jofer

In this case, it's not directly related. The inner core isn't what causes the Earth's magnetic field. It's convection within the liquid outer core that gives us a strong magnetic field. The inner core changing shape wouldn't necessarily cause changes in the Earth's magnetic field.

On a side note, https://www.science.org/content/article/scientists-probing-s... is a good pop sci article on the context.

The article kind of glosses over a key point about how all of this works and why "rotation" vs "shape changing" are difficult to distinguish. That's all because of anisotropy of seismic velocity in the inner core.

In other words, sound (seismic waves) travels much faster in one direction than the other through the inner core. That's true of most rocks to some degree, and it implies that the crystalline iron in the inner core is mostly aligned in a similar direction. But that is at the core (pun intended) of all of this.

So the "fast direction" has subtly changed over time based on the data we have. That's the "the Earth's inner core rotates differently than the rest" part. But we're mostly basing that on travel times in each direction (it's more complex than that - more in a bit). The differences "more fast stuff" and "less slower stuff" are hard to distinguish precisely, though they can be distinguished because of effects that occur at the boundary between different velocity + density bodies. It's also harder because the outer core is liquid and removes a key source of information coming from wave interactions at those boundaries (shear waves).

This is basically doing a lot of clever reprocessing of old data to carefully look data after corrections for the moderately-well-constrained rotation of the inner core. Rotation of the inner core can't explain all of the differences, so another thing that might cause it is changes in the shape of the boundary between the inner and outer core. It's also possible it's noise, though presumably the authors investigated that part carefully (haven't read the scientific article, but the primary author is a very well known person in the field, so the analysis is likely very sound).

There are always alternate explanations, though. Changes in shape on the order called for here do need an explanation via geological processes. Kilometer scale changes in a decade are difficult to immediately explain, though not impossible. I have no doubt the analysis is sound, but from a geologic perspective, this (and previous) work raises a lot of interesting questions.

Just as an FYI: "leaflet" is the name of a popular web mapping JS library. https://leafletjs.com/

May or may not be relevant for you (and is definitely a very different field/product), but if you're releasing client libraries/SDKs of any sort, it might be good to be aware of. There's a large ecosystem of plugin libraries named "leaflet-foo" or "foo-leaflet" etc in addition to the "main" one. If you start releasing any libraries to work with your app (even if they're not JS), you'll likely want to be aware of and work around naming collisions for library names.

Either way, looks nifty!! I love the approach and we need more people willing to do something like this that competes with google docs / etc, but does so by targeting a specific use case / niche / etc and not by trying to do everything.

I'd hesitate to call this a "cookbook". It's closer to a coffee-table photography book. It's somewhat more art than instruction, though it is both. It's inspiring, for sure, but is very different than most.

I also thought it had been out of print for years, but it seems I'm wrong... Perhaps this is news because it's no longer out of print? Or was I just misremembering?

It's pretty typical, actually. 50% is about the minimum that major universities take out of a grant you get as a researcher at the university.

It's nominally to fund general facilities, etc. At least at public universities, it does wind up indirectly supporting departments that get less grant money or (more commonly) just general overhead/funds. However, it's not explicitly for that. It's just that universities take at least half of any grant you get. There's a reason large research programs are pushed for at both private and public universities. They do bring in a lot of cash that can go to a lot of things.

This also factors a lot into postdocs vs grad students. In addition to the ~50% that the university takes, you then need to pay your grad student's tuition out of the grant. At some universities, that will be the full, out-of-state/unsupported rate. At others, it will be the minimum in-state rate. Then you also pay a grad student's (meager) salary out of the grant. However, for a post-doc, you only pay their (less meager, but still not great) salary. So you get a lot more bang for your buck out of post-docs than grad students, for better or worse. This has led to ~10 years of post-doc positions being pretty typical post PhD in a lot of fields.

With all that said, I know it sounds "greedy", but universities really do provide a lot that it's reasonable to take large portions of grants for. ~50% has always seemed high to me, but I do feel that the institution and facilities really provide value. E.g. things like "oh, hey, my fancy instrument needs a chilled water supply and the university has that in-place", as well as less tangible things like "large concentration of unique skillsets". I'm not sure it justifies 50% grant overhead, but before folks get out their pitchforks, universities really do provide a lot of value for that percentage of grant money they're taking.

See also: https://en.wikipedia.org/wiki/Australasian_strewnfield (debris over 10% to 30% of the Earth's surface from an impact ~788,000 years ago). Smaller, for sure, but very recent and large enough for "nuclear winter" type scenarios.

