Estimating vegetation structure and composition using UAV-based imagery
Ryan Christopher Blackburn, Smithsonian Conservation Biology Institute, blackburnrc@si.edu (Presenter)
Qiongyu Huang, Smithsonian Institution, huangq@si.edu
Ginger Allington, Cornell University, gra38@cornell.edu
Motzer Nicole, Montana State University
Grasslands provide a host of unique agricultural, cultural, and ecosystem services. Unfortunately, grasslands are also one of the most threatened and understudied ecosystems globally. The ability to quantify spatial multiscale responses of herbaceous plant communities to emerging threats is imperative to manage for ecosystem resilience. Remote sensing with unmanned aerial vehicles (UAVs) provides data that can help build towards multiscale grassland monitoring. Photogrammetric structure from motion processing uses UAV-based imagery to construct 3D point clouds which provide structural information about plant communities. The aim of this study was to evaluate the consistency of modeling herbaceous vegetative height across different grassland types in Arizona, USA. We chose four distinct grassland community types across a gradient of elevation (1636 - 2591 MASL). At each location, UAV surveys were conducted over three randomly chosen sites. We measured mode vegetative height a across ten 0.25 m2 quadrats (N =120) within each site. UAV-based point clouds for each quadrat were used to create predictor variables from spectral and structural data. Spectral predictor variables included the mean and standard deviation of six RGB-based vegetation indices, and structural predictor variables included height statistics. This resulted in 145 potential predictor variables for the prediction of mode vegetative height. Boruta feature selection removed 43 predictors prior to random forest estimation. Through 10x10 fold cross-validation, we estimated a mean RMSE of 4 cm and a mean R2 of 0.62. These results suggest that UAV-based imagery can accurately quantify vegetation height across surveyed areas. Future work will explore the upscaling of UAV surveys to satellite imagery for landscape level herbaceous height predictions. The fusion of these datasets will allow both scientists and practitioners to better understand the multiscale complexities of grassland dynamics.
Poster: Poster_Blackburn_1-11_116_35.pdf
Associated Project(s):
Poster Location ID: 1-11
Presentation Type: Poster
Session: Poster Session 1
Session Date: Tue (May 9) 5:00-7:00 PM
CCE Program: LCLUC