Eyes on Earth Episode 66 – Exotic Annual Grasses

JOHN HULT:

Hello everyone, and welcome to another episode of Eyes on Earth. We're a podcast that 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. I'm your host for today, John Hult. The rangelands of the western United States are changing more quickly than many other parts of the lower 48. Miles upon miles of the area or semi-arid landscapes in states like Idaho, Montana and Nevada are now carpeted by fire fueling invasive grasses. Cheatgrass is the most prevalent, which is troublesome for several reasons. First off, it greens up and browns down really quickly, leaving a layer of tinder-like vegetation. In many areas, it fills in the formerly barren spaces between thicker bunchgrasses and sagebrush, which in turn helps fires move rapidly from fuel source to fuel source. After a fire, cheatgrass proliferates easily, choking out native grasses that need more time to establish themselves. EROS scientists are among those who've stepped up to document the decades long spread of invasive grasses in the west. Stephen Boyte leads a project at EROS that maps exotic annual grasses early and late each season, using data from Landsat and European Space Agency Sentinel-2 satellites. His team recently released a dataset showing change over time from 2016 through 2020. Steve is with us today to talk about the genesis and evolution of the project, long with Devendra Dahal, a scientist and contractor at EROS who's helped push the project forward. Matt Reeves is among the users of that data. Reeves as a research ecologist with the USDA Forest Service, who built a tool called Fuelcast that pulls together information on precipitation, temperature, and a handful of other data points to produce monthly forecasts of biomass across the United States.

Matt, Steve, Dev, welcome to Eyes on Earth. Thank you, John.

STEVEN BOYTE/MATT REEVES:

Thank you, John

DEVENDRA DAHAL:

Thank you.

HULT:

Let's get started here by talking about this exotic annual grass dataset. Tell us about this product. What data do you use to build these maps, and what do users see when they download the data, when they sort of open the box?

BOYTE:

We started developing these datasets about a decade ago. And we previously used a MODIS based product at 250 meters, spatial resolution and 2020. We shifted to using 30-meter spatial resolution data, which is called Harmonized Landsat-Sentinel data.

HULT:

Just to break in here quickly, we're talking about a relatively big difference in spatial resolution. You're almost down to that sort of field scale when you get to the 30-meter resolution, is that fair?

BOYTE:

Yes, that's fair to say. The 250-meter resolution is more of a broader landscape tool. whereas the 30 meter, while it's still a moderate spatial resolution, it gives much finer data and it's more useful for more people. We also use geophysical data to drive our models as independent variables: elevation data, soils, data, and land cover.

HULT:

You're not just looking at the Landsat-Sentinel data, as you just explained. What's the importance of using these additional data sets? How does that benefit your project, and ultimately the users?

DAHAL:

The different variables gives us the different aspects of the landscapes. We use elevation, and we use aspect and slopes. Those (tell) us how the different grasses like cheatgrass and other invasive species drives in those landscapes. And similarly, we also use different soil types and those kinds of datasets, has their different properties. So we can drive our model more accurately to get the output we need.

HULT:

So it's a matter of adding the data sets that you need to make sure that you're as accurate as possible. And I want to ask another question about that too, because, you have a couple of different datasets, right? You have the early season and late season estimates, and then now you have this time series product. Can you talk a little bit about each of those products and the uses that we might have for each?

BOYTE:

The early estimates are mainly to give land managers, other scientists, other researchers, an early season, look at potential, fine fuel abundances. The historical datasets allow you to make comparisons to what characteristics are driving each year's cheatgrass or exotic annual grass abundance.

HULT:

Almost like the difference between weather and climate. The early season is kind of what's going on right now, and the time series, the sort of history product, gives you a chance to see those trend lines and maybe dig a little deeper into what's going on.

BOYTE:

I think that's a good analogy. Okay. Well,

HULT:

Okay, well, let's turn to Matt now. Let's talk to somebody who uses this data. Matt, you've done a lot of work with exotic grasses over the years. Can you tell us why it's important to understand where the exotic grasses are in the west.

REEVES:

The importance of exotic annual grasses is in two different places. The first is kind of what Steve had mentioned, and a little bit from Dev, as well. And that would be, we want to understand where the invasive annual grasses exist from a purely fire behavior potential. Those can create various significant difficulties for fire. The second reason that I think it's important to understand is that there are unique challenges in restoring lands where these invasive annual grasses have got a foothold. So by monitoring and understanding, when we do see invasions, you know, that kind of early detection idea, when we do see new invasion somewhere, it would be very helpful if we could go on attack those new invasions with our toolbox and try and get a handle on it. 

HULT:

This is a relatively new product, right? This is, this is not something that was available 10 years ago. How did you deal with this in the past, Matt, before we had this spatial data twice a year?

