Skip to main content
U.S. flag

An official website of the United States government

Eyes on Earth Episode 38 – Time Series Analysis with Landsat

Right-click and save to download

Detailed Description

In the past, remote sensing scientists looked for change on the Earth’s surface primarily by comparing one Landsat image to another. Today, open access to Landsat data, high-performance and cloud computing capabilities and sophisticated algorithms can be used to scan the entirety of the archive for change, enabling researchers to learn more about how the landscape shifts over time. On this episode of Eyes on Earth, Landsat Science Team member Dr. Curtis Woodcock shares his thoughts on time series analysis, the future of remote sensing and his hopes for the Landsat program.




Public Domain.


Hello everyone. Welcome to this episode of Eyes on Earth. Our podcast discusses our ever-changing planet with people here at EROS and across the globe who use remote sensing to monitor and study the health of Earth. I'm your host, Steve Young.
Today's guest is Dr. Curtis Woodcock, a professor of remote sensing in the Department of Earth and Environment at Boston University. Dr. Woodcock is a member and co-leader of the 2018-2023 Landsat Science Team (LST), as well as a member of the NASA Land Cover and Land Use Change program. In 2016, he was honored by NASA and the Department of the Interior with the William T. Pecora Award for his achievements in environmental remote sensing.
Recently, Dr. Woodcock made a fascinating presentation to the Landsat Science Team about using almost 50 years of consistently calibrated Landsat images in time series that enable us to analyze changes in land cover over time. He discussed the application opportunities that arise out of time series, too, as well as issues and challenges that remain.
Here to expound on all that is Dr. Curtis Woodcock. Welcome.
Thanks. Good to be here.
So Curtis, let's start with the seminar to the Landsat Science Team where you trace the evolution of the land cover change monitoring from your early days as a researcher to the present. What do you see as the key evolutionary transitions, and talk especially about today's strong emphasis on the use of Landsat and time-series approaches?
It's easy to say free data. You know, in 2008, we got free access to the data in the Landsat archive and started using the data we needed rather than the data we could afford. And that's hugely important. Just to give you an example, at the time we got access to the archive, only about 7 or 8 percent of the images had ever been collected had been processed and used. People had to be incredibly selective and only pick the very best images, which mostly meant no clouds. So, the taxpayers had spent billions of dollars to collect massive amounts of data that we, users, couldn't afford to get. Now we use virtually all the data, sorting through each image from whatever good observations are available and making use of those. So, that's been a sea change. Two other points. Before the free data era, the USGS used to process images individually for users, and according to each user's sort of specification. So each one was kind of hand-crafted. And those methods changed and evolved over time. So, if you got images from two different time periods, it was really hard to compare them because they had been processed differently. Now we have Collections, or this idea where the entire archive gets processed using the exact same methods. And so now, it's very easy to get images from different time periods and compare them. And so the movement to Collections has been huge. And one important part of that is also sort of the level and quality of the processing of the data. There have been pretty dramatic improvements in the last sort of 10 years. And they all support the ability to study large areas more effectively and more efficiently. We've improved the registration of the images so that they line up better in time. And then the ability to, in an automated way, find the clouds and shadows and snow, and things like that, which tend to confuse your analysis. And atmospheric correction ... all of those are huge. And so EROS deserves a lot of credit and USGS because the quality of the processing this data has gone way up in addition to making the data freely available.
Give us some examples of how the power of time series studies, with all this consistent Landsat data going back 30, 40, 50 years, tell us how that's impacting the world today, these studies.
I guess you can sort it into sort of two broad areas. There's sort of science or understanding a changing world. For example, how are the planet's ecosystems responding to a changing climate? You have to have observations over a long period of time to be able to detect subtle trends. One example is patterns of greening and browning at ecosystems at high latitudes where the climates have been most pronounced. These long consistent time series allow you to detect whether places are in fact getting greener as a result of climate change, or in some places, getting browner. The other place where there are big changes is in natural resource management. I think many of us got into this field to improve natural resource management. And so, for example, studying the health of ecosystems and trying to do that in something approaching near real time. Here in New England, the last few years we've had outbreaks of gypsy moths. And, we're now able to provide much better detail about both the magnitude of the damage, more spatial detail about the location of the damage, and provide it in a more timely fashion, sometimes within just a few weeks or a month, rather than having to wait months or years to get the information to the pest management people. So, I would say science at one end and natural resource management at the other are the primary beneficiaries.
