Upgrading ‘Built-in, Foundational Layer’ 

Annual NLCD, produced at the USGS Earth Resources Observation and Science (EROS) Center, is part of a larger suite of land cover mapping and monitoring data produced by the Multi-Resolution Land Characteristics (MRLC) consortium, a group of federal agencies that coordinate and generate consistent and relevant land cover information at the national scale.

The new “Annual” part of NLCD comes in response to the needs of people who use NLCD data. As Earth Observation Applications Coordinator for the USGS National Land Imaging Program, it’s Zhuoting Wu’s job to know what kinds of Earth observation products are valued most by federal agencies. 

Through a survey, Wu discovered: “NLCD is the most widely used observation product we surveyed. People use it pretty much for everything. It goes into models or applications as a built-in, foundational layer.”

Terry Sohl, Chief of the Integrated Science and Applications Branch at EROS, agreed. “The user community is so extensive,” he said. “There are so many federal agencies that absolutely rely on it, whether it’s the Bureau of Land Management, whether it’s the Environmental Protection Agency for regulatory concerns, whether it’s Fish and Wildlife for habitat management, or whether it’s Health and Human Services. It's hard to find an agency that does not use NLCD.”

However, in the federal survey from Wu, users did express the desire for annual updates produced more quickly.

In the meantime, another EROS-led land cover project arose to provide annual land cover and change information stretching back to 1985. However, Land Change Monitoring, Assessment and Projection (LCMAP), first released in 2020, did not contain as much detail about land cover types as NLCD, especially in urban and forested areas.

Wu said users found NLCD useful for its classification detail and LCMAP for its frequency, but “a combination of the two really gets the needs met.” That combination is Annual NLCD.

Media

Sources/Usage: Public Domain. View Media Details

Evolution of NLCD Leads to ‘Touching Every Landsat Pixel’

Work on the original NLCD product began well before high performance computing and cloud computing could provide automation. Processes have changed since Annual NLCD team member Jon Dewitz spent two years leading field work nearly 20 years ago to figure out which land classes should be labeled where. 

“Making a land cover map from scratch is very different than developing an algorithm,” Dewitz said.

That hard-earned information proved foundational to the progression of NLCD, however; processes for each data release grew more automated over time. “This has been a gradual evolution,” Dewitz said. “It’s another magnitude of effort to produce Annual NLCD because are we touching every Landsat pixel.”

That “magnitude of effort” might be stating it mildly. The number of Landsat pixels processed for Annual NLCD numbered 295 trillion, from a total of 310 terabytes of Landsat data used.

The task of creating Annual NLCD required new methods involving a lot of research and development, along with engineering. 

One improvement that helped the team produce Annual NLCD in just two years was the ability to process the vast amount of imagery in the cloud alongside the Landsat data. “That is enabling us to make things faster,” said Jesslyn Brown, the Annual NLCD project manager, compared to previously having to move the imagery to a supercomputer for processing.

Deep Learning Key to Development 

Deep learning is another technological advantage the team leveraged for processing. Deep learning is a type of artificial intelligence that uses large amounts of data and, like the human brain, learns to recognize patterns—in imagery, for example—to solve problems or make predictions. This was especially important for Annual NLCD because datasets that helped with past NLCD land cover decisions didn’t go as far back as 1985. 

Media

Sources/Usage: Public Domain. View Media Details

“We had to rely a lot more on the spectral imagery and also on deep learning to do a better job of inferring what’s happening in the Landsat imagery,” Dewitz said. “Deep learning really did a great job of linking all of that data together.”

EROS Center Director Pete Doucette has long been an advocate for the use of data science to help solve scientific challenges. “Annual NLCD is blazing the trail as among the first generation of operational products at EROS that incorporate deep learning methods to improve performance,” Doucette said. “And I believe that we’re just getting started with where we can take machine learning methods at EROS.”

Rylie Fleckenstein, the Research and Development (R&D) technical lead for Annual NLCD and a contractor at EROS, looked at previous methods for producing NLCD and LCMAP to help determine the new Annual NLCD process—“moving away from the hand editing, so to speak, and incorporating algorithms or different approaches to automate the process.”

That production process included a change detection component, like LCMAP had, to determine where and when change had occurred on the landscape, and also a classification component to determine the type of land cover in an area. Some refinement was necessary in areas with trickier or inaccurate classifications. 

