Monitoring canopy-scale autumn leaf phenology at fine-scale using unmanned aerial vehicle (UAV) photography

Investigation of deciduous forest phenology is significant for understanding temperate deciduous forest ecosystems. Fine-scale perspectives on vegetation phenology have recently been carried out with the development of near-surface sensors, particularly uncrewed aerial vehicles (UAV). However, the capability of UAV-derived indices for canopy-scale phenology monitoring remains under-studied. In this study, a DJI Phantom 4 Multispectral (P4-M) UAV was used to evaluate the abilities of drone-derived indices for autumn canopy phenodates (the start (SOF), middle (MOF) and end (EOF) of leaf fall) monitoring in a temperate deciduous forest on the Loess Plateau. Using 104 drone acquisitions acquired in one year, a UAV-derived vegetation index (normalized difference vegetation index, NDVI) and two color-channel indices (green and red chromatic coordinate, GCC and RCC) were explored for tree-crown scale phenophases estimation. It was found that NDVI exhibited the best matches with visual interpretation in canopy autumn phenodate capture (correlation coefficient (R) = 0.69 for EOF). NDVI and GCC showed a higher consistency (R ≥ 0.51, root-mean-square-error ≤ 10.44 days and |bias| ≤ 6.71 days) in extracting the phenometrics than RCC, except for the MOF and EOF of evergreen trees. RCC obtained the earliest SOF due to its sensitivity to red leaves but failed to track canopy phenology, especially EOF, in mountainous areas because of its inability to eliminate illumination effects. Generally, NDVI showed the highest potential for estimating autumn tree-crown scale phenometrics, followed by GCC and RCC. Moreover, topography indirectly impacts vegetation phenophases by controlling vegetation type and composition. The results of this study indicate the considerable potential of UAV-derived indices, particularly NDVI and GCC, for autumn canopy phenology monitoring at a fine scale.