Mapping Dominant Tree Species over Large Forested Areas Using Landsat Best-Available-Pixel Image Composites

. Remotely sensed image composites that are pixel based rather than scene based are increasingly feasible to use over large areas and fine spatial resolutions. For large jurisdictions that utilize remotely sensed imagery for ecosystem mapping and monitoring, pixel-based composites enable a wider range of applications, at higher quality. The goal of this study was to model spatial distributions of 6 tree species over a large forested area of Saskatchewan, Canada (>39 million ha) at 30-m spatial resolution using a multiyear Best-Available-Pixel (BAP) Landsat composite. We tested the influence of the BAP composite on the resultant maps by comparing species composition and configuration for areas where imagery was from a single sensor, year, and day of year, to areas with variable composite characteristics. Model error rates ranged from 0.09% to 0.24%, area under the curve values approaching 1, and met ecological expectations. The BAP composite was found to have little effect on model outcomes, with composition and configuration values in nonreference areas being similar for all species but one, which had an unexpected configuration. Moreover, sensor, year, and day of year were similar for reference and nonreference blocks for all species. Results indicate that Landsat BAP image composites are useful for generating large-area maps of tree species distributions.