Separating measurement from definition: an object-based workflow for national forest mapping

Arnt Kristian Gjertsen, Shivesh Kishore Karan, Knut Bjørkelo

Forest is defined rather than observed, and definitions are written to serve different management objectives. In Norway, four authoritative datasets place forest between 35 and 43% of the mainland because each encodes a different rule. Categorical forest maps inherit this problem, since each serves only the definition it was built on. We present a documented object-based workflow that separates measurement from definition. Very high-resolution imagery, Sentinel-2, and a laser-derived canopy height model are combined with the existing national masks in a spatial database, and the result is segmented in two steps into tree crowns, tree points derived from them, and forest-stand segments carrying height, crown cover, species, and site index. A rule-based expert system then classifies the tree points. Applied to mainland Norway, the workflow produced 2.72 billion tree crowns and 57.8 million forest-stand segments covering 144,755 km2. Because structure is stored as attributed objects rather than a single class, thresholding one dataset reproduces forest areas from 15.8 to 44.7% and brackets all four national estimates. Independent interpretation of 702 sampled points gave a commission error of 0.6% for the accepted tree points. The resulting dataset is called Skogmaske.