A late hello and introduction!

[Autobrecciation]

I keep intending to write a hello thread, but I'm really a lurker at heart. So about me:

• Grad student / Teaching Assistant - M.S. Geographic Information Science - thesis work is on applications of Neural Network classification and super resolution image processing, looking at impact of agriculture and wildfire on ground bird populations in the sagebrush steppe of ID / MT / WY. Did my undergrad in Geology, so I know something about rocks and earth processes (Not a ton though, I decided to specialize in the maps and computers side of Earth sciences).
• First Brandon Sanderson book was Mistborn 1. Wanted to read it before The Gathering Storm so I'd know what to expect. Like many of you I was hooked.
• Big reader growing up, got busy with school so read maybe 1-2 books a year for several years. This year I got a kindle, and with the lockdown I've had more free time to read. Did my complete Cosmere reread from May-August. Just finished Dawnshard this weekend.
• Favorite Cosmere series - Stormlight, followed closely by Mistborn Era 2.
• Favorite non-Cosmere series - this is a tougher question so I'm going to write it out as more of a discussion. The Wheel of Time is kind of foundational to my reading, but just due to massive amount of rereads of the series over the years it was coming out, I don't know if I could read it again, so maybe its not my favorite anymore? I know book 3 is never coming out, but The Kingkiller Chronicles I've read twice, still feel like I could read again. This year I read The Kings of the Wyld and I loved it. It bought me back to weekends playing D&D with my friends, some of the absurd things we got up to. The Expanse novels I read mostly (First 4) before this year at the slow rate, and I think the TV adaptation has been phenomenal. I hope that the WoT show is as good. Plus the last of the 9 book series comes out next year.
• What am I here for? Mostly to continue lurking. With the release of Dawnshard and Rhythm of War I'll probably be browsing discussion threads, maybe commenting a bit. Felt like I should put something in the introduce yourself section as a reference and a courtesy before commenting.

So hello to all! Ask me about cool rocks or fossils you found and I'll try to tell you something interesting about them.

[Chasmgoat]

HI, welcome to the shard! um... rocks? I found some rocks in Kansas with dozens of seashells (maybe they were seashell fossils) covering the edges.

What allomantic metal would you want to burn?

[Autobrecciation]

Good question. I think I'd love the freedom of pseudo-flight via steelpushing.

[I Am A Fish]

Hello, and Welcome to the Shard!

Who's your favorite Sanderson character?

[Scout_Fox]

Welcome to the Shard Auto! What a thorough introduction so happy to have you here with us! 

Although I don't think I fully understand what you are studying that sounds like a remarkably interesting project! In what ways are you using Neural Network classification and super resolution image processing to aid in that? Has there been anything completely surprising to come out of it yet?

Getting a Kindle has been one of the best things for SA books in general (my arms still hurt ) and quarantine good choice! Sounds like you are right on track with that Cosmere and Dawnshard read in prep for RoW well done! I would have to agree also on that SA favorite as well. Every time I read them there's just something special 

What's the coolest rock fact you know? And I'm guessing your favorite character would be Rock? And although lurking is entirely fine I'd say it would be the perfect time to hop into some of the discussion threads if you wanted!

 

[Autobrecciation]

Favorite character - I really love Steris. Every character I've ever played in D&D is like some insanely over-prepared because half the fun of character creation is buying gear! Think like, if you had Steris, but with Wayne's tendency to half plan ridiculous scenarios and absurd disguises.

Um, cool rock fact. So there is this layer between the earths crust and the mantle, called the moho. There are places in the world where there has been so much deformation that the moho (And parts of the mantle) have been thrust up, and are now exposed on the Earth's surface. If you ever see a green rock with red speckles (Eclogite), you can think to yourself, that might have formed in the mantle. It's not 100% but its a good guess.

