Hindcasting tree heights in tropical forests using time-series unmanned aerial vehicle imagery

Abstract

Tree heights are pivotal for assessing forest ecology and carbon management, but conventional field methods are time-consuming and labor-intensive, making them impractical for regional monitoring and long-term repeated measurements. Retrieving surface height (digital surface model, DSM) from point cloud data using a low-cost unmanned aerial vehicle (UAV) has been a prevailing technology. With the availability of a high spatial resolution digital terrain model (DTM) from lidar (light detection and ranging), tree heights can be derived by correlating field height measurements and the differences between DSM and DTM. Moreover, tree growth over time can be modeled with the availability of historical field data. In this study, we utilized canopy heights acquired by UAV in 2014 from a secondary broadleaf forest and broadleaf/conifer plantations in a mountainous region of central Taiwan. The mean tree heights (MTH) at the plot scale were assessed by comparison with ground data. We then hindcasted MTH from 2010–2013 by calibrating the UAV data with historical field MTH, and the relationships between MTH growth and topography and climate were investigated. We found that the performance of a UAV in MTH derivation was satisfactory and that the model could explain 74% of the variation. The hindcasting analysis revealed that the broadleaf and conifer species grew 0.98 m y−1 and 0.65 m y−1, respectively, during the observation period. Topography can only explain a small portion of the data variation in tree growth. However, broadleaf trees grew faster for the north, northeast and east nominal aspect classes than those on the southwest, west and northwest sides, which may indicate the negative effect of solar radiation. Climate analysis demonstrated that both vegetation types responded similarly to climate. Monthly precipitation may facilitate the growth of MTH from May to October but had a negative impact in November and December, possibly due to relatively low air temperatures, rainfall and lower solar radiation retarding photosynthesis. The monthly mean temperatures in spring (March, April) and summer (July, September) in the major growing seasons had a positive effect on tree growth. By contrast, the mean temperatures in May and November-December were negatively related to MTH growth, which could be explained by phenology and unfavorable growing conditions, respectively. This study demonstrates the feasibility of utilizing a UAV for forest management and the potential for investigating the effects of topography and climatic attributes on tree growth. © 2020 Elsevier B.V.