You know, I've heard you talk about such things as, you know, integrating other sensors in with Landsat. And I know that's part of a discussion about what's going to happen with Landsat in the future, the use of data from other sensors. And you've talked about topography. You've talked about improvements in removing clouds and shadows. If we succeed in all this, what is the potential for improving methods to monitor land use, land cover and land change in the future?
I guess I'd like ... this may sound funny, but I think we're still in our infancy concerning the use of time series of images, you know, to monitor change. And let me make fun of myself a little bitLet me  and give you an example. Some of the primary tools used in remote sensing, one is image classification, which is trying to sort the observations into different surface categories ... crops, grasslands, forests or whatever. People have been actively researching new methods of doing image classification since about the late 1960s and are still coming up with better ways of doing it. And, for sort of the two-date change detection paradigm, which has been dominant since the 1980s, same thing. People are still coming up with better ways of taking two images and comparing them through time to look for change. For time series analysis, we're using lots of observations. We've been at this for maybe 10 years, and 10 years sounds like a long time, but relative to sophistication of image classification and these two-date change detection methods, I think we're still kind of in our infancy. There are new ideas coming out all the time, and so I think we're a long way from having really settled on what the best way to do these kinds of things is, and that makes it exciting. There's a lot of progress to come.
As we spend a lot of time tweaking and trying to improve the Landsat archive to achieve consistent surface reflectance, surface temperature, cloud/shadow screens, all of that in all the imagery, why is that important. Why is it important to have all that included in the Landsat archive?
It's really about finding change. And if you're going to compare images through time to find change, there has to be consistency in what's being measured. Differences in images between dates could be as simple as differences in a sun angle at the time the data was collected. And so, you have to try to correct all that and take all of those sort of extraneous things out of the data, and distill it down to surface reflectance, which is a consistent surface characteristic. And it's not simple. Having the Landsat program do it ... makes everything easier for everybody downstream, if you want to think of it that way. Clouds and shadows, and I would throw snow in there sometimes ... in essence the observations effected by clouds and shadows and snow really are like noise. And if you include them in your analysis, all they do is confuse your analysis, which is a polite way of saying they ruin what you're trying to do. And so, the better we can screen out these bad observations or noisy observations, the better it is for everybody downstream as well.
I get the sense that sometimes with remote scientists, changing their ways is difficult. And if you're use to downloading data into your computer, and now the suggestion is, you need to go to the cloud, I mean, is that a hard transition for people to make?
Yes and no. In some ways, it's a big relief because trying to download and store and process the data locally for many people can be quite hard. At major universities, like Boston University, we have fabulous high-performance computing capabilities. But still, the volumes of data are a major issue. There is a bit of a learning curve and there are some needs to adapt, but people are doing it, and that's just going to do nothing but increase with time.
Well, the Landsat Science Team has advocated for the adoption of the Analysis Ready Data model for providing global Landsat time series data. What is ARD, and how has that concept progressed?
ARD is simply moving the data out of the original images that they come in, which is kind of an artifact of the path of the satellite over the surface of the Earth. And organizing it into a tiling system that makes it easier for a user to get all the observations ... rather than live with the data being organized based on where the satellite overpass goes, why not organize it so that its around ... makes it easy for people to get the data for just the study area that they want to use. The USGS has produced a good dataset, Analysis Ready Data for the U.S. But we've been trying to convince the USGS to go global with it, without much luck to date, I have to admit. But we'll keep trying. It's important because it makes easier for more people to study large areas more effectively than is possible now. At the Science Team, we're sort of all about advancing the science, but also about advocating for the broader community of users. On this one to date, to be honest, we're failing the community, but we'll keep after it. 
