Content area
Abstract
The transparency and shape of tree crowns in urban areas play a critical role in applications such as solar energy potential estimation, urban microclimate modeling, and urban landscape planning and architecture [1–3]. Modeling the parameters of tree crowns necessitates high-resolution three-dimensional data such as LiDAR data. However, the availability of high-resolution LiDAR data is limited and not widely available. Therefore, modeling the crown parameters over large spatial extents such as an entire city is limited by data availability. However, many government agencies provide comparatively lower resolution Aerial LiDAR data over large spatial extents including urban areas [4]. These aerial LiDAR datasets offer a great potential to model the crown attributes of trees over large spatial extents.
The research objective of this dissertation is to improve the modeling capabilities of tree crown attributes from widely available medium-resolution aerial LiDAR data. Research objectives include: (i) to model the seasonal transparency of trees from aerial LiDAR data to be applied in solar energy potential estimation over urban areas (ii) to model the shape of tree crowns at an individual tree level from aerial LiDAR data (iii) to compare the performances of Individual Tree Delineation algorithms on urban study areas.
Existing research has extensively explored the application of aerial LiDAR data in estimating solar energy potential over urban sites such as building rooftops [5]. However, they have considered only the shadows cast by the terrain and buildings and ignore features, such as vegetation and their seasonal transparency variations. We need an approach to improve the solar energy potential estimates by modeling the seasonal transparency of tree crowns present in a study area. To improve the solar energy potential estimates, we modeled the transparency of tree crowns by computing a light penetration factor for trees from field photographs.
There are numerous studies attempting various techniques for reconstructing three dimensional models of trees from high resolution LiDAR data [6,7]. However, they do not scale down to comparatively lower resolution aerial LiDAR data. We need an approach to model the 3D structure of real-world tree crowns from aerial LiDAR data that is not cost-prohibitive. To this end, I developed a deep learning model that could match individual trees in aerial LiDAR to a pre created library of 3D models of real-world trees created using Structure from Motion from field photographs.
There are many studies that have developed Individual Tree Delineation (ITD) techniques from LiDAR data [8,9]. However, they were developed and tested for forested areas. These algorithms may not fare well with urban areas where these algorithms may get confounded with the complexities of manmade objects near trees. We therefore tested the performance of three ITD algorithms over two different urban settings.
