Predicting forest inventory attributes using airborne laser scanning, aerial imagery, and harvester data

Atte Saukkola; Timo Melkas; Kirsi Riekki; Sanna Sirparanta; Jussi Peuhkurinen; Markus Holopainen; Juha Hyyppä; Mikko Vastaranta

doi:10.3390/rs11070797

Predicting forest inventory attributes using airborne laser scanning, aerial imagery, and harvester data

Atte Saukkola, Timo Melkas^*, Kirsi Riekki, Sanna Sirparanta, Jussi Peuhkurinen, Markus Holopainen, Juha Hyyppä, Mikko Vastaranta

^*Corresponding author for this work

Not published at VTT

Research output: Contribution to journal › Article › Scientific › peer-review

30 Citations (Scopus)

Abstract

The aim of the study was to develop a new method to use tree stem information recorded by harvesters along operative logging in remote sensing-based prediction of forest inventory attributes in mature stands. The reference sample plots were formed from harvester data, using two different tree positions: harvester positions (XYH) in global satellite navigation system and computationally improved harvester head positions (XYHH). Study materials consisted of 158 mature Norway-spruce-dominated stands located in Southern Finland that were clear-cut during 2015-16. Tree attributes were derived from the stem dimensions recorded by the harvester. The forest inventory attributes were compiled for both stands and sample plots generated for stands for four different sample plot sizes (254, 509, 761, and 1018 m²). Prediction models between the harvester-based forest inventory attributes and remote sensing features of sample plots were developed. The stand-level predictions were obtained, and basal-area weighted mean diameter (Dg) and basal-area weighted mean height (Hg) were nearly constant for all model alternatives with relative root-mean-square errors (RMSE) roughly 10-11% and 6-8%, respectively, and minor biases. For basal area (G) and volume (V), using either of the position methods, resulted in roughly similar predictions at best, with approximately 25% relative RMSE and 15% bias. With XYHH positions, the predictions of G and V were nearly independent of the sample plot size within 254-761 m². Therefore, the harvester-based data can be used as ground truth for remote sensing forest inventory methods. In predicting the forest inventory attributes, it is advisable to utilize harvester head positions (XYHH) and a smallest plot size of 254 m². Instead, if only harvester positions (XYH) are available, expanding the sample plot size to 761 m² reaches a similar accuracy to that obtained using XYHH positions, as the larger sample plot moderates the uncertainties when determining the individual tree position.

Original language	English
Article number	797
Journal	Remote Sensing
Volume	11
Issue number	7
DOIs	https://doi.org/10.3390/rs11070797
Publication status	Published - 1 Apr 2019
MoE publication type	A1 Journal article-refereed

Funding

This research was funded by the Ministry of Agriculture and Forestry of Finland (grant number 173/03.02.02.00/2016), Academy of Finland (grant number 272195) and Tekes (grant number 3165/31/2011), in the form of the projects: “Prediction of Forest Inventory Attributes by Using Remote Sensing and Information Measured by Harvester”, “A common and centralized data warehouse based on data collected by forest machines”, “Centre of Excellence in Laser Scanning Research”, and Data to Intelligence (D2I) research program (Forest Big Data). The APC was funded by a discount voucher of MDPI.

Keywords

Cut-to-length (CTL) harvester
Forest planning
K-Most similar neighbor
LiDAR
Tree positioning
Wood procurement

Access to Document

10.3390/rs11070797

Cite this

@article{52f3614aa79548fc91986f324239f833,

title = "Predicting forest inventory attributes using airborne laser scanning, aerial imagery, and harvester data",

abstract = "The aim of the study was to develop a new method to use tree stem information recorded by harvesters along operative logging in remote sensing-based prediction of forest inventory attributes in mature stands. The reference sample plots were formed from harvester data, using two different tree positions: harvester positions (XYH) in global satellite navigation system and computationally improved harvester head positions (XYHH). Study materials consisted of 158 mature Norway-spruce-dominated stands located in Southern Finland that were clear-cut during 2015-16. Tree attributes were derived from the stem dimensions recorded by the harvester. The forest inventory attributes were compiled for both stands and sample plots generated for stands for four different sample plot sizes (254, 509, 761, and 1018 m2). Prediction models between the harvester-based forest inventory attributes and remote sensing features of sample plots were developed. The stand-level predictions were obtained, and basal-area weighted mean diameter (Dg) and basal-area weighted mean height (Hg) were nearly constant for all model alternatives with relative root-mean-square errors (RMSE) roughly 10-11% and 6-8%, respectively, and minor biases. For basal area (G) and volume (V), using either of the position methods, resulted in roughly similar predictions at best, with approximately 25% relative RMSE and 15% bias. With XYHH positions, the predictions of G and V were nearly independent of the sample plot size within 254-761 m2. Therefore, the harvester-based data can be used as ground truth for remote sensing forest inventory methods. In predicting the forest inventory attributes, it is advisable to utilize harvester head positions (XYHH) and a smallest plot size of 254 m2. Instead, if only harvester positions (XYH) are available, expanding the sample plot size to 761 m2 reaches a similar accuracy to that obtained using XYHH positions, as the larger sample plot moderates the uncertainties when determining the individual tree position.",