Then, what are the possibilities in your mind for a seamless global multi-sensor ARD that integrates all Landsat data, the European Space agency's Sentinel-1 and Sentinel-2 platforms, and possibly others. What are the possibilities for that?
The possibilities are great. The more satellites you have, the shorter that near real time becomes, and the faster people will get information about the land they manage and care about. That information will lead to better management, reduced illegal logging in the Tropics, go all the way down the list. The possibilities are sitting there, and they're ripe. I'm hoping the USGS will take the leadership role in this, but if not, somebody else is going to do it. It's inevitable.
YOUNG: Dr. Woodcock, can you say something about your current research activities and what you're focusing on to advance land change monitoring?
We're trying to produce land cover change products globally over the 2000s. And so, all the same things that we've been doing in specific countries or regions, we're now trying to do globally. Operationalizing it is difficult, and there's a lot of variability in the quality and amount of satellite data available in the archive from place to place. There's a lot of variability in the availability of high-resolution imagery, which we use for reference data to tell whether or not we're getting things right. It's fun, and it's exciting, and I think the results we're going to produce are going to be quite novel and beneficial to the community. But at the moment it's all about trying to take these exciting sort of local scale demonstration ideas and build it into something that's a little more operational and can be done globally.
You've served as the co-leader of the Landsat Science Team for the past 15 years. What are the biggest changes or advances that have occurred in your tenure?
First let me say, it's been an honor to serve and to be honest, it's been fun. I guess the biggest change I would say has been the relationship between the user community and EROS and the USGS. When we started, there was very little attention paid to the needs of the users, whereas now the USGS and EROS have worked really hard to make the satellite data products available to the community in the best way they possibly can. That's boosted the whole industry. The other funny example I'll give you is LGAC. You know, one way to look at what the science team does is, we're constantly suggesting things that will be hard to do to the USGS and then expecting them to do them. And one of those was to try to produce a truly global consolidated archive. A lot of the data was downlinked to the international receiving stations. And a lot of the data still live there. Very little of it flowed back to the U.S. archive. When we realized this very early on in the terms of the science team, we said, that data's invaluable because it can't be reproduced in any other way. And so they did. And it took seven or eight years to go to each individual receiving station around the world, ask them to send all their data, and then figure out how to read it all because it's always in different formats. And very patiently and very thoroughly, the USGS did this. It's a tremendous benefit. It more than doubled the size of the archive in terms of historical imagery. And so, in essence, it recovered more than half the value of the whole program as far as I'm concerned. But it was a long tedious, hard job. And, so that's one of the best things I've seen out of the time on the Landsat science team.
YOUNG: What do you think the challenges are facing the Landsat program? Do you worry that the long-term continuity of Landsat will ever be in jeopardy and fall through the cracks?
Oh yeah, of course I do. Fortunately, we've had two satellites functioning most of the time in the 2000s. But continuing that legacy is harder than you think. The satellites are typically designed with sort of a five-year life expectancy, or what they call a design life. But we had 14 years between Landsat 7 and Landsat 8. And now it looks like it will be eight years between Landsat 8 and Landsat 9, and no real end to this pattern in sight. If you look at the value of the Landsat program over time, and it's been incredibly successful, really the strength of it has been continuity, that there have been consistent observations over long periods of time. Jeopardizing that would just be foolish. But just to assume it's going to continue I think is silly. We have to keep reminding people over and over and over again.
YOUNG: We've been talking to Dr. Curtis Woodcock, a professor of remote sensing in the Department of Earth and Environment at Boston University about using stacks of consistently calibrated Landsat images in time series as part of the analysis of changes in land cover over time. Thanks for joining us Dr. Woodcock.
Oh, you're welcome. It's been fun. Thank you.
YOUNG: We hope you come back for the next episode of Eyes on Earth. This podcast is a product of the U.S. Geological Survey, Department of the Interior. Thanks for joining us. 

Show Transcript