Mapping "Every Inch [or cm] of the Seafloor"

[Astronuc]

I came across an article in Scientific American while browsing the news app on my smartphone. The article was originally published with the title "Every Inch of the Seafloor" in Scientific American 327, 2, 40-47 (August 2022)
doi:10.1038/scientificamerican0822-40

Summary: https://www.scientificamerican.com/...-is-finding-surprising-structures-everywhere/

Oceanographers are fond of saying that we know more about the moon's surface than we do about Earth's seafloor. It's true. As of 2017, only 6 percent of the global seabed had been mapped, typically by ships with sonar instruments sailing back and forth in straight lines across a local section of sea.

But since then, nations have become eager to chart the seafloor within their own “exclusive economic zones,” which reach 200 nautical miles from their shores, in part to look for critical minerals they can scrape up using big mining machines. The other push is Seabed 2030—an effort to map Earth's entire seafloor by 2030, run jointly by the Nippon Foundation and the nonprofit General Bathymetric Chart of the Oceans.

The goal is to collect and stitch together mapping done by governments, industries and research institutions everywhere. Public release of previously private bathymetric data is helping to widen the areas plotted. And uncrewed, remotely operated vehicles fitted with sonar that can zoom around underwater for days at a time are speeding the pace of mapping. By June 2022 an impressive 21 percent of the world's seafloor had been charted. The more experts map, the more surprises they find—such as the three unexpected, unusual formations revealed here.

Hopefully someone will find Malaysia Air flight MH370

 

[Borg]

Hopefully someone will find Malaysia Air flight MH370

I've been reading the article and that was my first thought as well.

 

[berkeman]

I wonder how they plan to handle the sensitive/secret military stuff that they reveal. The submarine listening stations and communication cables come to mind (SOSUS and follow-on versions)...

https://en.wikipedia.org/wiki/SOSUS

 

[Vanadium 50]

Every inch seems to be a bit of an exaggeration, since there are 10¹⁷ of them. A million sensors, one measurement per second, and it's still centuries. But if you could, I am sure you would find all sorts of interesting things - planes, shipwrecks, craters, maybe more.

 

[Astronuc]

I think the resolution would be on the order of m or 10s of m, and perhaps that depends on depth. Maybe cms or dms in shallow water.

 

[anorlunda]

I wonder how they plan to handle the sensitive/secret military stuff that they reveal.

Like the one on Mars.

Maybe they'll find Jimmy Hoffa.

 

[anorlunda]

Perhaps the project will pay for itself in sunken treasures found.

 

[Baluncore]

There are things out there that will surprise us.
But there are also long-lost mysteries that may be solved.
https://en.wikipedia.org/wiki/København_(ship)

 

[Baluncore]

Every inch seems to be a bit of an exaggeration, since there are 1017 of them. A million sensors, one measurement per second, and it's still centuries.

There is a seismic technique that employs a surface scatter of geophones, recording naturally generated sounds, from which a 3D geology can be extracted.

Consider placing multiple sensor buoys on the ocean surface, each recording, with GPS position, precision synchronised clock, and a satellite uplink. Random animal or ship sounds, traveling over multiple paths, recorded by multiple buoys, can then be de-convoluted back to the original source event. The seafloor, texture, and any "embedded scatterers" can then be mapped in parallel, since the image is generated in transform space. The technique will also reveal 3D ocean density and current information. That makes the mapping of all the oceans a ten-year project, not a ten-century project.

 

[olivermsun]

There is a seismic technique that employs a surface scatter of geophones, recording naturally generated sounds, from which a 3D geology can be extracted.

Consider placing multiple sensor buoys on the ocean surface, each recording, with GPS position, precision synchronised clock, and a satellite uplink. Random animal or ship sounds, traveling over multiple paths, recorded by multiple buoys, can then be de-convoluted back to the original source event. The seafloor, texture, and any "embedded scatterers" can then be mapped in parallel, since the image is generated in transform space. The technique will also reveal 3D ocean density and current information. That makes the mapping of all the oceans a ten-year project, not a ten-century project.

This sounds difficult in practice. Buoy depth, 3D ocean density, and current information all fluctuate in time. Is there a concept for inverting for this many unknowns at once with the desired accuracy?

At a coarse resolution (~100 m to 1 km), global bathymetric mapping has been/is being done via satellite altimetry. The goal with multibeam mapping is to "see" much finer structures than that.

 

[Baluncore]

Is there a practical way to inverting for this many variables at once?

The trick is to not do it all at once, but in a series of stages, each extracting higher spatial frequencies from the data.

The advantage of a whale is that it produces a long chirp. That gives the differential range to all reflectors in a single record. The image construction capabilities advance with advances in computer processing. A single 3D seismic model is built up from multiple shots, or stimuli from different places, while the geophones remain fixed. Since the buoys will be drifting along a path, that is recorded with precision, the data can be applied to the model from the known location of the buoy. That is a very similar process to building up a record from airborne side-scan radar.

If you drop a grid on the oceans and scan those lines you will find most things by the end of the survey. But what is wanted is a map of the anomalies, the important differences. Big things like seamounts and abyssal plains will be mapped initially. Then anything on an abyssal plane will be seen by reflection vectors without any need to raster scan the entire area. Areas that require closer inspection would get more buoys at a later date, or an intensive ship based survey would be mounted.

As the framework of the model is constructed, new data is added, while the old records are reprocessed to extract more accurate and detailed information that was deliberately ignored in the earlier stages of construction. The spatial frequency of the model will rise, while the noise floor falls.

