Eyes on Earth Episode 53 - Remapping Canada's Fire History

JOHN HULT:

So, you're pretty sure that the lakes did not burn in any of these. Pretty sure there wasn't a fire in there? 

ROB SKAKUN:

It's a pretty easy thing to remove now days. We have geographic information systems and we have water maps. Back in the day we didn't have that luxury, so it's understandable that area burned included some of these non-flammable ground features.

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 John Hult. Mapping fire perimeters is important work. It guides post fire restoration efforts, fire mitigation strategies and helps us to keep track of trends in burn severity over time. An inaccurate fire perimeter can lead to inaccurate conclusions. Now, in the past, many of Canada's fire agencies relied on techniques like sketch mapping which estimate burned area without exact measurements. In the pre-satellite era pilots would sometimes fly over the edge of a fire while a passenger would record the fires extent. A recent study from Natural Resources Canada dove into the deep historical well of Landsat imagery to find out just how far off those conventional fire perimeters where. The difference was significant. The average area per year burned from 1950 to 2018 was 11% less than official estimates. One year was off by 1.4 million hectares according to their estimates. Here with us to talk about the study and its implications are two of its authors Rob Skakun and Ellen Whitman of Natural Resources Canada. Their study looks into not only the disparity in fire perimeters but also how Landsat can improve their nations approach to burn mapping in the future. Rob and Ellen welcome to Eyes on Earth.

SKAKUN:

Hi John. Thanks for having us.

ELLEN WHITMAN:

Yeah. Thanks.

HULT:

First off. For the benefit of our stateside audience and anyone else who is not in Canada, let's talk about Natural Resources Canada. What does your agency do and what is maybe the US equivalent of Natural Resources Canada?

SKAKUN:

Natural Resources Canada is a federal department. With branch departments in forestry, mining, earth sciences. Ellen and I work in the forestry sector. In the department called the Canadian Forest Service. Here at the Canadian Forest Service, we conduct research, as you can imagine, in all aspects of forestry. Wildfire science, insect and disease identification, climate change research and forest geomatics. So, myself, I work with the geomatics team as a remote sensing analyst, primarily working with Landsat data. And Ellen is a research scientist with fire science. 

WHITMAN:

We're quite parallel to the U.S. Forest Service in terms of what type of work we do. We do a lot of research that guides policy development, but the Canadian Forest Service is not a land manager. So National Forests like they exist in the United States, for example, are not the responsibility of the Canadian Forest Service. So parallel but a little bit different.

HULT:

You study the forests but don't manage the forests. So, sort of have to work hand in hand with the managers. Is that right?

WHITMAN:

Yeah. We collaborate really closely with land managers. In Canada, most land management is done by providences and territories. So, they generally are the ones that own public forest lands and are responsible for managing them and also for fighting fire, for example, on those lands as well. There's a few exceptions there. For example, Parks Canada, which is a national agency, also does land management. We also have the Department of National Defense and First Nations Reserves, as well, which are somewhat a federal responsibility. But, broadly speaking the forest service works really closely with our provincial colleagues in order to work with the people who are really responsible for land management.

HULT: 

Right. So you're sort of handing off information, research ... the kinds of work that you do. You hand that off to a lot of different interested parties across Canada.

SKAKUN:

Yeah, exactly. One of the things that I do is I work very closely with some of the provincial and territorial fire management agencies when it comes to fire mapping, where we're generating Landsat fire perimeters. That serves both their own mandate for reporting on area burned, but also our own mandate, where we have to report on carbon emissions.

HULT:

So you're getting into the next question here: I wanted to know how fire perimeters have been mapped in Canada in the past and maybe how the approach has evolved over time. Who maps fires in Canada? Is it a little bit of everybody pitching in on this? Is the approach consistent? Has it been consistent? Is it consistent now? Does everybody call you, Rob, or call you Ellen, and say "ok it's time to map these fires!" How's it go?

SKAKUN:

Well, it's really cool in Canada is that we have this really rich inventory of fire data. Back to the early 1900's for some parts of the country. But in our study we look at the fire data beginning in 1950. And that captures the reporting period for many of the fire management agencies. How the fire mapping really began in Canada and I'll start talking about the 1950's but this also was prior to the 1950s, was sketch mapping. And this was typically done from a fixed wing or a rotary aircraft. Imagine a pilot and a fire officer in a plane, and the pilot flies over the fire while the fire officer draws on a map sheet the extent of a burn scar. Now you can imagine this would be fairly tricky because you need to know reference landmarks on the ground to know where you are on that map sheet, and then subjectively draw your boundary of the fire. These boundaries would later be digitized from map sheet to electronic format. This approach was used for decades. Similar to this mapping approach but then starting to appear in the 70s and 80s was GPS delineation. This approach is similar in that an aircraft circles the boundary of the burn but is being tracked by the GPS system. The issue with the fire data that we have from both of these methods is that the mapped perimeter is often a broad delineation around the burn. And it's usually capturing the small water bodies and the unburned forest within that perimeter boundary. This leads to an overestimation bias of the burned area. Even today, some of these conventional approaches, they're actually still used. When satellite data became available in the 70s and 80s, Some of the agency's started to use Landsat and also very fine resolution data such as aerial photographs, and ultimately this finer resolution data provided a better delineation of the burn perimeter than what they were previously getting.

