This paper describes the development of a multisensor UAV-based imaging platform. A sensor package consisting of a LiDAR, imaging spectrometer, and RGB camera were integrated with an inertial navigation system. This package is currently being flown on a hexacopter. A number of data acquisition missions have been conducted with this platform, including the imaging of a small active rotational mass movement, a wetland estuary, and several vineyards.

. A study was undertaken to examine the spectral properties of trees infected by mountain pine beetle (Dendrotonus ponderosae, Hopkins) in central British Columbia. A monoculture pine forest infected with mountain pine beetle was flown in early autumn (September), with a repeat flight the following summer (July). High spatial/spectral resolution VNIR-SWIR and discrete LiDAR data were collected. Three states were recognized: healthy, previsual green attack, and red attack. Spectra from the September flight were sampled from positions derived from LiDAR-modeled treetops. Analysis focused primarily on 3 absorption features: pigment feature (685nm), and 2 water absorption features (970nm and 1200nm). Derivative- and depth-normalized continuum removed (CR) spectra were generated for all 3 classes. Results obtained suggest that consistent differences in the shape of the pigment absorption feature are observed among the healthy, green, and red attack samples. The results of the analyses of the water features were less conclusive. The 970nm feature did not show any separation, whereas the 1200nm showed significant separation with the derivative and CR curves. These results indicate that, although subtle, reflectance differences between healthy and green attack crowns might provide a means for early detection of mountain pine beetle attack.

Airborne and spaceborne imaging spectroscopy applied to measuring foliar chemistry has received considerable attention in the literature. Typically, results are based on data measuring all the reflective components that make up a given pixel. This introduces confounding variables that cannot be easily modeled. Spectral unmixing methods yield estimates of the percentage endmember coverage in each pixel. This methodology fails to provide spectra representing variations in these specific components and thus is not as accurate for inferring chemistry. We report on the integration of airborne LiDAR data with high resolution imaging spectroscopy. We compared laboratory-based leaf-level pigment modeling with results from airborne data. In this comparison two airborne datasets were generated; one representing spectra composed of all reflective elements within a forested plot, and a second representing the top of the dominant/codominant canopy. Empirical modeling indicated that there is an influence on the spectral reflectance recorded over a defined area from the lower canopy levels. This influence did not, however, add to our understanding of forest biology and structure.

The use of narrow band remote sensing techniques for vegetation analysis is in growing demand in the field of precision viticulture. Hyperspectral reflectance data were examined through a leaf-level study to assess the feasibility of discriminating grape variety over a month long period through leaf senescence. The study included eight different varieties of Vitis vinifera L., and encompassed the full extent of terroir effects. First order derivative spectra were generated, and one-way ANOVA with Post-hoc tests were performed. Results indicated that the discrimination between varieties was possible through the visible and NIR regions. The greatest differences were observed between red and white grape varieties, and to a lesser extent amongst the white grape cultivars. Differences were attributed to leaf pigment content and leaf cell structure.

Foliar pigment concentrations have the potential to provide information regarding the physiological status of vegetation. Since foliar pigments cause wavelength specific absorption, these spectral regions and metrics derived there from, have been applied to estimate pigment concentrations. Some literature suggests that foliar attributes not related to chlorophyll concentrations influence the reflectance-pigment relationship. To investigate the appropriateness of these relationships across species with different pigment types and potentially different mesophyll cell structure, a dataset was collected throughout fall senescence. This dataset provided a wide range of pigment levels for five dissimilar tree species. Regression models were generated and compared. This data determined that the widely applied red edge position is sensitive to different tree species. Not only were different model coefficients found but occasionally different functions. The continuum removed and depth normalized left area appears to be a more robust alternative to REP for estimating chlorophyll.

This study used foliar chemistry samples as calibration data to address the use of high spatial and spectral resolution hyperspectral and LiDAR data to model and predict foliar chlorophyll. We used linear multiple regression models to derive three relationships: total plot reflectance only, total plot integrated with LiDAR structure, and top of canopy reflectance. Results of the modeling suggest that nonfoliar reflectors degrade the results of the modeling and that the use of LiDAR-defined structural descriptors do little to help resolve this. The top of the canopy with the highest S/N yielded the best results. Preliminary analysis of LiDAR-related canopy structure yields some clues into the relationships with reflectance.

This paper describes a new framework to the collection and fusion of multisensor airborne LiDAR and hyperspectral data. We describe a data fusion philosophy that provides a spatially precise positioning of hyperspectral data based on discrete first and last return LiDAR data. Three dimensional objects defined by the LiDAR data are then used to sample optimal spectra for subsequent analysis. The sampled spectra retain their positioning metadata and so can be mapped back into geographic space for further analysis. While the paper presents this philosophy within the context of a species classification, other analytical analysis can be performed.

LiDAR (Light Detection and Ranging) is currently being used to extract the biophysical characteristics of forests. LiDAR can provide extensive information about tree canopies; pulses reflected back to the sensor can represent understory vegetation as well as partial tree canopies below the dominant trees. Canopy structure can yield valuable clues regarding the biodiversity, and processes affecting the ecology of forest stands. In addition structural information can provide insight into other processes such as fire behaviour and the distribution of fuels. . The objective of this paper is to provide an algorithm to identify and delineate partial tree crowns underneath dominant canopies. The study area is located in the Greater Victoria Water District, west of Victoria, British Columbia, Canada. Nine plots were chosen to represent the study area. A complete census was conducted in the summer of 2005 to provide information about height of living crown (HLC), tree height, diameter at breast height (DBH), and tree dominance (based on the criteria: suppressed, intermediate, co-dominant and dominant). Using an algorithm previously developed by the Hyperspectral and LiDAR Research Group (Department of Geography, University of Victoria), treetops and canopies of the dominant trees must first be identified and delineated. Then, using the HLC as a threshold for identifying canopy height, it is possible to remove the LiDAR points representing the dominant tree crowns from the dataset. With this newly developed algorithm, it is possible to identify and delineate understory tree canopies.

The fusion of active LiDAR and passive optical hyperspectral data allows us to characterize the forest environments in ways that have not been possible previously with only one data source. This paper describes an airborne platform configured to collect data from multiple sensors simultaneously. Data from the platform have been applied to describe forest environments both in terms to species and structure. Integration of the data yields information and characterization of forest environments than has been possible in the past.