Canopy Reflectance Model Inversion in Multiple Forward Mode: Forest Structural Information Retrieval from Solution Set Distributions

Remote estimation of canopy structure is important in forestry and a variety of environmental applications. Multiple Forward Mode (MFM) look-up table (LUT) inversion of canopy reflectance models is one approach for obtaining forest canopy biophysical-structural information (BSI). MFM provides inversion results from models that are not invertible directly, and has advantages in terms of software requirements, model complexity, computational demands, and provision of physically-based BSI output. Proper handling of MFM-LUT parameterization and inherent uncertainty in the inversion procedure at the critical final BSI retrieval stage is essential, and is the theme of this paper. Three approaches are presented for deriving BSI from MFMLUT multiple solution sets: reflectance equality (REQ), nearest spectral distance (NSD), and spectral range domain (SRD). These approaches were validated at a Rocky Mountain test site, for which SRD corresponded best with field data, with RMSE 0.4 m and 0.8 m obtained for horizontal and vertical crown radius, respectively. Recommendations for selecting MFM inversion approaches are provided for future applications. Introduction Remote estimation of forest canopy structure is important in forest inventory and plays a key role in forest fire modeling, forest management, carbon estimates, and climate change studies (Hall, 1999; Franklin, 2001; UNFCCC, 2004; Patenaude et al., 2005). Forest stand characteristics obtained by remote sensing image analysis and modeling (e.g., canopy dimensions, stand density, fraction of cast shadow) can be related to a number of important biophysical variables including canopy volume Canopy Reflectance Model Inversion in Multiple For ward Mode: For est Str uctural Information Retrieval fr om Solution Set Distributions S.A. Soenen, D.R. Peddle, C.A. Coburn, R.J. Hall, and F.G. Hall and bulk density (Riano et al., 2004), stem volume (Pilger et al., 2003), biomass (Fournier et al., 2003), and leaf area (Peddle et al., 2004). These biophysical parameters are important in afforestation, reforestation, and deforestation contexts in countries committed to sustainable development and international carbon reporting (Brown, 2002) as well as more generally for monitoring and change detection applications (Gong and Xu, 2003). Further, remote sensing is recognized as an important approach to derive biophysical-structural information (BSI) required as Essential Climate Variables (ECV; see UNFCCC, 2004) in international global climate change agreements (e.g., the Kyoto Protocol; UNFCCC, 1997) because of opportunities to obtain systematic, repetitive information at local to global scales with archival imagery dating back to baseline years (e.g., Kyoto in 1990) that is critical in carbon accounting and policy compliance (Rosenqvist et al., 2003; Patenaude et al., 2005). Canopy reflectance models link airborne and satellite image spectral response with the biophysical and structural composition of a forest canopy. These models are comprehensive in describing explicitly and quantitatively the main parts of the system being measured by a remote sensing instrument, that is, the bidirectional reflectance of forest canopies as a function of canopy structure, illumination and viewing positions, surface geometry, landscape component spectral properties, and sub-pixel scale abundance (Strahler, 1997). Compared to other methods, most canopy reflectance models have an explicit physical basis and thus, a fundamental advantage over traditional empirical methods that are mired in a purely statistical domain. For example, empirical relationships between canopy structural parameters and vegetation indices are influenced by factors including: inconsistency over varying cover types, mixed pixel problems, limited spectral dimensionality (typically only two bands), non-comprehensive biophysical characterization (e.g., saturation at higher LAI), an inability to account for variability in the following: canopy cover, illumination and canopy geometry, leaf optical properties PHOTOGRAMMETRIC ENGINEER ING & REMOTE SENS ING Ap r i l 2009 361 S.A. Soenen, D R. Peddle, and C.A. Coburn are with the Department of Geography, University of Lethbridge, 4401 University Drive West, Lethbridge, AB, T1K 3M4, Canada ([email protected]). R.J. Hall is with the Canadian Forest Service, Northern Forest Centre, 5320 122 Street, Edmonton, AB, T6H 3S5, Canada, and the Department of Geography, University of Lethbridge, AB, Canada. F.G. Hall is with NASA Goddard Space Flight Center, Code 614.4, 8800 Greenbelt Rd., Greenbelt MD 20771. Photogrammetric Engineering & Remote Sensing Vol. 75, No. 4, April 2009, pp. 361–374. 0099-1112/09/7504–0361/$3.00/0 © 2009 American Society for Photogrammetry and Remote Sensing 361-374_07-019.qxd 3/19/09 1:27 PM Page 361