2016 data theme: Remote sensing with reproducible workflows
The Institute was composed of three weeks of online activities followed by a week-long in-person course. Read on for more details about the inaugural Data Institute.
Pre-institute activities: 2016 Data Institute participants were asked to complete a series of online activities for three weeks prior to the Institute that provided the foundational knowledge for everyone to succeed in the in-person portion. Topics included how NEON collects data as well as reproducible workflow tools and techniques.
In-person course: The in-person phase of the Institute included guest lecturers and hands-on data-intensive activities, as well as individual/group activities and projects. The following topics were covered:
- Introduction to hyperspectral remote sensing
- Using & accessing the HDF5 file format
- Remote sensing uncertainty
- Data fusion (LiDAR with hyperspectral data)
In addition, several of the presentations at the Data Institute are available to view (see the bottom of this page).
2016 Faculty & Presenters
The Data Institute had three core faculty:
- Naupaka Zimmerman, University of Arizona, with expertise in coding, open science and informatics
- Kyla Dahlin, Michigan State University, with expertise in remote sensing and ecological modeling
- Leah Wasser, NEON, with expertise in remote sensing, coding, and reproducible workflows
Additional guest lectures included:
- Lindsay Powers, H5 Group – HDF5 data structure
- Chris Crosby, UNAVCO/Open Topography – LiDAR remote sensing
- David Schimel, NASA Jet Propulsion Lab – remote sensing, open science, ecology
NEON project scientists presented on the following topics (you can see some of their video presentations at the bottom of this page):
- David Hulslander – Remote sensing data processing
- Tristan Goulden – Remote sensing theory & Hyperspectral remote sensing
- Nathan Leisso – Introduction to NEON AOP data collection and processing
- Courtney Meier – NEON in situ field measurements.
- Keith Krause – NEON full waveform LiDAR
Participants came from institutions in the USA, Canada and the Netherlands. While 70% of the participants were graduate students, the Data Institute also attracted an undergraduate student, post-docs, and university research staff and faculty.
Participant were interested in using remote sensing data to answer a wide range of questions from wanting to be able to characterize forest structure and composition to using time series to detect vegetation disturbance patterns to from remote sensing data.
According to NEON science educator, Megan Jones, “Participants really appreciated the opportunities to work with data in small-group settings and the emphasis of using reproducible science methods. The science theme for 2016 was use of remote sensing data, but this was taught along with reproducible science methods including the importance of well documented code, version control and collaborative tools like GitHub, and quick sharing of results using RMarkdown and knitr.”
At the end of the Institute, participants presented group projects illustrating the use of reproducible workflows with remote sensing data. The skills learned are applicable to remote sensing data from any source, however, all participants were allowed to use NEON remote sensing data as well as their own data sets. According to Robert Paul from the University of Illinois at Urbana-Champaign, “The course offered a comprehensive overview of best practices for managing and analyzing remote sensing data, and how to make data analysis workflows well-documented, collaborative, and reproducible.”
Sarah Graves, from the University of Florida, said, “The NEON Data Institute gave us the tools to work with novel ecological data. With our own knowledge of the domain combined with NEON data and tools, we are in a position to ask novel ecological questions that will advance the field of ecology beyond what has been traditionally possible.” Jeff Atkins of Virginia Commonwealth University added, “Ecology increasingly depends on "big data" and remote sensing and scientists need the skills necessary to work with this data and to inform their hypotheses. NEON does an amazing job at helping scientists learn how to work with and use a suite of data and data products.”
Exploring the relationship between functional traits and spectral reflectance for Ordway Swisher Biological Station, FL
Sarah Graves, Jeff Atkins, Kunxuan Wang, and Catherine Hulshof de la Pena
We calculated plot-level foliar N content and functional diversity from in situ data. These metrics were related to mean plot reflectance and a spectral diversity metric from a PCA transformation.
Describing landscape-level phenology with MODIS vegetation index time series
Robert Paul, Jeff Stephens
This workflow detects the length of time for NDVI and EVI to go from baseline to peak over the course of the year. Each pixel is classified with a value reflecting the length of time in the year for NDVI and EVI to reach peak greenness.
Characterizing the forest using trees: how do forest characteristics vary with respect to disturbance history at Soaproot Saddle
Subtitle: Attempting species-level classification using Random Forest on LiDAR and imaging spectroscopy
Megan Cattau, Stella Cousins, Kristin Braziunas, Allie Weill
Towards individual tree crown segmentation with spectral indices
Enrique Montano & Dave McCaffrey
We attempted to implement an individual tree crown extraction algorithm, optimized with vegetation structure data from in situ plots. The ability to identify individual tree canopy with confidence will allow for comparison of spectral indices among individuals and across species.
Plant structure and function in complex terrain: Landscape controls and microclimatic consequences
Holly Andrews, Nate Looker, Amy Hudson
Pheno-topo-temp: We examined climate, topography, and vegetation interactions. Specifically, we assessed spectral and LiDAR-based properties of vegetation across topographic gradients of water availability and compared land surface temperature to NDVI.
Upscaling Structure for Soaproot Field Site, California
Cassondra Walker, Jon Weiner, Richard Remigio
We attempted to link vegetation indices to plot-level tree characteristics, and then upscale those indices to the landscape scale to predict structure that was derived from LiDAR.
Using HyperSpectral Imaging techniques to predict foliar nutrient concentrations