Latest LiDAR Articles from AGU 2010

Here are the latest papers / articles that have been presented as posters or orals at the AGU 2010 Fall meeting under "Active Remote Sensing Measurements of Vegetation 3-D Structure and Biomass: Assessing Accuracy and Sources of Uncertainty."

Follow me on twitter for the latest happening at the AGU 2010.

1. Effect of Ground Surface Reflectance on LiDAR Waveforms, Height Metrics and Biomass Estimation. B. D. Cook; J. Rosette; P. R. North; J. Rubio; J. Suárez.

Forest attributes such as aboveground woody biomass are commonly derived from LiDAR waveforms using empirical relationships with relative height metrics, i.e., percentiles of returned energy relative to the ground elevation. However, LiDAR waveforms and height metrics are strongly influenced by the reflectance properties of vegetation, soils, and litter at or near the ground surface, adding a level of uncertainty to biomass estimates. To estimate this range of uncertainty, we used FLIGHT, a ray tracing radiative transfer model that simulates single and multiple scattering, to simulate the effect of realistic ground cover types in a mixed, intermediate-aged forest near Howland, Maine, USA. Simulated cover types included sphagnum moss, lichen, leaf litter, bare soil, and snow, which were evaluated for forest canopy cover ranging from 27 to 98%. We discuss multiple scattering in feature waveforms (i.e., reflected energy above the ground peak) and attempts to reduce biomass uncertainty by excluding potentially confounding ground returns.

2. Estimating Above Ground Biomass using LiDAR in the Northcoast Redwood Forests. M. Rao; E. Stewart

In recent years, LiDAR (Light Intensity Detection Amplification and Ranging) is increasingly being used in estimating biophysical parameters related to forested environments. The main goal of the project is to estimate long-term biomass accumulation and carbon sequestration potential of the redwoods ecosystem. The project objectives are aimed at providing an assessment of carbon pools within the redwood ecosystem. Specifically, we intend to develop a relational model based on LiDAR-based canopy estimates and extensive ground-based measurements available for the old-growth redwood forest located within the Prairie Creek Redwoods State Park, CA. Our preliminary analysis involved developing a geospatial database, including LiDAR data collected in 2007 for the study site, and analyzing the data using USFS Fusion software. The study area comprised of a 12-acres section of coastal redwood (Sequoia sempervirens) in the Prairie Creek Redwoods State Park, located in Orick, CA. A series of analytical steps were executed using the USFS FUSION software to produce some intermediate data such as bare earth model, canopy height model, canopy coverage model, and canopy maxima treelist. Canopy maxima tree tops were compared to ground layer to determine height of tree tops. A total of over 1000 trees were estimated, and then with thinning (to eliminate errors due to low vegetation > 3 meters tall), a total of 950 trees were delineated. Ground measurements were imported as a point based shapefile and then compared to the treetop heights created from LiDAR data to the actual ground referenced data. The results were promising as most estimated treetops were within 1-3 meters of the ground measurements and generally within 3-5m of the actual tree height. Finally, we are in the process of applying some allometric equations to estimate above ground biomass using some of the LiDAR-derived canopy metrics.

3. Forest Biomass Mapping Using Lidar-derived Canopy Height Metrics at Maine in USA.
W. Huang; G. Sun

Forest biomass from regional to global level is important for underlying and monitoring the ecosystem responses to natural and human activities. Lidar provides the ability to directly measure canopy height index for aboveground biomass estimation. Our study site is located in Howland, Maine, United States. Data source consists of airborne medium footprint lidar data in 2009 and ground data from DESDynI field campaign in August 2009 and 2010. Canopy vertical structures are captured by the Laser Vegetation Imaging Sensor (LVIS) with entire return signal (i.e. in ~30 cm vertical bins). We first calculated height metrics (i.e. h10 to h100, totally 15 indices) by waveform decomposition using either Gaussian or numeric filter. Then, metrics were compared with RH indices at different levels: footprint of 20m diameter circle, squared plot of 25 x 25m, 50 x 50 m, 50 x 100 m and 50 x 200 m, respectively. At last, the biomass map was created. Height metrics from h50 to h80 show high correlation with biomass. Among them, h65 and h70 are the best, which is consistent with previous perspective that RH50 (or HOME, height of median energy) and RH75 have the best linear relationship with aboveground biomass. Comparison between h metrics and RH indices shows the latter one is better. In addition, both single and multi-variable linear regression model significant improvement with the increasing of field plot size.

