B41E-01
Rapid Retrieval of Forest Structure using an Under-Canopy, Upward-Scanning Lidar (EchinaR)
An under-canopy, upward-scanning, lidar field instrument, named EchidnaR, provides rapid, accurate, and automated measurements of forest stand structure, including tree diameters, stand basal area, stems per unit area, stand height, leaf area index, foliage profile, and foliage area volume density. We report results from scans at stands of varying characteristics at Harvard Forest (Massachusetts), Howland Experimental Forest (Maine), and Bartlett Experimental Forest (New Hampshire). Retrievals of stand characteristics are validated by ground measurements of tree stems in a large circular plot centered on the scan position; of canopy characteristics by LAI-2000 measurements, hemispherical digital photographs, and allometric calculations. The EchidnaR instrument, built by CSIRO and Ensis Australia, directs a horizontal 1064 nm laser beam with 5 mr divergence and a pulse rate of 2 kHz to a rotating mirror at 45° incidence to scan a vertical circle, recording data from +137° to –130° zenith angles and all azumuths as the instrument revolves 180° on a tripod mount. The return signal is sampled at 2 gigasamples per second, digitizing the full scattered waveform. The shape of the return pulse distinguishes readily between hard targets (tree boles, branches) and soft targets (leaves), and also measures the distribution of canopy gaps, including within-crown and between-crown gaps. The instrument is well-suited to ground sampling for calibration of airborne lidars, allowing accurate mapping of biomass and leaf area over large areas.
B41E-02 INVITED
Progress Towards Global Estimates of Forest Canopy Height from Geoscience Laser Altimeter System Waveforms
The vertical extent of waveforms collected by the Geoscience Laser Altimeter System (onboard ICESat - the Ice, Cloud, and land Elevation Satellite) increases as a function of terrain slope and footprint size (the area on the ground that is illuminated by the laser). Over sloped terrain, returns from both canopy and ground surfaces can occur at the same elevation. As a result, the height of the waveform (waveform extent) is insufficient to make estimates of tree height on sloped terrain, and algorithms are needed that are capable of retrieving information about terrain slope from the waveform itself. Early work on this problem used a combination of waveform height indices and slope indices from a digital elevation model (DEM). A second generation algorithm was developed using datasets from diverse forests in which forest canopy height has been estimated in the field or by via airborne lidar. Forest types originally considered in the development of the second generation algorithm included evergreen needleleaf, deciduous broadleaf and mixed stands in temperate North America, and tropical evergreen broadleaf forests in Brazil. Drawing on the work of our collaborators, we now include data from North America, European and Asian boreal forests, temperate needleleaf forests in Asia and North America, and tropical broadleaf forests in Central America, Africa and Southeast Asia. The resulting equations are the first step in creating global estimates of forest canopy height and aboveground biomass.
B41E-03
Using LiDAR Metrics to Characterize Forest Structural Complexity at Multiple Scales
Forest structure - the size and arrangement of trees and foliage - reflects a stand's history of initiation, growth, disturbance, and mortality. Because of this, studying the structure of forests can provide key insights into ecological processes, guides to silvicultural prescriptions to improve habitat, and assessments of forested landscapes. This study tested LiDAR metrics to characterize stands based on canopy structure. The study site was the 34,591 ha of forests in the Cedar River Watershed in western Washington State, USA. Stands ranged in age from <25 years old to >350 years old (including old-growth). Study sites spanned the western hemlock- Douglas fir (Tsuga heterophylla-Pseudotsuga menziesii), Pacific silver fir (Abies amabilis), and mountain hemlock (Tsuga mertansiana) forest zones. Eighty sample plots were used to ground truth the LiDAR data. A variety of structural indices were used to study canopy structural variations at the plot, stand, and landscape scales. The two most successful indices used the exposed geometry of the canopy surface: (1) the ratio of the canopy surface area to ground surface area (rumple index), and (2) the ratio of the volume beneath the canopy surface to maximum volume beneath the 95th percentile height (modified canopy volume method). These two indices integrated the spatial effects of tree heights, foliage distribution, and tree arrangement within 15m pixels. Variation between pixels revealed structural complexity at larger scales. Results: At the plot scale (~4 pixels), correlations with standard plot metrics (e.g., diameter at breast height) were similar to those reported by other studies. Comparison of structural complexity with age and height revealed a diversity of development pathways. The relationship between height and complexity allowed stands to be classified by the degree to which they have achieved their potential structural complexity, a new way to examine forest development. At the stand scale, the indices allowed spatial analysis of the patterns of structural variance. Analyzing these patterns would assist managers in designing forest management regimes based on natural stand complexity patterns. At the landscape scale, the indices allowed classification of stands based on structural complexity, which better differentiated stands than using only age or simpler structural element classifications. Old-growth stands, for example, were classified into four distinct groups based on their canopy structural complexity.