Yeah, ammonia leaks are much more nasty than methane or hydrogen leaks. Methane, especially in LNG form, is quite safe compared to ammonia. LPG is even more stable than LNG and requires lower pressures. With that said, hydrogen leaks are "fun" because large ones usually self ignite and burn with a hot but mostly invisible flame. But hydrogen itself isn't toxic. Similarly, methane and propane aren't directly toxic.

Basically, an ammonia leak will kill you. By itself. The others are only a problem if they're the right concentrations to ignite. That's a relatively high concentration and a larger leak. Much smaller leaks of ammonia are deadly.

It's still a good solution for some things, but it's a bad solution for consumer vehicles like cars for that reason.

Less advanced things have been labeled a national security risk.

It's currently quasi-illegal in the US to open source tooling that can be used to rapidly label and train a CNN on satellite imagery. That's export controlled due to some recent-ish changes. The defense world thinks about national security in a much broader sense than the tech world.

See https://www.federalregister.gov/documents/2020/01/06/2019-27...

This x1000. It's not about the org, really. It's about what's being changed.

Yes, a lot of large codebases have things that look wrong. They're not always wrong. Trying to clean up what look like cobwebs at the core of a large codebases often means removing things that are there for a reason. The reason is usually counterintuitive and could be documented better. But often only a few people can really evaluate the change.

A small PR is likely to be accepted quickly, and a large PR is likely to take awhile. No one gets upset by that, though.

The flip side is that a small PR to a critical part of the codebase is also likely to take awhile and be treated as "default to no". It's often hard to see that from the "outside", though.

With that said, trying to go the extra mile and make things clear, concise, and better documented than before goes a long way in getting an MR reviewed quickly.

Not all PhDs push boundaries of knowledge in quite that way. Often what you do was already known, but no one really figured out how to correctly apply it. Alternatively, it's often combining different things that were already well known, but no one had combined.

It's novel work that makes a PhD. Novel work is distinct from pushing the boundary of knowledge. Often what you do doesn't change what "humanity knows" in any way.

Both my wife and I have PhDs. Neither of us did things that look like the figures there. It's not a good mental model for what all PhDs mean, though it is a good model for some.

I combined fields and reinterpreted a ton of things that had already been done to draw very different fundamental conclusions about what was going on in a particular location. I put out alternative hypotheses for observations that had already been collected. It's not new knowledge at all and I didn't add to what we "know". I just added an additional hypothesis to the set of multiple working hypotheses that will hopefully be tested decades from now.

My wife worked on how to actually apply well-known methods in other fields to our field. Her work was half engineering, half field experiments. Lots of folks had been working with fiber optic strain gauges for decades. However, no one was using them to measure in-situ strain in rock masses yet (which has since become common). The application was broadly "known", but actually doing it and demonstrating that is novel.

My wife rather derisively calls this the "penis pimple model of science". It's a great article and a great set figures, but folks sometimes push it a bit too much, especially in grad school.

It has a lot of truth to it and it's been making the rounds for well over a decade. Unfortunately, sometimes it can set folks up to feel bad about themselves if what they do doesn't line up well with whatever is in vogue as the boundary of knowledge that's currently being pushed.

There's a _ton_ of value in more pragmatic parts of fields that focus on applications or combining relatively well-know parts of different sub-fields. Those parts of science often don't feel like you're pushing some "boundary". It's more layering on top of, filling in holes, and building up than building "out". Sometimes what you do is to use multiple things that lots of folks already knew, but the people who knew X and the people who knew Y didn't talk to each other, so no one thought about how to combine them. It can also be hard to get papers published because reviewers will consider one part obvious/well-known and the other part irrelevant because they come from one sub-field and not both.

It will often feel like you don't belong because your work doesn't look like these figures. Applied and interdisciplinary fields "feel" different. However, this type of integrative work can be among the most valuable parts of modern research.

Don't feel like what you do has to fit into the "penis pimple model" of science. There's also nothing wrong with pushing some known boundary of a field, either! Both are valid.

For folks asking what it does in general, see https://en.wikipedia.org/wiki/Image_registration

This method is meant for working with multiple images where something has significantly changed in a non-linear way. E.g. think of trying to produce a deformation map of clay that you've squeezed by taking lots of photographs during squeezing. You can also use it in the reverse sense and align multiple images where things are changing in complex ways. The fact that inverts for / is constrained by a continuous velocity field has some nice advantages compared to other methods. Or that's my bad summary of it, anyway.

As someone who uses this general class of methods pretty frequently (albeit in satellite imagery), I'm very curious to dig into this! Looks like a very understandable set of implementations!