REEVES:

Well, I think it's been a work in progress. First of all, we had looks at people in the field, trying to look for these invasions and make assessments of what they thought the fuel conditions would look like. But it was, I don't want to say haphazard, but certainly disjunct evaluations across the landscape and it wasn't complete. It was never comprehensive. It was never consistent. Different groups will evaluate the spread and abundance of cheatgrass differently and report it differently. And so in the past, I think it was a bit more of a piecemeal approach to dealing with the problem. Through time, we've improved that situation culminating in this type of fantastic work, which gives us a consistent look across the west, twice a year, that everyone has access to not just a somebody's database on a computer. So there's been kind of a real revolution in how we're able to understand where and how much invasive annual grasses are. 

HULT:

Let's stick with you for just a second and talk about Fuelcast. What goes into producing this? What does it tell us about what's happening or might happen on the ground, and who's looking at Fuelcast to get those answers?

REEVES:

The Fuelcast system derives from some of the same types of data that both Dev and Steve mentioned. So we have daily precipitation, we have the eddy, which is a NOAA drought product, and then we have weekly to monthly remote sensing feeds, which were previously MODIS, the 250-meter version, expedited MODIS, eMODIS, which has also a USGS product. But this year, we've pivoted to the Landsat stream, and so we're taking all these different indicators and we compare them to our rangeland production monitoring service information, which gives us yield data of rangelands from 1984 to 2021, and we compare these predictions to those yield observations, and we now are going to be making a monthly projections, or forecasts if you will, throughout the growing season. But Fuelcast starts the projections four months ahead of the peak. So if the peak has estimated in June in a given area, we subtract four months from that period, and the engine, the Fuelcast engine, starts and creates a model and makes the forecast. And there's one more component. It allows up to two years worth of antecedent moisture conditions, drought conditions, and remote sensing conditions to influence this month's projection of potential yield. Now, to your second question about the users, or people that have interests, I think there's two camps here. The first is as the name suggests: the fuel camp, the fuel and fire camp. So we have both a tactical and a strategic perspective here. The strategic perspective would be fire planners that use this information to say, it looks like, "Hey, this is going to be a big fuel year or not, and therefore we want to start thinking about some of the resources we might want to consider ahead of time to prepare for that." But there's also the tactical side, which is on an incident, say on a wildfire, people will want to have access to the latest fuel information they can get, to understand what the fire behavior implications might be. So that's the first camp that is using these types of data. The other one would be the agricultural loss programs, for example, having an interest in these types of data. The Farm Services Agency and its various types of insurance programs periodically looks at what the conditions are and what the conditions are forecast to be in order to start communicating with those producers on the public land side. That would be oftentimes permittees, or public land grazing permittees, to let them know, "hey, it looks like this is going to be a tough year, we understand that. So we're reaching out to you now to talk about some strategies to help you out.

HULT:

Just to put a real fine point on this, Matt, this exotic annual grass data allows you to be a little more specific, and give more useful information, really, to these users than you would've been able to do before.

REEVES:

Exactly. When we make our projections of yield, the annual grass maps go on top of our assessment of yield, and it's sort of a cookie cutter approach. If Steve and Dev tell us at a given area, "Hey, we think the cheatgrass is very abundant here," let's say it's 80% cover, pretty bad situation, what we do is we then partition our projection of yield to the invasive annual grasses, using those data that are coming on from USGS. So we offer a blended product, where invasive annual grasses are concerned-Fuelcast plus the USGS viewpoint of invasive annual grass abundance. Again, that's important for not only producers, but especially for those fire behavior analysts and managers. 

HULT:

And it keeps getting better. As I understand it, Steve and Dev, you are now able to tease out the species, whereas the first few, it was just exotic annual grasses. You weren't sure what kind of grass it was. Can either of you talk a little bit about how that was possible. It's my understanding that maybe some of the on the ground survey data improve that helped you to do that?

BOYTE:

The Bureau of Land Management has a database called Assessment Inventory and Monitoring, or AIM, so we refer to it as BLM AIM data. In the last couple of years, they've included in their database, the species. In 2021, we were able to develop specific species maps for both cheatgrass and Medusahead. Is there anything you'd like to add to that, Dev?

DAHAL:

Actually, AIM started their version 2 release in March of 2020. Before that, we did not have that information, so we could not break out these species in our maps. After that, we were able to do the cheatgrass, Medusahead and overall, the exotic annual grasses. 

HULT:

You also have native grasses in there too, isn't that right? So you can see where some species are that you might want to restore or preserve?

DAHAL:

We started this year including the Sandberg's bluegrass, that is native bunch grass-perennial, not annual, but we can predict annual percent cover.

HULT:

Right, so you have areas where it, it isn't something you want to get rid of, and that's potentially useful, as well. I want to talk about where this project might be going in the future. I know you were able to pull this together with the help of high-performance computing. What comes next? I mean, are we talking about cloud computing? If so, what's going to enable you to do that?

