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

Jeffrey W. Doser · 2025

Read Multivariate spatial models for small area estimation of species-specific forest inventory parameters by Jeffrey W. Doser (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-07118.txt.

Introduction

Jeffrey Doser and co-authors build a model that estimates forest traits for small areas, tree kind by tree kind. Their 2025 paper serves national forest surveys, which report well for big regions but thin out at county scale. The model works straight from plot-level records and handles quirks such as many zero counts and linked species.

Fake-data tests showed low bias and better accuracy than the standard survey math. Applied to the southern United States, the model mapped wood mass for the twenty most common tree kinds by county. Its maps tracked the standard ones closely, with a slight high lean. Nine in ten county values gained sharpness, which helps local forest planning.

Catalog record

CREATORJeffrey W. Doser
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2503-07118

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Multivariate spatial models for small area estimation of species-specific forest inventory parameters Jeffrey W. Doser Abstract 1 Department of Forestry and Environmental Resources, North Carolina State University, Raleigh, NC, USA 2 Department of Environmental Conservation, University of Massachusetts, Amherst, MA, USA. 3 Northern Research Station, United States Department of Agriculture Forest Service, St. Paul, MN, USA. 4 Department of Forestry, Michigan State University, East Lansing, MI, USA 5 Department of Statistics and Probability, Michigan State University, East Lansing, MI, USA Corresponding Author : Jeffrey W. Doser, email: [email protected]. Abstract 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, which is often required to inform wildlife habitat management, sustainable forestry certifications, or timber product assessments. 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 the design-based estimator and a non-parametric k-nearest neighbor (kNN) 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. Biomass estimates from the proposed model had high correlations with design-based estimates and kNN 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. 1 Introduction Monitoring forest resources across broad spatial and temporal scales is increasingly important for forest management to meet both economic and sustainability objectives. National forest inventories (NFIs) are designed to estimate forest parameters on regional to national scales using a large network of permanent inventory plots selected according to a probability sampling design ( McRoberts et al., 2010 ) . For example, the United States (US) Department of Agriculture Forest Service Forest Inventory and Analysis (FIA) program conducts the US NFI, which generates a valuable resource for quantifying forest health, timber resources, and forest landowner characteristics across broad regions of the US ( Bechtold and Patterson, 2005 ) . NFIs, like the FIA program, historically use design-based inference to estimate forest parameters, which generally provides sufficiently precise forest parameter estimates across large areas of forestland (e.g., US states) ( Westfall et al., 2022 ) . However, NFI programs across the world have experienced increased demand for estimates within smaller spatial, temporal, and biophysical extents than design-based inference and existing sampling intensities can reasonably deliver ( Breidenbach and Astrup, 2012 ; Prisley et al., 2021 ) . Methodological approaches that provide sufficiently precise estimates within small areas of interest using data from a larger sample are collectively referred to as small area estimation (SAE; Rao and Molina, 2015 ). Such techniques have seen increasing applications with NFI data in recent years (e.g., Hou et al., 2021 ; Cao et al., 2022 ; Finley et al., 2024 ; Shannon et al., 2024 ). Both design-based and model-based inferential paradigms have been employed in SAE applications with NFI data ( Dettmann et al., 2022 ) . Regardless of the inferential paradigm used, most SAE approaches seek to relate the variable of interest to auxiliary information from satellite and airborne remote sensing data through a statistical model. When a strong relationship exists between the auxiliary information and variable of interest, the precision of small area estimates can be substantially improved ( Särndal et al., 2003 ) . In addition to incorporating auxiliary information through regression-based frameworks, model-based SAE approaches can explicitly borrow information from outside the area of interest through the use of random effects. Random effect structures can range from simple, normally distributed effects to account for similarities in forest parameters across ecologically-similar areas (e.g., ecoregions), management units, or political boundaries (e.g., White et al., 2021 ), to more complex forms that explicitly acknowledge spatial and spatial-temporal dependencies in forest parameters through a spatial covariance structure (e.g., Finley et al., 2024 ; Shannon et al., 2024 ). Recent advances in model-based SAE approaches that accommodate spatial dependence have been shown to provide improved precision of forest parameters and forest owner characteristics at state ( Ver Planck et al., 2017 ; Harris et al., 2021 ) , regional ( Cao et al., 2022 ) , and national levels ( May et al., 2023 ; Shannon et al., 2024 ) . While considerable progress has been made in developing SAE methods suitable for forest inventory, a vast majority of proposed methods focus on estimating a single forest attribute such as total merchantable volume or aboveground biomass ( Dettmann et al., 2022 ) . Estimates of forest attributes by species or species class are often of interest due to individual species representing different forest products and values, providing unique benefits to wildlife, and serving as useful indicators of site quality and/or disturbance history ( Kershaw Jr et al., 2016 ) . While independent models can be fit for each individual species, such an approach fails to leverage dependencies between species-specific attributes, which can substantially improve precision of model predictions given the importance of species interactions (e.g., competition) in determining stand structure and dynamics ( Weiskittel et al., 2011 ; Clark et al., 2014 ; Tang et al., 2025 ) . Total biomass within a fixed area is limited by site quality, and as a result the biomass of one species generally cannot increase without a corresponding decrease in the biomass of another ( Clark et al., 2017 ) . Additionally, reliable small area estimates of species-specific forest inventory parameters are complicated by large amounts of zero-inflation and highly right skewed data distributions. For example, species tend to be absent from many plots, present at low abundances in many plots, and present at high abundances in only a few plots. In many cases, individual species are never observed in the small number of plots within an area of interest, which prevents the calculation of a standard error estimate using design-based estimation. Furthermore, even for small areas with some non-zero values of species-specific attributes, the distribution of values is typically highly right-skewed and close to zero, which calls into question the reliability of confidence intervals that rely on the central limit theorem. Thus, there is a need for model-based frameworks that deliver precise estimates of forest parameters by species and similar high-dimensional forest attributes. Here we develop a multivariate spatial model for SAE of species-specific forest inventory parameters. Our hierarchical Bayesian model simultaneously accounts for zero-inflation in individual species

Read the complete text