Combining unmanned aerial vehicle and multispectral Pleiades data for tree species identification, a prerequisite for accurate carbon estimation

Agbor Esong Effiom, L.M. Van Leeuwen, P. Nyktas, Jefferson Adetokunbo Okojie, Jana Erdbrügger

Research output: Contribution to journalArticleAcademicpeer-review

Abstract

Forest carbon estimation currently largely relies on remote sensing techniques in combination with field measurement. High-resolution images, which are commonly utilized for carbon estimation, are not readily available, and their cost prohibits communities from reaping the benefits of maintaining their forest under the UN reducing emissions from deforestation and forest degradation program. Our study explores the combination of readily available and relatively cheaper unmanned aerial vehicle (UAV) (4-cm resolution) and multispectral Pleiades (50-cm resolution) images for species classification robustness in view for carbon estimation through object-based image analysis. The images are resampled and used to evaluate the effect of combining multispectral Pleiades image on the accuracies of segmenting UAV images for tree crown projection area (CPA) estimation and species classification. RGB images from a UAV platform are processed in a photogrametric software and combined with the near-infrared band of a Pleiades image to get a UAV-Pleiades image composite. The images are segmented using the ESP 2 tool and the segmentation accuracy compared using a paired t-test. The segmented tree crowns are classified using random trees (RT), support vector machines (SVM), and maximum likelihood (ML) classifiers, and the classification accuracies of the three classifiers are compared using the McNemar's chi-squared test. Our study demonstrates a 93.5% accuracy of segmenting UAV-Pleiades image composite, which is significantly higher than the 84.8% accuracy of segmenting UAV images (p < 0.05). Also an 84% classification accuracy of UAV-Pleiades image composite is significantly higher than the 54% classification accuracy of the UAV images (p < 0.05). Of the three classifiers used, the classification accuracies of SVM and RT are significantly higher (p < 0.05) than that of the ML classifier. Given the significantly high accuracies observed from this study for tree CPA extraction and tree species classification, carbon/above ground biomass modeling is possible with significantly high accuracy using the combination of multispectral Pleiades and UAV images.

Original languageEnglish
Article number034530
Pages (from-to)1-20
Number of pages20
JournalJournal of applied remote sensing
Volume13
Issue number3
DOIs
Publication statusPublished - 28 Sep 2019

Fingerprint

carbon
image resolution
Pleiades
vehicle
aboveground biomass
image analysis
segmentation
deforestation
near infrared
remote sensing
software
degradation
cost
modeling

Keywords

  • ITC-ISI-JOURNAL-ARTICLE

Cite this

@article{29dfc25267174f82b34226d02024c86a,
title = "Combining unmanned aerial vehicle and multispectral Pleiades data for tree species identification, a prerequisite for accurate carbon estimation",
abstract = "Forest carbon estimation currently largely relies on remote sensing techniques in combination with field measurement. High-resolution images, which are commonly utilized for carbon estimation, are not readily available, and their cost prohibits communities from reaping the benefits of maintaining their forest under the UN reducing emissions from deforestation and forest degradation program. Our study explores the combination of readily available and relatively cheaper unmanned aerial vehicle (UAV) (4-cm resolution) and multispectral Pleiades (50-cm resolution) images for species classification robustness in view for carbon estimation through object-based image analysis. The images are resampled and used to evaluate the effect of combining multispectral Pleiades image on the accuracies of segmenting UAV images for tree crown projection area (CPA) estimation and species classification. RGB images from a UAV platform are processed in a photogrametric software and combined with the near-infrared band of a Pleiades image to get a UAV-Pleiades image composite. The images are segmented using the ESP 2 tool and the segmentation accuracy compared using a paired t-test. The segmented tree crowns are classified using random trees (RT), support vector machines (SVM), and maximum likelihood (ML) classifiers, and the classification accuracies of the three classifiers are compared using the McNemar's chi-squared test. Our study demonstrates a 93.5{\%} accuracy of segmenting UAV-Pleiades image composite, which is significantly higher than the 84.8{\%} accuracy of segmenting UAV images (p < 0.05). Also an 84{\%} classification accuracy of UAV-Pleiades image composite is significantly higher than the 54{\%} classification accuracy of the UAV images (p < 0.05). Of the three classifiers used, the classification accuracies of SVM and RT are significantly higher (p < 0.05) than that of the ML classifier. Given the significantly high accuracies observed from this study for tree CPA extraction and tree species classification, carbon/above ground biomass modeling is possible with significantly high accuracy using the combination of multispectral Pleiades and UAV images.",
keywords = "ITC-ISI-JOURNAL-ARTICLE",
author = "Effiom, {Agbor Esong} and {Van Leeuwen}, L.M. and P. Nyktas and Okojie, {Jefferson Adetokunbo} and Jana Erdbr{\"u}gger",
year = "2019",
month = "9",
day = "28",
doi = "10.1117/1.JRS.13.034530",
language = "English",
volume = "13",
pages = "1--20",
journal = "Journal of applied remote sensing",
issn = "1931-3195",
publisher = "SPIE",
number = "3",

}

Combining unmanned aerial vehicle and multispectral Pleiades data for tree species identification, a prerequisite for accurate carbon estimation. / Effiom, Agbor Esong; Van Leeuwen, L.M.; Nyktas, P.; Okojie, Jefferson Adetokunbo; Erdbrügger, Jana.