B41E-04
Ground-return Identification of Airborne LiDAR Data in a Forested Area Using Gaussian-fitting Models
Separating ground returns from non-ground returns is a key step for LiDAR-derived Digital Terrain Model (DTM) construction in many forest LiDAR-applications. Many algorithms are designed to identify ground returns in LiDAR data by using filters to identify the LiDAR return with the local lowest elevation. However, these methods are not always reasonable for creating an accurate DTM because of: (1) the Gaussian distribution of ground-reflected LiDAR elevations, even in an ideal flat area, caused by the noise of LiDAR instrument; and (2) the ground-slope and roughness influence on LiDAR data, especially in mountainous areas. Theoretically these conditions will lead to error or underestimation of the ground elevation (and DTM). A statistical method for determining ground elevations based on Gaussian-fitting models is developed in this research. The whole study area (1km by 1km) is divided into 200 by 200 small windows (or samples) and each of samples covers 5m by 5m area. For each sample, all LiDAR elevations within it are computed to form a single or bimodal Gaussian distribution depending on the ground-cover class (bare earth or forested surface, respectively). The ground cover classification is achieved by a supervised classification of the LiDAR intensity data. The ground elevation for each sample is assigned a value based on the mean of the ground-represented Gaussian distribution. More than 100 field observations will be sampled in Fall 2007 for an accuracy assessment. The field observations will compare field- measured ground elevations with LiDAR-derived elevations.
B41E-05 INVITED
Tropical Forest Vegetation Profiles and Biomass from Multibaseline Interferometric SAR at C- band
Interferometric synthetic aperture radar (InSAR) involves the reception of SAR signals at two spatially separated ends of a baseline. The resulting phase and coherence observations from InSAR are both sensitive to the vertical structure of vegetation. However, multiple InSAR observations—more than one phase-coherence-pair—are needed to estimate parameters describing vertical structure. Multiple observations can be made with different baselines, polarizations, or frequencies. This talk reviews why InSAR is sensitive to vertical structure. It then describes an experiment in the tropical forests of La Selva Biological Station in Costa Rica in which 12-14 baselines were used to estimate vegetation vertical profiles at C-band. Calibration of the InSAR phases and coherences with nearby pastures was essential for interpreting the data for vegetation, rather than surface, characteristics. Relative density profiles from primary, secondary, and selectively logged forests will be shown along with profiles from abandoned pastures. Field methods used to validate the profiles involve measuring individual tree dimensions, and the production of field profiles will be described and compared to InSAR profiles. Lidar profiles will also be shown for comparison. Functions of the InSAR profiles will be used estimate biomass of 30 stands
B41E-06
SAR, InSAR and Lidar studies for measuring vegetation structure over the Harvard Forest
In this paper we discuss ongoing studies of utilizing repeat-pass interferometric SAR, full waveform lidar, and radar polarimetry over the Harvard Forest in Massachusetts, for the purpose of characterizing vegetation three- dimensional structure. Polarimetric L-band Repeat-pass InSAR data is available over the region from the Japanese Space Agency's ALOS/PALSAR instrument, with a repeat period of 46 days. In 2003, the Laser Vegetation Imaging Sensor (LVIS) also flew over the area and provided extensive mapping of the regions true- ground surface topography and canopy height. The combination of these observations will provide a powerful combination for exploring the ability of the fundamental data types for estimating characteristics about the vegetation structure.