Influence of Field-Based Species Composition and Understory Descriptions on Spectral Mixture Analysis of Tree Species in the Northwest Territories, Canada

Abstract

Tree species information is a fundamental component of forest inventories that is challenging to obtain in northern boreal forests because inherently open stands with individual trees might be clumped or widely distributed and contain multiple tree species. These challenges result in a mixed pixel problem that was investigated using Multiple Endmember Spectral Mixture Analysis (MESMA), a technique for identifying the type and proportions of forest components (e.g., sunlit canopy, background vegetation, shadow) in remotely sensed imagery at the subpixel scale. This was tested using Landsat Thematic Mapper (TM) imagery from a study area in the Northwest Territories (NWT), Canada. How the dominant tree species was described in the field, and how the spectral properties of understory vegetation was measured, were both important factors that affected species discrimination. Landsat TM was most sensitive to dominant species defined by the fraction of total basal area of the dominant and codominant trees. Classification accuracies of 90% and 63% for very open and medium density forest stands, respectively, were achieved when understory diversity and abundance information was incorporated into the definition of the background component. It was concluded that this approach warrants further investigation for northern boreal forest species classification.