LiDAR & Point Cloud Processing for Forestry

Objective

Develop and implement automated workflows for forest inventory and biomass estimation using LiDAR point cloud data, with a focus on individual tree detection and attribute prediction.

Methods

Canopy Height Prediction

• Processed and normalized LiDAR data using RStudio and the lidR package
• Extracted plot-level metrics including height percentiles and density metrics
• Built regression models validated against field-measured data
• Generated wall-to-wall canopy height maps

Individual Tree Segmentation

• Applied Li2012 and Dalponte2016 algorithms for tree crown delineation
• Optimized segmentation parameters for mixed conifer forests
• Evaluated accuracy against reference data
• Extracted tree-level attributes (height, crown diameter, position)

LiDAR Data Processing

• Loaded and pre-processed LAS files using the lidR package
• Performed quality checks and filtering of noise points
• Generated Digital Elevation Models (DEMs) and Digital Surface Models (DSMs)
• Merged LiDAR with RGB imagery for enhanced visualization

Technical Highlights

# R code example for LiDAR processing and tree segmentation
library(lidR)
library(terra)
library(sf)

# Load and filter LiDAR point cloud
las <- readLAS("forest_lidar.las", filter = "-drop_z_below 0 -drop_z_above 60")

# Normalize point cloud heights
las_norm <- normalize_height(las, knnidw())

# Generate digital terrain model (DTM)
dtm <- grid_terrain(las, algorithm = tin())
plot(dtm, main = "Digital Terrain Model")

# Generate canopy height model (CHM)
chm <- grid_canopy(las_norm, res = 0.5, algorithm = p2r())
plot(chm, main = "Canopy Height Model")

# Individual tree segmentation using Dalponte algorithm
ttops <- find_trees(las_norm, lmf(ws = 3))
las_trees <- segment_trees(las_norm, dalponte2016(chm, ttops))

# Extract tree metrics
tree_metrics <- crown_metrics(las_trees, .stdtreemetrics)
head(tree_metrics)

# Visualize segmented trees
plot(las_trees, color = "treeID", size = 3)
                        

Outcomes

• Achieved high accuracy in canopy height prediction (R² = 0.84)
• Successfully segmented individual trees in complex forest structures
• Created precise forest attribute maps for ecosystem monitoring
• Developed methods for carbon stock assessment at landscape scales
• Generated Digital Terrain Models (DTMs) and Canopy Height Models (CHMs)

Key Findings

The analysis provided several important insights for forestry applications:

• LiDAR-derived height percentiles are strong predictors of forest biomass
• The Dalponte2016 algorithm outperformed Li2012 for coniferous tree segmentation
• Combining LiDAR with multispectral imagery improves species classification
• Point density significantly impacts the accuracy of individual tree detection
• Terrain complexity must be considered when normalizing LiDAR point clouds

Significance

This project demonstrates the power of LiDAR technology for detailed forest inventory and monitoring. The developed workflows enable cost-effective forest assessment at landscape scales, supporting sustainable forest management, carbon accounting, and ecological research. The methods can be applied to various forest types and integrated with other remote sensing data for comprehensive ecosystem analysis.