Eyes on Earth Episode 105 – Imagery Challenges of Alaska

JON DEWITZ

Between the different types of species that call Alaska their home and aren't found in other places in large populations, and then you think about the size of the state—when you realize Alaska is a quarter the size of the U.S., you throw in the added expense and hardship of mapping in Alaska, and people really underestimate the amount it takes to map Alaska well as far as funding goes, I appreciate the support that the USGS gives for Alaska work. I personally wish there was a bit more understanding about the difficulties and the expense of mapping Alaska well.

SHERI LEVISAY

Hello, everyone and welcome to another episode of Eyes on Earth, a podcast produced at the USGS EROS Center, which celebrates its 50th anniversary this year. 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 Sheri Levisay, and I'll be hosting today's episode, which focuses on the joys and trials of remote sensing in Alaska. Three EROS researchers well versed in interpreting Landsat imagery from Alaska are here to share their insights: Donna Viola, formerly of the Land Change Monitoring, Assessment and Projection project; Jon Dewitz, the coordinator for the National Land Cover Database at EROS, whom you heard in the intro to the podcast; and Neal Pastick, a physical scientist at EROS. 

Let's start with the stereotypes. It's cold in Alaska, which means significant snow cover over a good percentage of the state. Does the snow cause problems for getting clear Landsat images?

DONNA VIOLA

So I think that actually depends on what aspect of the landscape you're interested in. So one of the early questions I looked at when I joined the LCMAP project was how we could use their surface change detection algorithm to characterize changes within more persistent regions of snow and ice. So, for some context, the published LCMAP products handle snow and ice observations by ignoring them completely, which works fine in most places in the conterminous U.S., where snow is sort of a limited seasonal phenomena. But for Alaska, where snow and ice are much more prevalent, I tried incorporating snow and ice observations to identify changes in the spatial extent of glaciers and perennial snowpack. So, as you might imagine, most snow and ice related changes over the last several decades is in the form of loss like glacier retreat. So what I would look for is pixels that transition from permanently snow and ice to seasonally snow ice, so you still expect to see some sort of seasonal snow adjacent to a retreating glacier, but then it's rock or dirt in the summertime. So I do think there's a scientific value in the snow observations, at least for certain applications.

DEWITZ

Alaska is one of the most troublesome locations to get good imagery, for a lot of reasons. The first is the short growing season. That growing season tends to be end of May, beginning of June till the end of August, beginning of September, depending upon how far north or south you are in the state. Along with that because it's warm, that tends to be one of the cloudiest parts of the year. So between the short growing season, the clouds and it being snow covered the rest of the year, we're very lucky if we get a single Landsat image that's cloud free over two years rather than just a single year.

LEVISAY

Which would you say is more of a factor, snow or clouds or equally so?

DEWITZ

Well, they both have an effect. Clouds are more pervasive and harder to really plan for in the imagery. With snow, you have a time frame that you can't use any imagery gathered, so it's just a different problem. Clouds are really, really just more annoying because they're random and they just appear, especially when you don't want them to.

NEAL PASTICK

Yeah, clouds can obscure the surface from optical satellite sensors like Landsat. However, certain set spectral bands, like those in the short-wave infrared, can penetrate clouds to some extent, such as cirrus clouds. But trying to develop any type of mosaic composite on a yearly basis using one Landsat sensor or multiple sensors can be problematic. It takes, sometimes, two to three years to get clear images across the entire state of Alaska.

VIOLA

Clouds are definitely a factor when it comes to acquiring good images of Alaska. Unfortunately, that's not something we can really plan around. But there are some algorithms that help with figuring out which pixels within a Landsat scen are cloudy once the image has been acquired; for example, the Landsat Analysis Ready Data uses an algorithm called FMask. It's pretty good at what it does in most landscapes. But there are some limitations. The algorithm tends to misclassify colder surfaces as clouds, which impacts high elevation areas even within CONUS. But of course that's a much more prevalent issue up in Alaska. I've also seen some similar cloud misidentification in things like sediment-rich rivers like the ones that are sourced by glacial meltwater. Which, again, pretty common across Alaska, so in fact we're not only competing with the higher cloud density in general, the algorithms that we use can make it look as if there are even more clouds than there actually are.

LEVISAY

Does the angle lf acquisition matter with the Landsat satellites’ polar orbits?

