LAUSR

Abstract Monitoring forest ecosystems requires accurate and up-to-date information on the type and extent of forest depletions, which may exist but are not always open and transparent. The ever growing, freely accessible Landsat archive provides data to derive such information; however, the manipulation of raw imagery can constitute a barrier to those lacking remote sensing expertise. The Landsat-based, global dataset of annual forest loss (version 1.0, Hansen et al., 2013) makes such information readily available. While the accuracy of this dataset has been assessed at the global scale, its applicability for reliable local monitoring of forest harvesting has not yet been validated. Here we undertake such an exercise in a temperate forest in Atlantic Canada. We used a census, polygon-based approach to comprehensively assess thematic, temporal and structural accuracy. We vectorized Hansen's forest loss raster for the 8,520 km 2 of public lands in the Miramichi River basin (13,496 km 2 ), which yielded 9299 polygons of 1 ha minimum size. Then we used the provincial forest harvest inventory as reference. User's and producer's accuracies were 81% and 82% based on area, and 86% and 85% based on polygon counts. Detection probability decreased with decreasing cutblock size and harvest intensity. From all Hansen polygons, 85% had the correct harvest year and 88% structurally matched one or more reference polygons either alone or together with other Hansen polygons. After the validation, we used the Hansen dataset to derive trends for the entire basin. Mean annual harvest rate was 0.92 ha/km 2 /year between 2000 and 2012. Most of the catchments around the western headwaters of the Miramichi River underwent intensive harvesting, underscoring the need of further monitoring. Our results indicate that the Hansen dataset could be used as a surrogate harvest layer for temperate forests where clear-cutting is common and fire is rare.