Advancing vegetation monitoring with virtual laser scanning of dynamic scenes (VLS-4D): Opportunities, implementations and future perspectives

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Virtual Laser Scanning (VLS) is an established and valuable research tool in forestry and ecology, widely used to simulate labelled LiDAR point cloud data for sensitivity analysis, model training and method testing. In VLS, vegetation has traditionally been modelled as static, neglecting the influence of vegetation dynamics on LiDAR point cloud representations and limiting applications to mono-temporal analyses. In this review, we formalise VLS-4D, a framework that extends traditional VLS by using dynamic (i.e., 4D: 3D + time) input scenes. This advancement has opened new avenues for research on vegetation monitoring. We outline key concepts for representing dynamic scenes in LiDAR simulations, review technical implementations, and present innovative VLS-4D applications. We find that current simulation frameworks suitable for vegetation applications do not yet fully support dynamic scenes. While LiDAR time series of vegetation growth can be generated from static scene snapshots, simulating the effects of vegetation movement during a scan remains a challenge. We group the reviewed applications of VLS-4D into three key methodological areas: i) investigating LiDAR data acquisition and vegetation movement effects, ii) testing and validating new methods for change detection and analysis, and iii) generating labelled training data for machine and deep learning. We recommend that future efforts focus on extending the functionality of current LiDAR simulators and increasing the availability of open-source tools for modelling dynamic vegetation to enable more realistic simulations. Used as a complement, not a replacement, to real data, VLS-4D has the potential to significantly advance LiDAR-based vegetation monitoring by improving our understanding of point cloud representations, enabling reliable algorithm validation, and providing high-quality training data for deep learning.

Article activity feed