In: Journal of applied remote sensing, Vol. 13, No. 3, 034530, 28.09.2019, p. 1-20.

Research output: Contribution to journalArticleAcademicpeer-review

TY - JOUR

T1 - Combining unmanned aerial vehicle and multispectral Pleiades data for tree species identification, a prerequisite for accurate carbon estimation

AU - Effiom, Agbor Esong

AU - Van Leeuwen, L.M.

AU - Nyktas, P.

AU - Okojie, Jefferson Adetokunbo

AU - Erdbrügger, Jana

PY - 2019/9/28

Y1 - 2019/9/28

N2 - Forest carbon estimation currently largely relies on remote sensing techniques in combination with field measurement. High-resolution images, which are commonly utilized for carbon estimation, are not readily available, and their cost prohibits communities from reaping the benefits of maintaining their forest under the UN reducing emissions from deforestation and forest degradation program. Our study explores the combination of readily available and relatively cheaper unmanned aerial vehicle (UAV) (4-cm resolution) and multispectral Pleiades (50-cm resolution) images for species classification robustness in view for carbon estimation through object-based image analysis. The images are resampled and used to evaluate the effect of combining multispectral Pleiades image on the accuracies of segmenting UAV images for tree crown projection area (CPA) estimation and species classification. RGB images from a UAV platform are processed in a photogrametric software and combined with the near-infrared band of a Pleiades image to get a UAV-Pleiades image composite. The images are segmented using the ESP 2 tool and the segmentation accuracy compared using a paired t-test. The segmented tree crowns are classified using random trees (RT), support vector machines (SVM), and maximum likelihood (ML) classifiers, and the classification accuracies of the three classifiers are compared using the McNemar's chi-squared test. Our study demonstrates a 93.5% accuracy of segmenting UAV-Pleiades image composite, which is significantly higher than the 84.8% accuracy of segmenting UAV images (p < 0.05). Also an 84% classification accuracy of UAV-Pleiades image composite is significantly higher than the 54% classification accuracy of the UAV images (p < 0.05). Of the three classifiers used, the classification accuracies of SVM and RT are significantly higher (p < 0.05) than that of the ML classifier. Given the significantly high accuracies observed from this study for tree CPA extraction and tree species classification, carbon/above ground biomass modeling is possible with significantly high accuracy using the combination of multispectral Pleiades and UAV images.

AB - Forest carbon estimation currently largely relies on remote sensing techniques in combination with field measurement. High-resolution images, which are commonly utilized for carbon estimation, are not readily available, and their cost prohibits communities from reaping the benefits of maintaining their forest under the UN reducing emissions from deforestation and forest degradation program. Our study explores the combination of readily available and relatively cheaper unmanned aerial vehicle (UAV) (4-cm resolution) and multispectral Pleiades (50-cm resolution) images for species classification robustness in view for carbon estimation through object-based image analysis. The images are resampled and used to evaluate the effect of combining multispectral Pleiades image on the accuracies of segmenting UAV images for tree crown projection area (CPA) estimation and species classification. RGB images from a UAV platform are processed in a photogrametric software and combined with the near-infrared band of a Pleiades image to get a UAV-Pleiades image composite. The images are segmented using the ESP 2 tool and the segmentation accuracy compared using a paired t-test. The segmented tree crowns are classified using random trees (RT), support vector machines (SVM), and maximum likelihood (ML) classifiers, and the classification accuracies of the three classifiers are compared using the McNemar's chi-squared test. Our study demonstrates a 93.5% accuracy of segmenting UAV-Pleiades image composite, which is significantly higher than the 84.8% accuracy of segmenting UAV images (p < 0.05). Also an 84% classification accuracy of UAV-Pleiades image composite is significantly higher than the 54% classification accuracy of the UAV images (p < 0.05). Of the three classifiers used, the classification accuracies of SVM and RT are significantly higher (p < 0.05) than that of the ML classifier. Given the significantly high accuracies observed from this study for tree CPA extraction and tree species classification, carbon/above ground biomass modeling is possible with significantly high accuracy using the combination of multispectral Pleiades and UAV images.

KW - ITC-ISI-JOURNAL-ARTICLE

UR - https://ezproxy2.utwente.nl/login?url=https://doi.org/10.1117/1.JRS.13.034530

UR - https://ezproxy2.utwente.nl/login?url=https://library.itc.utwente.nl/login/2019/isi/vanleeuwen_com.pdf

U2 - 10.1117/1.JRS.13.034530

DO - 10.1117/1.JRS.13.034530

M3 - Article

VL - 13

SP - 1

EP - 20

JO - Journal of applied remote sensing

JF - Journal of applied remote sensing

SN - 1931-3195

IS - 3

M1 - 034530

ER -