Eyes on Earth Episode 54 - National Land Cover Database 2019

JOHN HULT:

Hello everyone and welcome to another episode of Eyes on Earth. Weíre 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. Iím your host, John Hult. The National Land Cover Database, or NLCD, was the first and remains the most well-known set of satellite-based land cover mapping products released by EROS. It sorts the each 30-by-30 meter plot of land in the United States into a land cover class, such as cropland, pasture, high-intensity developed, deciduous forest and the like. It also includes information on impervious urban surfaces, forest canopy cover and more. NLCD is about to release its newest data. It maps conditions on ground for calendar year 2019, and adds new layers that map features like the Nationís wind energy. The land cover maps for NLCD 2019 can be compared with data from seven other years reaching back to 2001, to understand landscape change in the U.S. For more background on NLCD, be sure to check out Episode 3 of Eyes on Earth. For today, letís talk about the new release with NLCD manager Jon Dewitz, a research physical geographer for EROS who has been involved with NLCD for around 20 years. Thatís a while. Jon, thatís a while.

JON DEWITZ:

It is.

HULT:

Welcome to Eyes on Earth, Jon. Itís great to have you.

DEWITZ:

Well thanks, John. Itís great to be here.

HULT:

Letís do a real quick refresher about what NLCD is. When did it start, what does it do, who uses it and why?

DEWITZ:

It started with a consortium of folks that came together around (the) early 90s. At that time, satellite imagery was still only available for purchase. It wasnít for free like it is now. Mapping was very expensive, so these federal partners came together to purchase this imagery as a group, and that led to the MRLC consortium, the Multi-Resolution Land Characteristics Consortium.

HULT:

Okay, so thatís how it starts. And then where did you go from there? What does the National Land Cover Database do? Whatís the mission?

DEWITZ:

The mission is to map the United States. That includes Alaska, Hawaii and all the territories. 

HULT:

Okay, and who uses land cover data?

DEWITZ:

Believe it or not, everybody uses land cover data in some way or another. For us, the things that we map, it helps monitor the amount of developed area, which leads to surface runoff, which can lead to pollution. It helps monitor burned areas, forest fires, the amount of forest thatís in an area. That can be used for carbon sequestration, things like that. We have folks on both sides of the spectrum, from ecologists who are interested in endangered species, to say, gas and oil groups that are interested in finding areas that least disrupt endangered species or other species. 

HULT:

So thatís the history, thatís whoís using it. The MRLC consortium goes back to, when did you say? 1999?

DEWITZ:

Itís going to be mid-90s.

HULT:

Okay, somewhere in the mid-90s. So that comes together, you put out NLCD 1992 as the first one? Right?

DEWITZ:

It is.

HULT:

And then 2001 after that?

DEWITZ:

Yeah.

HULT:

And 2006?

DEWITZ:

Yeah.

HULT:

And ë11?

DEWITZ:

Yeah.

HULT:

í16 and now í19.

DEWITZ:

Right. 

HULT:

So thereís our, our quick refresher for what it is and whatís available. Now, weíre looking at NLCD 2019. But letís talk about í16 for just a second because it was just a couple years ago and it was a big step for the program. Talk to us about how that laid the foundation for more frequent updates.

DEWITZ:

To talk about 2016, let me give you a bit of a history on 2001, which was our first release based on our newer definition types. Those newer definition types more closely match to what the satellite naturally captures. And in 2006, we did basically a change detection. Same for 2011. For 2016, we went and we remapped all of this, from 2001 to 2016, and we added mapping every two and a half years. This allowed us to do succession and trajectory, which really helps to improve both the accuracy and the consistency of the land cover.

HULT: Just briefly, succession and trajectory, youíre talking about Ö

DEWITZ:

The natural state of vegetation. So letís say thereís a forest fire in 2002, right? In 2001, we still had it mapped in forest. In 2003, youíre going to see that that forest disappeared. What can happen is if you donít follow natural succession, you can get things that donít make sense. So you might see that in 2004 we had it mapped as shrub, and 2006 is grass. Well, that doesnít make sense. Hopefully we wouldnít do that, but it does happen once in a while. So with succession and trajectory we can figure out, okay, if a forest disappears, the next natural state is a grassland, which will regrow to a shrubland, which will then regrow to forest. 

HULT:

Right, right, so it made more sense over time. Succession and trajectory allows you to sort of string things together in a more logical way, and I suppose improve each mapping product as you move along. 

