The NEON Airborne Observation Platform (AOP) has a suite of instruments for remote sensing, including both lidar and hyperspectral imaging. Most researchers use one or the other to study questions related to plant biodiversity, land cover, or forest structure. But what happens when you combine them?
A recent study published in Global Ecology and Biogeography explores how lidar and hyperspectral remote sensing data could be fused to improve estimates of plant biodiversity in temperate forests.
Lidar vs. Hyperspectral Imaging: What They Do
The study was led by Dr. Aaron Kamoske, then a Ph.D. candidate at Michigan State University (MSU), and Dr. Kyla Dahlin, an associate professor at MSU in the Department of Geography, Environment, and Spatial Science. Kamoske has since taken a position as an Adaptive Management Analyst for the U.S. Forest Service. The paper is based on Kamoske’s Ph.D. dissertation.
Kamoske, Dahlin, and their coauthors wanted to see whether combining lidar and hyperspectral data can provide more accurate maps of plant biodiversity at large scales. The two technologies are both used for remote sensing but provide different types of data.
- Lidar (Light Detection and Ranging) uses a pulsed laser to scan the landscape. With each pulse, some of the light is reflected back from leaves, branches and other structures. The resulting data provides a detailed 3D map of not only the canopy top but also the internal structure of vegetation cover. Lidar is used to study questions around forest structure, such as the density of leaf cover at different heights (strata), the height of trees and shrubs that make up the forest canopy and understory, and the diversity of plant types based on their structure.
- Hyperspectral imaging uses a high-resolution imaging spectrometer that collects measurements of reflected sunlight across a broad range of wavelengths (380 to 2500 nm), extending far past the visible light spectrum for human vision (380 to 700 nm). Data are collected in very narrow bands (~5 nm), allowing for highly precise measurements of the spectra of reflected light. Scientists can use these data to identify chemicals (such as chlorophyll, for example) based on their unique spectra reflectance. Hyperspectral imaging can be used to extrapolate data about a landscape, such as canopy nitrogen and water content, and the presence of lignin, chlorophyll and other chemicals. In turn, this provides valuable information about forest health, productivity, and cycling of nutrients and carbon. Hyperspectral data can also be used to measure plant biodiversity using the unique spectral patterns associated with different vegetation.
Three Ways to Look at Plant Biodiversity
The study examines how lidar, hyperspectral, and combined data perform on three different measures of plant biodiversity:
- Taxonomic biodiversity is the basic measure of biodiversity that most of us think of: how many different species are in an area, and what is the relative abundance of each?
- Phylogenetic biodiversity is similar to taxonomic biodiversity but is focused on understanding how species present in an area are related to each other. Is the ecosystem dominated by only broad-leafed deciduous trees, or does it feature a mix of tree types that are more distantly related? Are the individual species closely related (for example, many oaks but few other species), or are many different taxa represented?
- Functional biodiversity focuses on functional aspects of ecosystem biodiversity, such as variation in canopy height, leaf area index and leaf chemistry. Rather than looking at the specific species present, it looks at the functions different types of plants fill in the ecosystem. Functional biodiversity is correlated with higher resiliency in ecosystems; when there are many different plants filling different roles and niches, the ecosystem as a whole is better able to adapt to change.
They used NEON AOP data from five eastern temperate forest sites – HARV, SERC, MLBS, ORNL, and TALL. The authors were interested in how biodiversity varies both within a single forested site and across larger spatial scales. They compared predictive models for different measures of biodiversity using just lidar, just hyperspectral imaging, and both data products combined. The models were then compared to woody vegetation data collected by NEON field staff to see how well they performed.
Kamoske explains, “We wanted to start working towards frameworks for what kinds of biodiversity we can map using remote sensing data. There is a lot of interest in satellite remote sensing data from space, like NASA’s spaceborne lidar, GEDI, and the proposed Surface Biology and Geology hyperspectral mission, to calculate things like forest health, nutrient levels, or forest structure. But we’re not really there yet in terms of bringing it all together to look at different aspects of biodiversity.”
The results show that the model combining both lidar and hyperspectral data performs better on all three measures of biodiversity. Lidar data showing structural diversity proves to be particularly important for predicting taxonomic and functional biodiversity. “For taxonomic and functional biodiversity, about 65% of the model can be explained by structure,” Kamoske says. “But when you get to phylogenetic biodiversity, things get more interesting. For example, why are there dozens of oak species in the Alabama site but only a handful in the Massachusetts site? For these kinds of questions, understanding site history turned out to be more important [in predicting biodiversity] than the lidar data.”
Dahlin says, “Combining hyperspectral and lidar data from the NEON AOP system is a great way to look at vegetation in a bunch of different dimensions. And the data are well documented, consistently collected, and publicly available across all the sites. We could not do this work without NEON.”
Opening the Door to Large-Scale Biodiversity Measurement
The results of the study could help improve models of biodiversity created using remote sensing data - including satellite data - over large spatial scales. Dahlin and other co-authors plan to carry the work forward and look at how the frameworks perform in different kinds of ecosystems. “The main conclusion from this paper is how important structural diversity was in predicting different biodiversity metrics in this one particular biome: eastern temperate forests,” she says. “If you can imagine scaling this up to a fully continental or even global scale, the results might be different. Different measures might prove to be more important for different ecosystem types, like grasslands.”
Understanding how different measurements of plant biodiversity are changing both locally and globally will be increasingly important in coming years as ecosystems are stressed by climate change, invasive species, and human activity. “Whenever you start thinking about things like global climate goals or ecosystem productivity, biodiversity becomes important,” Kamoske states. “For example, more diverse ecosystems are more resilient and are able to sequester more carbon. However, we don’t really know how these different dimensions of biodiversity respond to local and global changes. We often think about things globally: we know this ecosystem is important, and we want to protect it. But we need to be able to have a more nuanced and localized picture of biodiversity at the forest, stand, or even plot level for on-the-ground conservation and management goals. This work shows that hyperspectral data alone may not be able to give us that nuance. A combined approach that includes things like topography, structure, and land use history could help us build better frameworks.”
Dahlin is most interested in functional biodiversity, especially as it relates to questions around nutrient cycling and carbon uptake. She sees remote sensing data as a way to screen for biodiversity hotspots that may warrant additional study and fieldwork. She says, “Hopefully, we’ll someday be able to map plant biodiversity globally, which will allow us to better understand how our world is changing. Aaron’s dissertation is one step in that direction.”