The resulting new release contains a suite of six products associated with land cover and change: 

• Land Cover: The predominant land cover class
• Land Cover Change: The change between one year and the next
• Land Cover Confidence: The probability value for the land cover class
• Fractional Impervious Surface: The amount of area covered by artificial surfaces like pavement or concrete
• Impervious Descriptor: The differentiation between roads and other artificial surfaces
• Spectral Change Day of Year: The timing of a significant change in Landsat data  

Team Met Challenges During ‘Intense Two Years’

Brown estimated about 30 people have been involved in producing Annual NLCD. That includes scientists and engineers involved in the research and production stages, and also those collecting reference data to check for errors and validate the results.

Dewitz praised the team for all they accomplished in the two-year timeframe. “The R&D team was challenged and pushed, and they performed wonderfully,” he said. 

The engineering side had to do much of their work while R&D was still going on. “Thankfully we have an excellent engineering team,” Dewitz said. “They worked in pieces and did kind of a hybrid engineering process.”

Sohl, the EROS science chief, thinks the infrastructure developed to produce Annual NLCD should be helpful for other science projects, too. 

“This has been an intense two years,” Sohl said. “I'm just so proud of the team. They have worked so hard, and they performed a minor miracle in terms of completely revamping the methodology and moving all of the technology into the cloud. Now that we have this infrastructure set up, it really facilitates the next level of improvements for Annual NLCD.”

Improvements Helpful for Heat and Flooding Studies

Annual NLCD is national in scope, but on a local level, it fills the need that cities or other entities have for detailed and accurate land cover information that spans decades.

George Xian, a research physical scientist at EROS, is grateful that Annual NLCD has arrived so he can start using it in his urban heat island work. He is in the midst of expanding his study of trends in changing average surface temperatures and hotspot locations from 50 to 300 U.S. cities.

This type of information is important for cities to know because they can develop plans to help residents cope during periods of extreme heat, which can cause illness or death in vulnerable populations.

For the 50-city study, Xian and his colleagues needed the annual land cover data beginning with 1985 that LCMAP provided, but also the more detailed information about paved surfaces, concrete and rooftops—collectively called impervious surfaces, which typically retain more heat—contained in NLCD. “We had to use a so-called hybrid way to integrate NLCD and LCMAP to gather the data for this four-category urban area and also annual change,” Xian said.

Media

Sources/Usage: Public Domain. View Media Details

For the expanded study with more than 300 cities from 1985 to 2023, Xian said, “we can use Annual NLCD to directly define our urban categories into four categories. We can study their variations and their variation impact to the urban heat island. We can directly pull the data into our algorithm and use it. We don't need to regenerate the data.”

Ryan Corcoran is looking forward to using Annual NLCD as well. He serves as the Coordinated Needs Management Strategy (CNMS) team lead at Niyam IT, which is part of the Advancing Resilience in Communities joint venture that provides planning, engineering and mapping support for FEMA’s Zone 1.  One aspect that Corcoran and his colleagues work with involves checking whether flood studies of river and coastal areas remain valid after a period of time, or whether conditions have changed and require a new study.

In the past, Corcoran said they have had to use multiple data sources, including NLCD, for baseline watershed information and to assess annual changes. 

“We are excited about the upcoming expansion of the NLCD. It will make it easier for us to calculate baseline watershed imperviousness and land use changes using a single dataset,” Corcoran said. “The availability of this extensive data is critical, as we sometimes validate flood studies that date back to the 1970s. Increased data availability allows us to better evaluate flood risk, especially when validating older flood studies.”

More Access to Annual NLCD Data

Annual NLCD users have more options to access the data than before. The data is still available on the MRLC website, but it also has been added to the cloud and to the USGS EROS data access site EarthExplorer. 

“We’re trying to respond to people’s requests for data in all kinds of different ways,” said Brown, the Annual NLCD project manager.

The data will be updated more frequently, too. “In the past, it’s usually taken over a year, if not more, to do an NLCD update,” said science branch chief Sohl. “We’re setting the stage where, by the middle of every year, we’re going to have an update for the previous year.”

Annual NLCD is providing more useful information more quickly for the people relying on it—which, as it turns out, might be most of us, with NLCD’s history as a key source of data woven into the background of society.   

“Annual NLCD represents the next generation of highly accurate mapping information that keeps pace with evolving user needs,” said EROS Director Doucette. “Annual NLCD products will become increasingly relevant toward assessing land use and land condition. They provide key change indicators for understanding environmental interactions and consequences. These are the kinds of things that decision makers ultimately want to know.”