Forest structure estimation using SAR, lidar, and optical data in the Canadian Boreal forest

Michael L. Benson, The University of Michigan, mlbenson@umich.edu (Presenter)
Leland E. Pierce, The University of Michigan, lep@umich.edu
Kathleen M. Bergen, The University of Michigan, kbergen@umich.edu
Kamal Sarabandi, The University of Michigan, saraband@umich.edu
Kailai Zhang, The University of Michigan, kailaiz@umich.edu

One of the most fundamental new technical challenges of a DESDynI space-borne mission is the fusion of the several sensor modalities - LiDAR, SAR, InSAR, and Optical - in order to accurately estimate desired 3D vegetation structures and biomass parameters in areas where the sensors overlap, and to extrapolate them over continuous areas where lidar data is absent.

The objective of this paper is to use measured datasets in conjunction with our sensor forward models to develop and validate an estimation algorithm that fuses various remote sensing technologies with a minimal amount of ground information and yields an accurate estimate of forest structure, including biomass, canopy height, and tree species.

Several recent studies have focused on assessing accuracies of forest structure estimates using two or more sensors including LIDAR, radar, and Visible-Infrared (VIR). These studies, based primarily on empirical analysis, have produced rms error assessments and interpretations of sensor utilities, and have also suggested development of physical-based models as a needed advancement. One of the first attempts at fusion of four modalities (VIR, SAR/INSAR, and LIDAR) used the following features: height from LIDAR, vegetation community from VIR, INSAR heights, and SAR powers. This study used the empirical Bayesian approach and demonstrated that using this approach, multiple features quickly resulted in poorly-estimated multi-dimensional density functions. In a step towards physical or model-based methods, Moghaddam [2002] used AirSAR and Landsat TM to estimate foliage mass. Their method is a combination of Bayesian and model-based, with training areas used to develop forward models instead of probability density functions, yielding an improvement in the error over VIR-alone from about 30\% to 15\%. Recently Kimes studied fusion of lidar with multi-angle and used an optical model.

Extrapolation of LIDAR heights using VIR has been used in Hudak [2002], where an empirical relationship between the two was used with kriging and cokriging. The study concluded that the spacing of the LIDAR data needed to be 250 meters or less for an accurate extrapolation. SRTM (INSAR), Landsat (tasseled-cap), and a canopy density layer were used to extrapolate LIDAR heights, using non-spatial regressions, resulting in rms errors of 3 meters.

Working in the Boreas Southern Study Area, this paper will explore the potential to extract height and biomass parameters from the fusion of Lidar, SAR, and optical data. Comparisons will be made to known ground truth data as well as to estimates of both the canopy height and biomass made using the SIR-C instrument as a part of the BOREAS study project.

Presentation Type:  Poster

Session:  Global Change Impact & Vulnerability   (Tue 11:30 AM)

Associated Project(s): 

• Sarabandi, Kamal: VEGEX3D: LiDAR-SAR/InSAR Extrapolation and Simulation Models for Retrieving Vegetation 3D Structure and Biomass ...details

Poster Location ID: 113