Applying an efficient K-nearest neighbor search to forest attribute imputation

Andrew O. Finley; Ronald E. McRoberts; Alan R. Ek

Treesearch

Applying an efficient K-nearest neighbor search to forest attribute imputation

Informally Refereed

Authors:

Andrew O. Finley, Ronald E. McRoberts, Alan R. Ek

Year:

2006

Type:

Scientific Journal

Station:

Northern Research Station

Source:

In: Cieszewski, Chris J.; Strub, Mike, eds. Second international conference on forest measurements and quantitative methods and management & the 2004 Southern Mensurationists meeting. Hot Springs, AK: Athens, GA. 4-10.

Download (PDF 308 KB)

Download (PDF 308 KB): https://research.fs.usda.gov/download/treesearch/18453.pdf

Abstract

This paper explores the utility of an efficient nearest neighbor (NN) search algorithm for applications in multi-source kNN forest attribute imputation. The search algorithm reduces the number of distance calculations between a given target vector and each reference vector, thereby, decreasing the time needed to discover the NN subset. Results of five trials show gains in NN search efficiency ranging from 75 to 98 percent for k = 1. The search algorithm can be easily incorporated into routines that optimize feature subsets or weights, values of k, distance decomposition coefficients, and mapping.

Citation

Finley, Andrew O.; McRoberts, Ronald E.; Ek, Alan R. 2006. Applying an efficient K-nearest neighbor search to forest attribute imputation. In: Cieszewski, Chris J.; Strub, Mike, eds. Second international conference on forest measurements and quantitative methods and management & the 2004 Southern Mensurationists meeting. Hot Springs, AK: Athens, GA. 4-10.