Evaluation of semi-empirical BRDF models inverted against multi-angle data from a digital airborne frame camera for enhancing forest type classification

摘要: Abstract Forest mapping based on remote sensing data usually relies purely spectral information ignoring that the observed signal may be substantially influenced by angular effects. On other hand, it is known for forest canopies variation of reflectance with sun-view-geometry significant. The study examines different approaches to extract spectro-directional from airborne imagery and explores its use in type classification. images were acquired a standard aerial survey. They taken common forward side overlap result each point ground several view directions. To obtain directional might useful classification, two widely used semi-empirical models bidirectional distribution function (BRDF) tested, i.e., Rahman–Pinty–Verstraete model (RPV; 3 4 parameters) (Rahman et al., 1993) RossThick-LiSparse (RTLS; 5 (Wanner 1995). For sample plot, observations at least 10 directions available. inverted against applying look-up table approach. estimated parameters 1) directly as explanatory variables 2) estimate factor (BRF) selected sun-view-geometries, referred modeled BRFs. BRFs also served classification Random Forests classifier was used. Both proved provide can successfully outperforming conventional (single-angle) multi-spectral data. 3-parameter version RPV found most effective. Compared dataset, overall accuracy increased 72% (kappa: 0.64) 85% 0.81), when this variables. 4-parameter 5-parameter RTLS fitted more accurately than corresponding versions. However, accuracies obtained significantly lower those models. All tested yielded higher parameters. Again, highest RPV-3P (overall accuracy: 92%; kappa: 0.89). varied sun-view-geometry. These differences could partly explained sampling scheme models' capability angles not observed.