WHITMAN:

The provincial and territorial and Parks Canada fire management agencies very generously and consistently provide their map perimeters to Natural Resources Canada. And that's been consolidated into a national data set called the Canadian National Fire Data Base. But in addition to that there is also a second product that Rob is specifically responsible for, where he uses those remotely sensed more high-resolution options like Landsat to produce perimeters, which creates a product called National Burned Area Composite, which is a little bit more refined but is heavily derived from that initial reporting for the provincial and territorial agencies.

HULT:

Was there a concern that the perimeters were inaccurate and were having an impact on the decisions that were being made?

WHITMAN:

I wouldn't necessarily frame it as a concern so much as, we were all aware that they were inaccurate. But I would absolutely say that the agencies are well aware of that too, especially for these older years where satellite data that wasn't necessarily available. But we wanted to be able to be a little bit more confident in the types of analyses we could do with them. So, for example, if people are interested in looking at time series analysis of area burned, which is something that is being discussed quite a lot with climate change. We wanted to leverage this amazing really extensive data set to understand what the scale of the error was associated with those older methods. Some of these years that had a really extensive burn historically before we had high quality mapping of perimeters. That over reporting of area burn in an early period could sort of dampen your trend. If your trend was increasing over time because you have these false spikes earlier on in your time series.

HULT:

You're saying that everybody kind of knew that there were inaccuracies built in, that's kind of what you expect based on the reality of where these perimeters came from, but you're looking at time series analysis and trying to figure out something like carbon: impact to carbon, how much carbon has been released etc, etc. There would be spikes that maybe weren't actually related to what actually happened. But were more related to the inaccuracies. Or the possibility that was in there.

WHITMAN:

Yeah, I think you nailed it. It's not necessarily a surprise that these areas exist. But it's more like we wanted to understand better what they are and what effect they could have on things. Like national carbon reporting or analyses of how much something has changed over time.

HULT:

That's a great point and that's something I wanted to ask you about: why does this matter? Why is this important? And it sounds like one of the issues is that it would affect your estimation of carbon emissions. What else would be affected by this? Why else would this matter? Does it matter on the ground to have an accurate fire perimeter from 1972?

WHITMAN:

Canada is responsible for national carbon emissions reporting, and that includes both natural and human caused wildfires. But it also affects other things as well. For species at risk, people use area burn estimates to determine target disturbance levels or alternatively to target policies that would protect animals that were negatively affected by area burned. And if those historical estimates are based on these maps that we have these known issues with, it can actually really affect policy which can have super long-term down scale affects, like the closing of an area to forest harvesting or a policy of suppressing wildfire where maybe that may not be appropriate.

HULT:

Rob, anything you want to add to that one there?

SKAKUN:

Yeah, well Ellen kind of lead me on there to talk a little bit more about our Landsat perimeters and how this relates to carbon accounting. So, it actually goes back a number of years, actually. It was the early 2000s when the Canadian Forest Service started creating fire perimeters from Landsat, and this was just to improve the area burn mapping, where fire management agencies had mapped using sketch mapping and the aerial GPS methods. These map perimeters from Landsat since we started creating them, they feed each year into CAS - carbon accounting system. And that is used for the annual reporting on emissions from forest wildfires. So again, the reason why we wanted to use Landsat is just having this  better data leads into the better estimates of area burn and better inputs to more accurately model carbon emissions. What's interesting even today is we still continue to create these perimeters from Landsat because some of the agencies still apply some of the conventional methods. But when we actually started doing this work back in the early 2000s, we actually couldn't remap all the conventional perimeters of the agencies, and that was largely due to cost restraints. And we had to target say, the largest fires that we could remap because they had the largest impact on carbon emissions. But once the open data policy came later on in the 2000s, we were able to retroactively update more of the conventional fire perimeters dating all the way back to 1986. So for our study, this is how we generated our calibration, or this is how we were able to obtain our calibration data, because we had already Landsat fire perimeters that were created for carbon accounting that we're now going to use for our study to create an area base adjustment model. So from 1986 to 2018 we collected all of our paired samples of Landsat and conventional perimeters, developed an area base prediction model that we then applied back through time.

HULT:

Let's talk about that a little bit: the study itself. What did you learn going back in time? I threw a few numbers out there in the beginning. What else did you learn?

SKAKUN:

The first thing that we observed was that there was an over estimation of 40% of the average perimeter area between a Landsat and the potential perimeter. So, on average. 40% was being over estimated.

HULT:

So, just like generally speaking, Landsat vs. conventional, on average, chopped 40% off.

SKAKUN:

That's exactly right.

HULT:

Wow.

