Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data

David A. Coomes, Michele Dalponte, Tommaso Jucker, Gregory P. Asner, Lindsay F. Banin, David F. R. P. Burslem, Simon L. Lewis, Reuben Nilus, Oliver L. Phillips, Mui-How Phua, Lan Qie

Research output: Contribution to journalArticle

43 Citations (Scopus)
3 Downloads (Pure)

Abstract

Tropical forests are a key component of the global carbon cycle, and mapping their carbon density is essential for understanding human influences on climate and for ecosystem-service-based payments for forest protection. Discrete-return airborne laser scanning (ALS) is increasingly recognised as a high-quality technology for mapping tropical forest carbon, because it generates 3D point clouds of forest structure from which aboveground carbon density (ACD) can be estimated. Area-based models are state of the art when it comes to estimating ACD from ALS data, but discard tree-level information contained within the ALS point cloud. This paper compares area-based and tree-centric models for estimating ACD in lowland old-growth forests in Sabah, Malaysia. These forests are challenging to map because of their immense height. We compare the performance of (a) an area-based model developed by Asner and Mascaro (2014), and used primarily in the neotropics hitherto, with (b) a tree-centric approach that uses a new algorithm (itcSegment) to locate trees within the ALS canopy height model, measures their heights and crown widths, and calculates biomass from these dimensions. We find that Asner and Mascaro's model needed regional calibration, reflecting the distinctive structure of Southeast Asian forests. We also discover that forest basal area is closely related to canopy gap fraction measured by ALS, and use this finding to refine Asner and Mascaro's model. Finally, we show that our tree-centric approach is less accurate at estimating ACD than the best-performing area-based model (RMSE 18% vs 13%). Tree-centric modelling is appealing because it is based on summing the biomass of individual trees, but until algorithms can detect understory trees reliably and estimate biomass from crown dimensions precisely, areas-based modelling will remain the method of choice.
Original languageEnglish
Pages (from-to)77-88
Number of pages12
JournalRemote Sensing of Environment
Volume194
Early online date28 Mar 2017
DOIs
Publication statusPublished - 1 Jun 2017

Fingerprint

lasers
laser
Scanning
Carbon
Lasers
carbon
Biomass
tropical forest
tropical forests
biomass
tree crown
canopy gap
old-growth forest
Ecosystems
canopy gaps
tropical montane cloud forests
old-growth forests
carbon cycle
basal area
ecosystem service

Keywords

  • Allometry
  • Aboveground carbon density
  • Biomass estimation
  • Image analysis
  • LiDAR
  • Object recognition
  • Power-law
  • Tree delineation
  • Tropical forests

Cite this

Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data. / Coomes, David A.; Dalponte, Michele; Jucker, Tommaso; Asner, Gregory P. ; Banin, Lindsay F.; Burslem, David F. R. P.; Lewis, Simon L.; Nilus, Reuben; Phillips, Oliver L.; Phua, Mui-How ; Qie, Lan .

In: Remote Sensing of Environment, Vol. 194, 01.06.2017, p. 77-88.