As far as my thesis work... we have this bird that has declined in population over the last 100 or more years in the western US, except in this small pocket in Idaho, where its successful enough that its still considered a game bird. Old scientific papers from the 50s lay out that the decline in the bird population is due to increased agriculture, but don't present evidence other than the declining population. Intuitively, it makes sense that if you remove the habitat it should adversely affect the bird population. My predecessor used the National Land Cover Data set to examine the change in land cover, and correlate that with the population statistics, but the results were inconclusive (no correlation between agriculture increase from 2001 to 2018 and the bird counts for those years). Their conclusions is that there isn't enough agriculture change in the last 17 years to really capture its real correlation with the population.

So I will be using neural network to classify satellite imagery, which will allow us to look at a longer time scale. My hypothesis coming in to my masters is that it isn't agriculture (alone), but wildfire that is decimating the birds habitat and thus the population.

So super resolution processing is essentially when people on TV say "Enhance" and make images better. Turns out this is possible with a neural network and good training data. So the second half of my research was planning and flying a drone to take pictures over some bird habitat that had burned a few years ago. I will be looking at the imagery collected there to see how the vegetation is recovering. Since I have the data to train a neural network, we want to see how well we can increase the resolution of satellite imagery for the area. Hypothetically we could use this process to examine the area using satellite data back before a wildfire, to get better resolution images essentially back in time. This is useful because satellites are generally collecting data over years, but old satellite imagery has relatively poor resolution. Landsat (the series) has had 30 meter pixels since the 70s, but Sentinel-2 has 10-20 meter pixels which is pretty good. The drone data I collected has ~7 centimeter (0.07 meter) pixels.

Sorry for the essay there.

TLDR; Over 100 years, the local bird population has gone down, so we ask why? Some folks think farming, I think fires (and maybe farming too). The plan is to use a computer to figure out whats in old pictures (1980's) and compare to new pictures (2020), then compare the change (In farms and fires) to bird population and see how well it matches. Finally, I will use drone pictures matched to satellite pictures to train a computer, then the computer will use old satellite picture to make fake drone picture, and see how that matches to reality.

[Honorless]

The Doors of Stone will release when the Aleu fall nameless from the sky, friend. Welcome to the Shard! 

[Koloss17]

Hey @Autobrecciation! If you were to establish a university that has a course on the study of magical creatures and where to find them, but in Boy Scout form, which world of the sandoverse would you choose?

[Autobrecciation]

Drominad system. Dusk as the reluctant, crotchety scout master with a heart of gold and wisdom.

[The Storming Stormfather]

Welcome to the Shard, Autobrecciation! Which character (non-Mistborn series) do you think would have the best chance to take down the Lord Ruler?

[Autobrecciation]

Just take him out? Chiri-Chiri. Bring down / co-opt his church and raise himself to God-Emperor status? Taravangian.

[Scout_Fox]

@Autobrecciation wow... THAT... was something.

Ok first on the rock fact: Moho is actually a sick name lol. Dang I didn't even know that could happen! Is it bad that it has been thrust up or is that just something that happens? Green rock red speckles, alrighty well if I ever do see one I'll be sure to ping you 

Ok now on your thesis: Huhhhh that's interesting that they could come to that conclusion if the evidence for it really can't be proved. Seems mighty convenient to blame it on agriculture cuz no one wants to not eat! And that's actually really cool that you are continuing this research!

Ohhh interesting, interesting. So wildfire naturally or human caused (accidents and global warming)? And neural network is like machine learning right with the layers of nodes and then they like process data and send it to the next layer and eventually come to result?

Yknow honestly thought this wasn't a thing that's super interesting. I just saw an ad for this too with Adobe I think it was advertising that with photoshop. So you'll be doing essentially the same thing right taking satellite data and trying to upscale it so you can make comparisons? Ohhh and the drone is to get like "base" accurate images and then you will run the satellite images through along with this to train the neural network to upscale them to that fashion and then apply that to new stuff? So it is machine learning right just teaching it to upscale using drone images as reference. That's actually really smart that sounds awesome!! 