 

[olivermsun]

This sounds like a neat concept, but I wonder: how many drifting buoys do you envision needing over a 10-year-period?

 

[Baluncore]

This sounds like a neat concept, but I wonder: how many drifting buoys do you envision needing over a 10-year-period?

I would start the experiment with about 250, one dropped every 1 km on a line crossing the ocean current. That initially sweeps a 250 km wide field as the buoys disperse.

Since ocean currents circulate, while surface winds confuse things, the number needed will be the sum of the radius of the oceans, in km. The buoys may be distributed and possibly collected later by out-of-work fishing boats. GPS targeted collection, becomes a traveling salesman problem, where the buoy density will be the greatest.

It would not surprise me if the later funding and computer processing came from a military budget, because the first pass of the data would reveal valuable dynamic data, that might be selectively redacted by the funding agency.

 

[Vanadium 50]

Consider placing multiple sensor buoys on the ocean surface

Sure. Let's consider the data volume.

We want 10¹⁷ outputs. The job is essentially inverting a 10¹⁷ x 10¹⁷ matrix, but of course actually doing it this way would be silly.

The average ocean depth is about 4 km, so the relevant surface area is at least twice that, probably more. A circle of radius 8km has ~10¹² points. So we really have not a 10¹⁷ p`robl`em, but 100,00010¹² problems. That's a lot better.

Inverting a matrix depends on many things, especially how sparse the matrix is. I am going to use $O(n^{2.3})$ as the matrix inversion time, which is probably optimistic. I am also going to assume this takes only one opp, which is crazy optimistic. That gets me to 4 x 10³² ops, or on a hypothetical exaflop machine, millions of years. And of course you need to do this more than once to avoid transients like you mention.

I'm afraid this is a case where "we'll just take care of it in software" doesn't work.

Now, I kind of already knew the answer. This basic idea is being done with seismic information collected by buys, funded by the Japanese government, They have an interest in early detection of tsunamis and this is one way to do it. One can work out the sort of inverse problem you describe to look for seismic hot spots on a near-exaflop scale computer.

And it works! They can see three - Yellowstone, Hawaii, and Erebus. That gives you kind of an idea of the resolution- tens or hundreds of miles. They don't see the Ring Of Fire: it's too big and diffuse.

 

[Baluncore]

I'm afraid this is a case where "we'll just take care of it in software" doesn't work.

You have proved that your technique, based on your assumptions, will not work. But that does not preclude other methods that factorise the problem differently.

 

[olivermsun]

What assumptions do you deem essential to the floating hydrophone method working?

An explicit assumption seems to be that a set of Lagrangian surface floats will circulate around the ocean basin in a roughly swath-like fashion. This appears to be a non-starter, regardless of the method of inversion. There have been many, many drifter studies, and drifters typically don't cooperate in this way in the open ocean, let alone in coastal areas.

There seem to be numerous other implicit assumptions required for the concept to work. One is availability of whale calls and other ambient sources within useful range to build statistics for the drifting buoys would. On the engineering side, endurance and data recording/transmission capability of the buoys would be huge limitations, especially if the method hinges on the ability to keep extract higher-resolution data at later times.

Why is this method expected to be better than waiting for Seabed 2030 to aggregate multibeam bathymetry over most of the ocean floor?

 

[Vanadium 50]

You have proved that your technique, based on your assumptions, will not work. But that does not preclude other methods that factorise the problem differently.

Well, I think I have proven that a similar problem solved by very smart people quicky gets limited by available computing.

The problem isn't mysterious: If I want 10¹⁷ outputs, I need no fewer than 10¹⁷ inputs. Large )OK, gigantic) computing awards are of order 10²⁴ operations, so you have of order ten million operations per number. That's not realistic - and if someone thinks it is, they can put together a toy with only a million points and solve it on their PC. It should only take minutes,

 

[CGandC]

Sounds great but jeez, aren't they going to leave something for future generations to discover?

 

[Oldman too]

Here is an example of what might be hiding. It seems likely that the cretaceous-Paleogene boundary might have had a little help besides the Chicxulub event. https://www.science.org/doi/10.1126/sciadv.abn3096

Abstract from the paper.
"Evidence of marine target impacts, binary impact craters, or impact clusters are rare on Earth. Seismic reflection data from the Guinea Plateau, West Africa, reveal a ≥8.5-km-wide structure buried below ~300 to 400 m of Paleogene sediment with characteristics consistent with a complex impact crater. These include an elevated rim above a terraced crater floor, a pronounced central uplift, and extensive subsurface deformation. Numerical simulations of crater formation indicate a marine target (~800-m water depth) impact of a ≥400-m asteroid, resulting in a train of large tsunami waves and the potential release of substantial quantities of greenhouse gases from shallow buried black shale deposits. Our stratigraphic framework suggests that the crater formed at or near the Cretaceous-Paleogene boundary (~66 million years ago), approximately the same age as the Chicxulub impact crater. We hypothesize that this formed as part of a closely timed impact cluster or by breakup of a common parent asteroid."

 

[pbuk]

Every inch seems to be a bit of an exaggeration, since there are 10¹⁷ of them.

Yes it's a ridiculous exaggeration, the standard of editing in Scientific American is not what it used to be. The resolution the featured study is aiming for over the majority of the ocean floor is 400m x 400m which is a cool 250 million times worse than the "every inch" claim in the headline.

See https://seabed2030.org/faq

 

[olivermsun]

Covering every (square) inch of the seafloor with depth soundings is itself a huge achievement. It doesn't have to mean mapping every inch individually.