Eyes on Earth Episode 84 - Hurricane Disturbance Mapping

LAWSON:
Hello, everyone, and welcome to another episode of Eyes on Earth. Our podcast focuses on our ever-changing planet and on the people at EROS and across the globe who use remote sensing to monitor the health of Earth. My name is Jane Lawson, and I'll be hosting today's episode, where we're talking about how satellite imagery can help with disaster monitoring in near real time. A newly developed system using the Harmonized Landsat Sentinel-2 data set and artificial intelligence was put to the test when Hurricane Ian tore through Florida in late September 2022.
By comparing pre-storm and post-storm imagery, the system quickly flagged anomalies statewide such as brightness, which could indicate exposed sand or bare land after a hurricane. Harmonized Landsat Sentinel-2 data is available through the Land Processes Distributed Active Archive Center, or LPDAAC for short, located at EROS. Our guest today is here to talk about the development of this new monitoring system.
Zhe Zhu is an assistant professor at the University of Connecticut, where he serves as the founding director of the Global Environmental Remote Sensing Laboratory. He is also a member of the Landsat Science Team, and he previously worked as a Land Change Scientist at EROS. Zhe, welcome to Eyes on Earth.

ZHU:
Thank you, Jane.

LAWSON:
Let's first lay the groundwork for talking about Hurricane Ian by discussing your background a bit. At Boston University before you came to EROS, you and Curtis Woodcock developed an algorithm that uses all available Landsat data to detect change in classification of land cover called Continuous Change Detection and Classification or CCDC. And that was published in 2014. What need did this algorithm and variations of it since fill in the remote sensing community?

ZHU:
One of the major needs I think this algorithm filled in is how to use this really, really dense time series. As we all know, that Landsat data is open and free policy started. And after the cloud caused shadow detection getting into a more mature phase, there are lots of observations available for the same location and how to better use it.
At that moment, I don't think that too many people have ever thought of this question before because there's never been so much data available at this high resolution. And CCDC actually I think it's one of the first approach trying to linking all those data sets together into a time series model. And using this model, we're able to not only capture what has been changed, but also to capture the seasonal and also the temporal information throughout the time series data set.
And there are a lot of variations of this algorithm since it was published. And different kinds of application has been applied. For example, forest degradation, coastal tidal wetland change, and also recently it has been used for even monitoring those nighttime lights change at a much larger scale. So I think that's mostly what has been done since 2014.

LAWSON:
Excellent. So let's jump ahead in time now to almost today, and tell us about the development of the University of Connecticut's Disturbance Watcher and that intended purpose.

ZHU:
So this is actually is a very, very new development here at Univ of Connecticut. What we're trying to do is we're trying to make this watcher to monitor the entire United States on a sub weekly basis. So every two or three days we update this map at a high resolution, a 30 meter resolution for every land pixel in the US.
And so far we already finished multiple places like in California, the northeast part of the country, in Florida. They're also doing something in the Centeral Plains. Now the places we're going to expand for the CONUS wide. One of the first events we targeted is Hurricane Ian, and this watcher is not able to achieve this goal if there's no code algorithm, no SCC.
So SCC is another more near-real-time approach is adapted from the code by this is called a stochastic continuous change section algorithm. Which is able to getting the data online, online change the action approach, getting the data and then getting the results much, much faster than the code approach and achieving similar accuracy. And also another thing I want to emphasize is the availability of harmonized Landsat Sentinel 2 data sets. That's a huge difference as compared to us what we can do versus now consider.
Right now we have Landsat 8, Landsat 9, Sentinel 2A, 2B all together. We're able to get up 30 meter resolution data in less than three days. That's a big, big improvement in terms of the temporal frequency of those medium resolution datasets. And another thing I want to emphasize, this harmonized data set created actually by NASA, they are also making them in a more timely domain. Basically, previously, I think just a few months ago, we need to wait for like a week before the data is collected and we're able to get the data.
Now, it's only like less than two days when the data is collected and we are able to get this dataset. So this greatly reduced the latency for this kind of near real time monitoring system.

