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. 

Sandy beaches and coastlines are constantly changing. These areas are vulnerable because of sea level rise and changes in land use practices. Eyes on Earth is going to spend a few episodes on how Landsat is helping researchers study these coastal changes. For the first episode in this series, we’re talking to Sean Vitousek, research oceanographer at the USGS Pacific Coastal and Marine Science Center. Sean, can you start by telling us what your study looked at?

SEAN VITOUSEK:
Yeah. So we did a numerical modeling study of California beaches across the whole state, and we just tried to use past data to see if our models could be trained on that data and then validated with new data coming in and then make a long-term prediction of where the shoreline is going to be in 10, 20, 30, 40, 50, 60, 70, 80, a hundred years into the future. Are we going to have some beaches left? Are we going to have no beaches left? What’s going to happen?

ADAMSON:
Who cares if beaches are changing over time?

VITOUSEK:
Yeah, great question. I would say in general, if beaches were about in the same position as they always were, but they just kind of eroded and accreted back and came to the same position, then we probably wouldn’t be studying them. The reason that we do care about where they are in the future is that a lot of beaches are eroding. That means they’re decreasing in width over time, so in some places where you used to have a beach 20-30 years ago, there’s almost no beach. And as sea level keeps increasing, beaches generally only have one way to go, which is erode. So the main reason that we run this study is to try to see what beaches are most vulnerable in the future and potentially what we can do about mitigating some of the erosion that we’re expecting.

ADAMSON:
Yeah, after a beach erodes, what does it look like? Is it just sort of rocky shoreline that’s left?

VITOUSEK:
Either rocky shoreline or you just have water right up against the cliff, right up against a wall, right up against a beach house or something like that. So basically just where you once had a beach, you now have open ocean. Beaches provide sort of a buffer against storms and wave-driven flooding. So if you lose that protective buffer, you end up being able to flood all the inland infrastructure and sort of the ecosystem much more often and much more frequently. 

A good example in California is Highway 1. Highway 1 in a lot of locations runs right along the coast and they have to close sections of Highway 1 during the winter time because with the loss of the beaches, it’s just much more easy for waves to overtop that area and attack the cliff and things like that. So those beaches can really serve an important role for not only people but, you know, shorebirds and marine mammals and other things of the ecosystem there.

ADAMSON:
So let’s talk about the data that you’re using. I understand Landsat is a part of this. Is there anything else? How does Landsat work together with what you’re using?

VITOUSEK:
So we use basically about any shoreline, historical shoreline observation that we can. We ran a previous study that relied on lidar data, you know, aerial lidar, this laser that you put on a plane and it scans the elevation of the coastline. And that lidar data you can say you’re elevation contour is mean sea level and you can get at a historical shoreline position sometime in the past. So our previous studies used lidar data and lidar data is fantastic. It’s high resolution. The only problem is you can’t fly lidar that much. You usually get a lidar survey once every year, once every two years, something like that. So our previous study for a given location in Southern California, where the study took place, we maybe had around 5 to 10 lidar surveys of historical shoreline positions going back 20 years. 5, 10, 15 data points is good, but it would be nice to have more. And the thing that really just completely opened up the modeling possibilities tremendously was the integration of satellite data. Satellites just—they’re always flying over the earth, taking pictures, and because new machine learning models came out where you can efficiently extract and accuracy extract where the shoreline position was from that image, it opened up our world in terms of, previously we had five or ten data points and now for anywhere on the California coastline, because it’s nice and sunny, there’s not a lot of clouds impacting the imagery, we have 500 to a thousand observations dating back 40 years. So all of this amount of data was a tremendous asset to improve the calibration and the validation of our models. So we now know that we have a much more accurate prediction of where the shoreline was and potentially where it’s going to be in the future. The nice thing about satellites is just the temporal resolution is just unmatched. It can get an observation about every week or even less out of Landsat and Sentinel, and with Planet you can get an image almost every day. So there’s kind of a trade-off between what the temporal resolution that you’re interested in and the spatial resolution that you’re interested in. And for our application, the temporal resolution was just phenomenal because beaches can change pretty drastically by 5, 20, 15 meters over the span of about a week. And so capturing that temporal resolution was really an asset to this study.

