Weeding the West: Monitoring Invasives using NASA Earth Observations

Category: Identifying Invasive Species Extent & Critical Species Habitat
Project Team: Southwest U.S. Ecological Forecasting
Team Location: NASA Langley Research Center – Hampton, Virginia

A multispectral classification using Bands 2-7 and Band 9 for two Landsat dates. Classes in red highlight grassland areas susceptible to cheatgrass. Image Credit: Southwest U.S. Ecological Forecasting Team

A multispectral classification using Bands 2-7 and Band 9 for two Landsat dates. Classes in red highlight grassland areas susceptible to cheatgrass. Image Credit: Southwest U.S. Ecological Forecasting Team

Authors:
Ryan Avery
Katherine Landesman
Jordan Vaa
Timmera Whaley
Dakoyta Greenman

Mentors/Advisors:
Dr. Kenton Ross (NASA Langley Research Center)

Past/Other Contributors:
Emily Gotschalk (Center Lead)
Tyler Rhodes (Center Lead)

Abstract:

The southwestern United States spans six states, more than 55 national parks, and a wide range of ecosystems, historical landmarks, and culturally significant landscapes. Of these parks, Bandelier National Monument in New Mexico (NM), Big Bend National Park (Texas), Glen Canyon National Recreation Area (Arizona, Utah), and Valles Caldera National Preserve (NM) are threatened by three particularly problematic invasive plant species: cheatgrass (Bromus tectorum), ravenna grass (Saccharum ravennae), and giant reed (Arundo donax). Currently, park management uses field observations to monitor these species, which requires a significant investment in time, effort, and money by the National Park Service (NPS). The NPS is interested in mapping and predicting the presence of invasive species by using NASA’s Earth observations. To this end, the Southwest U.S. Ecological Forecasting team created classified species distribution maps using Moderate Resolution Imaging Spectroradiometer (MODIS), Landsat 5 Thematic Mapper (TM), and Landsat 8 Operational Land Imager (OLI) data for the years 2000, 2008, and 2016. This project also used vegetation and topographic indices, as well as field data to predict invasive species presence using a Species Distribution Model (SDM) for each national park area and generated likelihood maps of species presence/absence.

Previous story | Main Page | Next story

6 Comments

Daryl Ann Winstead (Mekong River Basin Agriculture) 18-08-2016, 12:33

Really enjoyed the video! Did the team use the in situ data for any accuracy assessment? Thanks in advance!

Reply
Ryan Avery 17-09-2016, 11:01

Hi Daryl! Thanks for your comment. For our classification of giant reed in Big Bend National Park, we did set aside 10% of our in situ validated pixels for accuracy assessment. The rest of the in situ data was used to train the Random Forest algorithm used to classify giant reed. However in Bandelier and Valles Caldera, the amount and type of in situ data we had available didn’t lend itself to carrying out a full accuracy assessment. We instead tried to show the difference in the classified extent of classified grasslands based on differing amounts of in situ data used to validate our unsupervised classifications in the area.

Reply
Brian Woodward 16-08-2016, 02:25

Awesome, SW Eco Forecasting! What a great geographic impact from one project. And great to see the fusion of in situ data and remote
sensing.

All my best,
Brian Woodward

Reply
Ryan Avery 17-09-2016, 11:02

Thank you Brian we appreciate that! It was nice meeting you in D.C.!

Reply
Sara Lubkin 13-08-2016, 19:45

Did the 30 meter Landsat resolution work well for you?

Reply
Ryan Avery 17-09-2016, 11:08

Hi Sara, thanks for your question!

Performance varied depending on the weed we were trying to sense and the location. In Valles caldera and Bandelier, it is difficult to say since the amount of in situ data we had for each year was pretty sparse and was in point form (as opposed to a coverage polygon), thus making it difficult to validate and assess the accuracy of our classifications with confidence. But in Big Bend NP in Texas, a classification incorporating Landsat’s 15 meter panchromatic band performed miles ahead of a classification just using the 30 meter bands. The panchromatic classification was able to distinguish thin patches of riparian vegetation from the river line and shrubland but was not able to distinguish different types of riparian vegetation as well as Sentinel 2A data (which is 10-20 meter resolution)

Reply

Leave a Reply