LAWSON:
Do you want to tell us more specifically about the work with Hurricane Ian? The process and the results of using this Disturbance Watcher?

ZHU:
So this hurricane, yeah, and we know it has causing lots of devastating damage to the entire state of Florida. And even just before it's trying to touch the Florida, we know it's going to be a big event. We know the time is going to be the most important value for maps like that. So we actually prepared time series analysis, time series data for the Harmonized Landsat and Sentinel 2 datasets before it's reaching the state of Florida and when it's actually landed in Florida.
We are just waiting for the first image. And basically when we get the first clear sky image, we put that image into the analysis ready data, and the data actually can predict what the future date looks like based on the past. And when we get that first the clear sky image, we are able to compare the predicted with the observed. And then using the high performance computing center capability to provide that this state wide hurricane damage map almost in real time.

LAWSON:
Did everything go as you predicted, or did you learn anything new from this process?

ZHU:
One of the most important things we learned through this process is that near real time is just, not just about algorithm. There are a lot of moving parts. You have to figure them out, and you don't want to hear any kind of surprise. For example, the high performance computer we use at University of Connecticut actually didn't work well. They were having some problem when the hurricane actually landed Florida.
So we actually waited, I think, for a day or two, at least a day, to have the computer ready to be able to be used. So this is something - this is a latency we never expected when we're writing the algorithm, right. But in reality, we have to count every kind of moving parts. And anything, if they are not right in its original expected condition, it's going to causing latency to your final results. And consider our map is trying to be used for the people, for rescue, for damage, so any minutes - the earlier the better, right.
We want to reduce this latency as much as possible. Right now, it's about three days. About three days. Just not good enough, right. And it's possible if the delay of that hardware is more than five days, then the map is not time relevant anymore. So that's something I want to say. Yeah, every moving parts we need to making sure they're in the right place, in the right condition.

LAWSON:
There's an urgency to this because you're trying to save lives and property. 

ZHU:
Right.

LAWSON:
 How would you ultimately like to see this Disturbance Watcher used then?  By what types of groups, for what kinds of disasters and in what areas? You mentioned CONUS, so across the continental United States.

ZHU:
So this Disturbance Watcher, we want to build it for a generic purpose. Not just for forest disturbance, not just for one type of disturbance type. For example, the hurricane. We want to do it for all, so for all kinds of land disturbance types, including landslides, earthquake, hurricane and even some flooding and also some anthropogenic change like construction, forest harvest, forest plantation and stress, disease, fires.
So we want to include everything in this Disturbance Watcher and for all the land surface type. And the ideal hope is that we can expand this CONUS wide and we can also extend to Alaska if we want and to provide this useful information. And have you ever used an application called, I think, the earthquake monitor by the USGS. Basically if you feel like something, it looks like earthquake, you go to that website. And this is something we want to provide. If something happen, if some disaster or some kind of disturbance happens nearby you, or any parts of the US you're interested,
this is website you're going to look for.

LAWSON:
That seems incredibly useful. How does the Disturbance Watcher then fit in with your area of study as a member of the Landsat Science Team? So your area of study is toward near real time monitoring and characterization of Landsat surface change for the conterminus U.S.

ZHU:
So you already mentioned that two components, major components, in this team work. First one is the monitoring, near real time monitoring. The second one is characterization. And this near real time actually fill in this monitoring component. So anything happens on that land surface, we want to see that. We want to identify that in a near real time mode. And we also have another component to actually characterizing. Like, 
what's the land cover before that? What's the agent or the driver of this disturbance? That's another work. We also are working on that. And hopefully this publication that for CONUS wide characterization of this land disturbance for the past like 30 or 40 years.

LAWSON:
As someone who has worked with the Landsat archive on land change, including the Land Change Monitoring Assessment and Projection, or LCMAP, program during a stint at EROS, and also forest and coastal wetland research, can you describe your perspective of the significance of the Landsat program?

