TOM ADAMSON:
Hello everyone and welcome to another episode of Eyes on Earth, a podcast produced at the USGS EROS Center. Our podcast focuses on our ever-changing planet and on the people here at EROS and across the globe who use remote sensing to monitor and study the health of Earth. My name is Tom Adamson.

Here at EROS near Sioux Falls, South Dakota, we are about as far from the ocean as you can get in North America. So it may seem odd that we’re talking about coastal studies, but that’s the value of Landsat satellite data. It collects data all over the globe, and for quite a long time period. This episode is the second in a series about how Landsat is helping researchers study coastal changes. 

A dataset called Digital Earth Australia coastlines documents three decades of coastal change mapping the entire coast of Australia. It uses freely available Landsat data. Robbi Bishop-Taylor, coastal Earth observation scientist at Geoscience Australia, is here to tell us more about this dataset. Robbi, can you please first just give us an overview of what this dataset is?

ROBBI BISHOP-TAYLOR:
Digital Earth Australia coastlines is a coastal change dataset developed at Geoscience Australia. And so what it does is basically provide a snapshot of coastal change across our entire coastline, reaching back through time to the late 80s. And so this is all using Landsat data. The dataset contains a range of different layers, so at a sort of very lowest level we have what we call annual shorelines that represent the typical position of the shoreline, corrected for tide for each year from 1988 onwards. And then from there we sort of zoom out a little bit and we get a dataset of coastal change, which gives us sort of quite detailed rates of change, so meters that the coastline has changed over time.

And then all the way at the very top level, we have a sort of a continental hotspots view that highlights areas that are rapidly eroding or rapidly growing through time. And so the idea is that our users can basically zoom into this map and kind of look at either the entire continent or zoom into their sort of individual beach and check how their coastlines have changed through time.

ADAMSON:
Now, that’s a lot of coastline and a lot of Landsat imagery going back to the late 1980s. How do you deal with that much data?

BISHOP-TAYLOR:
Yes. So everything we do is based around a piece of software called the Open Data Cube. And so this is something that was originally developed in Australia in I think around 2011 basically to solve the problem that we had a huge amount of Landsat data on tape. That’s the physical tapes in storage that we needed to manage and organize so that they could be available to people. And so the Open Data Cube originally was called the Australian Geoscience Data Cube and it was essentially a database management system to let people access and analyze Landsat data in a consistent way. So that’s now become an open source package, so this is now something that’s used by people all around the world. And so that sort of lets us have a really powerful way of accessing data for any location in Australia through time. 

And so we use that in combination with cloud processing, so on Amazon Web Services. And this lets us really scale up our workflows incredibly efficiently, and so the whole coastline dataset, 35 years of data across the entire continent of Australia, can be done in about four hours using Amazon. So we can really sort of scale up these Earth observation pipelines to really large scale and do some exciting science that would just not be possible without it.

ADAMSON:
So a big part of this is making it available to the public as well.

BISHOP-TAYLOR:
Yeah, definitely. So everything we do is open source. All of our code is open source. All of that data and that’s just been really, really important for letting people both kind of check our methods. Having open code is great for sort of making the processes you do transparent, but also just allowing people to use it for their own applications much more easily. The ease with which we can just sort of access 30 years of data and just chuck away scenes that we don’t like because of things like cloud cover. In the past, those individual scenes would have been thousands of dollars worth, but yes, it completely changes how you do things when you can just access it at such scale and be able to filter it and get rid of the stuff you don’t want easily.

ADAMSON:
Can you even use a portion of an image that’s cloudy but keep the clear part?

BISHOP-TAYLOR:
Yeah, definitely. So that’s actually really important to all of the methods we do. We, rather than doing things based on scenes or tiles, now we try to do things at the pixel level as much as we can. So basically we pull out the best pixels rather than the best images and kind of use the best of what we’ve got. And that is really important for something like coastal change because a coastline is kind of a continuous feature that goes around the whole coastline; if you try to do that based on images or tiles, things get sort of broken up and choppy. So yeah, we try to do things on the pixel level, which lets us generate these really nice smooth continuous datasets.

ADAMSON:
Things that happen at the pixel level, we’re talking about 30-meter pixels in Landsat data, and I’m also thinking of specific things that happen on the shoreline like high tides and low tides. The shoreline really changes from moment to moment, so how do you account for that kind of small-scale change in a large study like this?

