Classifying forestland from model-generated tree species habitat suitability in the Western Ecoregion of Nova Scotia, Canada

Abstract

Forestland classification is central to the sustainable management of forests. In this paper, we explore the possibility of classifying forestland from species–habitat–suitability indices and a hybrid classification of modeled data. Raster-based calculations of species–habitat–suitability were derived as a function of landscape-level descriptions of incident photosynthetically active radiation (PAR), soil water content (SWC), and growing degree-days (GDD) for southwestern Nova Scotia, Canada. PAR and SWC were both generated with the LanDSET model and GDD from thermal data captured with the space-borne MODIS sensor. We compared the distribution of predicted forestland types with the natural range of target species as found in the provincial permanent sample plots (PSPs). Reasonable agreement (≥50% accuracy) existed between some forestland types (e.g., red maple – white birch – red oak and balsam fir – red maple) and PSP-based assessments of species presence–absence. Agreement was noticeably lower for other forestland types, such as sugar maple – beech – yellow birch (<50% accuracy). This discrepancy is attributed to forest-forming factors not directly addressed by the model, e.g., forest succession, stand interventions, and disturbance. Their addition in the model could change the dynamics of tree-species preference in southwest Nova Scotia and is worth examining. True model inaccuracies accounted for about 0.3%–15.0% of the total reported error.