Canopy Mapping

A significant portion of my consulting work has focused on remote sensing analysis in the forestry sector, with a particular emphasis on mapping canopy cover. This page provides a brief overview of my approach to canopy mapping, past projects I’ve contributed to, and my research in the field.

Legend

This sample dataset shows the results of canopy mapping, individual tree detection, crown segmentation, and classification. Each tree is represented by a delineated polygon of its crown. The map overlays high-resolution orthophotos and a lidar-derived canopy height model, both of which were use as inputs in the process.

Methodology

Benefits

Mapping tree canopy helps local governments understand and manage urban forests. These maps are key components of urban tree canopy assessments , which provide the information needed to protect tree health, plan drainage, detect pests, and enhance both biodiversity and community life.

Data Sources

Canopy mapping can be performed at scales ranging from city-wide surveys to individual trees. High-resolution imagery combined with lidar typically provides the most precise and detailed maps and captures subtle variations in canopy structure that coarser data often miss.

The canopy mapping process — **Steps in the canopy mapping process**

1. *Input layers*: Lidar point cloud (and derivatives) and orthographic imagery.

2. *Intermediate layers*: Spectral, structural, and textural indices derived from the input data.

3. *Output layer*: A specialized AI model integrates the intermediate layers to produce a pixel-based canopy map.

Further analyses, such as tree-level detection, delineation, and classification (see below), are performed using this canopy map.

Individual Tree-Level Analysis

Mapping out the extent of an urban tree canopy is often only the first step. Individual trees can be detected and delineated using a variety of specialized tools and algorithms (ex.: ForestTools). This, in turn, enables a wide range of analyses and insights.

Measurements like stem height or crown volume can be extracted from lidar. Trees can be classified based on their spectral or structural characteristics. Groupings could include health status or whether they are coniferous or deciduous. In some cases, it is even possible to achieve classification at the species level.

Examples of individually delinated and classified trees — Left: an orthographic aerial image. Centre: individually outlined trees overlayed on a canopy height model. Right: trees coloured according to their class
(red = dead, blue = coniferous, green = deciduous).

Previous Work

Since 2013, I have produced high-resolution canopy maps in the following communities in Canada and the United States.

Area Mapped (km²)	XXX
Year(s) Mapped	XXX

Future Direction

The tools available for canopy mapping are evolving rapidly. While my previous work relied heavily on traditional machine learning techniques for classifying imagery and lidar data (ex.: segment-based classification using randomForest), these approaches are being gradually supplanted by more sophisticated tools.

Neural networks are able to incorporate deeper contextual information and be applied to a wider range of complex, heterogenous environments. I'll be using this page to discuss and share my work in developping models for canopy mapping in the future.