ZHU:
It's so important. I think one of the most important thing is it's open and free. Without that, I don't think any work from the LCMAP or the coastal tidal wetland we're doing is possible. And another thing I think is extremely important is this analysis ready data. So for a long time, I still remember when I was a student at the Boston University, I spent a few weeks to download 500 image.
You have to click twice for one image. And then I spent one month to install atmospheric correction algorithm and run the algorithm to get surface reflectance and to run the Fmask to get clouds and shadows, snow detection. So to just get the data ready for you to do the analysis can take you several months for one path and row. But what EROS has created is analysis ready data, entirely changing the game.
We can just download data and do the analysis. It's a big thing. And another significance of this Landsat program is almost, if you look at the literature, almost every important major medium resolution global or large scale work, you can see Landsat showing up. Sometime  you can see Sentinel-2, but a lot of time you're going to see Landsat used along with Sentinel-2, or just Landsat.
Because it's the only data set that's covered global scale at the medium resolution and has 50 years of time from Landsat 1 all the way to the current Landsat 9. I just attended this Pecora 202(2), and we celebrate the 50 year anniversary for Landsat program. And it's a huge success. I would say if you rank Landsat as the second Earth's Observation Satellite, I don't think there's another satellite can rank the first.
It just has too much huge impact on Earth observation. And I'm so glad it's, it's this continuity and also Landsat Next is going to continue its success and getting more exciting and even more frequent, more hyperspectral and high resolution data set. Landsat is going to - not only a long term success, but also the future for remote sensing.

LAWSON:
Thank you for sharing that perspective. Do you want to share any notable memories of your time of EROS or any connections or experiences here that have remained important to you?

ZHU:
Sure, there are a lot of connections at EROS. And a lot of people - when I talk about it, I may cry sometimes. Like Tom Loveland, he's the person who got me into EROS, and I worked with him. Such a nice person. He's like the academic father to me. Always caring my research and also gave me the great vision for this, for my research. And there are a lot of other people I want to mention like Jess Brown, Heather, Qiang Zhou, Suming Jin, George Xian - there's so, so many people there. And if I have I want to name I'd give out the name and if I finish with things like 15 minutes but all those people
are still keep very close connections. And we are still working closely on papers, project on the LCMAP and now NLCD on the urban detection and deep learning, machine learning method. I think EROS is a big Earth observation research hub, and lots of exciting research is going on here at EROS. And I just feel like I'm so fortunate to have worked at EROS for two years and know all those great people there.
It's my opportunity, a great opportunity and an honor to be able to have those two years there.

LAWSON:
Thank you for sharing that. Is there anything else you'd like to add today?

ZHU:
I think, just wait for our product. The near real time monitoring capability is going to be available for the CONUS wide, hopefully at end of this year. So should be soon. We are going to have a actually CONUS wide product available end of this year, or early next year. So if you have any kind of disturbance, you can go to that product.
Another thing is that we already have a CONUS wide disturbance characterization platform, basically to tell you all the disturbance happened in the past 35 years, 36 years from 1985 to 2020. So not only tell you it has been disturbed, but it'll also tell you the agent of the disturbances - is it mechanical, is it fires, agriculture, is it forest, it's, it's construction.
So a lot of categories you are able to see.

LAWSON:
And is there a name for the disturbance product that shows change over time?

ZHU:
We haven't get a name for the system yet, but if you go to my website at GERS Lab, there's a product on the upper right. And if you see there the multiple image, if you click those image, you're able to see this platform. We put all the platform and the products link there.

LAWSON:
Great. And we'll be watching for the CONUS Disturbance Watcher. Thank you Zhe for joining us for this episode of Eyes on Earth where we talked about satellite monitoring of natural disasters. And thank you to the listeners. Check out our EROS Facebook and Twitter pages to watch for our newest episodes. And you can also subscribe to us on Apple Podcasts.
Google Podcasts.

This podcast.

This podcast.

This podcast. 

This podcast is a product of the US Geological Survey, Department of Interior.