Remote sensing of boreal forest terrain: Sub-pixel scale mixture analysis of land cover and biophysical parameters at forest stand and regional scales

Increasing concentrations of atmospheric carbon dioxide and other greenhouse gases have focused attention on the global carbon cycle. Predicted climate change scenarios indicate the release of large stores of organic carbon in boreal forest regions could have profound ecological, cultural and economic impacts on agricultural, boreal and Arctic tundra zones. Remote sensing provides the only comprehensive information to monitor such large tracts of land, however, conventional NDVI vegetation index approaches have been shown to be unreliable for extracting required biophysical parameters such as biomass, leaf area index and productivity. In this research. spectral mixture analysis (SMA) and geometric-optical reflectance models provide sub-pixel scale forest information such as sunlit canopy, background and shadow fractions which yield improved biophysical estimates when compared to NDVI. This was validated first for individual forest stands using the NASA C scOVER data set from the Superior National Forest, Minnesota USA. Best results were obtained from shadow function using a spheroid based reflectance model with corrections for mutual shadowing and solar zenith angle variations. Following this, a regional scale methodology was implemented in the Boreal Ecosystem Atmosphere Study (B scOREAS) which coupled canopy reflectance models, spectral mixture analysis, and a powerful evidential reasoning classifier into an integrated, physically based land cover and biophysical algorithm (the M{dollar}oplus{dollar}P software package). Field spectrometer data processed to end-member reflectance and stand level tree geometry were input to canopy optical models to produce spectral trajectories of reflectance and forest scene components over a full range of stand densities. These trajectories were input to the new M{dollar}oplus{dollar}P software to produce land cover and sub-pixel scale outputs for predicting biophysical parameters. Improved classification accuracies and biophysical estimates were obtained compared to conventional approaches, with a potential shown for estimating tree height and stem diameter.