ADAMSON:
Can you say more about how the medium resolution of satellites complements that higher resolution of lidar?

VITOUSEK:
It’s really nice to have a range of resolutions, you know from centimeter scale GPS measurements or centimeter scale structure from motion measurements. Like, if you had aerial photos of a location. Generally speaking those high resolution measurements, you can’t make as often. So it’s nice to have some very high resolution measurements to sort of compare satellites to, and if you increase the resolution a little bit, you’re looking at sort of a more medium resolution satellite imagery like coming from Planet or from WorldView or from Sentinel or other missions coming. Those are really helpful because you can compare the accuracy of those medium resolution imagery to the high resolution lidar measurements or the centimeter scale measurements from structure from motion or something like that and see how accurate do those measurements do. On the other hand, if you’re looking at a complicated feature like the shoreline, there are other factors in the error of mapping that feature that may be a bottleneck. Like, how much the tide is fluctuating, or how much the waves swashing up and down the beach are present and are affecting the shoreline position. So there’s a lot of complicated questions about all this that we’re trying to nail down with the aid of these multiple datasets and datasets for ground truthing. Another really helpful asset in order to understand this complexity that we’re seeing in the satellite imagery is we do a lot of beach surveys with the GPS mounted on an ATV. We’ve surveyed this beach, Ocean Beach at San Francisco, every month for the past 20 years. And so that’s really a nice complementary dataset to really understand the accuracy and the sort of time scales involved in satellites versus more traditional measurements of beaches.

ADAMSON:
OK, so there’s sort of a test case where you can validate the accuracy overall. You can’t go to every point on every beach in the state every day.

VITOUSEK:
Right.

ADAMSON:
Some people would love to go to the beach every day. But you just can’t.

VITOUSEK:
Yeah. No, absolutely. I mean that I think has really been the main benefit of satellites is, like you mentioned, in coastal science, we have a few super sites where they survey the heck out of the beaches. We have a really great comprehensive understanding and wonderfully long datasets of how beaches change at specific sites like Ocean Beach in California, Torrey Pines in San Diego where Scripps Institution of Oceanography is, the field research facility in Duck, NC, and internationally, some sites like Narrabeen in Australia, Truc Vert in France. You know, there’s beaches that have been surveyed over and over and over again, and it provided a tremendous amount of data for a theoretical understanding of how certain beaches behave. Satellites come into play because they just they allow us to be almost everywhere around the world. And see the changes taking place. There are some beaches and islands that are just so remote that you just almost can’t get out there. So understanding how those beaches change from satellites is really the only thing you can do.

ADAMSON:
Studying the coastline of California, is that something that could be done without Landsat?

VITOUSEK:
Studying the coast of California could be done without Landsat, but I don’t think it could be done well. When we did this sort of study in 2017, the previous version of it we didn’t have Landsat. We didn’t have that level of data. So we tried to make a prediction based on relatively limited sets of data, and that prediction was OK, but it could have been much better. I think going forward in the future we’re going to need it. We’re going to need to have that level of data to make a good prediction because to me a good prediction is just you have a model, you show how that model has been calibrated and validated well across real behavior that has been observed over a large scale. And then the only thing that’s giving you confidence in the future prediction is, well, how well did it do in the past? So that’s really the only way that I think we can predict coastal change going forward in the future is by using a lot of data coming from satellites. And in this setup, I do believe that Landsat will always play a very critical role because Landsat is really the only means of getting 4 decades of data. You know, with Sentinel you can get almost a decade. With Planet, you can get, well, it’s getting towards almost a decade, but to look at long-term changes, you’re going to need Landsat because Landsat is the only thing that’s going back 4 decades. Conventional wisdom really says, you know, if the data are easy to access and free to access, then they’re going to be used widely. So I think Landsat and Sentinel will always play a really important role in this sort of research, you know, because they have such long archives, because they’re free and easy to access. There’s a lot of information sitting in those spectral bands that is waiting to be uncovered. You know, we use bands like near-infrared and shortwave infrared, but other spectral mixture models, you know, might be really interesting to see. You know, what can we get out of those bands beyond just sort of the infrared and the RGB.

