The widespread degradation of natural ecosystems requires cost-efficient restoration techniques that minimize risk and consider context-specific restoration conditions. However, meeting these demands can be difficult because information on ecosystem-level factors controlling vegetation and continuous spatial data on species interactions are often lacking. Using airborne LiDAR data from a Hawaiian dry forest, we delineated crowns and assessed the 3D structure of more than 700,000 shrubs and trees. We used Random Forest machine learning to assess the relative importance of resource availability, environmental conditions, and disturbance regimes on canopy density. We then modelled and scaled up plant–plant interactions (i.e. potential nursery effects) to landscape units using a LiDAR-derived Canopy Coalescence index (CC). We used the relative importance of ecosystem factors, canopy cover, and CC to prioritize landscapes for restoration. Here, we demonstrate a methodological framework that prioritizes landscapes in need of restoration (i.e. planting woody species) using two ecological perspectives: (1) ecosystem-level controls on remnant woody vegetation, and (2) potential interactions between established canopies and seedlings (e.g. nursery effect, competition). Our results highlight the heterogeneous nature of ecosystem-level drivers affecting forest structure along elevation gradients. Consequently, the degree of potential nursery interactions between established canopies and seedlings at the landscape-scale was context-specific. Synthesis and applications. Our study provides a methodological approach that prioritizes landscapes for restoration by identifying the main controls on tree spatial distribution and by inferring the favourable conditions for seedlings. This approach can guide land managers to define cost-efficient restoration strategies for large ecological areas.