DEWITZ:

It does. And it can also fill in some of the gaps from the imagery. There are many times in a year weíll only get one clear image of a leaf-on area. If we donít have that, sometimes we have to say ìthis is what the succession and trajectory is showing. This is what it should be this year.î And we can go to the following one and say ìOh, itís following the right pattern,î even if we donít have that in-between signal to get a map from it. 

HULT:

Sure, sure. So it helps you get to where youíre going and have more confidence. 

DEWITZ:

Yes.

HULT:

But succession and trajectory, this comes in NLCD 2016? Youíre able to do this? 

DEWITZ:

Yes.

HULT: 

Okay. And thatís because you go all the way back and you sort of go back and make all of the maps make sense together, you sort of integrate them in a way that makes them more comparable?

DEWITZ:

Right. So, you know, itís like stacking blocks on top of each other. If you have seven blocks, representing seven years, take the middle one out, the rest donít stand. And thatís kind of what we were doing. We still did very accurate maps before 2016, but with succession and trajectory, we can stack that land cover and make sure that itís doing the correct things, and it increases our accuracy, as well.

HULT:

Okay, so 2016 was a big step forward, and also in 2016, you added some interesting new layers. You added an impervious surface descriptor layer, right? You have the impervious surface layer, which is the %age of a pixel, of a 30-by-30-meter plot thatís covered with impervious surfaces. You have that, and within that, in 2016, you added a layer of descriptions. You made it even more complicated Ö

DEWITZ:

Or simpler, depending upon who you are.

HULT:

Okay, okay. Well, to my mind, I think ìwell, thereís more roads there, you have oil pads mapped in this descriptor layer Ö thatís interesting that you call it simpler. Tell me how.

DEWITZ:

So if you look at a map of developed impervious surface, itís simply a range of values from zero to 100 %. If youíre looking at an interstate going down, say, in the prairie, all you see is that zero to 100 %. You know it might be 40 % in some places, it might be 30 %, but you donít know that itís a road. Your computer has no way to tell that itís a road, and roads really make a large difference in determining the difference between growth in an urban area, and somebody just repaving a road. 

HULT:

It gives you more detail on whatís actually there, whatís actually sort of forming the rest of the map?

DEWITZ:

Right.

HULT:

Okay. Well, letís get into the details on NLCD 2019. Whatís new and different? What did you do differently in making 2019, and what should users expect when they open the box?

DEWITZ:

For the user, it should be transparent that itís the same as the 2016 release. The one difference that they will see is that there will be added years of impervious surface. For the 2016 release, we were only ably to classify every five years because of time constraints. Weíve matched up our impervious surface products to our land cover products so that we have a corresponding land cover and impervious for every single year. Other than that, users should see exactly the same look and feel. There will be a few more impervious descriptor layers. Weíve added things like wind turbines, but our goal is a consistent map, and the same thing, the same experience for our users. Behind the scenes, we put a lot of work into streamlining how quickly we can do this. We went from a scene-based classification, which is Landsat path and row, to a composite, where we take each single pixel and put it together. And using each single pixel to make an image allows us to fill in some of those areas that are perpetually cloudy, and use those few good pixels that we have. It also reduces the time to make those path/row images, so we could put out a newer land cover product in a shorter time frame. 

HULT:

You have explained this to me in the past, the difference between a composite and a scene-based, you sort of compared it to like a digital camera looking at one picture versus 100 and picking the best? Is that kind of the idea?

DEWITZ:

Thatís kind of the idea. Scene-based is like taking a picture of your family, right? You have seven pictures, and you have seven people in your family, and each time, somebody has their eyes closed in each picture. So you have to pick the best one, and thereís always something wrong with one of those pictures. With a composite, itís like taking the best individual snapshot of each person in that snapshot and putting them together into a single best picture. 

HULT:

Translate that over to land cover mapping, and doing this gave you more to choose from to fill in the gaps, and you were able to pull this together much more quickly. 

DEWITZ:

Thatís true.

HULT:

So weíre looking at a year and a half or so, two years, between when the data was collected and when it appears as a land cover product?

DEWITZ:

Yeah, weíre right at a year and a half now. 

HULT:

Well, how come you canít just do land cover tomorrow from data you got today?

