Home List Map Upload Request Area
Welcome to CoherentKarst! Here you will find leads identified from LiDAR data that can be exported by the user and then updated when they are ground truthed. It is my hope that sharing this database amongst trusted individuals who are actively ridgewalking will yield new discoveries and reduce redundant efforts.
Much of the data presented here was produced by a modified version of Alex Fischer’s python script (an explanation of which can be found at his website, caves.science), which is in turn a modified version of the program of Zhou et. al. (https://doi.org/10.1016/j.cageo.2016.02.021). The modified version of the Fischer script used by Steven Rehbein utilizes the same digital elevation model (DEM) filling model to locate sinkholes, but then filters out features that are not associated with carbonate rock and features that are associated closely with mining and roadways. It is also capable of batch processing, such that many tif files can be processed at once to allow for analysis of a large areas or entire mountain ranges. Importantly, the script also renders an image of each feature located in the DEM to allow for easy screening such that any point here at least reasonably appears to be a sinkhole. The Rehbein python package will be available here soon.
The “Upload” page will accept (a) geojson files produced with Fischer’s script and (b) kmz’s with embedded images produced with Steven Rehbein’s modified script. When uploading a geojson file, images of the DEM at each point will be rendered and the user required to screen each point. Following screening, the data will be uploaded and live on the page. Please try to use a unique prefix when uploading to avoid confusion. The kmz’s produced with the Rehbein script can be uploaded as is as long as they have been screened locally using a simple screening tool. If you aren’t interested in doing the processing yourself you can submit a request for an area which Steven will process, screen, and upload!
On the subject of screening: it is possible using both python programs to look for shallow features in the DEMs—such as features in the 0.5–2.0 meter range—that are in reality deep caves. However, such a search can return tens to hundreds of thousands of points which makes manual screening daunting or impossible. For such searches, a simple machine learning model trained on our DEM image is available to automate classification which works quite well. Reach out to Steven if you have an area you want analyzed for shallow features and we can take a look using this approach.
Users should be aware that the processing of the LiDAR point cloud data to a DEM can produce artifacts. Some of these artifacts can appear very sinkhole-like in the DEMs. In general, any feature that appears very point- or single pixel-like or that is exceptionally deep (say, deeper than 10 m) should be treated as suspect. If you find a point that interests you but will require a serious effort to visit please reach out before visiting so we can take a look at the actual point cloud data to double check the feature isn’t an artifact. Efforts are underway to produce DEMs from the point cloud data that more closely align with our use.
Take a look at the points, their images, export some and get out there and check them! And when you return, please update me and the website with your findings.
Finally, I want to thank individuals who have contributed to this effort, including: Alex Fischer, whose script helped produce the data presented here; Jason Ballensky, whose database of caves served as a model for this database; and Pete Johnson and Philip Schuchardt, whose record keeping and tireless LiDAR lead checking inspired both this database and my own ridgewalking efforts.
Reach out to me with any questions or concerns at steven.m.rehbein@gmail.com.