Predicting tree diameter using allometry described by non-parametric locally-estimated copulas from tree dimensions derived from airborne laser scanning

Qing Xu, Bo Li, Matti Maltamo, Timo Tokola, Zhengyang Hou

Research output: Contribution to journalArticle

Abstract

Biomass inventories that employ airborne laser scanning (ALS) require models that can predict tree diameter at breast height (DBH) from ALS-derived tree dimensions, as ALS can usually not directly measure DBH due to scanning angle, inadequate point density and canopy obstruction. Although some work has been done in using correlation as a measure of dependence to describe the linear relationship between variable means, none has investigated the copula-based measure of dependence for the prediction of DBH from ALS-derived height and crown diameter. Following the application of a locally-estimated copula method to 79 sample plots in eastern Finland, we compared the performance of the copula method with a baseline local regression (LOESS) model and an ordinary least squares (OLS) model. We found that the copula method outperformed the OLS model by decreasing 30% of the root-mean-squared error (RMSE). The copula method performed slightly better than the LOESS model for the original sample, but the results of the bootstrap samples showed that the variance in RMSE was sixteen times lower in the copula method than the LOESS model, suggesting that the copula had a more consistent and robust model performance across the 10,000 bootstrap samples. Moreover, while the LOESS model only predicts the conditional mean of the response variable, the copula method can also predict median and other quantiles.

Original languageEnglish (US)
Pages (from-to)205-212
Number of pages8
JournalForest Ecology and Management
Volume434
DOIs
StatePublished - Feb 28 2019

Fingerprint

allometry
tree and stand measurements
lasers
laser
least squares
methodology
sampling
tree crown
Finland
method
canopy
prediction
biomass

Keywords

  • Copula
  • Individual tree detection
  • Marginal distribution
  • Nearest neighbour
  • Quantile regression

ASJC Scopus subject areas

  • Forestry
  • Nature and Landscape Conservation
  • Management, Monitoring, Policy and Law

Cite this

Predicting tree diameter using allometry described by non-parametric locally-estimated copulas from tree dimensions derived from airborne laser scanning. / Xu, Qing; Li, Bo; Maltamo, Matti; Tokola, Timo; Hou, Zhengyang.

In: Forest Ecology and Management, Vol. 434, 28.02.2019, p. 205-212.

Research output: Contribution to journalArticle

@article{3a0330118ebf42d7a35276d5a44056a8,
title = "Predicting tree diameter using allometry described by non-parametric locally-estimated copulas from tree dimensions derived from airborne laser scanning",
abstract = "Biomass inventories that employ airborne laser scanning (ALS) require models that can predict tree diameter at breast height (DBH) from ALS-derived tree dimensions, as ALS can usually not directly measure DBH due to scanning angle, inadequate point density and canopy obstruction. Although some work has been done in using correlation as a measure of dependence to describe the linear relationship between variable means, none has investigated the copula-based measure of dependence for the prediction of DBH from ALS-derived height and crown diameter. Following the application of a locally-estimated copula method to 79 sample plots in eastern Finland, we compared the performance of the copula method with a baseline local regression (LOESS) model and an ordinary least squares (OLS) model. We found that the copula method outperformed the OLS model by decreasing 30{\%} of the root-mean-squared error (RMSE). The copula method performed slightly better than the LOESS model for the original sample, but the results of the bootstrap samples showed that the variance in RMSE was sixteen times lower in the copula method than the LOESS model, suggesting that the copula had a more consistent and robust model performance across the 10,000 bootstrap samples. Moreover, while the LOESS model only predicts the conditional mean of the response variable, the copula method can also predict median and other quantiles.",
keywords = "Copula, Individual tree detection, Marginal distribution, Nearest neighbour, Quantile regression",
author = "Qing Xu and Bo Li and Matti Maltamo and Timo Tokola and Zhengyang Hou",
year = "2019",
month = "2",
day = "28",
doi = "10.1016/j.foreco.2018.12.020",
language = "English (US)",
volume = "434",
pages = "205--212",
journal = "Forest Ecology and Management",
issn = "0378-1127",
publisher = "Elsevier",

}

TY - JOUR

T1 - Predicting tree diameter using allometry described by non-parametric locally-estimated copulas from tree dimensions derived from airborne laser scanning

AU - Xu, Qing

AU - Li, Bo

AU - Maltamo, Matti

AU - Tokola, Timo

AU - Hou, Zhengyang

PY - 2019/2/28

Y1 - 2019/2/28

N2 - Biomass inventories that employ airborne laser scanning (ALS) require models that can predict tree diameter at breast height (DBH) from ALS-derived tree dimensions, as ALS can usually not directly measure DBH due to scanning angle, inadequate point density and canopy obstruction. Although some work has been done in using correlation as a measure of dependence to describe the linear relationship between variable means, none has investigated the copula-based measure of dependence for the prediction of DBH from ALS-derived height and crown diameter. Following the application of a locally-estimated copula method to 79 sample plots in eastern Finland, we compared the performance of the copula method with a baseline local regression (LOESS) model and an ordinary least squares (OLS) model. We found that the copula method outperformed the OLS model by decreasing 30% of the root-mean-squared error (RMSE). The copula method performed slightly better than the LOESS model for the original sample, but the results of the bootstrap samples showed that the variance in RMSE was sixteen times lower in the copula method than the LOESS model, suggesting that the copula had a more consistent and robust model performance across the 10,000 bootstrap samples. Moreover, while the LOESS model only predicts the conditional mean of the response variable, the copula method can also predict median and other quantiles.

AB - Biomass inventories that employ airborne laser scanning (ALS) require models that can predict tree diameter at breast height (DBH) from ALS-derived tree dimensions, as ALS can usually not directly measure DBH due to scanning angle, inadequate point density and canopy obstruction. Although some work has been done in using correlation as a measure of dependence to describe the linear relationship between variable means, none has investigated the copula-based measure of dependence for the prediction of DBH from ALS-derived height and crown diameter. Following the application of a locally-estimated copula method to 79 sample plots in eastern Finland, we compared the performance of the copula method with a baseline local regression (LOESS) model and an ordinary least squares (OLS) model. We found that the copula method outperformed the OLS model by decreasing 30% of the root-mean-squared error (RMSE). The copula method performed slightly better than the LOESS model for the original sample, but the results of the bootstrap samples showed that the variance in RMSE was sixteen times lower in the copula method than the LOESS model, suggesting that the copula had a more consistent and robust model performance across the 10,000 bootstrap samples. Moreover, while the LOESS model only predicts the conditional mean of the response variable, the copula method can also predict median and other quantiles.

KW - Copula

KW - Individual tree detection

KW - Marginal distribution

KW - Nearest neighbour

KW - Quantile regression

UR - http://www.scopus.com/inward/record.url?scp=85058521380&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85058521380&partnerID=8YFLogxK

U2 - 10.1016/j.foreco.2018.12.020

DO - 10.1016/j.foreco.2018.12.020

M3 - Article

VL - 434

SP - 205

EP - 212

JO - Forest Ecology and Management

JF - Forest Ecology and Management

SN - 0378-1127

ER -