SKAKUN:

There's about three factors that contribute to that. One being again, the conventional does this broad delineation around the burn scar, and that leads to an over estimation. The other being that water. Small water bodies within that burn perimeter are often not removed, again leading to over estimation. The third being the unburned forest, which would be your residual forest that doesn't get burned by the fire but is still being captured within that burn perimeter. One of the other things that we did observe was that larger fires resulted in larger area differences. That wasn't too much of a surprise. But the percent change in their difference was greater for the smaller fires compared to the larger. So imagine a fire that's five hectares-very small fire, five hectares-but the agency mapped it to be ten hectares. The area of difference is only five hectares. Which again is quite small. But the percent change is 100%. Now if we compare that to say some of our larger fires. If it's a 50,000 hector fire, a 100% change would be a 100,000 hectare fire. We never saw an area difference of 100,000. It would be like 10,000 or 20,000. When we looked at the distribution of the data from small to large fires. The data was highly correlated. This set us up to then develop a model that we could use to apply back in time.

HULT:

Ellen, did you want to add anything to that one?

WHITMAN:

Yeah, once we had developed that model that Rob just did a great job of describing the training data that went into it, we essentially just made a regression that corrected the estimated fire area burned to the level that was presumably more accurately reported by the Landsat area burn. So essentially it is adjusting what was reported using these various conventional methods to match Landsat. Once we made that adjustment, we were then able to look at averages in individual fires and create these annual time series of adjustments to area burned. We found some pretty interesting stuff. Rob spoke very clearly about how these large fires contributed to very large overestimates of area burned, and that scales across large fire years as well. So typically, these fire years that really stand out as a very extreme fire year with a very high area burned also, tend to be very large fires creating that really extensive fire year. So then when we had fire years that were already identified as the most severe fire years, they also tended to be those years that had that dramatic accumulative overestimate of area burned, which led to some of those numbers which you reported in the intro. Like in one year a reduction of 1.4 million hectares. That's kind of an extreme case. More often we saw that it was closer to say an 11% reduction in area burned for an average year across that time series. And the other thing that is worth highlighting within this is that as we moved through the time series there were more and more fires in that Canadian National Fire data base that didn't actually require any adjusting. So we applied these modeled calibration adjustments to fires that had been mapped with conventional methods. But we didn't adjust fires that had been mapped with aerial methods. So whether it was reported by the agency that they used a different satellite sensor, or if they had flown it, for example, with really high-resolution aerial photography, we didn't adjust those fires. You can see in our data that as time goes on agencies more and more are shifting towards these really successful and high-resolution methods for mapping their own fires. More and more the percent change seemed to decline in the later years of the time series.

HULT:

You talk in the paper a little bit about how this can be useful in the future. How do you see this being used in the future? Is this kind of a road map for someone else who wants to pick it up and run with it and if they're running, which way are they running? What are they going to do with it, do you think?

WHITMAN:

The first step for us honestly will be to make this data a permanent home, and so that will probably be including some of these modeled adjusted area burned estimates as an additional field in some of the national data. So whether that is the Canadian National Fire Data Base or the National Burned Area Composite, we haven't totally worked that out yet, but our initial step is just making it so that people can use these estimates. Personally, I really would love to see this being put into some research to look at trends in area burned over time, because that's really, I think the strength of these data, and it's a major question in Canada. There's been some excellent research done by some other colleagues from the forest service and also lots of academics as well, suggesting that there absolutely have been increases in area burned over time. And these adjustments could then be used to increase the confidence of exactly what that trend is, and I hope, look at how it varies spatially as well. We do see these patterns where climate change is increasing temperature pretty much everywhere, but the seasonality of that is quite variable within Canada. We also have a lot of areas that are getting quite a lot wetter at the same time. And so, in order to provide a really strong foundation to examine these trends within these data set, it could really enhance our confidence in understanding how it's affecting fire in Canada. That hopefully could lead to some new policy angles and guide our management going forward as well.

HULT:

Rob, you mentioned this before, but I want to bring this up again to put a finer point on it: how important are the Landsat program and the USGS open data policy to work of this nature? Would you have been able to do this work without Landsat and open data?

SKAKUN:

The Landsat open data policy has been huge. For Canada being such a large country, it is amazing that we can get spatial coverage from Landsat from coast to coast to coast, sating all the way back to 1984 with the thematic mapper sensor. Having that data cataloged, we've been able to use it to a thirty year plus time series, using very fine resolution data to get these very precise estimates of area burned. Without the Landsat program, I don't see how we could have made a study like this possible. The other thing too is that with Landsat, we can map fires that sometimes agencies miss. For various reasons, an agency may not map a perimeter for a fire whether they couldn't get out to fly to it. Or a resource issue or an area they don't manage. We can now capture those areas with Landsat. It serves as a gap fill. There are many good reasons why Landsat has been very beneficial to our study and to the Canadian Forest Service for fire mapping.

HULT:

We've been talking with Rob Skakun and Ellen Whitman of Natural Resources Canada about how satellites like Landsat can improve fire perimeter mapping. Rob and Ellen, thank you for a fascinating conversation.

WHITMAN:

Thanks, John

SKAKUN:

Thank you, John

HULT:

Be sure to drop in for the next episode of Eyes on Earth. You can find us on our website at usgs.gov/eros or you can find us on Apple podcasts or Google podcasts. This podcast is a product of the US Geological Survey/Department of Interior.