Back to Forest & Ecosystem Analysis

Overview

This project investigates whether LiDAR-derived structural metrics alone can support dominant tree species classification in the Petawawa Research Forest, Ontario, as part of Ontario's transition to an Enhanced Forest Inventory (EFI). ALS metrics were extracted at ground plot locations, explored for species separability, and evaluated against a classification framework. The analysis focuses on three species — White Pine, Red Pine, and Red Oak — and provides recommendations to the Ontario Ministry of Natural Resources on the operational viability of LiDAR-only species mapping.

Study site map showing the Petawawa Research Forest boundary in Ontario, Canada

Objectives

• Extract and process LiDAR (ALS) metrics at field plot locations
• Explore species distribution and data quality across the study area
• Identify the most informative ALS features for species classification using mutual information
• Evaluate whether LiDAR alone is sufficient for operational EFI species mapping
• Provide recommendations on integrating multispectral data if needed

Data Sources & Tools

Layer	Description
als_metrics.tif	2018 wall-to-wall ALS metrics raster — 67 bands at 25 m resolution
plot_locations.gpkg	Ground plot locations across the Petawawa Research Forest
plot_data.xlsx	Dominant species recorded at each ground plot
sp_codes.csv	Species name and code lookup table

2018 airborne LiDAR data coverage across the Petawawa Research Forest

Spatial distribution of dominant tree species across ground plot locations in the study area

Library	Purpose
geopandas	Reading and handling spatial plot location data
pandas	Data wrangling and tabular analysis
numpy	Numerical operations and array handling
rasterio	Reading multi-band ALS raster metrics
matplotlib	Data visualization
scikit-learn	Mutual information feature selection

Methods

ALS metrics were extracted at each ground plot location by reading the 67-band raster and aligning it with plot coordinates. Plots with missing LiDAR data were removed. Species with fewer than 15 observations were filtered out, leaving White Pine, Red Pine, and Red Oak for analysis. Mutual information scoring (mutual_info_classif) was used to rank ALS features by their correlation with species class labels. Box plots were generated for the top-ranking features to visually assess class separability. Results were interpreted in the context of EFI operational requirements.

Bar chart showing the number of ground plots per dominant species across the full dataset

For the code used in this analysis, click here

Outputs

Box plots of top LiDAR metrics (d32_34, p10, d26_28, vci_0.5bin) comparing class distributions across White Pine, Red Pine, and Red Oak

Key Findings

• Many species had insufficient plot observations for robust model training, limiting classification to three species
• Mutual information analysis identified a small subset of ALS metrics with meaningful correlation to dominant species
• LiDAR alone showed limited species separability, suggesting multispectral data integration would improve classification accuracy for an operational EFI workflow

Skills Learned

• Reading and extracting values from multi-band rasters using rasterio
• Spatial data handling with geopandas
• Data cleaning including NaN filtering and class frequency filtering
• Feature selection using mutual information (scikit-learn)
• Visualizing class distributions with box plots and bar charts using matplotlib
• Interpreting LiDAR structural metrics in a forest inventory context