PASTICK

It does. The high latitude of Alaska means the Landsat satellites see at a steep angle. This can distort features, affect illumination, but on the flip side, the polar orbit of Landsat satellites also offers frequent revisits, which is beneficial for satellite monitoring of Alaska's landscape.

DEWITZ

The angle definitely matters in general, but during the season that we're looking at Alaska, that part of the year, those two or three months, the sun angle is minimized. As the year goes on increasingly into the spring or into the fall, that sun angle does become a much, much larger problem. The shadows from the mountains—and Alaska has very high relief across large parts of the state—that shadow definitely plays a large role in what we can and cannot map across the state.

LEVISAY

Or is it an issue because of the darkness several months each winter?

PASTICK

Darkness during the winter months in Alaska does pose a challenge. Landsat, being an optical sensor, needs sunlight to capture images. Consequently, it can't acquire useful data during the polar night.

VIOLA

So the way that the LCMAP change detection algorithm looks at the land change over time is by fitting models to at least a full year of observations in order to capture the natural variability across seasons. So when you're missing a big chunk of seasonal data, the model-fitting procedure doesn't always know what to do across these big data gaps, and I've seen annual data gaps as long as five months, you know, particularly as you get farther north, you know, towards like the Northern Slope of Alaska. So yeah, definitely an issue.

DEWITZ

So the time that the sun is above the horizon and shining where we're used to it here in the Lower 48 at that same sort of sun angle, that time frame again is compressed into those few months in the summer.

LEVISAY

So during the winter, do you still acquire imagery, but just find that most of it isn't helpful?

DEWITZ

Landsat 8 is the first satellite that acquires imagery at night at the later hours of the day. Certainly, if it's not light out, that imagery isn't useful for those of us mapping vegetation, so it's the first time we've really had a chance to acquire, let's say, night-time imagery. It certainly isn't helpful for mapping vegetation. But previously it was never done across Alaska because there's just no reason.

LEVISAY

What is the best window for acquiring imagery in Alaska?

PASTICK

Generally, the best window for acquiring imagery in Alaska is typically the summer months or growing season months, when there's less cloud cover. So that's generally what individuals will use is peak growing season Landsat imagery to make seasonal composites for the state, depending on what they're actually trying to investigate.

LEVISAY

We talked about some of the problems. How do you solve them?

PASTICK

Good question. Without going into too many details, remote sensing scientists often resort to innovative solutions, such as fusing observations from multiple sensors or using spatial and temporal gap-filling procedures. But a really easy example would be using historical data from the same year and the previous year. You're making the assumption that the landscape hasn't changed over that time. Which is a fairly naive assumption, so there's more sophisticated methods such as model-based approaches for interpolating observations. Basically, if you have two points, you're trying to draw a line between those points that best fits the phenology of the landscape. Another example would be we can use the synthetic aperture radar data available from satellites like Sentinel 1, which can provide insights even in cloudy conditions or during the night, which really complements Landsat’s capabilities. The same can be said for making use of other broadband multispectral data acquired from Sentinel 2 or Planet Labs. Combining all these data fusion techniques really allows us to obtain the best possible insights from remote sensing in these challenging conditions.

VIOLA

So one thing that I think can be really helpful is taking advantage of big data and sort of scouring the Landsat archive for as many clear observations as you can. So, full disclosure, I very, very rarely look at individual Landsat scenes in Alaska or anywhere anymore. Instead, I look at time series data for individual pixels that pull out all observations from Landsat 4 through 8, filter to the best of our ability to remove clouds and shadow, and even snow for cases where that's not what I'm trying to look at. And while a graph of time series pixel data doesn't always look as pretty as a full scene, showing genetic landscapes, spatial variability, it does have a lot of scientific power. It helps to maximize usable data. So, for example, for any given pixel observation, I don't care if the next pixel over more than 90% of the scene that it came from this cloud cover. I could still use the data from that individual pixel.

LEVISAY

I guess that's the value of the time series. That's the whole point of Landsat versus any other source.

VIOLA

Yeah. And especially since, like, you know, 19 in the early, like the ‘80s and ‘90s are even pretty spotty, for Alaska at least. But at least since Landsat 7, you know, we have the data density to really able to do some powerful things with it.

