Fall 2005

Sometimes a Great Notion:
Visualizing Forest Fragmentation

by Zoë Hoyle

It may seem pretty obvious that American forests are being split apart by roads, houses, and strip malls, but, until recently, it was difficult to visualize the extent and magnitude of forest fragmentation in the United States. Until, that is, Southern Research Station (SRS) researcher Kurt Riitters and his collaborators started applying a method called “moving windows” to landscapescale analysis.

Riitters and three fellow landscape ecologists—James Wickham, Timothy Wade, and John Coulston—have come up with a deceptively simple method to make fragmentation of the landscape visually apparent. The four—Riitters and Coulston from the SRS Forest Health Monitoring unit in Research Triangle Park, NC, and Wickham and Wade from the nearby Environmental Protection Agency (EPA) National Exposure Research Laboratory—are using moving windows to produce information about forest fragmentation for highlevel clients such as the Montréal Process, the H.J. Heinz Center, and the European Commission—as well as for regional planners in the Southeast.

A Window Opens

Riitters and Wickham started looking at ways to visualize landscape patterns more effectively in the 1990s, when they both worked at the Tennessee Valley Authority Landscape Ecology Project in Norris, TN. By 1995, when they published a landscape atlas of the Chesapeake Bay watershed (with the help of Wade, then at the Desert Research Institute in Reno, NV), the two were well aware of the limitations of the approaches used to assess forest fragmentation. Analysts would start out with landcover maps generated from Landsat satellite images that were divided into millions of tiny squares called pixels (short for “picture elements”). They would count the pixels in a given area for each type of landcover and assign a number, or fragmentation value, to represent all of the pixels in the area. This aggregating method meant throwing away original information from the images, as well as removing the ability to make comparisons.

For the 1995 atlas, Riitters and Wickham introduced the idea of using moving windows to recover this lost information. “When you make a map of forest fragmentation using moving windows, it’s like focusing a camera so that the detailed patterns stand out,” says Riitters. “But the main benefit is that each pixel now has its own unique fragmentation value, a number that describes its particular context.” This is especially important when looking at the effect of fragmentation on forest inhabitants. “When you start looking at fragmentation in relation to habitat, you need to be able to look at context,” says Riitters. “The small block of forest that might provide adequate habitat for one species becomes problematic when you move out to the larger scale and realize that it is an island or part of a checkerboard. The pattern of forest loss is just as important as the amount lost.”

Enter the Montréal Process

In 2000, when Riitters arrived in Research Triangle Park to take the position of deputy program manager at the SRS Forest Health Monitoring unit, Coulston, then a North Carolina State University (NCSU) employee specializing in landscape assessment techniques, was already working at the unit through a cooperative agreement between SRS and NCSU. Meanwhile, Wade and Wickham had also moved to the area to work for a new landscape ecology project at the EPA, so the team was in place when the Montréal Process came looking for someone to develop forest fragmentation information for the United States.

The Montréal Process, an international framework for measuring progress towards sustainable forest management, developed criteria and indicators to assess the condition of the world’s temperate and boreal forest, with “fragmentation of forest types” as one indicator of biological diversity. In the United States, scientists recommended using what Riitters calls “the old method” to describe fragmentation for the continental United States. Riitters and his collaborators volunteered to provide the information, using the old aggregating method to come up with data for a preliminary assessment.

To generate the data, the team used a relatively new resource, the 1992 landcover maps generated from the National Land Cover Data (NLCD) database. Produced from satellite imagery, the maps are the result of years of work across multiple U.S. agencies. “Lots of people make landcover maps,” says Wickham. “You can go to each of the States and get satellitegenerated landcover maps, but they don’t fit together. The 1992 NLCD map was the first effort to make a national landcover map that goes across boundaries to give consistent information.”

Riitters and his colleagues couldn’t stop with just generating data from the NLCD map using the old method. Their preliminary assessment included a discussion of the limitations of the method, and posed a series of research questions that they then set out to answer. “We used the old method they suggested to provide data for a preliminary assessment,” says Riitters. “Then we generated the information using our moving windows, and did something they hadn’t thought could be done. We provided them the means to analyze fragmentation at multiple scales.”

How Do Moving Windows Work?

The moving window technique wasn’t new—it had been used before in medical image processing. “New was the notion of applying this method to landcover maps,” says Riitters. “In medical image processing, moving windows are used to standardize images and make them more continuous. We took it in the other direction, pulling out patterns in the landscape that you can’t really see any other way.”

In the moving window approach, a square window representing a unit of interest—for instance, the home range of a particular species—is moved across the landcover map one pixel at a time. As the window is moved, a fragmentation index is calculated within the window and used to generate a new map. This is putting it simply: there’s a lot more involved than counting pixels. Making maps means dealing with the Geographic Information System (GIS) software used to analyze and map the spatial data the team generates. “A lot of what we do, in terms of presentation, relies on GIS software, “Riitters adds. “Tim Wade is our go-to guy in terms of what can be done visually.”

The result is a richer source of information about forest condition and health. “We used to describe landscapes by what was contained in them,” says Riitters. “By providing overlapping information, moving windows allow you to analyze context. Imagine you are standing on one of the forest pixels. This method allows you to examine the pattern elements all around you, and to use different windows to make comparisons.”

For example, an ecologist can look at how forest fragmentation affects an animal species by making the window the size of the animal’s home range, applying the window to the NLCD map, and then comparing the result to the species’ needs for forest cover. This can be done for multiple species in the same region by using separate windows for each. This allows ecologists to evaluate a specific area in terms of the forest cover requirements of multiple species at one time and provides a more realistic picture of habitat availability.