Research output: Contribution to journalArticle

Coomes, DA, Dalponte, M, Jucker, T, Asner, GP, Banin, LF, Burslem, DFRP, Lewis, SL, Nilus, R, Phillips, OL, Phua, M-H & Qie, L 2017, 'Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data', Remote Sensing of Environment, vol. 194, pp. 77-88. https://doi.org/10.1016/j.rse.2017.03.017
Coomes, David A. ; Dalponte, Michele ; Jucker, Tommaso ; Asner, Gregory P. ; Banin, Lindsay F. ; Burslem, David F. R. P. ; Lewis, Simon L. ; Nilus, Reuben ; Phillips, Oliver L. ; Phua, Mui-How ; Qie, Lan . / Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data. In: Remote Sensing of Environment. 2017 ; Vol. 194. pp. 77-88.
@article{da33689f18854af18651b926f7f8987a,
title = "Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data",
abstract = "Tropical forests are a key component of the global carbon cycle, and mapping their carbon density is essential for understanding human influences on climate and for ecosystem-service-based payments for forest protection. Discrete-return airborne laser scanning (ALS) is increasingly recognised as a high-quality technology for mapping tropical forest carbon, because it generates 3D point clouds of forest structure from which aboveground carbon density (ACD) can be estimated. Area-based models are state of the art when it comes to estimating ACD from ALS data, but discard tree-level information contained within the ALS point cloud. This paper compares area-based and tree-centric models for estimating ACD in lowland old-growth forests in Sabah, Malaysia. These forests are challenging to map because of their immense height. We compare the performance of (a) an area-based model developed by Asner and Mascaro (2014), and used primarily in the neotropics hitherto, with (b) a tree-centric approach that uses a new algorithm (itcSegment) to locate trees within the ALS canopy height model, measures their heights and crown widths, and calculates biomass from these dimensions. We find that Asner and Mascaro's model needed regional calibration, reflecting the distinctive structure of Southeast Asian forests. We also discover that forest basal area is closely related to canopy gap fraction measured by ALS, and use this finding to refine Asner and Mascaro's model. Finally, we show that our tree-centric approach is less accurate at estimating ACD than the best-performing area-based model (RMSE 18{\%} vs 13{\%}). Tree-centric modelling is appealing because it is based on summing the biomass of individual trees, but until algorithms can detect understory trees reliably and estimate biomass from crown dimensions precisely, areas-based modelling will remain the method of choice.",
keywords = "Allometry, Aboveground carbon density, Biomass estimation, Image analysis, LiDAR, Object recognition, Power-law, Tree delineation, Tropical forests",
author = "Coomes, {David A.} and Michele Dalponte and Tommaso Jucker and Asner, {Gregory P.} and Banin, {Lindsay F.} and Burslem, {David F. R. P.} and Lewis, {Simon L.} and Reuben Nilus and Phillips, {Oliver L.} and Mui-How Phua and Lan Qie",
note = "Acknowledgement This project was supported by a grant through the Human Modified Tropical Forests programme of NERC (NE/K016377/1). We thank members of the NERC Airborne Remote Sensing Facility and NERC Data Analysis Node for collecting and processing the data (project code MA14-14). David Coomes was supported by an International Academic Fellowship from the Leverhulme Trust. Lindsay Banin contributed field allometry data which were collected during her PhD at University Leeds, supported by NERC and a RGS Henrietta Hutton Grant. Oliver Phillips, Simon Lewis and Lan Qie provided census data collected as part of an ERC Advanced Grant (T-Forces). Thanks to Richard Bryan Sebastian for collecting field data on delineated trees.",
year = "2017",
month = "6",
day = "1",
doi = "10.1016/j.rse.2017.03.017",
language = "English",
volume = "194",
pages = "77--88",
journal = "Remote Sensing of Environment",
issn = "0034-4257",
publisher = "Elsevier Inc.",

}

TY - JOUR

T1 - Area-based vs tree-centric approaches to mapping forest carbon in Southeast Asian forests from airborne laser scanning data

AU - Coomes, David A.

AU - Dalponte, Michele

AU - Jucker, Tommaso

AU - Asner, Gregory P.

AU - Banin, Lindsay F.

AU - Burslem, David F. R. P.

AU - Lewis, Simon L.

AU - Nilus, Reuben

AU - Phillips, Oliver L.

AU - Phua, Mui-How

AU - Qie, Lan

N1 - Acknowledgement This project was supported by a grant through the Human Modified Tropical Forests programme of NERC (NE/K016377/1). We thank members of the NERC Airborne Remote Sensing Facility and NERC Data Analysis Node for collecting and processing the data (project code MA14-14). David Coomes was supported by an International Academic Fellowship from the Leverhulme Trust. Lindsay Banin contributed field allometry data which were collected during her PhD at University Leeds, supported by NERC and a RGS Henrietta Hutton Grant. Oliver Phillips, Simon Lewis and Lan Qie provided census data collected as part of an ERC Advanced Grant (T-Forces). Thanks to Richard Bryan Sebastian for collecting field data on delineated trees.

