-
Notifications
You must be signed in to change notification settings - Fork 14
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
ISSUE-148 Add option to burn features from multiple vectors onto raster #183
Conversation
Nice work... please keep us updated on the trial and error approach with the alt_locator. Wouldn't it just be great if we were able to search via proximity to some geographic feature? |
I tried to come up with something more efficient than trial and error: calculating the latitude and longitude of the dams/reservoirs using the Field Calculator in QGIS. However, I found that the bounding box on the alt_locator had some issues so this approach is still pretty inaccurate. I have to cross-reference the general location with Google Maps and then find a nearby image in the alt_locator (and even then, the landmarks shown in the alt_locator don't line up with Google Maps so it's difficult to find corresponding locations and gauge their proximity to dams and reservoirs). Long story short - I've been playing around with a couple of images that I think have mining activity and are close to dams and reservoirs. Still working on confirming this, though. Regarding searching by feature @lewismc, that would be amazing. I haven't forgotten about the clustering approach you mentioned last week with centroids and searching more efficiently than pixel-by-pixel. I'll bring it up in the tagup tomorrow so I can flesh it out enough to create an issue. |
Got it, thanks |
Hi @Lactem I think I ended up figuring out a better workflow.
This enabled me to look at flight lines in the AVIRISNG alt_locator which capture data over significant water bodies and then go over to QGIS and plug in the location name to see if there was any overlap with GRaND. I ended up getting the following results. Please start with the latter (ang20190623t220015). It is much smaller than the first one. |
@lewismc Thanks for this. I didn't see it until just now and I've already tried another 2 files (1 was in close proximity but didn't have mining activity after I classified it). I'll try these files out today. |
ack |
@lewismc The download link is not present from the map. I ended up finding it from the database view, but it's currently broken. I browsed around the FTP server and noticed that there's simply no data uploaded from 2019 (perhaps the plan is to release it at the end of the year?). Anyway, I found the 2017 data while browsing, so I'm currently classifying the larger image you mentioned with #191. So far I'm surprised because the RGB output is in black and white only, which makes me worry that the final classified product will be incorrect -- but I'll let it finish running and see what happens. This is not directly related, but I noticed that Google Fusion Tables is being taken down in less than a month (link). Just thought I'd throw that out there since it looks like the alt_locators depend on it. |
I emailed my colleagues Sarah Lundeen and Winston Olson-Duvall at JPL with the request for the data... we should have it soon.
Great
We'll the RGB available from the alt_locator is also greyscale so I wonder if the scene was messed up... there's only one way of finding out.
I've never even used fusion tables, I knew nothing about it. I'll let the AVIRIS team know when they respond with the scenes. Thanks |
Host: avng.jpl.nasa.gov We recommend connecting with a client like Filezilla or Cyberduck. You can also see these files in many browsers (like Chrome) using the linksftp://avng.jpl.nasa.gov/pub/LMcgibbney |
Thanks so much for running this scene, the results are unfortunate. Please try with the new scene I posted above. |
@lewismc I'll download and classify this tonight, thanks. |
Here's a thought... although this is bad for providing a use case, it is in reality a REALLY good thing.! |
@lewismc Haha that's a good point, it's good that it's so hard to find mining activity near reservoirs. I'll report any new findings as they arise. |
Any luck with those new scenes ? |
No luck just yet. While this is not specifically for coal mining, from my understanding the tailings from potash products are just as toxic to the environment. This whole paper is extremely long, and section 3.3.3.1 is what strikes me as the most useful to us:
Also section 3.4.3.1:
Also section 3.2.6.1:
From my research, halite makes up 89%-94% of tailings. Also, although kieserite makes up a low percentage of tailings (and only for potash as far as I know), it's the second-most common mineral. I've read elsewhere that muscovite is often observed in high levels in tailings. Also, one major caveat is that this research studied mines in European countries, whereas we're classifying data in the U.S. and Canada and maybe Mexico. So I'm not sure how the minerals differ by region. Another problem is that these are not spectroscopy analyses so the only mineral I can add to the PROXY_CLASS_NAMES for mining correlation is "Kieserite KIEDE1.a crse gr NIC4cc AREF." Final issue -- I realized that the mineral example classifies minerals with spectral library v7, but the mining example uses v6. This means that there could have actually been mining activity in some of the previous scenes, but they would've been missed in classification because the names of mining-indicative materials vary by version. For future tests I'll use v7 across the board. Another potentially useful source -- I encountered a number of research papers that claimed to have analyzed the spectral characteristics of coal mining or other mining. However, I don't have access to all of these. @lewismc if you have access to Taylor & Francis, this one paper in particular looks promising. Finally, this publicly available source from a USC student is a very well written case study that overlaps with Pycoal a lot. The paper says they used 3 sulfates to identify mining activity: Jarosite (KFe3+3(OH)6(SO4)2), Alunite (KAl3(SO4)2(OH)6), and Muscovite (KAl2(AlSi3O10)(F,OH)2). I should be able to search for these in the Spectral library and find their exact names. |
The following is very relevant
|
Hi @Lactem please rebase with master to resolve conflicts. |
@lewismc Please note that the failing tests seem to have been introduced from the master branch. I also updated the Spectral library version to 7 and will test the new proxy classes momentarily. |
Thanks @Lactem I think this is nearly good to merge.
That's fine @mjn0898 is fixing that in #201
Excellent. Did you ever manage to run over the Fort McMurray flight lines? Also, can you please add the GRanD shapefiles to the test directory. I can't see them included in the pull request. Thanks |
Just added them but then realized the reason I didn't add them earlier: the reservoirs shapefile is over 50MB. Would you prefer I keep them or remove them and add a download link in the examples README? If we do it the latter way then nosetests won't be able to verify proximity functionality.
I've been working with ang20170811t212940, which is part of that case study, but I haven't had a chance to run the other scenes yet. I'm still testing out the proxy classes and figuring out an issue with their spacing and encoding. |
Yes. Please use the following functionality to exclude https://stackoverflow.com/questions/26545668/setup-py-sdist-exclude-packages-in-subdirectory
OK great thanks
Got it. Yes these are a bit finicky but I appreciate you adding more... this will add significantly to detection capability. |
Will do. After more manual inspection I'm pretty sure there's no mining activity in the scene I've been using. I added a lot of proxy classes, so I'll try classifying a different scene from the Fort McMurray campaign. |
@Lactem please push a trivial update to this branch so that TravisCI is rerun. I would like to see all tests pass before we merge into master. Thanks |
Update: I didn't see any of the proxy minerals present in ang20170811t212940 or ang20170811t202652. I even tried manually going into QGIS and removing all minerals that weren't proxies from the mineral-classified image, in order to double-check that the mineral image really was blank after trimming it down to only the proxies (it was blank). I looked at the JPEG preview of ang20170811t210417 and see something that looks like heavy mining activity (I think), so I'm currently downloading and classifying this third scene to see if I can verify the proximity functionality.
I'll push very soon. |
Closes #148 but is still a WIP. It seems to be running without errors on my subset image, but I haven't yet found a reasonably-sized hyperspectral image with close proximity to dams and reservoirs. I'm still looking around on https://aviris.jpl.nasa.gov/alt_locator/, but it takes time to download the large files and then classify them. So far I've been manually importing the files into QGIS and then adding the dams and reservoirs vector layers to see if there's anything nearby. I'll let you know when I find a good example that can demonstrate that streams/lakes/rivers are working with dams/reservoirs on the same image.