DEWITZ

That lack of imagery has a pronounced effect on mapping Alaska. Previous to 2000, there is little to no consistent Landsat imagery. There are single scenes here or there, but there is not enough of a Landsat record to do the mapping that we're currently doing over CONUS. Those problems can be resolved a lot of different ways. And all of them require more looks at the ground. The USGS is committed to getting more than one satellite acquiring imagery at the same time, so the more times we image, that is a huge help for trying to find that small bit of time between the cloud coverage while it's still green on the ground.

LEVISAY

We've just celebrated EROS' 50th anniversary, and I think it's fair to say that EROS has been obsessed with Alaska since the beginning, with pages and pages of reports devoted to the topic. What makes it so important to observe Alaska via remote sensing?

DEWITZ

There's a few reasons that Alaska is very important when you look at landscape of the United States. First off, it's the largest state, and it has some of the most varied species that survive there. And one of the reasons Alaska is really important is that it's one of the least-understood states in the United States as far as mapping goes. Because of the incomplete coverage through the Landsat record, because of the massiveness of the state and the inability to drive to most places in the state, mapping there becomes much, much harder. You have to fly to places just to get training data for other areas across the state that in the lower 48 we would just drive there, take a picture of and understand the landscape. Many places in Alaska have not had people there other than just flying over. So, given that trouble, given the lack of investment in high-res imagery, that's one of the other troublesome things about mapping in Alaska. Previous to 2015, there was little or no high-res imagery available for the state. In the lower 48, the USGS has a Digital Ortho Quarter Quad [Digital Orthographic Quarter Quadrangle] program where we've had high-res imagery since the ‘80s across the entirety of CONUS. That's not true for Alaska. The high-res imagery allows you to ... well, simple things like whether it's a tree or whether it's a small shrub, because spectrally they can be similar. In Alaska, we had nothing like that. No high-res imagery. So we had to rely on folks going out, doing field work, figuring things out through Landsat where the trees were, where they weren't, where the wetlands are. And when you compress that with that short growing season and the lack of imagery, it makes mapping them much, much harder across the state of Alaska. Knowing what's there, knowing the amount of carbon that's being stored, knowing when things green up every year, seeing glaciers or seeds, seeing when snow falls at the beginning and at the end of the season—those are very important things for mapping, for climate change and for carbon accounting.

VIOLA

Remote sensing in general is a very powerful tool for learning about places that are really hard to get to, and in particular for larger spatial areas. Really, so much of Alaska inaccessible due to complex landscapes, little infrastructure, to do that is really the only way to see what's going on in some of these areas, especially if you're interested in longer term monitoring.

LEVISAY

Have all of you spent time in Alaska over the years? Tell us a bit about what remote scientists can learn in the field in Alaska.

PASTICK

Oh, you're making me start to daydream. I've been conducting field research in Alaska the last 15 or so years. Field work really provides ground truth data, which helps us validate and calibrate remote sensing data and also helps scientists understand the context and conditions in which the images are captured. Simply put, any remote sensing scientists who are modeling worth their salt needs to get their boots on the ground to better appreciate the system they're trying to characterize.

DEWITZ

I have spent quite a bit of time in the, doing field work in Alaska. Dave Selkowitz was the original Alaska mapping person for NLCD. I was his sidekick for the first couple of years, but we mapped almost all of the field work together, flying through almost the entirety of the state, you see a lot of things that probably most Alaskans haven't even seen. We went to places that were outside of the routes to and from cities to gather field data, to gather knowledge about what's going on in the landscape for these generally unmapped areas. That involved a lot of flight time, a lot of coordination, and a much, much greater expense to do very basic things for mapping that we don't have to do in the lower 48.

LEVISAY

So you go during the summer, right? You don't go during the winter.

DEWITZ

We try not to go during the summer summer. We try to go later in the fall, where we have a lower chance of cloudiness, and we try and hit that first part in the spring where things are just greening up. The summer in Alaska is miserable for a couple of reasons. First off, it's hot. Second off, there's mosquitoes the size of large helicopters up there that just are not pleasant at all in the middle of the summer.

VIOLA

So I've actually never been to Alaska. I have spent my career studying cold places from computer screens in much, much warmer ones. Actually, all my grad school research sites were on Mars, which is even harder to get to than Alaska, but I would love the chance to go someday.

LEVISAY

Thanks for joining for this episode of Eyes on Earth. Check out our EROS Facebook and and Twitter pages to watch for our newest episodes. You can also subscribe to us on Apple Podcasts. This podcast. 

VARIOUS SPEAKERS

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