NCCS Systems Empower Boreal Forest Study Using Airborne and Satellite Data


This aerial image from a 2014 flight for Goddard’s Lidar, Hyperspectral, and Thermal Imager (G-LiHT) shows a portion of the Tanana Valley, Alaska, boreal forest. It features a variety of vegetation structure and patterns: taller dense conifers accompanied by sub-canopy shrubs (center); uniformly distributed, shorter conifers (left); and a broader distribution of forest heights, densities, and possibly age (right). Image from G-LiHT.

Vast boreal forests across portions of the Arctic and sub-Arctic form a belt of primarily conifers (e.g., spruce, fir, and larch) around Earth’s upper Northern Hemisphere. These forests are experiencing climate change at a faster rate than anywhere else on the planet. A key step in understanding and predicting interactions between climate and boreal vegetation is estimating the 3D structure of the forests.

Moving towards such a capability, NASA Goddard Space Flight Center (GSFC) scientists examined variation in surface elevations of boreal forest canopies—leaves, branches, and stems—captured in satellite-derived digital surface models (DSMs). They used several different types of satellite data acquisitions, including varying sun elevation angles and seasonal ground conditions. Instrumental to this study were NASA Center for Climate Simulation (NCCS) high-performance computing (HPC) resources.

The NASA boreal forest study focused on Tanana Valley, Alaska, an area the size of Iowa, and compared DSMs with reference airborne lidar data for that region. The scientists wanted to understand which of the forest canopy vertical components (top-, mid-, and lower-level) they were capturing with the variety of satellite acquisitions available to them in order to achieve consistent estimates of canopy surface elevations and forest structure.

(a) Study sites in Tanana Valley, Alaska, where reference lidar provided reference measurements of forest structure for coincident strips of high-resolution spaceborne image digital surface models (HRSI DSMs). (b) Aerial images acquired during the 2014 airborne survey highlight a variety of forest structure patterns including (from left to right) small groups of conifers, dense continuous canopy, dense canopy adjacent to standing dead and fallen trees, and heterogeneous canopy cover. Note: All technical figures are from Montesano, P.M., et al. through a Creative Commons license.

The DSMs are computed products of Maxar Technologies/DigitalGlobe Worldview satellite data, made available under a National Geospatial-Intelligence Agency (NGA) license agreement and stored on the NCCS Advanced Data Analytics Platform (ADAPT). Across the study region there were 565,921 satellite observations, all at sub-meter spatial resolutions.

The reference lidar data came from Goddard’s Lidar, Hyperspectral, and Thermal Imager (G-LiHT), initially gridded at 1-meter resolution. Lidar from small footprint airborne platforms provides “the most accurate and precise estimates of forest structure through the full vertical column—from the top of the canopy down to the ground,” said Paul Montesano, research scientist in GSFC’s Biospheric Sciences Laboratory.

Left: DigitalGlobe’s WorldView-3 satellite captures forest canopy imagery at resolutions spanning 30 to 55 centimeters. Illustration by DigitalGlobe. Right: NASA and U.S. Forestry Service collaborators in Fairbanks, Alaska, prepare for G-LiHT flights with a Piper Cherokee PA-32 airplane on July 14, 2014. In the bottom row and second from left is study co-author Bruce Cook. Photo provided by Earth Observatory and Doug Morton, GSFC.

Deriving the DSMs entails processing a pair of images collected at two different viewing positions, a technique called stereogrammetry. The NCCS Discover supercomputer performed most of the initial image processing.

The study team used ADAPT to process the input image pairs to create the DSMs, co-register the DSMs with the lidar, and make pixel-level comparisons of the lidar- and DSM-captured forest canopies. As part of their broader boreal forest research program, the researchers have been using 20 virtual machines (VMs) on ADAPT for the past few years. They also use ADAPT to store their output data and make select datasets available to other ADAPT users.

Representative canopy surfaces of HRSI DSMs in the boreal study domain.

The results of the Tanana Valley study help clarify which boreal canopy surfaces are representative of those captured with high-resolution spaceborne image (HRSI) DSMs. While lidar models the 3D structure much more thoroughly, the study shows that DSMs do provide useful forest structure information—resolving canopy surfaces in detail—and do so over broader spatial extents.

Examples of representative canopy surfaces of HRSI DSMs at three study sites show how the variations in both canopy cover and DSM type are associated with variations in the boreal canopy surfaces they capture.

“NASA HPC resources have allowed us to scale our research from the site level to regional, continental, and global extents,” Montesano said. “We can now examine forest structure patterns across entire biomes that are derived from data at meter-level scales. This allows us to more reliably account for the spatial heterogeneity in the forest characteristics of interest. This clear accounting of how forests vary may help explain differences in how forests continue to change.”

Montesano envisions several applications for the forest structure and texture information available in spaceborne DSMs: quantifying forest carbon and biomass density, examining the variation in forest patterns, and linking DSMs with other types of remote sensing data to provide inputs to predictive computer models.

Collaborators on the Tanana Valley study include Christopher Neigh, William Wagner, Margaret Wooten, and Bruce Cook, all of the GSFC Biospheric Sciences Laboratory.

More Information:
Montesano, P.M., C.S. Neigh, W. Wagner, M. Wooten, and B.D. Cook, 2019: Boreal Canopy Surfaces from Spaceborne Stereogrammetry. Remote Sensing of Environment, 225, 148–159, doi:10.1016/j.rse.2019.02.012.

Jarrett Cohen, NASA/Goddard Space Flight Center