Structure Analysis and Classification of Boreal Forests Using Airborne Hyperspectral BRDF Data from ASAS

A new approach is presented for deriving vegetation INTRODUCTION canopy structural characteristics from hyperspectral bidiSeveral climate simulation models indicate that the rectional reflectance distribution function (BRDF) data. northern latitudes will experience a significant temperaThe methodology is based on the relationship between ture increase as a result of increasing levels of atmospectral variability of BRDF effects and canopy geomespheric CO2 (Schlesinger and Mitchell, 1987). With try. Tests with data acquired with the Advanced Solidglobal warming, species composition, structure, and pheState Array Spectroradiometer (ASAS) over Canadian nology of boreal forests could potentially change their boreal forests during the BOREAS campaign show that carbon balance. Due to their large spatial extent and biovegetation structural characteristics can be derived from mass, such changes could ultimately impact the global the spectral variability of BRDF effects. In addition, the carbon cycle (Myneni et al., 1997). Thus, it is crucial that incorporation of both BRDF effects and hyperspectral we improve our understanding of the ecological function resolution data substantially improve the classification of the boreal biome and carefully monitor changes in the accuracy. Best classification results are obtained when extent and structure of this forest. Considerable success hyperspectral resolution and BRDF data are combined, in mapping forest cover characteristics has been achieved but the improvement is not consistent for all classes. For in recent years through the use of remote sensing techexample, adding BRDF information to hyperspectral data niques (Meyer et al., 1993; Hall et al., 1997; Steyart et increases the overall classification accuracy for a six-class al., 1997), but most of these studies relied on nadir-view fen site from 37.8% to 44.7%. The addition, however, remultispectral reflectance data only and have largely neduces the accuracy for the jack pine class from 43.6% to glected bidirectional reflectance effects. Field data of the 28.8%. These new findings provide evidence for imbidirectional reflectance distribution function (BRDF) of proved capabilities for applications of MISR and MODIS boreal forests have shown that the structure of a forest data. The spectral resolution of MODIS is expected to be canopy is related to its BRDF characteristics (Deering et sufficient to derive canopy structural information based al., 1999), and modeling studies have revealed a strong relationship between vegetation canopy architecture and on the spectral variability of BRDF effects, and for MISR the hot spot phenomenon (Qin and Xiang, 1994). Nevera significant improvement of classification accuracies can theless, BRDF data have not been well utilized for derivbe anticipated from the combination of nadir reflectance ing canopy structure parameters. Only very recently has and off-nadir data. Elsevier Science Inc., 1999 the information content of remotely sensed BRDF data been examined (Barnsley et al., 1997; Russell et al. 1997; Abuelgasim et al., 1996). Hyperspectral remote sensing *USRA/NASA Goddard Space Flight Center, Biospheric Sciences imagery and processing techniques have also been used Branch, Greenbelt for forest applications (e.g., Martin et al., 1998), but the †NASA Goddard Space Flight Center, Biospheric Sciences Branch, Greenbelt potential for combining both high spectral resolution and Address correspondence to St. Sandmeier, NASA Goddard Space BRDF information has not yet been explored, to our Flight Center, Biospheric Sciences Br., Code 923, Greenbelt, MD 20771. knowledge, mainly due to a lack of adequate data. E-mail: [email protected] Received 23 November 1998; revised 10 March 1999. In this study, we use airborne hyperspectral BRDF