I never knew that about the different satellite class resolutions too! Makes complete sense though. That whole looking back in time is awesome too I so hope that works! So basically try to get these images upscaled, then compare what the land looks like in 1980 and 2020 to see how much fires have affected it and then compare that to the bird population right? Neat stuff!

No problem for the essay lol super interesting to look at. So considering this does work what is the final goal? Like use this to advance forest management or as data on the result of global warming induced fires on bird populations? This sounds awesome! PM in future on how it's going I'm super interested in the results now 

[Autobrecciation]

Quote

Ok first on the rock fact: Moho is actually a sick name lol. Dang I didn't even know that could happen! Is it bad that it has been thrust up or is that just something that happens?

Don't think its bad, since it probably happens over millions of years. Just an interesting thing. And I think its pretty rare, but I'm sure its less rare than I think it is. Basically when you have subduction, part of the plate thats getting covered is also getting scraped, warped, and broken. Sometimes that brings up rocks from the mantle, which is cool. Kind of like like a snow plow picking up dirt from cracks in your driveway, even if its going over those cracks.

Quote

Huhhhh that's interesting that they could come to that conclusion if the evidence for it really can't be proved.

I think of it as just old timey scientists making assumptions. We can do better now.

Quote

So wildfire naturally or human caused (accidents and global warming)?

Wildfire is such a complex thing that I've only scratched the surface on. The main (direct) cause of the fires is generally lightning strikes, and with hotter dry periods, there is more dry fuel available; both of which are associated with patterns of extreme weather variability rooted in climate change. In the historic fire database I'm using there have been 831 fires since 1939 for my area. 218 are from before 1984, and the rest are after, so we see an increase in fire frequency. The cause of this (I think) that when an area burns it destroys the built up vegetation, and we get invasive grass (cheatgrass if you've heard of it) that grow quickly, out-competing more draught resilient native grasses. The cheatgrass also dies off early after spring, and tend to catch fire easier, and help the spread the fires between islands of fuel sources. So you have these areas of sagebrush, juniper, low brush, etc that get converted to cheatgrass, which creates a feedback loop.

Quote

And neural network is like machine learning right with the layers of nodes and then they like process data and send it to the next layer and eventually come to result?

Yeah, neural networks are just fancy machine learning algorithms. You create an architecture for the layers, feed in training and validation data, the computer optimizes the output based on making correct predictions (Like, lowest mean squared error, or low loss rates between iterations). Then you feed in some data that you know the answer to see how well it did (Kind of seperate validation). If you are happy with the results, then you can feed in data you want to make predictions with. The one I'm using is good for image processing, its called a "Convolutional Neural Network". They are supposed to be good at breaking down an image into constituent parts - recognizing edges and curves, which is good for object classification within an image. They also do some randomized selection of the input data, in an effort to make a more generalized model (Basically to make predictions accurately for data that is different from what the model was trained with).

Quote

So it is machine learning right just teaching it to upscale using drone images as reference. That's actually really smart that sounds awesome!!

It is. The thing that hopefully makes it useful is that the drone sensor has matching bands/colors to the satellite imagery, which is useful because the colors can tell us about plant health which has applications to agriculture and food security.

Quote

So considering this does work what is the final goal? Like use this to advance forest management or as data on the result of global warming induced fires on bird populations?

So one of my goals for the study was to look at wildfire recovery, and quantify how well fires recovered based on land management after the fact. So potentially Forest Service or USGS has an area that burned, one group had done like, drill seeding to restore native veg, one did seed drops from like a plane or something, you could tell the difference, then start doing the better treatment. As for the birds, since they are still (relatively) successful in this area, but have declined all over the west, if we can isolate whats different here, we could attempt to re-introduce them in areas that they are more likely to thrive in. Since they are popular game bird, that can (eventually) drive some of the local economy.

