LiDAR based prediction of forest biomass using hierarchical models with spatially varying coefficients

摘要: Many studies and production inventory systems have shown the utility of coupling covariates derived from Light Detection Ranging (LiDAR) data with forest variables measured on georeferenced plots through regression models. The objective this study was to propose assess use a Bayesian hierarchical modeling framework that accommodates both residual spatial dependence non-stationarity model introduction random effects. We explored using four datasets are part North American Carbon Program, each comprising point-referenced measures above-ground biomass discrete LiDAR. For dataset, we considered at least five specifications varying complexity. Models were assessed based goodness fit criteria predictive performance 10-fold cross-validation procedure. Results showed addition effects intercept improved in presence substantial dependence. Additionally, some cases, allowing either or all slope parameters vary spatially, via effects, further performance. In other instances, models but decreased performance—indicating over-fitting underscoring need for ability. proposed provided access pixel-level posterior distributions useful uncertainty mapping, diagnosing extrapolation issues, revealing missing covariates, discovering locally significant parameters.