Multivariate spatial models for small area estimation of species-specific forest inventory parameters

National Forest Inventories (NFIs) provide statistically reliable information on forest resources at national and other large spatial scales. As forest management and conservation needs become increasingly complex, NFIs are being called upon to provide forest parameter estimates at spatial scales smaller than current design-based estimation procedures can provide. This is particularly true when estimates are desired by species or species groups. Here we propose a multivariate spatial model for small area estimation of species-specific forest inventory parameters. The hierarchical Bayesian modeling framework accounts for key complexities in species-specific forest inventory data, such as zero-inflation, correlations among species, and residual spatial autocorrelation. Importantly, by fitting the model directly to the individual plot-level data, the framework enables estimates of species-level forest parameters, with associated uncertainty, across any user-defined small area of interest. A simulation study revealed minimal bias and higher accuracy of the proposed model-based approach compared to design-based estimator. We applied the model to estimate species-specific county-level aboveground biomass for the 20 most abundant tree species in the southern United States using Forest Inventory and Analysis (FIA) data. Model-based biomass estimates had high correlations with design-based estimates, yet the model-based estimates tended to have a slight positive bias relative to design-based estimates. Importantly, the proposed model provided large gains in precision across all 20 species. On average across species, 91.5% of county-level biomass estimates had higher precision compared to the design-based estimates. The proposed framework improves the ability of NFI data users to generate species-level forest parameter estimates with reasonable precision at management-relevant spatial scales.