BISHOP-TAYLOR:
So first of all, the key part of this study is being able to go back through time, so you’re building to sort of look at how our coastlines have changed over the last 35 years or more. And that was sort of the critical thing about this study because that sort of gives us a baseline of how our coastal systems are changing through time, which then helps us sort of compare future change to that. So Landsat was kind of essential to this process because no other satellite platform goes back that far. Our starting point was that Landsat was what we needed for this study.

But yes, the resolution is a real challenge, so there’s estimates that most coastal change globally is less than about half a meter per year, which does make Landsat with 30-meter pixels difficult. So one thing that we’ve implemented in this approach is that we’re using something called subpixel mapping to try and get more detail out of these 30-meter pixels than their 30-meter resolution. And so basically how that works is that normally you just draw a line between two pixels when you’re trying to separate them. This sort of compares two pixels, looks at little tiny differences in wetness between them, and then moves the shoreline in or out depending on how wet or dry each pixel is, and that that works really well. So even though it kind of feels a bit—it’s a sort of strange thing to get more detail out of a pixel than its actual resolution. We can generally get down to about 10-meter accuracy from 30-meter pixels, so that is a huge increase and makes it much more useful data to people who are looking at that really fine scale coastal change.

Tides are actually something that makes coastal change monitoring really difficult. So you can sort of imagine that the shoreline moves in and out as the tide goes in and out. If you have satellite photos taken at low tide and high tide, you can kind of get a really quite misleading idea of how the shoreline’s changing because you’re mixing up tide processes with actual coastal change. And in Australia we have some of the world’s largest tides. We have areas of the coastline which change about 40 feet in sea level between low and high tide.

ADAMSON
Forty feet up and down, you mean?

BISHOP-TAYLOR:
Yeah, exactly. 

ADAMSON:
Oh wow, OK.

BISHOP-TAYLOR:
If you combine that with a really low sloping beach, you can get change of up to 4 kilometers from low tide to high tide. 

ADAMSON:
OK, that’s pretty extreme.

BISHOP-TAYLOR:
Yeah. Yeah. So that’s a really important thing to account for, especially when you trying to do things consistently through time and across the coastline. So the nice thing about tides, though, is that they’re really predictable. So out of all natural processes, tides are probably the most predictable things you can get. So what we can do is that we can use what are called ocean tide models that basically are configured with all of the maths of these tides, and they let you basically choose a location and time period, and then estimate the tide for any time, any place in the world through time. So what we do with that data is that we collect all of our satellite images. We then use these ocean tide models to estimate the tide height for each image, and then we use that to basically as an extra dimension to filter our data. So we kind of know what images are high tide images, what images are low tide images. We can get rid of the ones we want and then sort of get a much more consistent set of images through time. And so that then means that the shorelines that we produce are sort of corrected for tide. We’re not getting the interference of that sort of huge change between low and high tide, and it means that we can then produce rates of change, which are actually comparable across large areas and through time.

ADAMSON:
It’s also a really diverse landscape. You’re talking about mapping the entire coastline of a continent. How do you know that the mapping is accurate?

BISHOP-TAYLOR:
Yes, I think that’s a really important thing. Say EO data products, products from remote sensing, are never correct everywhere. They’ve always got issues. They always got problems. They’re really, really useful. But there’s always sort of, you know, challenges and limitations to using them. So making sure that we documented them and worked out where the product works well and where it doesn’t was really important to this study. So we worked really closely with a whole bunch of stakeholders. So people who’ve been collecting data on our shorelines over decades and years and years manually with tape measures or with lasers or with drones.

So we sort of got all of the data they’ve been collecting this really, really amazing data and then use that as a validation set for this study. And we’re able to do this at pretty good scale. So Australia’s really lucky that we’ve had these people doing this amazing work for so long. So we were able to get really nice measured data for our beaches reaching back 30-40 years all across the coastline. I think we ended up using about 57,000 validation points, so a whole lot of validation data, a whole lot of really hard work that’s been done by these people all over the last few decades. And so we’ve used that to compare against our actual ones, look at where it’s accurate, look at where it’s not and sort of reveal these places where people can rely on it pretty well and areas where they should be careful. And so we use that to produce a whole range of different accuracy statistics that were published with the product. And we just found that that is a really valuable thing to help people make EO data products work for them because they’re not perfect, they’ve always got issues. But if you know what the issues are, you can sort of work around them and use it for the right applications.