4. Measuring Above Ground Biomass and Vegetation Structure in the South Florida Everglades Wetland Ecosystem with X-, C-, and L-band SAR data and Ground-based LiDAR. E. A. Feliciano; S. Wdowinski; M. Potts; S. Chin; D. A. Phillips

Worldwide, anthropogenic activities are disturbing and disrupting nutrient rich bio-diverse wetland ecosystems. Disturbance of the South Florida Everglades has been particularly acute, but difficult to quantify given its limited accessibility. Successful ecosystem monitoring requires the use of remote sensing. We used space-based Synthetic Aperture Radar (SAR) observations to estimate vegetation structure and above-ground biomass and track their changes over time. Our study leveraged three different SAR wavelengths that interact with different aspects of the vegetation. The short wavelength X-band (3.2 cm) signal interacts mainly with canopies; the intermediate wavelength C-band (5.6 cm) signal interacts with both canopies and branches; and the long wavelength L-band (24 cm) signal interacts with both the surface and lower portion of the vegetation. We used dual- and quadruple-polarization observations acquired from the TerraSAR-X, RadarSAT-2, and ALOS satellites. Different polarization data reflect radar signal interaction with different sections of the vegetation due to different scattering mechanisms.

In order to calibrate the multi-wavelength and multi-polarization SAR observations, we conducted field measurement in three vegetation communities: hammock, pine and cypress. Our ground measurements included both traditional forestry surveys and state-of-the-art Terrestrial Laser Scanning (TLS), a.k.a. ground based LiDAR surveys. A week long TLS survey was conducted in the Everglades National Park in the three calibrations sites using a Leica ScanStation C10 TLS instrument which utilizes a narrow, green (532 nm) laser beam. During this week we collected a total of 29 scans (33 GB of data). The TLS surveys provided centimeter resolution 3-D point clouds of the ground surface and below-canopy vegetation. Initial analysis of the data has provided detailed 3-D estimates of the vegetation structure and above ground biomass. A comparative analysis of the ability of the three bands of SAR to quantify above ground biomass in the different communities is presented. We also determine the essential bands needed to most efficiently estimate biomass. We find that the performance of SAR differs by community types. More rigorous data processing will provide important quantitative measures that will allow careful calibration of the remote sensing SAR data.

5. Modelling Sensor and Target effects on LiDAR Waveforms. J. Rosette; P. R. North; J. Rubio; B. D. Cook; J. Suárez

The aim of this research is to explore the influence of sensor characteristics and interactions with vegetation and terrain properties on the estimation of vegetation parameters from LiDAR waveforms. This is carried out using waveform simulations produced by the FLIGHT radiative transfer model which is based on Monte Carlo simulation of photon transport (North, 1996; North et al., 2010). The opportunities for vegetation analysis that are offered by LiDAR modelling are also demonstrated by other authors e.g. Sun and Ranson, 2000; Ni-Meister et al., 2001. Simulations from the FLIGHT model were driven using reflectance and transmittance properties collected from the Howland Research Forest, Maine, USA in 2003 together with a tree list for a 200m x 150m area. This was generated using field measurements of location, species and diameter at breast height. Tree height and crown dimensions of individual trees were calculated using relationships established with a competition index determined for this site. Waveforms obtained by the Laser Vegetation Imaging Sensor (LVIS) were used as validation of simulations. This provided a base from which factors such as slope, laser incidence angle and pulse width could be varied. This has enabled the effect of instrument design and laser interactions with different surface characteristics to be tested. As such, waveform simulation is relevant for the development of future satellite LiDAR sensors, such as NASA’s forthcoming DESDynI mission (NASA, 2010), which aim to improve capabilities of vegetation parameter estimation.

6. Reducing Uncertainty In Ecosystem Structure Inventories From Spaceborne Lidar Using Alternate Spatial Sampling Approaches. M. A. Lefsky; T. Ramond; C. S. Weimer

Current and proposed spaceborne lidar sensors sample the land surface using observations along transects in which consecutive observations in the along-track dimension are either contiguous (e.g. VCL, DESDynI, Livex) or spaced (ICESat). These sampling patterns are inefficient because multiple observations are made of a spatially autocorrelated phenomenon (i.e. vegetation patches) while large areas of the landscape are left un-sampled. This results in higher uncertainty in estimates of average ecosystem structure than would be obtained using either random sampling or sampling in regular grids. We compared three sampling scenarios for spaceborne lidar: five transects spaced every 850 m across-track with contiguous 25m footprints along-track, the same number of footprints distributed randomly, and a hybrid approach that retains the central transect of contiguous 25m footprints and distributes the remainder of the footprints into a grid with 178 m spacing. We used simulated ground tracks at four latitudes for a realistic spaceborne lidar mission and calculated the amount of time required to achieve 150 m spacing between transects and the number of near-coincident observations for each scenario. We used four lidar height datasets collected using the Laser Vegetation Imaging Sensor (La Selva, Costa Rica, Sierra Nevada, California, Duke Forest, North Carolina and Harvard Forest, Massachusetts) to calculate the standard error of estimates of landscape height for each scenario.
We found that a hybrid sampling approach reduced the amount of time required to reach a transect spacing of 150 m by a factor of three at all four latitudes, and that the number of near-coincident observations was greater by a factor of five at the equator and at least equal throughout the range of latitudes sampled. The standard error of landscape height was between 2 and 2.5 times smaller using either hybrid or random sampling than using transect sampling. As the pulses generated by a spaceborne laser are a valuable resource to be conserved, any strategy that decreases the number of observations required to develop large scale inventories with a given level of confidence should be pursued. Data fusion between lidar data and a spatially complete data source (e.g. polarmetric or interferrometric SAR) will also benefit from a spatially distributed sample of lidar as the average distance between any point and a lidar observation is greatly reduced. This study demonstrates that more flexible spatial arrangements of observations can result in estimates of average landscape height that have as little as one-third of the uncertainty of estimates made with an equal number of observations along transects. The method of sampling described here can be implemented by a technology, Electronically Steerable Flash Lidar, that can distribute observations in the patterns described here and simultaneously support transect sampling.

