Crown Structure Explains the Discrepancy in Leaf Phenology Metrics Derived from Ground- and UAV-Based Observations in a Japanese Cool Temperate Deciduous Forest

Atsuhiro Iio,Hiromi Mizunaga,Noviana Budianti

doi:10.3390/f12040425

Atsuhiro Iio, Hiromi Mizunaga + Show 1 more

Open Access

https://doi.org/10.3390/f12040425

Copy DOI

Abstract

Unmanned aerial vehicles (UAV) provide a new platform for monitoring crown-level leaf phenology due to the ability to cover a vast area while offering branch-level image resolution. However, below-crown vegetation, e.g., understory vegetation, subcanopy trees, and the branches of neighboring trees, along with the multi-layered structure of the target crown may significantly reduce the accuracy of UAV-based estimates of crown leaf phenology. To test this hypothesis, we compared UAV-derived crown leaf phenology results against those based on ground observations at the individual tree scale for 19 deciduous broad-leaved species (55 individuals in total) characterized by different crown structures. The mean crown-level green chromatic coordinate derived from UAV images poorly explained inter- and intra-species variations in spring leaf phenology, most probably due to the consistently early leaf emergence in the below-crown vegetation. The start dates for leaf expansion and end dates for leaf falling could be estimated with an accuracy of <1-week when the influence of below-crown vegetation was removed from the UAV images through visual interpretation. However, a large discrepancy between the phenological metrics derived from UAV images and ground observations was still found for the end date of leaf expansion (EOE) and start date of leaf falling (SOF). Bayesian modeling revealed that the discrepancy for EOE increased as crown length and volume increased. The crown structure was not found to contribute to the discrepancy in SOF value. Our study provides evidence that crown structure is a pivotal factor to consider when using UAV photography to reliably estimate crown leaf phenology at the individual tree-scale.

Highlights

Introduction published maps and institutional affilPlant phenology, which covers the annual developmental dynamics in plants, has long been recognized as a critical driver of ecosystem processes such as carbon assimilation and evapotranspiration.It has been suggested to be heavily influenced by climate [1]
Intra-species variations in the phenological transition dates were generally smaller than the inter-species variations, with the exception of Prunus grayana in the autumn
The start dates of leaf expansion (SOE) and end dates of leaf falling (EOF) derived from crown leaf cover determined by visual interpretation of Unmanned aerial vehicles (UAV) images (CLCUAV ) were associated with those derived from crown leaf cover based on ground observation (CLCground ) across all 19 studied tree species, with an accuracy of about six days (Figure 3)

Summary

Introduction

Introduction published maps and institutional affilPlant phenology, which covers the annual developmental dynamics in plants (e.g., from bud burst in the spring to leaf falling in the autumn), has long been recognized as a critical driver of ecosystem processes such as carbon assimilation and evapotranspiration.It has been suggested to be heavily influenced by climate [1]. Assessing and monitoring phenological dynamics are key requirements to improving our understanding of how plants respond to climate change and how this influences forest ecosystems. The leaf phenology of large trees (i.e., crown leaf phenology) has been traditionally evaluated with ground-based, visual observations, such as determining crown leaf cover (e.g., [2,3]). This approach is useful for individual tree-level monitoring, it is labor intensive, prone to human error and difficult to employ in tall, dense and multi-layered forests where an observer cannot see the upper part of the crown. Near-surface remote sensing, such as tower-mounted cameras (“phenocams”), provides the step for individual-level iations

Methods

Results

Discussion

Conclusion