Modeling structural attributes of Hyrcanian forests using spaceborne LiDAR, machine learning algorithms, and a hierarchical field data collection approach

Most conservation plans do not incorporate forest structural attributes, prioritizing species-based criteria instead. We integrated GEDI LiDAR canopy metrics with hierarchical field inventories and management-unit aggregation to model basal area (G, m² ha⁻¹), standing volume (V, m³ ha⁻¹), and tree density (N, trees ha⁻¹) at plot and compartment levels. Using 1195 field plots, we benchmarked five learners: linear regression, support vector regression, random forest, gradient boosting, and a neural network under cross-validation; we then applied a two-phase design for regional estimation and compared high-V areas with UNESCO core zones. At the plot-level, V reached R² ≈ 0.75 (RMSE ≈ 65–66 m³ ha⁻¹), G was moderate (R² = 0.47; RMSE ≈ 6.1 m² ha⁻¹), and N was weak (R² = 0.10; RMSE ≈ 266 trees ha⁻¹). At the compartment-level, V achieved R² = 0.82 (RMSE = 37.7 m³ ha⁻¹), G reached R² = 0.54 (RMSE = 2.5 m² ha⁻¹), and N remained difficult to predict even after aggregation. Linear and SVR models matched or exceeded tree-based and neural models for G and V. Regional means were V = 241.05 m³ ha⁻¹ and G = 22.33 m² ha⁻¹ . Of the 60 1 × 1 km cells with V > 500 m³ ha⁻¹ , only 8 were inside UNESCO core zones, indicating a protection gap. Collectively, a compact set of GEDI height metrics, coupled with routine inventories and unit-level aggregation, enables timely updates of forest thematic maps and supports zoning that prioritizes high-biomass stands in the Hyrcanian region. • GEDI LiDAR and machine learning model Hyrcanian forest structure. • Compartment-level modeling improves accuracy of structural-attribute estimates. • High-carbon forests occur mostly outside UNESCO-protected areas. • GEDI updates forest maps and strengthens zoning of high-biomass areas.