ADAMSON:
Right. And Landsat Next is a bit in the future yet, but the huge number of infrared bands that it’s going to have, I understand, are going to be pretty helpful for coastal studies as well.

VITOUSEK:
Yeah, absolutely. I think so.

ADAMSON:
Are there other studies that you have going on right now that are doing similar things?

VITOUSEK:
We almost about to publish a similar study, a shoreline modeling study like we did in California for the South Atlantic—make coastal predictions across the South Atlantic from about Miami, FL, through Delaware that used Landsat shorelines from you know back to the back four decades. We’re also working on a similar study in the Gulf of Mexico to map coastal change. In the Gulf of Mexico, using Landsat and modeling. And also in the Pacific Northwest, Oregon and Washington, working with some collaborators in Hawaii who are trying to do similar efforts in Hawaii and some collaborators in Australia who are trying to do similar shoreline mapping for the New South Wales state of Australia. So it’s definitely an exciting time. I think there’s a lot that can be done to just use the satellite data to understand coastal change and to try to make predictions. And that’s really the asset of satellites is you can get data everywhere, whereas before you really couldn’t. So I think satellites are really going to transform coastal science from a data poor field into a data rich field, and I think that’s really exciting.

ADAMSON:
You’re having to work with a lot of data and a lot of different types of data. It seems like it wasn’t too long ago that researchers weren’t able to use that amount of data because computers just didn’t work that fast. But that’s getting better, isn’t it?

VITOUSEK:
Yeah, absolutely. I think that the computing infrastructure in terms of cloud computing is getting better and better. The machine learning algorithms are getting better and better, so it’s really something that, almost 10 years ago, we just we didn’t have, and it’s really a new thing and an exciting thing. And on the accuracy side, I think we’ve done an OK job to assess the accuracy, but the accuracy as these methods become used and tested more and more often is only going to get better and better and better, so it’s an exciting time. I think another exciting thing is, trend is going to be more and more artificial intelligence, machine learning modeling techniques because those techniques are a way to have a dataset and a few list of important drivers like wave height or tide or water level or things like that and then just come up with a model by taking all that data, that training data, which is the satellite observations and trying to come up with a model. And satellites are really giving us all that training data, all that decades long training data of where the shoreline position was. So I think a real exciting trend will be the combination of satellites and machine learning methods. And I’m much more of a traditional data assimilation, coastal engineering type modeling person, but I see the future being a lot of really interesting data assimilation predictions, data simulation and AI/ML directions.

ADAMSON:
AI/ML is kind of one of those buzzwords I’ve heard a lot lately. Artificial intelligence and machine learning is what that’s all about. It’s not terribly new in terms of modeling things like this, right?

VITOUSEK:
Yeah, I have a colleague, Dan Buscombe, who’s kind of our resident artificial intelligence machine learning methods guy, and he describes it as just you’re teaching a computer to learn by example. So you’re providing all of these examples, which is just the data, coming out of satellites. And you’re just teaching a computer to sort of learn sort of those relationships within that data. And you’re right that it’s, you know, concept of neural networks have been around for quite a while, but the power of these methods has taken off tremendously in the past two decades, even in the past decade, because of the computer architecture GPUs that have developed to the point where you can train these massive models efficiently. Also having access to all this new data for training has really been the factor that has made it so that these AI models are just so powerful. And hopefully we can leverage that for some difficult, you know, science applications and science challenges related to sea level rise and climate change because we’re going to need all the help we can get to try to solve issues like sea level rise and coastal erosion. To me, it’s going to be the future of a lot of the work that we do, a lot of the science that we do.

ADAMSON:
I’d like to thank Sean Vitousek for joining us on this episode of Eyes on Earth. In our next episode about how Landsat helps with coastal studies, we’ll be talking with a scientist at Geoscience Australia about mapping the coastline of that entire continent. Check out our social media accounts to watch for that and our other 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.