DEWITZ:

Well, a lot of things go into making a land cover product. We talked about succession and trajectory, but, you know, one of the other things is we need that time in between the last pixel and the first pixel for a time frame, to kind of pull all those different things together in the imagery. It takes processing time, it takes knowledge, and it takes a lot of work, you know, just going through and making sure all the processes are working the way they should to get an accurate land cover map.

HULT:

Itís like 9 billion pixels, 9 billion plots? Something like that?

DEWITZ:

8.9, yeah.

HULT:

And I suspect that a plot of grassland and a plot of cropland look very similar in just a typical Landsat image Ö thereís a lot of processing thatís involved.

DEWITZ:

Itís very evolved. You know, if you think of what a satellite sees, wheat is a grass. Wheat comes up in the spring, just like many grassland areas, just like many pasture areas. And if you only have one image, itís very hard to tell the difference between the two. If you have multiple images at different time frames, that really helps to differentiate those different class types because they senesce at different times. Wheat will be Ö

HULT:

Theyíll brown down?

DEWITZ:

Theyíll brown down. So wheat will be brown mid-summer, and a grassland will stay green earlier and later in the year. 

HULT:

Letís talk specifically about the wind energy layer. Why was this included, and what does it take to do something like that, to add an extra layer to NLCD?

DEWITZ:

The wind energy layer was included because there has been a large growth in wind energy, and it does have an impact on the landscape, much the same way any other developed feature does. Wind energy has gone, from when we published 2016, to being fairly minor Ö itís really doubled in the last three or four years. And people want to know where thatís being developed. They want to know where wind turbines are, how many there are, and how that might have an effect on the landscape. Researchers can then combine that data with our other land cover types to find things that we may not think about.

HULT:

Thatís kind of the value of having a national-scale map is that you can see these things in context. Wind energy and oil pads and whatís changing in and around them, right?

DEWITZ:

Context is one of the most important things in geography and land cover. That ability to add different data layers to NLCD is really, probably, I think our biggest strength. We add many different data layers from our MRLC partners. Things like NASS crop types, LANDFIRE veg types, we used LCMAP imagery to start as the basis for NLCD 2019. We have many partners, including NOAA, the Forest Service that all make their own products, that we then integrate, (we) take the best pieces of those and make our wall-to-wall land cover.

HULT:

Wall-to-wall land cover thatís complete and consistent, produced with the help of a lot of your friends it sounds like.

DEWITZ:

Very talented friends, I might add. 

(LAUGHTER)

HULT:

Well, yes, these are EROS people, a lot of these folks, so we can say theyíre very talented. And youíre collaborators with MRLC. Letís unpack what you talked about there, you dropped a lot of acronyms. LCMAP, thatís Land Change Monitoring, Assessment, and Projection, that a project at EROS. 

DEWITZ:

It is.

HULT:

And LANDFIRE, thatís Landscape Fire and Resource Management Planning Tools, thatís also an EROS project.

DEWITZ:

Partly at EROS

HULT:

Partly at EROS, yeah, itís a partnership between several agencies, right, but the productionóthe mapping portion of itóis at EROS.

DEWITZ:

Right.

HULT:

So basically you have all these teams producing different land cover maps, and you collaborate and work together to produce the best versions of what youíre making.

DEWITZ:

Right.

HULT:

Talk to us about that. What kinds of things do you learn from these other talented people?

DEWITZ:

When you look at a land cover map, they tend to look similar. But when you dive into the classes and what theyíre mapping, theyíre very different products. LANDFIRE itself maps, I believe, over 120 different vegetation types. They categorize structure of these, and we donít do that much. Theyíre specialized to provide that for a reason. Same with, say, NASS. They have 100 different crop types. We only have two crop types. NASS uses us for their non-crop types. NOAA is an expert in wetland areas. The Forest Service, of course, as the name implies, knows a thing or two about forest. Each of these agencies has a specialized part that goes into our land cover, and we combine each of these specialized parts into a single, harmonious land cover. 

HULT:

NLCD and LCMAP and LANDFIRE, the EROS-based projects that youíre working with, youíre all putting out new data for 2019 this year, and for all of you thatís a pretty quick turnaround. Would you guys have been able to do this had you not been working together?

DEWITZ:

Not as well, and certainly, probably not in the same time frame. I think both of those are true.

HULT:

Which of the new features in NLCD 2019 will be the most useful to your users?

DEWITZ:

I think the most useful thing will be the completion of those impervious surface layers to match all of the land cover years. That really fills in and completes our database concept. We didnít have time for the last publication, for NLCD 2016, to do those, and thatís something we really wanted to do for this time.