ADAMSON:
OK, there’s still some groundwork that’s needed in a thing like this, just to help make sure that you’re accurate then.

BISHOP-TAYLOR:
It actually becomes more critical, that kind of groundwork, to give us confidence in our satellite remote sensing products actually becomes even more valuable. So yes, it’s definitely, this is not intended to replace that kind of on the ground measuring, but they sort of work together. We provide the scale that stuff provides the accuracy and the detail, but together they are really powerful tool. 

ADAMSON:
And you’re still, you know, creating this dataset much more quickly than you could possibly by just surveying on the ground anyway.

BISHOP-TAYLOR:
Yeah, definitely. So the limitation of these ground approaches is that they’re really expensive. They’ll hire someone to go out and measure things by hand or with a laser or a drone, but you can only do that on one or two beaches. And Australia has shorelines, which are 10-15 hours’ drive from the nearest big city, so it’s just not possible to do that at a scale that we need. So yes, the remote sensing fills that gap of the really big scale, so the really continental scale, top-down look at how our coastlines are changing, and then the other sort of manual stuff is better for those really intensely changing beaches that are in areas where everyone lives on the cities.

ADAMSON:
What are some surprising examples of change that you noticed?

BISHOP-TAYLOR:
Yeah. So there’s some really fascinating results came out when we first ran this thing. So we have some pretty extreme coastal changing areas. So an example of this sort of the biggest hotspot, there’s a place called Point Stewart in Northern Territories, so tropical, northern Australia. And so there we’ve got these sort of muddy mangrove shorelines that are eroding by about 15 meters per year, about 50 feet per year. So pretty, pretty crazy scale of change there, but then we also have the sort of the opposite. So on the other side of the country, we have a place called Twilight Cove, this is one of the ones that’s about 12 hours’ drive from the nearest city, and so it basically had no change. This beach was stable until about 1996, and then suddenly in 1996 it started growing by about 20 meters per year, and it’s grown by about almost half a mile since 1996. And we don’t really know why that’s happening yet. So it’s sort of, it’s revealed this incredible change of this beach that went from nothing to growing hugely. So yeah, this dataset, we don’t provide the answers of why it’s happening. We just provide the when and the where of the change, but it kind of gives coastal scientists a dataset that they can then jump in and understand these processes more.

But I guess the other kind of more macro level change was that, surprisingly, Australia is actually quite stable. So if you actually look at all the coastal change across the whole country, even though we have these areas of extreme change in some local areas, overall, most of our shorelines are stable, and it’s not really a difference between the amount of erosion and the amount of growth through time. And so this is quite interesting because it sort of hints that at the moment, we haven’t really hit this tipping point of sea level rise yet, but we’re sort of able to get this really big level view, something that we haven’t seen before.

And this gives us a baseline. So that as our shorelines start changing in the future, we can kind of look at this historical record and then sort of work out whether this really is a big change, whether it’s a small-scale change in history and sort of get a much richer context, a much richer history of how our shorelines are moving.

ADAMSON:
Yeah, you’ve got to have something to start with when you’re going to compare, well, we think this changed, well now we have something accurate to compare it to.

What is this dataset going to be useful for?

BISHOP-TAYLOR:
Yes. So we’re already seeing it being used quite widely by local councils and state government coastal managers. And so that is really the key use of it is that historical baseline. So giving these coastal managers a context of how their beach has changed through time and using that to evaluate future changes and how that links to its history. We’re also seeing it being used to understand drivers and causes of change, so things like changing wind directions, changing currents, sort of working out how that has been affecting our coastlines and then sort of how that might be affecting it in the future. And the other cool one is that we’re seeing it combined with citizen science drone data to develop these kind of indicators of coastal erosion, sort of like warnings of these areas of the coastline might be particularly vulnerable to future change. And that’s a really sort of fun combination because it’s again, that kind of big level remote sensing at large scale combined with really detailed data about individual stats from drones both coming together to give a better picture of how these systems are changing through time.

