Mapping the rhythms of tropical forest canopy deciduousness using unmanned aerial vehicles
Tropical forest canopy deciduousness (TFCD) is an important indicator of productivity and the response of forests to climate change. Even so, knowledge of the relationship between environmental conditions and the timing (onset, duration, synchrony) of deciduousness in the tropics remains lacking, especially on a species population level. Optical remote sensing is sensitive to canopy deciduousness and can be used to augment field surveys over large areas, but typically cannot observe crown or species level dynamics due to tradeoffs between spatial and temporal resolution. Advances in 'personal remote sensing' via consumer grade unmanned aerial vehicles (UAVs) offers the potential to overcome these tradeoffs over small extents (10's – 100's hectares) and we pose the question of how TFCD can be tracked using this new technology. Here we use UAVs equipped with point-and-shoot digital cameras to characterize TFCD at the Barro Colorado Island (BCI) 50 hectare permanent plot in Panama. A weekly time-series of UAV red-green-blue imagery and three-dimensional (3D) canopy height models was collected over the 2014-2015 dry season during peak deciduous activity. We measured the relative greenness of each crown in UAV images every week and examined the degree to which TFCD could be tracked as a function of greenness. We also compared greenness within and across species to assess synchrony in TFCD.
Preliminary analysis reveals that deciduousness associated with a reduction in greenness can be detected within UAV imagery, however there are several complicating factors. For example, for one dominant deciduous species, Cavanilesia platanifolia, there was a strong negative trend in greenness at onset of deciduousness (R2 =0.50), but the weekly coefficient of variation within crowns of the same species was typically greater than 50%. Other species showed similarly high variation in weekly crown greenness. Variability in the TFCD greenness signal may be due to natural and methodological phenomena, including within species synchrony, UAV image lighting differences, and 'false-positive' greenness signals of green understory vegetation that becomes visible when overstory crowns are deciduous. Further work will examine the extent to which UAV information on 3D canopy structure can be used to further partition the greenness signal to that of overstory only. Personal remote sensing with UAVs shows great promise for improving understanding of the phenology of TFCD within and across species and at the scale of individual trees. These techniques can be deployed across sampling sites to improve understanding of TFCD within different forest types and across landscape gradients.