PY - 2017/6/1

Y1 - 2017/6/1

N2 - Tropical forests are a key component of the global carbon cycle, and mapping their carbon density is essential for understanding human influences on climate and for ecosystem-service-based payments for forest protection. Discrete-return airborne laser scanning (ALS) is increasingly recognised as a high-quality technology for mapping tropical forest carbon, because it generates 3D point clouds of forest structure from which aboveground carbon density (ACD) can be estimated. Area-based models are state of the art when it comes to estimating ACD from ALS data, but discard tree-level information contained within the ALS point cloud. This paper compares area-based and tree-centric models for estimating ACD in lowland old-growth forests in Sabah, Malaysia. These forests are challenging to map because of their immense height. We compare the performance of (a) an area-based model developed by Asner and Mascaro (2014), and used primarily in the neotropics hitherto, with (b) a tree-centric approach that uses a new algorithm (itcSegment) to locate trees within the ALS canopy height model, measures their heights and crown widths, and calculates biomass from these dimensions. We find that Asner and Mascaro's model needed regional calibration, reflecting the distinctive structure of Southeast Asian forests. We also discover that forest basal area is closely related to canopy gap fraction measured by ALS, and use this finding to refine Asner and Mascaro's model. Finally, we show that our tree-centric approach is less accurate at estimating ACD than the best-performing area-based model (RMSE 18% vs 13%). Tree-centric modelling is appealing because it is based on summing the biomass of individual trees, but until algorithms can detect understory trees reliably and estimate biomass from crown dimensions precisely, areas-based modelling will remain the method of choice.

AB - Tropical forests are a key component of the global carbon cycle, and mapping their carbon density is essential for understanding human influences on climate and for ecosystem-service-based payments for forest protection. Discrete-return airborne laser scanning (ALS) is increasingly recognised as a high-quality technology for mapping tropical forest carbon, because it generates 3D point clouds of forest structure from which aboveground carbon density (ACD) can be estimated. Area-based models are state of the art when it comes to estimating ACD from ALS data, but discard tree-level information contained within the ALS point cloud. This paper compares area-based and tree-centric models for estimating ACD in lowland old-growth forests in Sabah, Malaysia. These forests are challenging to map because of their immense height. We compare the performance of (a) an area-based model developed by Asner and Mascaro (2014), and used primarily in the neotropics hitherto, with (b) a tree-centric approach that uses a new algorithm (itcSegment) to locate trees within the ALS canopy height model, measures their heights and crown widths, and calculates biomass from these dimensions. We find that Asner and Mascaro's model needed regional calibration, reflecting the distinctive structure of Southeast Asian forests. We also discover that forest basal area is closely related to canopy gap fraction measured by ALS, and use this finding to refine Asner and Mascaro's model. Finally, we show that our tree-centric approach is less accurate at estimating ACD than the best-performing area-based model (RMSE 18% vs 13%). Tree-centric modelling is appealing because it is based on summing the biomass of individual trees, but until algorithms can detect understory trees reliably and estimate biomass from crown dimensions precisely, areas-based modelling will remain the method of choice.

KW - Allometry

KW - Aboveground carbon density

KW - Biomass estimation

KW - Image analysis

KW - LiDAR

KW - Object recognition

KW - Power-law

KW - Tree delineation

KW - Tropical forests

U2 - 10.1016/j.rse.2017.03.017

DO - 10.1016/j.rse.2017.03.017

M3 - Article

VL - 194

SP - 77

EP - 88

JO - Remote Sensing of Environment

JF - Remote Sensing of Environment

SN - 0034-4257

ER -