keywords = "Cut-to-length (CTL) harvester, Forest planning, K-Most similar neighbor, LiDAR, Tree positioning, Wood procurement",

author = "Atte Saukkola and Timo Melkas and Kirsi Riekki and Sanna Sirparanta and Jussi Peuhkurinen and Markus Holopainen and Juha Hyypp{\"a} and Mikko Vastaranta",

note = "Publisher Copyright: {\textcopyright} 2019 by the authors.",

year = "2019",

month = apr,

day = "1",

doi = "10.3390/rs11070797",

language = "English",

volume = "11",

journal = "Remote Sensing",

issn = "2072-4292",

publisher = "MDPI",

number = "7",

}

TY - JOUR

T1 - Predicting forest inventory attributes using airborne laser scanning, aerial imagery, and harvester data

AU - Saukkola, Atte

AU - Melkas, Timo

AU - Riekki, Kirsi

AU - Sirparanta, Sanna

AU - Peuhkurinen, Jussi

AU - Holopainen, Markus

AU - Hyyppä, Juha

AU - Vastaranta, Mikko

PY - 2019/4/1

Y1 - 2019/4/1

N2 - The aim of the study was to develop a new method to use tree stem information recorded by harvesters along operative logging in remote sensing-based prediction of forest inventory attributes in mature stands. The reference sample plots were formed from harvester data, using two different tree positions: harvester positions (XYH) in global satellite navigation system and computationally improved harvester head positions (XYHH). Study materials consisted of 158 mature Norway-spruce-dominated stands located in Southern Finland that were clear-cut during 2015-16. Tree attributes were derived from the stem dimensions recorded by the harvester. The forest inventory attributes were compiled for both stands and sample plots generated for stands for four different sample plot sizes (254, 509, 761, and 1018 m2). Prediction models between the harvester-based forest inventory attributes and remote sensing features of sample plots were developed. The stand-level predictions were obtained, and basal-area weighted mean diameter (Dg) and basal-area weighted mean height (Hg) were nearly constant for all model alternatives with relative root-mean-square errors (RMSE) roughly 10-11% and 6-8%, respectively, and minor biases. For basal area (G) and volume (V), using either of the position methods, resulted in roughly similar predictions at best, with approximately 25% relative RMSE and 15% bias. With XYHH positions, the predictions of G and V were nearly independent of the sample plot size within 254-761 m2. Therefore, the harvester-based data can be used as ground truth for remote sensing forest inventory methods. In predicting the forest inventory attributes, it is advisable to utilize harvester head positions (XYHH) and a smallest plot size of 254 m2. Instead, if only harvester positions (XYH) are available, expanding the sample plot size to 761 m2 reaches a similar accuracy to that obtained using XYHH positions, as the larger sample plot moderates the uncertainties when determining the individual tree position.

AB - The aim of the study was to develop a new method to use tree stem information recorded by harvesters along operative logging in remote sensing-based prediction of forest inventory attributes in mature stands. The reference sample plots were formed from harvester data, using two different tree positions: harvester positions (XYH) in global satellite navigation system and computationally improved harvester head positions (XYHH). Study materials consisted of 158 mature Norway-spruce-dominated stands located in Southern Finland that were clear-cut during 2015-16. Tree attributes were derived from the stem dimensions recorded by the harvester. The forest inventory attributes were compiled for both stands and sample plots generated for stands for four different sample plot sizes (254, 509, 761, and 1018 m2). Prediction models between the harvester-based forest inventory attributes and remote sensing features of sample plots were developed. The stand-level predictions were obtained, and basal-area weighted mean diameter (Dg) and basal-area weighted mean height (Hg) were nearly constant for all model alternatives with relative root-mean-square errors (RMSE) roughly 10-11% and 6-8%, respectively, and minor biases. For basal area (G) and volume (V), using either of the position methods, resulted in roughly similar predictions at best, with approximately 25% relative RMSE and 15% bias. With XYHH positions, the predictions of G and V were nearly independent of the sample plot size within 254-761 m2. Therefore, the harvester-based data can be used as ground truth for remote sensing forest inventory methods. In predicting the forest inventory attributes, it is advisable to utilize harvester head positions (XYHH) and a smallest plot size of 254 m2. Instead, if only harvester positions (XYH) are available, expanding the sample plot size to 761 m2 reaches a similar accuracy to that obtained using XYHH positions, as the larger sample plot moderates the uncertainties when determining the individual tree position.

KW - Cut-to-length (CTL) harvester

KW - Forest planning

KW - K-Most similar neighbor

KW - LiDAR

KW - Tree positioning

KW - Wood procurement

UR - http://www.scopus.com/inward/record.url?scp=85063990283&partnerID=8YFLogxK

U2 - 10.3390/rs11070797

DO - 10.3390/rs11070797

M3 - Article

AN - SCOPUS:85063990283

SN - 2072-4292

VL - 11

JO - Remote Sensing

JF - Remote Sensing

IS - 7

M1 - 797

ER -

Predicting forest inventory attributes using airborne laser scanning, aerial imagery, and harvester data

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this