COS 43-3 - Intra- and interspecific variation in tropical tree phenology derived from drone images: Association with plant traits and response to interannual climate variation

Tuesday, August 8, 2017: 8:40 AM
D139, Oregon Convention Center
Stephanie A. Bohlman, School of Forest Resources and Conservation, University of Florida, Gainesville, FL, Sami Walid Rifai, Environmental Change Institute, University of Oxford, Oxford, United Kingdom, Helene C Muller-Landau, Smithsonian Tropical Research Institute, Panama, Panama and Jonathan Dandois, Center for GIS, Towson University, Townson, MD

Phenology is a key life history trait of plant species and critical driver of ecosystem processes. There is strong evidence that phenology is shifting in response to climate change in temperate ecosystems, but tropical forest phenology remains poorly quantified. A key challenge is that tropical forests contain hundreds of plant species with a wide variety of phenological patterns. To quantify phenology over a large number of individuals and species, we collected bi-weekly images from unmanned aerial vehicles (UAVs) over Barro Colorado Island, Panama. The objective was to quantify inter- and intra-specific responses of tropical tree leaf phenology to environmental variation and identify key environmental variables and physiological mechanisms underpinning phenological variation.

Between October 2014 and May 2016, we collected 35 sets of UAV images, each with continuous coverage of the 50-ha forest inventory plot. UAV imagery was processed to extract spectral and texture information for individual tree crowns, which was then used as input for a machine learning algorithm that predicted branch and leaf cover. We obtained the species identities of 2000 crowns in the images by linking the crowns to stem tags in the field, thus producing a time series of cumulative annual deciduousness for 65 species.


The machine learning algorithm successfully predicted the percentages of leaf, branch, and flower cover for each tree crown (r2=0.76 between observed and predicted percent branch cover for individual tree crowns). Deciduousness showed continuous variation among species rather than distinct phenological categories (ie, evergreen and deciduous), which are commonly used in physiological, ecosystem and modeling studies. Some species labelled as evergreen by expert-based classification had annual deciduousness higher than those labelled as deciduous. We found significant, positive relationships between species mean deciduousness and species’ leaf phosphorous, photosynthetic capacity and adult relative growth rate, suggesting that higher deciduousness is associated with greater resource acquisition. Liana cover altered the annual phenology signal, causing tree crowns to appear less deciduous in the dry season. Comparing May 2016 (during an El Nino drought) and May 2015 (a non-El Nino year with normal rainfall), mean deciduousness values for nearly all species was greater in 2015 but with differing levels of intraspecific variation. Thus variation within and among species may play an important role in species functional responses to climate variation.