BOYTE:

As we've increased the study area size and the data density, you're exactly right: We could no longer use virtual desktops or normal desktops. We had to put our code so it would work in a high-performance computing system. Going forward, in 2022, we want to test the ability of our programmers to develop code that'll work in AWS, or Amazon Web Services cloud, for efficiencies. We think that the AWS cloud (will be) probably even more efficient and faster than the high-performance computing systems are. And also, if we can get our data in what are called S3 buckets in the Amazon Web Services cloud world, people can access them easier and use them in the cloud. They don't have to download them. What we're trying to do is make our product more available to our user community. 

DAHAL:

If we're successful and processing in AWS, we are also planning to do map every two weeks or so instead of just two maps per year.

HULT:

So now there's a possibility, as we move forward, of seeing even more data releases because of this computing power that continues to improve. Matt, can I ask you to weigh in on this? Do you have any thoughts on the cloud computing, the high-performance computing, from your perspective as a person who works with data? How big a difference has that made in your world, as a person who brings these datasets together?

REEVES:

Well, it's made gigantic improvements, as both Steve and Dev mentioned. For example, we used to have to develop models for large areas, and then apply them at each pixel. This is a little bit in depth, but now for example, on the Google Earth Engine, we're able to model each pixel independently. So you're looking at tens of billions of models being run simultaneously. When we're talking about the 30-meter approach, that's something you wouldn't even attempt with anything other than some kind of supercomputing environment. And the other reason that this is exciting is, although it's kind of in the infancy, when you're operating in a cloud, Google on the Google side can talk to Amazon Web Services, and they can see each others' assets, or buckets as was referred to earlier, and trade back and forth in the cloud. And that will really open up a lot of opportunities. And one of the reasons this will become so important in the future, I think, from the public land management perspective is imagine if you're a manager, and you had instantaneous access to not only this work, but also a slew of others, right at your fingertips, in the cloud. No need to download. And a perfect example of use would be, I want to start up a targeted grazing program and I want to dial up all the resources in my pasture right now that can tell me something about the reasonableness, if you will, of going down the path of targeted grazing, where you identify a large flush of vegetation that happens, say, October, happened in August, and you want to permit livestock to go there and take advantage of those resources for fuel reduction, for animal performance, et cetera. When we start talking like that and connecting that kind of information through the cloud, that will make managers very happy, I think.

HULT:

Just a world apart from what you could do 20 years ago, if you were managing this, this rangeland environment.

REEVES:

That's right. It's going to continue to change things. And I think Steve and Dev both are probably old enough to remember just being on a laptop, crunching through a few rather small datasets was even cumbersome. But now, the game has totally changed. And,  you know, by the time I'm sure most of us retire, we won't even recognize the resources that are available, I think.

HULT:

I think I'd like to ask each of you now if you have any closing thoughts?

BOYTE:

Invasive annual grasses, or exotic annual grasses have been problematic in the United States, and especially in the Western United States, for a hundred+ years now. The problem is not going away, with climate change, with increase in fires, increases in fire sizes, there's just going to continue to be a need for data like we produce for people like Matt. It's really nice to sit here and listen to him and talk about how this is important to his modeling and his work process, that's very gratifying. This isn't a problem that's going away anytime soon. And it probably is just going to get worse. We're going to try to stay at the cutting edge of producing data that the user community can use an access easily, just like Matt just said, bringing the land managers or even the ranchers along technologically, we have to be cognizant of their technical abilities too. So we will keep making our data available on sciencebase, even if it is available in an S3 bucket on Amazon Web Services. 

DAHAL:

I would echo what Steve said. It was so good to know that somebody is using the data in a really good way, and they're appreciating it. This feedback will give us energy and motivation to improve and go beyond what we do now, and do more in the future.

REEVES:

I will just add briefly that this work is the tip of the technology spear in terms of rangeland modeling. One of the next steps that I'll be looking to take will be to leverage this kind of information and others, to be able to work with managers and our constituents, like the producers that operate in our public land allotments, to say, "how can we make better information from this giant mound of data from these different types of efforts?" So for me, that will be the next thing, is to begin to understand how to convert the data into the most useful information that we can.

HULT:

We've been talking to Steve Boyte, Devendra Dahal, and Matt Reeves about how satellite-derived data on invasive grasses can help land managers and firefighters in the Western U.S. Steve, Matt, Dev, thank you for joining us.

BOYTE/DAHAL:

Thank you for having us, John. Appreciate it.

REEVES:

Thank you for including me as well. 

DAHAL:

Thank you very much, yeah.

HULT:

And thank you to the listeners for joining us as well. You can find all our shows on our website at usgs.gov/EROS. That's U-S-G-S dot G-O-V, forward slash E-R-O-S. You can also follow EROS on Facebook and Twitter to find the latest updates, or you can find our shows on Apple or Google podcasts.

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