B32A-01 INVITED
The remote characterization of vegetation using Unmanned Aerial Vehicle photography
Unmanned Aerial Vehicles (UAVs) can fly in place of piloted aircraft to gather remote sensing information on vegetation characteristics. The type of sensors flown depends on the instrument payload capacity available, so that, depending on the specific UAV, it is possible to obtain video, aerial photographic, multispectral and hyperspectral radiometric, LIDAR, and radar data. The characteristics of several small UAVs less than 55lbs (25kg)) along with some payload instruments will be reviewed. Common types of remote sensing coverage available from a small, limited-payload UAV are video and hyperspatial, digital photography. From evaluation of these simple types of remote sensing data, we conclude that UAVs can play an important role in measuring and monitoring vegetation health and structure of the vegetation/soil complex in rangelands. If we fly our MLB Bat-3 at an altitude of 700ft (213m), we can obtain a digital photographic resolution of 6cm. The digital images acquired cover an area of approximately 29,350sq m. Video imaging is usually only useful for monitoring the flight path of the UAV in real time. In our experiments with the 6cm resolution data, we have been able to measure vegetation patch size, crown width, gap sizes between vegetation, percent vegetation and bare soil cover, and type of vegetation. The UAV system is also being tested to acquire height of the vegetation canopy using shadow measurements and a digital elevation model obtained with stereo images. Evaluation of combining the UAV digital photography with LIDAR data of the Jornada Experimental Range in south central New Mexico is ongoing. The use of UAVs is increasing and is becoming a very promising tool for vegetation assessment and change, but there are several operational components to flying UAVs that users need to consider. These include cost, a whole set of, as yet, undefined regulations regarding flying in the National Air Space(NAS), procedures to gain approval for flying in the NAS(FAA Certificate of Authorization), and training(remote control piloting, UAV-specific instruction, FAA ground school and testing, FAA observer procedures, FAA medical Class 2 exam, and a private pilot's license). The relevance and need of all these to developing a UAV capability will be explained. While working through the necessary requirements above, we have also learned that we need to know how to handle extremely large and easily acquired data sets as well as to develop tools to orthorectify and mosaic individual UAV images for analysis.
B32A-02
The contribution of vegetation cover and bare soil to pixel reflectance in an arid ecosystem
The heterogeneity of vegetation and soils in arid and semi-arid environments complicates the analysis of medium spatial resolution remotely sensed imagery. A single pixel may contain several different types of vegetation, as well as a sizeable proportion of bare soil. We have used linear mixture modeling to explore the contribution of vegetation cover and bare soil to pixel reflectance. In October, 2006, aerial imagery (0.25 m spatial resolution) was acquired for our study sites in the Jornada Experimental Range, southern New Mexico. Imagery was also acquired from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) for June and November, 2006. These data corresponded with pre- and post monsoon conditions. Object-based feature extraction was used to classify the aerial imagery to shrub, grass and bare ground cover classes. Percent cover was then calculated for each cover class. Visible-near-infrared and shortwave infrared ASTER reflectance data from both dates were combined into a single 18-band dataset (30 m spatial resolution). A vector overlay from the classification results of the aerial imagery was used to define pure endmember pixels in the ASTER imagery. Estimates of the proportions of shrub, grass and bare ground cover from the linear mixture modeling approach were compared with cover calculated using feature extraction from the aerial imagery. The results indicate that reflectance in ASTER pixels is likely to be a linear combination of the cover proportions of the three main cover types (shrubs, grass, bare ground). However, noticeable outliers in the relationship between cover calculated from each method, indicate there may be other variables that affect the accuracy with which we can estimate cover using linear mixture modeling.
B32A-03
Comparison of forest structure between two regions in Brazilian Amazon using LiDAR data
Characterizing the landscape dynamics of forested areas, including changes in forest structure caused by natural gap disturbances or human activities, is a key source of information to support studies of ecological processes such as secondary succession, the CO2 cycle, maintenance of tree diversity and community dynamics, and for development of Reduced Emissions from Degradation and Deforestation (REDD) mechanisms under the UN Convention on Climate Change. Airborne lidar (Light Detection And Ranging) sensors have been demonstrated to be a useful tool to quantify canopy structure complexity by directly obtaining measurements of key forest structural characteristics, such as canopy height, distribution of intercepted surfaces, crown height and width, and quantity of aboveground biomass. Airborne lidar data was collected for seventeen sites (4000 ha) in the Manaus (AM) and Santarém (PA) regions of Brazil in June of 2008 at data densities between 3.7 and 9.0 shots per m2. We have created high resolution digital terrain models and canopy surface models for these sites and are using them to compare relationships between stand height and biomass in each region. We are also investigating the correlation between species composition and forest canopy structure. These two objectives depend on a third: a method for describing these forests' complex canopy structure. In this work, we present an analysis of the association between multiple canopy structure metrics, including those based on height and height variability, the vertical profile of lidar returns, and canopy volume. New indices of canopy structure, such as the height, density, and mean surface area of emergent trees are introduced and compared to these existing approaches. A comparison between the regions is ongoing to analyze the forest structure variability within and among these study sites.
