Scaling and sampling dependencies of forest canopy height mapping towards jurisdictional biomass reporting using airborne LiDAR and small-area estimation

Consolidated airborne laser scanning (ALS) programs, satellite imagery and spaceborne structural measurements have enabled major advances in canopy height mapping that translate towards the forest carbon biomass arena. However, we must carefully evaluate the cost of using fine-grained canopy height products to predict biomass under calibration models scoped at the scale of inventory plots. In this study, we estimated biomass using field plots and ALS metrics before predicting biomass over a jurisdiction of ~ 15,500 km² in Spain using 10 m, 25 m, 44 m, and 100 m as prediction scales. We altered the scale of ALS-based biomass predictors in 10 sub-jurisdictions intensively surveyed by the Spanish National Forest Inventory (NFI) before estimating mean and total biomass using three options: (i) traditional NFI design-based (DB) estimation, (ii) a model-based (MB) approach using scale-varying canopy height metrics from ALS and NFI plots, and (iii) an small-area estimation (SAE) implemntation designed for sub-jurisdictional domains. Higher uncertainties - relative standard errors (SE) - were found for DB, particularly at sub-jurisdictional and stratum levels. We observed a consistent increase in uncertainty for MB estimation from the finest 10 m scale up to 100 m. In MB estimation, the maximum relative bias reached 11% for 10-m predictions compared to the baseline estimate at the NFI sampling native resolution. The bias associated with the prediction scale ranged from + 5% (25 m) to -8% (100 m). The mean biomass estimates for SAE generally ranged between DB and MB but at lower uncertainty to the former, especially as the NFI sampling becomes scarcer and not enough for solid inference of biomass mean. The SEA statistics helped to disentangle biomass comparisons between ALS-based inference and the traditional NFI estimation that do not incorporate remote sensing data.