Model-assisted domain estimation of postfire tree regeneration in the western US using nearest neighbor techniques

David L.R. Affleck, George C. Gaines

Research output: Contribution to journalArticlepeer-review

Abstract

Many nations administer national forest inventory programs for unbiased estimation of forest attributes over broad spatial and temporal extents. However, management and conservation decisions often demand reliable estimates for finer spatiotemporal domains. In the western US, wildfire activity is expanding and postfire regeneration must contend with a warmer, drier climate. We evaluate the potential of K nearest neighbor (KNN) strategies for estimation of stocking across postfire measurements of Forest Inventory & Analysis plots in 11 western US states, and subsequently for model-assisted (MA) estimation of stocking over domains defined by aggregations of burned areas within individual states and 4-year periods. In particular, we develop and evaluate a form of constrained KNN that allows for unbiased MA domain estimation under simple random sampling by drawing only on measurements external to a domain of interest. KNN strategies based on geographically, radiometrically, and climatically proximate measurements are found to provide more accurate estimates of stocking at the plot level than domain means. Applying the selected external KNN strategy also reduced standard errors of MA domain estimates by 16% over direct domain estimators, but bias correction introduces substantial variability over synthetic estimates. Further applications of the external constraint imposed on KNN are discussed.

Original languageEnglish
Pages (from-to)981-995
Number of pages15
JournalCanadian Journal of Forest Research
Volume53
Issue number12
DOIs
StatePublished - Dec 2023

Keywords

  • forest inventory
  • forest regeneration
  • model-assisted estimation
  • small area estimation
  • wildland fire

Fingerprint

Dive into the research topics of 'Model-assisted domain estimation of postfire tree regeneration in the western US using nearest neighbor techniques'. Together they form a unique fingerprint.

Cite this