B32A-04
Mapping Potential Ivory Billed Woodpecker Habitat using Lidar and Hyperspectral Data Fusion
Multisensor fusion is a powerful approach towards characterizing forest structure for effective management of wildlife habitats. The rediscovery of the Ivory Billed Woodpecker in 2005 reinforced the need to map and conserve suitable habitat for the previously thought extinct bird. In this study we fused waveform lidar and hyperspectral data to map potential habitat for the woodpecker along the Lower Mississippi Valley of Arkansas. Laser Vegetation Imaging Sensor (LVIS) data was processed to produce high-resolution forest structure maps. We used multiple endmember spectral mixture analysis (MESMA) to map stressed and dead vegetation from the Airborne Visible Infrared Imaging Spectrometer (AVIRIS) data. LVIS and AVIRIS maps were fused to identify habitat hot-spots based on historical records of habitat preferences of the bird. Results indicate several small hotspots in the bottomland hardwood forests, but very few large and continuous patches qualify as potential woodpecker habitat. Results from this study are expected to aid search efforts for the woodpecker and also provide useful insights into lidar fusion for large scale habitat mapping.
B32A-05
Integration of Lidar and Hyperspectral Remote Sensing to Examine the Influence of Tree Species Arrangements on Site Estimates of Biophysical Variables, LAI, fPAR, and GPP
In mixedwood environments, the spatial arrangement of tree species and species-groupings can have a significant impact on stand-level estimates of biophysical structure and physiological variables such as above-ground biomass, leaf area index (LAI), the fraction of photosynthetically active radiation absorbed by the canopy (fPAR), and foliar biochemistry. This heterogeneity can influence carbon exchange measured on flux towers, depending on the location of the flux footprint (which varies according to micrometeorological conditions such as wind speed, wind direction, relative humidity, and temperature). As a result, site-level estimates of LAI, fPAR, and GPP derived from coarse-resolution remote sensing (e.g., MODIS), and used for global modeling, do not necessarily coincide with flux tower measurements. This represents a serious challenge for carbon exchange monitoring over mixedwood environments. We used highly detailed airborne discrete lidar and hyperspectral data to map biomass, basal area, LAI, fPAR, and canopy biochemistry for a boreal mixedwood environment in northern Ontario, Canada. Distinct spatial patterns are evident and influence the total photosynthesis for the area. We demonstrate a decrease in tower-measured GEP when influenced by a large homogenous patch of black spruce within the tower footprint and the challenges of representing this within at a coarse-resolution pixel. Under certain conditions, site heterogeneity may introduce error when validating or parameterizing global remote-sensing-based models with flux tower data.
B32A-06
Predicting Biomass in Temperate Hardwood and Mixed Forests Using VHF Radar, Interferometric SAR and Discrete-Return Lidar Data
Our working hypothesis is that biomass models derived from radar variables or radar-lidar synergy will provide greater accuracy in hardwood and mixed forests than those derived using lidar variables only. To provide the data to test this hypothesis, BioSAR, profiling lidar (PALS), imaging lidar, GeoSAR, and field data were collected over hardwood and mixed forests of Appomattox-Buckingham State Forest, Virginia, USA in 2007 and 2008. We are evaluating BioSAR, GeoSAR and lidar data for their ability to provide reliable biomass estimates both individually and collectively. Lidar derived variables include canopy height, crown diameter, and distributional parameters (mean, standard deviation, percentiles, etc.). GeoSAR variables include the canopy height model produced by differencing the X- and P-band interferometric heights. The BioSAR variables consist of the cross-sectional areas from six different frequencies. Field data consist of a network of 64 fixed plots and 256 variable radius plots on which standard forest measurements were made. Both parametric (best subsets regression) and non-parametric (regression tree) approaches are being used for variable selection and model evaluation.
B32A-07
Allometric constraints to inversion of canopy structure from remote sensing
Canopy radiative transfer models employ a large number of vegetation architectural and leaf biochemical attributes. Studies of leaf biochemistry show a wide array of chemical and spectral diversity that suggests that several leaf biochemical constituents can be independently retrieved from multi-spectral remotely sensed imagery. In contrast, attempts to exploit multi-angle imagery to retrieve canopy structure only succeed in finding two or three of the many unknown canopy arhitectural attributes. We examine a database of over 5000 destructive tree harvests from Eurasia to show that allometry - the covariation of plant form across a broad range of plant size and canopy density - restricts the architectural diversity of plant canopies into a single composite variable ranging from young canopies with many short trees with small crowns to older canopies with fewer trees and larger crowns. Moreover, these architectural attributes are closely linked to biomass via allometric constraints such as the "self-thinning law". We use the measured variance and covariance of plant canopy architecture in these stands to drive the radiative transfer model DISORD, which employs the Li-Strahler geometric optics model. This correlations introduced in the Monte Carlo study are used to determine which attributes of canopy architecture lead to important variation that can be observed by multi-angle or multi-spectral satellite observations, using the sun-view geometry characteristic of MODIS observations in different biomes located at different latitude bands. We conclude that although multi-angle/multi-spectral remote sensing is only sensitive to some of the many unknown canopy attributes that ecologists would wish to know, the strong allometric covariation between these attributes and others permits a large number of inferrences, such as forest biomass, that will be meaningful next-generation vegetation products useful for data assimilation.
B32A-08 INVITED
Gradient Analysis: A new paradigm in vegetation modeling
Vegetation remote sensing has commonly depended on a-priory definitions of vegetation communities that do not necessarily account for scale or specific applications. These classification schemes convolve species relationships and provide discrete representation of data that is inherently continuous in nature. We introduce an analytical hierarchy, where a continuous gradient of vegetation occurrence, structure, or suitability is the foundation, and all subsequent levels in the hierarchy are derived from this gradient. By starting with a continuous measurement of vegetation a coherent down-scaling strategy can be developed, thus avoiding many statistical and aggregation issues. This analytical framework allows for integration of ecological theory including niche, adaptation, and meta-populations. We use a random forest niche model and Lidar derived structural variables to demonstrate vegetation gradients. We then introduce a few simple landscape metrics for analyzing gradients. Finally, we demonstrate how this data can be integrated into analysis addressing climate change and habitat relationships.