Introduction Karst Topography is an important landscape to understand due to the amount of destruction it can cause to our infrastructure ID: 535279
Download Presentation The PPT/PDF document "Using LiDAR Data to Automatically Deline..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Using LiDAR Data to Automatically Delineate Sinkholes in Southeastern Minnesota
Nate Green and Jacob Hartle
Forest and Natural Resource Management 3262 Slide2
Introduction
Karst Topography is an important landscape to understand due to the amount of destruction it can cause to our infrastructure.
Former methods for studying sinkholes have been manual or with aerial photography.
Our goal is to find a suitable method to identify sinkhole remotely with the use of LiDAR data.
LiDAR is a useful and effective tool for identifying landscape and in our case sinkholes, due to the multiple returns a LiDAR pulse goes through after it is sent out from the sensor.
The key return of the pulse for our study is the last pulse, as this will show the shape and contours of the land and this will show depression in our area of interest. Slide3
Background
What is Karst topography?
Karst topography is a landscape that is formed from soluble rocks that are dissolved easily such as limestone and gypsum by underground water flows.
In order to form karst landscape there needs to be vegetation to supply organic acids for weathering, joints in the ground (water channels to allow for weathering in the rock), and a highly basic ground rock (calcium carbonate) (Christopherson, 2012, pg. 380).
In addition many other studies have been performed for identifying sinkholes most recently using LiDAR data.
There have been many studies conducted on trying to automate the delineation of sinkholes using LiDAR data.
One example is the study conducted by Alexander et al. (2013). They were trying to identify sinkholes with LiDAR to make the process more efficient and cost effective. Their focus on identifying the sinkholes was for land use planning.
Another example comes from Miao et al. (2013) and in their study they were trying to find a way to automate delineation of sinkholes. The authors mention many factors that need to be considered for sinkhole detection such as, “... Sinkhole size, map scale, contour interval, and slope of the ground...” (Miao et al., 2013, pg. 545). It is because of these many factors that they prove automation is difficult.Slide4
Example of a Sinkhole
To give a more current example of the damage that a sinkhole can do here is the Corvette Museum sinkhole that occurred on February 13, 2014.
The sinkhole is estimated to be around 60-70 ft in diameter and 40 to 50 ft deep.
The length of time it took to form the sinkhole after the ground gave way was about one minute and occurred in the morning of February 13, 2014.
Corvette Museum, Bowling Green Kentucky
Image Source:
http://www.usatoday.com/story/news/nation/2014/02/12/corvette-museum-sinkhole/5417171/
Youtube link to surveillance video of the Museum the night it happened:
https://www.youtube.com/watch?v=IukDWhf7U9ISlide5
Importance
Looking at the Corvette Museum sinkhole again we can see the amount of damage that sinkholes can cause to human infrastructure.
In addition to infrastructure costs there is also a human and safety cost with the possibility of sinkholes forming at any time.
Therefore identifying sinkholes is of utmost importance in order to be able to detect potential problems before they occur and solve them before they can cause damage to our infrastructure. Slide6
Study Area
Southeastern Minnesota is a hotspot for karst topography
Since we generally do not know when these areas will collapse, it is important to identify where sinkholes are in respect to populated areas.
Following this idea we decided to find an area in Minnesota that showed characteristics of showing Karst topography and we found it in Olmsted County.
The sinkholes in this area are all well documented and many of them are ground truthed (Jeff Green, personal communication, November 7, 2014).
The images on the next slide show where there are areas of karst topography and our areas of interest within Olmsted County.
Olmsted County, Minnesota
Image Source: Google EarthSlide7
Study Area Cont.
Map of Karst Lands in Minnesota
(Rahimi, M., & Alexander Jr, E. C., 2013, pg. 471).
Map of Karst Features in Olmsted County
Image Source: Jeff Green Slide8
Methods
By subtracting two DEMs (one the ultra accurate LiDAR derived product) from one another the resulting output image should display the sinkholes.
USGS 10m DEM
LiDAR derived filled DEM
Subtraction of DEMs
Order of the operation, positive Z-values
Unintuitive, but allows user to more easily see the locations of sinkholes on a black background.
Conversion of output DEM to vector data
Cell values must first be truncated to integer values.
Accuracy assessment using sinkhole pointfile
How many sinkhole points fall into sinkhole polygons?
Select by location, sinkhole points as target file, sinkhole polygons as source layer.
Creation of Arc Model for automated process
Adding the World View Imagery in ArcMap as a backdrop to give context
Slide9
Methods Cont.
