Publication Details

Multivariate inference for forest inventories using auxiliary airborne laser scanning data

Publication Toolbox

  • Download PDF (1.0 MB)
  • This publication is available only online.

Year Published

2017

Publication

Forest Ecology and Management

Abstract

National forest inventories have a long history of using remotely sensed auxiliary information to enhance estimation of forest parameters. For this purpose, aerial photography and satellite spectral data have been shown to be effective as sources of information in support of stratified estimators. These spectral-based stratifications are much more effective for reducing variances for forest area-related parameters than for parameters related to continuous attributes such as volume and biomass. For variables related to the latter attributes, stratified estimators using airborne laser scanning auxiliary data are much more effective, but are less effective than model-assisted estimators using the same auxiliary data. For inventory applications, however, stratified estimators using the same stratification for all response variables are naturally multivariate, whereas model-assisted estimators are not. A consequence is that multiple, univariate applications of model-assisted estimators cannot ensure compatibility among estimates of inventory parameters related to variables such as forest area, growing stock volume, and tree density. The objectives of the study were twofold: (1) to optimize a multivariate, k-NN approach for simultaneously predicting multiple forest inventory variables; and (2) to compare multivariate model-assisted generalized regression estimators using optimized k-NN predictions to post-stratified estimators with respect to inferences in the form of confidence intervals for multiple forest inventory parameters. The analyses included use of airborne laser scanning data as auxiliary information and the multivariate k- NN technique for prediction in support of the model-assisted estimators. The study area was in north central Minnesota in the USA and is characterized by both lowland and upland forest types interspersed with wetlands and lakes. The first primary result was that the optimized k-NN technique in combination with a model-assisted estimator produced compatible multivariate estimates of population means for six inventory parameters. Second, variances for the multivariate model-assisted estimators were smaller by 23%–35% than variances for a post-stratified estimator. These results warrant serious consideration of this approach for operational implementation by national forest inventories.

Citation

McRoberts, Ronald E.; Chen, Qi; Walters, Brian F. 2017. Multivariate inference for forest inventories using auxiliary airborne laser scanning data. Forest Ecology and Management. 401: 295-303. https://doi.org/10.1016/j.foreco.2017.07.017.

Last updated on: June 28, 2018