“Fragmentation is scale dependent,” says Riitters. “If you are a small creature requiring 50 percent forest cover over a small range, an area may not seem so fragmented, but if you are a large animal with a home range of 100 to 120 square miles, there may be very little suitable habitat in that same area. A larger window size, especially when applied to a rapidly developing area such as the Southeast, will reveal more fragmentation.”

The limitation to this sort of analysis is the lack of scientific knowledge about how much forest cover different species need. But that need not put a damper on the process. “It’s not just about species, but about the forest,” interjects Wickham. “Forests themselves have a scale, and we can use this method to look at what happens to them at different scales in relation to climate, sprawl, and other issues. Think about it: if a forest stops functioning as a forest, you can say a lot about impacts on forestdependent species.”

“The beauty of the method is that it generates fragmentation information that can be fed into many kinds of process models,” adds Coulston.

“Using the same input data makes it possible to compare results from models that operate at different spatial and temporal scales.”

How Fragmented are U.S. Forest?

What Riitters and his colleagues have found at the national scale is pretty dramatic. “While forest is still dominant where is occurs, the pervasiveness of fragmentation and roads in the continental United States is stunning,” he says. “Many, if not all, aspects of the forest ecological condition are potentially at risk.”

Results from the team’s work published in the 2002 State of the Nation’s Ecosystems report by the H.J. Heinz Center and included in the 2003 USDA Forest Service Report on Sustainable Forests showed that almost three-quarters of all U.S. forests were within 500 feet of a forest edge, with very little interior forest left along major rivers or near urban or agricultural areas. And this did not take into account the over 4 million miles of roads that crisscross the U.S. countryside.

Map showing fragmentation across the United States
(Tim Wade, EPA)

Because landcover maps derived from satellite imagery do not adequately portray the effects of roads, Riitters and his colleagues took the next step of superimposing road maps onto landcover maps. “Ecological impacts from roads may be the rule rather than the exception in most of the continental United States,” says Riitters. “We found that 20 percent of all U.S. forestland is within 400 feet of a road, and the proportion increases rapidly as you move back from roads. Our analysis showed that over 80 percent of forestland is within 0.6 mile (3,200 feet) of a road, and only 3 percent of forestland is more than 3.1 miles from a road.”

The impact of roads on forest ecosystems extends well past the roads themselves. “Effects from roads—what we call influence zones—stretch tens to hundreds of yards from the roads themselves,” Riitters adds. “Water drainage patterns are disrupted and sediment loads to streams increased. While roads act as a barrier for some species, they serve as excellent corridors for the introduction of others, such as nonnative plant species.”

The fragmentation information Riitters and his colleagues produced for the national assessments also did not identify hotspots, places where ecological impacts are most likely, or where particular types of forests are most at risk.

Coulston began working on this problem, applying methods developed for human epidemiology studies to the forests of the Eastern United States. “I took Kurt’s fragmentation maps and used a technique that allows me to zoom in to find areas where fragmentation is significantly more prevalent,” says Coulston. “The mathematical calculation I used was originally designed to search medical records for hotspots of human diseases.”

Results published earlier this year show hotspots covering 20 percent of the Eastern United States, with most of the hotspots associated with human land use concentrated in the Piedmont of the Carolinas and in the upper Great Lakes region. However, the types and causes of fragmentation associated with hotspots varied with geography. “This tells me that how we look at and manage this type of fragmentation should be tailored to local conditions,” says Riitters. “For instance, if the goal is to restore ecological function to a perforated forest, management could use information from our studies to fill holes in specific areas.”

A New Map Adds Another Dimension

In 1993, several Federal Agencies— including the USDA Forest Service— formed a consortium to develop the dataset for the 1992 NLCD map used by Riitters and his colleagues to generate information about fragmentation. In 1999, a second generation consortium was formed to develop the next national landcover database, the NLCD 2001, from satellite images taken in 2000. When the new map is completed, Riitters and his team will generate new landcover pattern data that will feed into the 2007 Heinz Center report and the 2008 assessment of the sustainability indicators developed by the Forest Service for the Montréal Process.

“The information from the 2001 map will add the dimension of temporal change,” says Wickham, who is a key player in the current NLCD process. “Because the 1992 and 2001 datasets are compatible, we will be able to produce snapshots of how forest loss happens over time in the United States. We can look at the hotspots Coulston and Riitter identified to see if they are growing or diminishing— or if hotspots have cropped up in new places.”

Researchers and planners will be able to use these fragmentation comparisons and the moving windows technique to inform regional and local planning. At the EPA, Wickham has already been using Riitters’ maps to look at water issues. The relationship between landcover and water quality is not linear; relatively small losses of forest can result in much larger declines in water quality. EPA regional planners are also interested in looking at how certain land use decisions affect forest fragmentation over a large scale, and at the cumulative impact of local land use changes on air and water quality and wildlife habitat.

“At a more local level, planners could use multiple-scale windows to look at the broader impact of their local decisions,” says Riitters. “How does allowing a use that removes forest on the 25-acre level affect what happens in terms of fragmentation at the 2,500-acre level? How does preserving a specific 25-acre parcel contribute to a regional network of forest habitat?”

Having comparison data will not only make it easier to show people what is happening to forests at the national, regional, and local levels, it may even help individuals make decisions about their own land. “We can take it down to a very fine level, even to the 2-acre lot you plan to build your house on,” says Riitters. “With moving windows, you could analyze how clearing different parts of your property will affect forest fragmentation or the connectivity of habitat at a larger level.”

For More Information:
Kurt Riitters at 919–549–4015 or kriitters@fs.fed.us

Back to top