On a personal note, my goal coming into my masters was to learn as many tools as I could, to be versatile and have a useful toolbox. I wanted to learn some programming and data science, to learn to plan and pilot UAS missions. Right now all my data is collected, and I'm getting ready to write my thesis between now and the spring. Hopefully I will graduate in May, do some work helping new students plan and fly missions over the summer while I look for a job. I picked a good and bad time to graduate tho, because once the COVID vaccine is out I'm sure everyone is going to be out looking.

[Scout_Fox]

1 hour ago, Autobrecciation said:

Don't think its bad, since it probably happens over millions of years. Just an interesting thing. And I think its pretty rare, but I'm sure its less rare than I think it is. 

Oh yeah ok so just something happening over time cool. And yeah that does sounds quite rare and likely really interesting to get to see!

6 hours ago, Autobrecciation said:

Wildfire is such a complex thing that I've only scratched the surface on. The main (direct) cause of the fires is generally lightning strikes, and with hotter dry periods, there is more dry fuel available; both of which are associated with patterns of extreme weather variability rooted in climate change. In the historic fire database I'm using there have been 831 fires since 1939 for my area. 218 are from before 1984, and the rest are after, so we see an increase in fire frequency.

Yeah that makes compete sense there's probably a million factors that go into it. Oh and interesting so the lighting is igniting it but climate change causing drier climates creating greater amounts of dry fuel for the fire is exacerbating this. Oh and no way wow that is a massive increase! Yeah there's some proof right there dang!

6 hours ago, Autobrecciation said:

You create an architecture for the layers, feed in training and validation data, the computer optimizes the output based on making correct predictions (Like, lowest mean squared error, or low loss rates between iterations). The one I'm using is good for image processing, its called a "Convolutional Neural Network". They also do some randomized selection of the input data, in an effort to make a more generalized model

That's honestly so crazy that we can do that now. It makes complete sense, just the fact that it is actually possible is awesome! Interesting as well there is already a sort of guide for what works best for the different test scenarios as well. That separation of the images is exactly what you need right so it'll define the separate parts. Ohhh and the randomization makes total sense cool that is awesome! Yeah that way you can apply the method you developed to very different data sets cool stuff!

6 hours ago, Autobrecciation said:

It is. The thing that hopefully makes it useful is that the drone sensor has matching bands/colors to the satellite imagery, which is useful because the colors can tell us about plant health which has applications to agriculture and food security.

That's mighty convenient. I honestly woulda been less surprised if you said it had opposite colors or something just because when is anything ever easy like that  That's awesome though for that color recognition because then you can easily label and analyze the images!

6 hours ago, Autobrecciation said:

So potentially Forest Service or USGS has an area that burned, one group had done like, drill seeding to restore native veg, one did seed drops from like a plane or something, you could tell the difference, then start doing the better treatment. As for the birds, since they are still (relatively) successful in this area, but have declined all over the west, if we can isolate whats different here, we could attempt to re-introduce them in areas that they are more likely to thrive in. 

That would be so awesome actually! Like indirectly studying something just as important so they could make even more informed decisions about which works the best for restoring the landscape. And yeah if you could find that factor that's creating that difference it would most definitely be like saving the species there in a way!

7 hours ago, Autobrecciation said:

On a personal note, my goal coming into my masters was to learn as many tools as I could, to be versatile and have a useful toolbox. Right now all my data is collected, and I'm getting ready to write my thesis between now and the spring. I picked a good and bad time to graduate tho, because once the COVID vaccine is out I'm sure everyone is going to be out looking.

Honestly that sounds like exactly where you want to be! Positioning yourself with all of those skills and the firsthand experience that came with them will make you all of the more qualifed for wherever you go from there. Good luck finishing it off as well it sounds like a really awesome project, both the skills it has taught you and the work that you are doing has applications way beyond just this as well. And oof, true yeah but hopefully it works out for you as good as it can be!