It's usually the exact opposite for this sort of thing. You can't do this with natural language. Traditional computer vision is well suited to it and works with some tweaks. "Modern" techniques for it require collecting insane amounts of training data for simple things. You can't just throw transfer learning at this because it's a lot different than standard photographs that models are trained on. The old school methods are faster and more reliable for a significant number of problems in the geospatial world. And you still need a lot of deep expertise no matter what.

It's not that, it's breathlessly proclaiming that techniques that have been standards for decades are "groundbreaking AI". The hyperbole makes it impossible to get at anything, and if you accurately propose a time tested solution at work these days, it gets dismissed because it's "not AI". So now standard computer vision methods that aren't AI in any way are getting proclaimed as "AI". It's quite annoying, as least from the perspective of someone who does more or less this exact thing (geospatial analysis and data processing of various types) for a living.

Folks won't let you use the right tool for the job anymore unless you make wildly hyperbolic claims about how groundbreaking it is and claim it's cutting edge AI.

The situation is bad for everyone. There's nothing wrong with using the right tool for the job and accurately describing it. I'm tired of having to inaccurately describe methods to be allowed to use them. E.g. claiming a Hough transform is "deep learning" so folks won't immediately dismiss it and demand I use some completely incorrect approach to a simple problem.

This is super useful, but it's a bit disappointing to see map digitization called "AI".

I mean, sure, these are methods broadly in the computer vision realm and that gets referred to as "AI" sometimes. But at the end of the day, this is "find all unfilled black circles of a specified diameter on these images". It's amenable to (and has been done by) traditional computer vision methods for a long time. There are certainly a lot of cases where a CNN type approach can perform better than traditional computer vision and there are always improvements to make.

However, I think it's a bit odd to treat this type of use case as some sort of AI breakthrough that wasn't possible or wasn't frequently done in the past.

Why can't normal standard work have a press release? Why do we need to play pretend and add buzzwords just to make things sound "cool"?

...But that's just me being a bit bitter, perhaps...

Klong is a different language than Python. Python using numpy would look broadly similar for the first few examples, yes. However, this is a way of executing a different language on numpy arrays. Array languages are a different paradigm. Klong is an array language, while Python + numpy allows some array paradigms, but isn't an array language.

Notice how they're _defining_ the "sum" operation there. Instead of being something builtin, they defined "sum" as {+/x}.

{+/x} is the interesting part. That's Klong. It's not being able to define a "sum" operation, it's that "sum" can be expressed as {+/x}. That's very different than both R and python.

First off, major kudos and this is very very cool work. It's also a great article and really, everyone should go read it.

Beyond "just" better string arrays, my favorite side effect of this is efficient NaN support in string arrays. The article talks about this a lot, but I had already started this comment before fully reading the article :p

I mean, sure, the old approach was object arrays, and you can do it there because each element is an independent object, but they're super inefficient. This both makes things efficient _and_ has a really cool side effect of supporting something that had become common partly as an accident of the old object array approach - NaNs in arrays of strings.

This is really really really useful work and it's _super_ cool!!

Yes, but that's among the most expensive ways of approaching the problem. In regions where it's feasible, evaporation approaches are far cheaper, which is why they're still the most widely used approach. In humid regions with a lot of precipitation like the gulf coast, though, you have to take more expensive and energy-intensive approaches to concentrate things.

Those are expensive and are avoided when possible. Brine ponds are cheap if you can use them. But with that said, yeah, evaporation ponds don't work especially well on the Gulf Coast.

There are a lot of things in deep subsurface brine. It really varies.

First and foremost, here are definitely lots of other salts. It is brine, after all. You produce a lot of halite (salt), gypsum, calcite, and all kinds of other evaporite minerals.

There are all kinds of things in smaller concentrations, though.

What comes out of a oil/water separator would need lots of additional processing before going to something like an evap pond. It's relatively hazardous stuff for a lot of reasons other than oil (e.g. it can be rather radioactive). It typically goes through quite a bit of additional processing unless it's being immediately reinjected.

You physically can't remove the overburden for this. The Smackover is at a depth of multiple kilometers in most of these areas.

It's mining brine. I.e. the "mines" are basically deep water wells.

The limestone itself doesn't have any lithium. It's the water in the pores in the limestone that is relatively concentrated in lithium.

In most of these cases, you're already producing brines from the smackover formation as a part of existing oil and gas production, but the brine is being re-injecting after oil is separated from it. The idea is that it's better to keep those and evaporate them down for lithium production.

That does require large evaporation ponds, generally speaking, but it's not strip mining.

There wouldn't be any core for this. It would be a holdout of the brine samples used in training. The thing that would be being produced is brine, so lithium concentrations in brine samples are the validation dataset as well. In other words, this is spatial interpolation.

