Quantification of central hardwood forest structure and aboveground biomass with low-density airborne lidar data

Gang Shao, Purdue University


Sustainable forest management requires forest inventory information at the individual tree level. Lidar-based algorithms developed for individual tree detection commonly focus on finding individual tree tops to discern tree positions. However, the deliquescent tree forms (broad flattened crowns) in temperate hardwood forests make these algorithms ineffective, particularly when tree density is high. Moreover, lidar-based aboveground biomass (AGB) estimation models for temperate hardwood and hardwood-dominated mixed forests are scant due to difficulties in estimating strong site-to-site variations in height and diameter growth rates within complex, multi-species, and often multi-aged hardwood forests. In this study, I developed a point-cloud-based algorithm and a raster-based algorithm to detect individual trees using relatively low density lidar data (2.3 pts m-2) in two dense temperate hardwood forests in Indiana, USA. Experiments revealed the point-cloud-based algorithm with a stepwise framework improved the time efficiency of 3D approaches and produced promising hardwood tree detections with 93% accuracy. By incorporating a clustering-based concept into the raster-based algorithm, the accuracy of this 2D hardwood tree detection approach increased from 82% to 89%. Both the point-cloud- and raster-based algorithms were able to explain around 80% structural variance in horizontal and vertical phases. I also developed a robust statistical model to estimate aboveground biomass of temperate hardwood forests across gradients of site productivity in southern Indiana, USA. This multiplicative nonlinear regression model incorporated lidar-derived metrics and soil features and collapsed soil-based site productivity in two classes, high (HPS) and low (LPS); the use of this site productivity in the lidar-based models resulted in estimations with a R2 value of 0.81, which explained 14% more variance than the best fit multiplicative models without site productivity included. Moreover, I found that the relationship between AGB and lidar-based metrics was nonlinear on LPS and nearly linear on HPS, indicating that site productivity is critical to include in lidar-based biomass model for the temperate hardwood and mixed forests.




Shao, Purdue University.

Subject Area

Forestry|Natural Resource Management|Engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server