Optimizing forest stand aggregation in fragmented stands using graph convolutional networks: A case study in Japan

• Introduces a GCN-based method for forest stand aggregation, enabling stand-level features—such as slope, road proximity, and adjacency—to interact through the graph structure and generate aggregated feature values that reflect spatial dependencies. • Constructs two types of connectivity models—selective and full connection—to represent different spatial relationship assumptions and evaluate their effects on aggregation outcomes. • Offers a flexible, data-driven approach to stand aggregation, capturing complex spatial interactions tailored to Japan’s fragmented forest ownership and diverse terrain. • Provides a foundation for precise harvesting and long-term forest planning, as the resulting aggregated stand features can inform sustainable and economically viable operations in highly fragmented stands. Japan’s forest sector faces persistent challenges due to fragmented land ownership and an increasingly skewed age-class structure, undermining not only management efficiency but also economic viability and long-term sustainability. This study proposes a novel approach to forest stand aggregation by integrating Geographic Information Systems (GIS) with Graph Convolutional Networks (GCNs), enabling a data-driven modeling of spatial interactions among forest stands. Unlike traditional rule-based methods that rely on fixed thresholds and static adjacency rules, the GCN framework allows stand-level attributes—such as slope, road proximity, and spatial adjacency—to influence each other through a graph structure. This interaction generates new aggregated feature values that reflect underlying spatial dependencies. Using forestry data from the Nishikawa area, two types of stand connectivity models—selective and full connection—were constructed to represent different spatial contexts. Results show that the selective model emphasizes road accessibility, while the full model better maintains age-class continuity and internal cohesion. This GCN-based aggregation method offers a flexible and scalable solution to stand consolidation, forming a foundation for more precise and terrain-aware harvesting plans in future forest management strategies.