HULT:

Why do have impervious surfaces as a separate layer? Why is that important enough to map?

DEWITZ:

If you think about soybeans, itís either soybeans or itís not soybeans. But with developed, there can be a range of developed intensities, and thatís what weíre really mapping. Thereís a large difference between something thatís 75 % developed, which covers 75 % of the landscape, and 5 % developed, which covers only 5 %. That equates to a much larger amount of surface runoff, impact on the environment, need for water management, al of those things come into play. 

HULT:

Youíre kind of the only land cover product that does that nationwide, provides that level of detail on impervious surfaces, is that right?

DEWITZ:

Thatís right.

HULT:

Because Ö like you talked about LANDFIRE. LANDFIRE has hundreds of vegetation classes, but they donít have this much detail within the boundaries of an urban area. And LCMAP has data going back 30 years for every year, but thereís just one developed class, right?

DEWITZ:

Right.

HULT:

Okay. So you kind of fill in the gaps between the different products.

DEWITZ:

The impervious surface was a gap that we saw when we started NLCD 2001, and it really has become really widely used. Thatís our two specialties: our first one is combining all of these other specialized land covers, and the second is our impervious surfaces. 

HULT:

How do people find NLCD data? Where do they go, what kinds of tools are available for researchers and GIS professionals, and just the general public? 

DEWITZ:

All of our data is freely available at mrlc.gov. We have a variety of tools, for the very sophisticated users to folks who are completely unfamiliar with GIS. Folks can download our entire map in one piece. They can go into our viewer, pan through our land cover and find the small piece that they want and download it, and then we also have a newly published tool called the EVA toolóenhanced visualization and analysis toolóthis was made for folks who just want to know whatís happening in their local area. They just go in and click on their county, and you can find out whatís changed from any year to any year, get a printout for that and get some relevant information on what that means to the landscape. 

HULT:

The EVA tool is not really a mapping product. A lot of this stuff is a mapping interface and you have to zoom in and find your area, but here you can just type in your county and itíll give you charts.

DEWITZ:

It will. We made this tool for folks who donít want to do GIS or just want a quick answer to a question and donít want to take the time to do the GIS themselves. 

HULT:

How do we know that thatís accurate, those county figures? Because you have to translate it, I suppose, from pixels to square miles. Is it square miles? Is that what the change is calculated in?

DEWITZ:

It is in square miles. 

HULT:

How accurate is NLCD, and how do people know that they can trust this information?

DEWITZ:

NLCD publishes an accuracy assessment with every release of our products. So for NLCD 2016, we just released our accuracy assessment for that, and weíre around 93 percent accurate. 

HULT:

So about 93 percent accurate, and thereís documentation behind this, so if thereís a question about how close it is, you can go back and check.

DEWITZ:

Absolutely.

HULT:

But you did say that youíre not perfect. How many of those 9 billion pixels are completely off?

DEWITZ:

Well, it would be 7 percent, according to that. The interesting thing is, if you look back, weíve increased our accuracy every time weíve published. Thatís because we build on our previous products. So the increase in accuracy from our 2011 to our 2016 release, we classified an area the size of California correctly. 

HULT:

Thatís the improvement. Well, that seems pretty good. Is it good?

DEWITZ:

Thatís great. 

(LAUGHTER)

HULT:

Okay. Just wanted to check. Any closing thoughts? Anything youíd want to leave people with about NLCD or just land cover in general?

DEWITZ:

Land cover in general, whether people realize it or not, is in their every day lives. When you pull up that map on your phone, thatís derived land cover. It gets you from one point to another. And if you pan out just a little bit, you can see those developed areas in that Google map that youíre using. That is something new. Ten years ago, people didnít really have that much exposure to maps, at least digital maps. And as people get familiar with these, theyíre looking for knowledge, and NLCD is one of the places they come to understand a little bit more. 

HULT:

Weíve been talking to Jon Dewitz about the National Land Cover Database. Thank you for joining us, Jon.

DEWITZ:

Thank you for having me.

HULT:

And thank you, listeners, for joining us. Be sure to drop in for the next episode of Eyes on Earth. You can find us on our website at usgs.gov/eros, that usgs.gov/e-r-o-s, or by finding us on Apple Podcasts or Google Podcasts. All of our past episodes are there, including episode 3 on the National Land Cover Database.

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