7. Sensitivity of LIDAR Canopy Height Estimate to Geolocation Error. H. Tang; R. Dubayah

Many factors affect the quality of canopy height structure data derived from space-based lidar such as DESDynI. Among these is geolocation accuracy. Inadequate geolocation information hinders subsequent analyses because a different portion of the canopy is observed relative to what is assumed. This is especially true in mountainous terrain where the effects of slope magnify geolocation errors. Mission engineering design must trade the expense of providing more accurate geolocation with the potential improvement in measurement accuracy. The objective of our work is to assess the effects of small errors in geolocation on subsequent retrievals of maximum canopy height for a varying set of canopy structures and terrains. Dense discrete lidar data from different forest sites (from La Selva Biological Station, Costa Rica, Sierra National Forest, California, and Hubbard Brook and Bartlett Experimental Forests in New Hampshire) are used to simulate DESDynI height retrievals using various geolocation accuracies. Results show that canopy height measurement errors generally increase as the geolocation error increases. Interestingly, most of the height errors are caused by variation of canopy height rather than topography (slope and aspect).

8. Validating LiDAR Derived Estimates of Canopy Height, Structure and Fractional Cover in Riparian Areas: A Comparison of Leaf-on and Leaf-off LiDAR Data. L. A. Wasser; L. E. Chasmer; A. Taylor; R. Day

Characterization of riparian buffers is integral to understanding the landscape scale impacts of disturbance on wildlife and aquatic ecosystems. Riparian buffers may be characterized using in situ plot sampling or via high resolution remote sensing. Field measurements are time-consuming and may not cover a broad range of ecosystem types. Further, spectral remote sensing methods introduce a compromise between spatial resolution (grain) and area extent. Airborne LiDAR can be used to continuously map and characterize riparian vegetation structure and composition due to the three-dimensional reflectance of laser pulses within and below the canopy, understory and at the ground surface. The distance between reflections (or ‘returns’) allows for detection of narrow buffer corridors at the landscape scale.

There is a need to compare leaf-off and leaf-on surveyed LiDAR data with in situ measurements to assess accuracy in landscape scale analysis. These comparisons are particularly important considering increased availability of leaf-off surveyed LiDAR datasets. And given this increased availability, differences between leaf-on and leaf-off derived LiDAR metrics are largely unknown for riparian vegetation of varying composition and structure. This study compares the effectiveness of leaf-on and leaf-off LiDAR in characterizing riparian buffers of varying structure and composition as compared to field measurements.

Field measurements were used to validate LiDAR derived metrics. Vegetation height, canopy cover, density and overstory and understory species composition were recorded in 80 random plots of varying vegetation type, density and structure within a Pennsylvania watershed (-77.841, 40.818). Plot data were compared with LiDAR data collected during leaf on and leaf off conditions to determine 1) accuracy of LiDAR derived metrics compared to field measures and 2) differences between leaf-on and leaf-off LiDAR metrics. Results illustrate that differences exist between metrics derived from leaf on and leaf-off surveyed LiDAR. There is greater variability between the two datasets within taller deciduous and mixed (conifer and deciduous) vegetation compared to shorter deciduous and mixed vegetation. Differences decrease as stand density increases for both mixed and deciduous forests. LiDAR derived canopy height is more sensitive to understory vegetation as stand density decreases making measurement of understory vegetation in the field important in the validation process. Finally, while leaf-on LiDAR is often preferred for vegetation analysis, results suggest that leaf-off LiDAR may be sufficient to categorize vegetation into height classes to be used for landscape scale habitat models.

I am tweeting the AGU 2010 events. Follow me on twitter @talksmart.