Silvilaser 2019 - Poster Presentations »
Measuring tree heights in Eucalyptus stands in Brazil using Airborne Laser Scanning (ALS), accounting for error-in-variable induced bias
In the last two decades, the use of LIDAR (Light Detection and Ranging) has evolved from research to operational applications in forestry, demonstrating its use in even-aged and native forest inventories. Lidar sensors estimate a three-dimensional distribution of the forest canopies as well as the underlying topography, allowing for high-resolution topographic maps and detailed estimates of vegetation height, area cover, and canopy structure. Airborne Laser Scanning (ALS) is an active remote sensor used to characterize the surface of the earth producing a cloud of georeferenced points. This three-dimensional point cloud corresponds to a sample of the objects intercepting the laser light beam along the flight path returning a signal to the light sensor. Being a sample, the cloud of points is not exempt from errors or imprecisions along the way; therefore, models tailored to predict stand attributes need to be calibrated and corrected for such artifacts. The goal of this study is to develop an analysis framework to accurately estimate Eucalyptus heights by using Airborne Laser Scanning (ALS), solving the inherent bias present in Lidar calibrated models to improve the precision of forest stock estimation. The study location is a farm, called Santa Cruz do Sertãozinho, which is in the municipality of São Luiz do Paraitinga, São Paulo State, Brazil. The data was collected in 2012 when the Eucalyptus stands were 6 years old. We derived a Digital Elevation Model (DEM) from the data allowing for the derivation of a tree canopy model. Stand attributes like individual tree canopy size and height were isolated from the data using a watershed algorithm. We combined these individual metrics with ground-based measurements using regression analysis. To account for the inherent error in the dependent variables, the error-in-variable simulation extrapolation (Simex) algorithm was used to correct trees heights extracted from ALS based on the inventory plots. The results showed that height extraction from LiDAR data suffered from underestimation. The Simex algorithm, however, allowed for an improved unbiased estimation of tree heights, which brings a novelty to this type of modeling analysis. The unbiased estimation of tree height configures an essential step towards a better assessment and understanding of the forest stock not previously addressed in other research.