ADAMSON;
It seems like that’s the kind of thing that we couldn’t do even just a decade ago because computer processing couldn’t handle that much data if there’s people contributing a lot of data. Well now computers can handle that.

BISHOP-TAYLOR:
Yeah, the things you can do with parallelizing processes using super computers and in the cloud, it just it’s a game changer in the kind of analyses you can do.

ADAMSON:
Is this always a work in progress?

BISHOP-TAYLOR:
Yes. So this is very much designed as an operational product. So the idea is this will be updated every year going forward, and it has, I think we’ve gone through, I think we’re approaching our 4th annual update in a few weeks actually. So this is yeah, every single year as we get new satellite data coming in, we rerun this thing and we generate new annual shorelines and then new rates of change through time. Yeah, very much intended to be a live dataset that will track how our coastlines are changing as they evolve.

ADAMSON:
That’s really cool. And Landsats 8 and 9 are still operating and will operate we believe for a few more years. And then Landsat Next will come along, and we’re just going to keep having this Landsat record to help you map the coastline there.

BISHOP-TAYLOR:
Yeah, definitely. Landsat Next is super exciting as well for us, probably less for resolution. The really exciting thing for us for Landsat Next will be the temporal resolution. So getting more observations per year. Some of the challenges we deal with with coastal analysis is based on the time that satellites come over, often you don’t get pictures of the high tides or the low tides or the mid tides, and so with something like Landsat Next, we’re sort of hoping we’ll get a whole lot more observation density through time and really start to capture some of these coastal processes we can’t currently see with the current Landsat data.

ADAMSON:
Yeah, with Landsat Next it will be a repeat of every six days on every spot on Earth and on every piece of coastline in Australia versus every eight days, which is what we have now with Landsat 8 and 9. Of course, that’s even better if Landsat 8 and 9 are still operating at the time.

BISHOP-TAYLOR:
Yeah, it’s crazy amounts of data. That’s super exciting, but also slightly scary. There’s things that we’re able to do with that, like sort of shrinking our time period down to less than a year to sort of looking at much, much more rapid coastal change. And that just opens up looking at event-based processes. So looking at shorelines before and after storms, the impacts of different coastal developments, that kind of thing. So yeah, very exciting stuff.

ADAMSON:
Definitely a work in progress, but you can’t just coast either, you’re going to have to keep innovating once we have new types of data.

BISHOP-TAYLOR:
Yeah, future coastal products will probably look very different from this one. I think there’ll be things that we can do with Landsat Next that no one’s thought about yet, which is pretty cool.

ADAMSON:
But keep it consistent with the previous database because that’s what people want to see too.

BISHOP-TAYLOR:
Yeah, yeah, definitely. I think that that would be critical. So finding a way of using that data in cool ways, that doesn’t negate the value of the historical archive of Landsat, cause I think that’s super important to keep.

ADAMSON:
Are there other studies that you’re working on now?

BISHOP-TAYLOR:
Yes. So we’ve actually just published a new product called Digital Earth Australia Intertidal. And so what this does is we combine Landsat with Sentinel-2 data. So we’re sort of using data from both satellite platforms. And then whereas DA Coastlines is kind of just drawing a line on the map for each year, this is so taking it the next step and kind of turning it into a 3D model of the shoreline for every year, and kind of every pixel along the shoreline will have an elevation value, which is really useful because it lets us do things like kind of compare coastal change three dimensionally through time to sort of subtract 1 time step from another and look at where sediments moving or decreasing or increasing. Yes, that’s a new one that we’ve just published and we’re working on a paper for that at the moment. But it’s complementary. So it’s less years. It only covers the years on from the Sentinel-2 record. Now from 2015 on, but it does give you a sort of a rich, for those years it gives you kind of a richer three-dimensional view from combining both Landsat and Sentinel-2 data.

ADAMSON:
I’d like to thank Robbi Bishop-Taylor for joining us on this episode of Eyes on Earth. Find our previous episode in this series about mapping beaches on the California coastline, and check out our social media accounts to watch for all future episodes. You can also subscribe to us on Apple and YouTube podcasts.

VARIOUS VOICES:
This podcast, this podcast, this podcast, this podcast, this podcast is a product of the U.S. Geological Survey, Department of Interior.