What's the point? Individual urban tree species identification using a fusion of LiDAR point-clouds and 4-band imagery

Keywords: Remote sensing, deep learning, urban forest, planning and management, health assessment
Abstract Type: Virtual Paper Abstract

Abstract

Urban trees mitigate heat, clean air and water, and provide residents with an improvement in general well-being. However, trees are threatened by climate change, disease, invasive species, and land-use change. Today, most urban tree inventories rely on time-intensive visual assessments deployed unevenly and are generally expensive to execute. The County of Los Angeles manages at least 800,000 public trees, with inventory costs using traditional methods exceeding $5 million. Currently, no department within the County has a cost-effective or efficient way to know where and when to plant or manage trees, and this is quite common among local governance structures around the world. We identify the best approach to urban tree species identification to help ease the burden of manually managing local tree stock.

The ability to inventory urban trees using remote sensing will provide the County with a faster, cheaper, and more efficient way to measure and monitor trees throughout the region. In our model, we overlap the GPS location of street trees with multi-source remote sensing imagery and use deep learning algorithms to develop critical tree metrics: Genus and species identification, and canopy structure. Our model also identifies:

- Trees vulnerable to changing climate (native/non-native);
- Trees vulnerable to drought and natural disasters (fire, tree fall, mortality); and
- Tree assessment of vulnerable populations (seniors, low-income, and immigrant communities).

Application of our model gives scientists and stakeholders the technical know-how to update urban forest inventories and retrieve critical tree health assessments to deploy their management resources more effectively.