Homing In On the Range

Ecologically diverse, economically valuable, and simply beautiful rangelands stretch across the American West. These areas are grazed by livestock and wildlife. They are important resources for mineral, energy, and agricultural production. And they are explored by tourists and daytrippers who enjoy nature and outdoor adventure.

Half of the Western United States is covered in such rangelands, and it’s the job of the U.S. Geological Survey (USGS), the Bureau of Land Management, and other agencies to routinely assess their condition so that resource managers and landowners can make responsible decisions about their use and care.

One aspect of such surveys is classification and mapping of vegetation. Knowing what types of vegetation are found across the rangelands helps land managers make decisions about where and when to let livestock roam, how to prevent wildfires, how to manage invasive species, and other matters. 

USGS scientists employ various survey methods, including satellite observations, to map rangelands. One of their primary tools is the USGS/NASA Landsat series of satellites, which are equipped with multi-band sensors that offer broad, regular land coverage, strong vegetation detection, and a deep archive going back more than 50 years. But there are also known challenges and limits to Landsat's sensors, which can make it hard to precisely identify certain types of vegetation, especially in complex terrain or geology. 

To see if they could improve their land cover maps, USGS researchers recently tested the combination of NASA imaging spectroscopy data and a key USGS vegetation mapping product. Imaging spectroscopy sensors—also called hyperspectral sensors—observe Earth's surface through many more spectral intervals of light than multi-band sensors. This means they can distinguish more subtle differences in vegetation, potentially increasing the accuracy of maps.

For the study, the researchers integrated imaging spectroscopy data from NASA’s Earth Surface Mineral Dust Source Investigation (EMIT) sensor with USGS’s Rangeland Condition Monitoring, Assessment, and Projection (RCMAP) model. They found that hyperspectral data improved mapping accuracy by at least one third in the study area.

“Imaging spectroscopy data can significantly improve vegetation classifications that can help refine assessments of wildfire-prone areas, invasive species encroachment, and rangeland ecosystem health,” said Cole Krehbiel, a USGS scientist and leader of NASA’s Land Processes Distributed Active Archive Center (LP DAAC), which archives and distributes EMIT data. “This study shows the data have broad ecological and land-management value, and could shape the next generation of ecosystem monitoring.”

Testing RCMAP With Imaging Spectroscopy Data

The current version of RCMAP uses artificial intelligence to analyze Landsat imagery and then identify and map shrubs, grasses and forbs, dead plant litter, trees, sagebrush, or bare ground across rangelands.

“The primary problem we have is accurately separating shrubs from grasses. And then, secondly, separating annual grasses from perennial grasses,” said Matthew Rigge, a USGS ecologist who leads the RCMAP project. “The RCMAP models also struggle in places with really dark volcanic soils, making it hard to get a good vegetation cover estimate.”

Imaging spectroscopy data can eliminate some of those limits by providing modelers with more detailed imagery—especially in the 900-1250 nm and 1500-1780 nm spectral regions that Landsat does not cover. These wavelengths enable important measurements related to the water content of vegetation, plant stress, and non-photosynthetic vegetation. The data can also improve accuracy when identifying vegetation in the aftermath of a fire and for distinguishing shrub‑grass mixtures—two of the most challenging aspects of rangeland mapping. This information could improve wildfire management through better predictions about the presence of herbaceous types of plants that light and burn easily and about the total fuel loads across a patch of land.

To assess the value of imaging spectroscopy data for rangeland mapping, Rigge and his team combined data from EMIT, which observes Earth's surface from the International Space Station (ISS), with observations from Landsat. They trained their RCMAP model to integrate the EMIT and Landsat data and examine a challenging area of Montana and Idaho known for significant elevation changes and complex geology along the Continental Divide.

“We picked a beautiful part of the world that’s hard to classify because the geology and topography are complex, and there’s lots of perennial grasses that grow together in clumps that cast a shadow similar to a shrub,” said Rigge. “You can't distinctly see the shadow on a Landsat pixel, which then makes the model struggle to differentiate grasses and shrubs.”

The RCMAP model runs made with Landsat-EMIT data were compared with actual field measurements and with high-resolution imagery of the area to see how well they resolved the challenging terrain and vegetation.

“EMIT data improved our results by 30 percent,” said Rigge. “No single thing we’ve done in RCMAP has improved our accuracy that much before.”

The researchers found that RCMAP’s revised maps of the region were especially good for herbaceous vegetation, litter, and sagebrush cover, each of which is critical for wildfire fuel modeling and land health assessments. And yes, Rigge noted, RCMAP with Landsat-EMIT showed improvement at distinguishing bunchgrasses from shrubs.

As an additional step, the study team tried adding synthetic aperture radar (SAR) data from the Sentinel-1 satellite to the RCMAP model because SAR is useful for detecting height differences and other changes across areas. The researchers found that SAR also improved RCMAP’s capabilities by about 7.5 percent.

Overall, the study demonstrated the potential for significantly enhancing classification accuracy by moving beyond traditional optical imagery. However, EMIT and many existing SAR datasets have limited spatial and temporal extent, so such products are not ready for regular operations. For instance, because the orbital tracks of the space station do not span the entire Western United States in a consistent pattern, there can be time and spatial gaps in EMIT's coverage. 

The study results suggest it could be good to equip future land-observing satellites with imaging spectroscopy sensors geared for rangeland monitoring. And in terms of SAR data, the powerful new NASA/ISRO (Indian Space Research Organization) Synthetic Aperture Radar (NISAR) satellite can detect large- and small-scale vegetation changes and could have applications for RCMAP and its mapping products. 

For now, the work with EMIT data will inform the design, algorithms, and operational workflows for future imaging spectrometers. When the day comes that scientists have complementary and regular imaging spectroscopic, SAR, and multispectral data, they will have complementary tools for supporting the care and monitoring of rangelands and other natural resources.

Editor's note: USGS’s RCMAP project and NASA's LP DAAC are both housed at the USGS Earth Resources Observation and Science Center.