Model
Created in ArcMap using the Model Builder
Automates the adding of the DEM files, filling of the DEM, subtracting of the two DEMs, and the steps needed to convert the output raster into a vector file
Due to the need to use the GUI in ArcCatalog, creation of the DEMs from the LiDAR pointfile is not able to be automated
Due to the need to use the GUI in ArcMap for selecting the sinkhole points in sinkhole polygons, this step could not be automated Slide10
Results
The sinkholes appeared in our difference raster image.
For our western AOI we ended up identifying 20 out of 66 sinkholes and that is 30% accurate for our LiDAR derived DEM and 15 out of 66 sinkholes (22%) for the 10m DEM.
In addition the end result for our northern AOI was 61.6% or 151 out of 245 sinkholes were identified for the LiDAR derived DEM and 51 out of 245 (20%) for the 10m DEM. .
Clearly the northern AOI had a much better conversion rate and this is why it had a better result.
Attempt at automation
We created a model in ArcGIS and with it we were able to semi-automate the process
The model allowed us to: import the DEM files, fill the DEMs, subtracts the DEMs and make that output into integer values.
In addition, it converted the now integer based rasters into polygons, which were used to identify the sinkholes.
It is semi-automatic due to the limitations of a model such as not being able to open ArcCatalog (we used this to clip the 10m DEM file as it covered the Continental United States and it would crash if we opened it directly in ArcMap) and it does not allow a user to select a hard number of points in polygons due to a lack of a selection in ArcGIS.Slide11Slide12Slide13Slide14Slide15Slide16Slide17Slide18Slide19Slide20Slide21Slide22Slide23Slide24Slide25Slide26Slide27Slide28Slide29Slide30Slide31Slide32Slide33Slide34Slide35Slide36Slide37Slide38Slide39
Improvements
If time allowed, we would have tried to get different images and perhaps looked at custom software for cloud correction if it is available for students. Ultimately this is to make the images more aesthetically appealing for the final product.
Another objective we would have attempted if there was more time, would be to have more automation for the identification of sinkholes. We were able to essentially semi-automate the process and there were some limits involved with what a model could do for us.
If we had more time for this project we would have chosen an object based classifying scheme to identify sinkholes due to the fact that sinkholes are individual objects in the landscape.
In addition if we used the object-based path we could develop rule sets that may be able to automate the process for any area if the rule set is valid or at the very least our study area.
Using Python, we could make this project more automated and more dynamic than just using an Arc model. Also it works well with iterations and it would be carrying out this process multiple times depending on the amount of karst topography in the area and files for analysis.
Some of the other studies have incorporated segmentation of hydrology in their study areas and it does work to an extent, but it can lead to oversegmentation, which can be a problem in and of itself. Slide40
Conclusions
Overall, the project was a success to an extent.
We were able to identify sinkholes in both of our study areas, but there was a limit to how many we could identify based off the results after converting the sinkholes points into polygons.
Our results show that resolution of your data really matters when trying to identify sinkholes as they vary in size and this will allow or negate them from being seen in the result image, thus affecting the accuracy in the end.
Our end result is comparable to other studies done in this field because trying to find a way to automate sinkholes is a difficult task due to the presence of water nearby and the hydrography of the area can affect the presence of sinkholes.
Slide41
Works Cited
Alexander, S. C., Rahimi, M., Larson, E., Bomberger, C., Greenwaldt, B., & Alexander Jr, C. (2013).
Combining LiDAR, aerial photography, and Pictometry® tools for karst features database management.
Christopherson, R. (2012).
Geosystems: An introduction to physical geography
(8th ed.). Upper Saddle River, NJ: Prentice
Hall.
Corvette Museum Sinkhole [Image]. (2014). Retrieved from the USA Today website:
http://www.usatoday.com/story/news/nation/2014/02/12/corvette-museum-sinkhol
e/5417171/
Green, Jeff. (2014, November 7th). Telephone interview.
Miao, X., Qiu, X., Wu, S. S., Luo, J., Gouzie, D. R., & Xie, H. (2013). Developing
Efficient Procedures for Automated Sinkhole Extraction from Lidar DEMs.
Photogrammetric Engineering & Remote Sensing
,
79
(6), 545-554.
Rahimi, M., & Alexander Jr, E. C. (2013). Locating sinkholes in LiDAR coverage of a glacio-fluvial karst, Winona
County, MN.Slide42
Questions/Comments