COS 281-3 - Detecting an invasive species in grasslands using Planetscope cubesat time-series

Thursday, August 18, 2022

4:00 PM – 4:15 PM EDT

Presenting Author(s)

NR

Ny Aina M. Rakotoarivony

Ph.D. student
Oklahoma State University
Stillwater, Oklahoma, United States

Coauthor(s)

HG

Hamed Gholizadeh

Oklahoma State University
Stillwater, Oklahoma, United States
MF

Michael Friedman

American International College of Arts and Sciences in Antigua
St. John, United States
NM

Nicholas A. McMillan

Oklahoma State University
Stillwater, Oklahoma, United States
WH

William M. Hammond

Agronomy Department, University of Florida, United States
KH

Kianoosh Hassani

Oklahoma State University, United States
AS

Aisha V. Sams

Oklahoma State University
Stillwater, OK, United States
MC

Makyla D. Charles

Oklahoma State University, United States
DG

DeAndre Garrett

Oklahoma State University, United States
OJ

Omkar Joshi

Oklahoma State University, United States
RH

Robert G. Hamilton

The Nature Conservancy, United States
SF

Samuel D. Fuhlendorf

Oklahoma State University, United States
AT

Amy M. Trowbridge, PhD

Assistant Professor
University of Wisconsin
Madison, Wisconsin, United States
HA

Henry D. Adams

School of the Environment, Washington State University, United States

Background/Question/Methods

Invasive species can have negative economic and environmental impacts on habitats they invade and grasslands are one of the ecosystems that have been significantly impacted. However, identifying invaded grassland regions is challenging due to their large spatial extent. Remote sensing with its continuously increasing spatio-temporal resolution can play a central role in mapping invasive species. Our ultimate goal was to develop an operational remote sensing approach to detect invasive species in grasslands. Specifically, we asked 1) whether we can detect an invasive legume known as sericea lespedeza (Lespedeza cuneata) remotely and 2) if so, when is the best time of the growing season for mapping its spread. We collected field-based plant functional traits and airborne hyperspectral data from The Nature Conservancy’s Tallgrass Prairie Preserve, OK, U.S. We identified functional traits that distinguish sericea from native species and used our hyperspectral data to select vegetation indices (VIs) that are known to be highly associated with these traits. We then applied the selected VIs to PlanetScope cubesat time-series data and used the VIs in a machine-learning algorithm to map sericea during the growing season. Finally, we identified the best times of the growing season for mapping sericea based on its detection accuracy.

Results/Conclusions

Our results showed that sericea has significantly higher nitrogen content, carotenoid, chlorophyll, and canopy height (p-value < 0.05) compared to native species. Based on this knowledge, we selected a total of five VIs to differentiate sericea from native species. To test the validity of our approach, we first applied our machine-learning approach to the VIs obtained from our mono-temporal hyperspectral data (collected in August 2020) and obtained an overall classification accuracy of 92.11% on our validation data. Applying the same approach to PlanetScope multispectral/multi-temporal data suggested that the best times to remotely detect sericea in the Tallgrass Prairie Preserve are during the peak of the growing season, around Mid-August (overall classification accuracy of 86.36%), and towards the end of the growing season in late September (overall classification accuracy of 75%). We were not able to assess the performance of our approach in July due to cloud cover. These results showed that remote sensing data with coarse spectral resolution but fine spatial and temporal resolution can provide promising opportunities for remote detection of grassland invasive plants.