Bridge Internship
Program

Project: Exploring the potential for automated species identification of tropical plants from hyperspectral and 3D scanning data

You are here

Project title
Exploring the potential for automated species identification of tropical plants from hyperspectral and 3D scanning data

Mentor's name(s)
Helene Muller-Landau, Senior Scientist

Contact information: mullerh@si.edu
 

Location of internship. 
Barro Colorado Island, Panama. 

Project summary
Tropical forests account for a majority of terrestrial carbon stocks and biodiversity, and it is thus critical to understand how they are being affected by global climate change. Their high plant diversity offers the potential for high resilience to anthropogenic global change because species vary widely in their responses to environmental variation and the most negatively affected species will invariably become less common. However, this very diversity presents a tremendous challenge to our ability to understand tropical forest function today and to predict how it will respond to global change, as it means interspecific variation is critical. We currently lack the abundant, species-specific data needed to quantify this interspecific variation and capture it in models. Ground-based studies are small-scale and contain inadequate sample sizes of most plant species, and remote sensing data provide large sample sizes but cannot currently distinguish individual species in diverse tropical forests.
Technological advances in hyperspectral imaging, laser scanning, and artificial intelligence now offer the potential for automated species identification of individual plants using remote sensing, smartphones, and/or other tools. This would enable large-scale, species-specific data collection in tropical forest and empower a wider range of researchers, students, and citizen scientists to contribute. However, to realize this potential we need to collect high-quality training data and analyze these data to develop appropriate machine learning algorithms and determine which types or combinations of data are most useful for species identification.
This project aims to build the basis for automated species identification of woody plants in diverse tropical forests. The specific objectives are:
(1) collect and publish a multidimensional high-quality training dataset including hyperspectral reflectance and three-dimensional structure of whole plants, leaves, bark, fruits and flowers of > 150 woody plant species in central Panama;
(2) quantify within-species variation and between-species distinctiveness in these data;
(3) evaluate the ability to distinguish taxa for each data type and combination of types, and identify the most promising approaches for automated species identification.
The successful completion of the proposed research will lay the groundwork for large-scale collection of taxon-specific data encompassing many co-occurring plant species in tropical forests. Findings on the utility of spectrometer reflectance will be of particular importance for efforts to scale up tropical biodiversity monitoring from large plots like Smithsonian ForestGEO to larger landscapes with airborne and space-borne hyperspectral sensors.

Skills required:

The ideal candidate has a bachelor’s degree in a relevant field, ability to conduct tropical forest field work in rugged terrain, strong organizational skills, ability to work well with team members from diverse backgrounds, strong quantitative skills including programming experience, strong English oral and written communication skills, and good Spanish communication skills. 

Back to Top