Put another way, this is pretty similar to the interpolation approaches that would normally be used for datasets like this in the world of mineral exploration. Kriging/co-kriging (i.e. gaussian processes) is the more commonly used approach in this particular field due to both the long history and the available hyperparameters for things like spatial aniostropy.

However, kriging is really quite difficult to use with non-continuous inputs. RF is a lot more forgiving there. You don't need to develop a covariance model for discrete values (or a covariance model for how the different inputs relate, either).

I am very happy to see that the article _immediately_ explains how this is different from long-existing np.float128/np.longdouble.

I was scratching my head for a bit there... It's also good to see that custom dtypes have come a long way. This is a great application of a custom dtype, and that wouldn't have been possible a few years back (well, okay, a "few" to me is probably like 8 years, but still).

Edit: Apparently that's the airborne equivalent of Tanager, not Tanager. (Same instrument design, but one is on a plane and one just launched into space not-too-long-ago.)

That's around flaring, which is a bit different. Energy companies are very likely to buy the same data. Detecting methane leaks is a _good_ thing for them, both from an "avoiding fines" perspective and also from a "this is infrastructure we _want_ to fix" perspective.

Banning routine flaring is a very good thing that needs to happen in more places. You _do_ still need to flare. There are lots of time periods where it will be required for safety reasons. But currently, it's common to simply flare methane that's produced instead of trying to use it. Methane can't be easily transported, and you need a pipeline to a populated area to use it unless you build expensive LNG facilities or slightly less expensive facilities to reinject it back into the subsurface. So remote oil fields are designed to flare off the methane that's produced alongside oil production, often for vast quantities of methane. That's "routine flaring". It's better (both from a safety perspective and a greenhouse gas perspective) than directly releasing it. However, it's far better to reinject it back into the reservoir (or another reservoir) or otherwise find some use for it than to flare it.

Routine flaring is used quite simply because regulators allow it. If you change the regulations, then companies will take the more expensive route or develop other resources. If you don't, then they're more or less legally required (read: shareholders _will_ have grounds to dismiss the CEO) to take the legal and much cheaper route of flaring methane that can't easily be sold. Can you really justify to shareholders that you're going to spend an extra several tens of billions USD to do something that isn't required and that your competitors aren't and that won't increase profits at all? The regulatory environment has to change for that to happen, but it's a patchwork and not some global thing. The EU has been leading there.

But detecting flares (even "hidden" ones) is _much_ easier than detecting methane leaks. Methane leaks are pretty damned insidious and hard to find. That's a big part of why they're so common. Hyperspectral imaging is _really_ damned cool, and while I'm certainly biased, the Tanager satellite they used there is really really neat.

Yeah, but it has to fit down a well. In other words, ideally it would be a maximum of 6 inches in diameter (and significantly less if it's going to be widely useful for solution-mined salt caverns).

The "drilled shaft" is the only entrance, and while those are typically huge by well standards (solution mining needs to circulate water, so you have to have room for nested casings/pipes), they're still dramatically smaller than what's described here. You typically have a "hanging string", which is a suspended pipe within the pipe for injecting water with flow around the outside of it back up the well (either to extract gas or for solution mining). You want something that fits down the inside of that so you don't need to pull the hanging string for a sonar survey. It's a big pipe, but you're not going to fit a literal mini-blimp that's a couple feet long down it.

With that said, I do wonder if these have other mining applications. It's not good for solution mining, but it sounds like it could be useful for remote inspection of dangerous levels of conventional underground mines. Conventional mine shafts and levels have to be big enough for people and equipment to fit in, and this seems reasonably sized for that application.

Honestly, I'd recommend some of the NA beers that are being brewed these days over the hop sodas / hop waters. Hop water has no balance, and hops are all about being balanced out with something sweet/malty. NA beer has come a _long_ way from the days of O'doul's. Sierra Nevada Trail Pass is a personal favorite of mine when it comes to good, hoppy, non-alcoholic beers. Lots of classic "C" hop flavors (i.e. somewhat piney hop varieties from the NW US).

But if you don't like beer and just want to get a sense of what hops are about, look at some of the options from Hoplark or similar. Those are single-hop hop waters that focus on showcasing the flavor of a specify variety. They're a good sense of the flavors, but I still find hop waters in general to be unbalanced and a bit hard to drink.

I finally live somewhere where it's possible to grow hops (albeit it not ideal), and I _really_ want to experience brewing with hops straight off the vine! I've always been jealous of y'all in the Pacific NW in that regard (hard to beat it as a beer region). Fresh hops are amazing.