Tree height estimates in boreal forest using Gaussian process regression

Teemu Mutanen, Laura Sirro, Yrjö Rauste

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

Abstract

Tree height is one of the continuous essential climate model variables. This work applies Gaussian process regression to tree height estimation. Estimates are produced in a coniferous boreal forest area in Southern Finland. GP regression produces both estimate and predictive variance for each pixel. Results show that Gaussian process regression produces estimates which are as good as the ones produced by existing methods. The root-mean-square error from GP regression is lower, estimates have less bias and maximum value is closer to the actual maximum. Predictions can be applied and ingested to carbon modelling where estimates are either direct input parameters or they serve for validation the parameters computed by the models.
Original languageEnglish
Title of host publicationGeoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International
PublisherInstitute of Electrical and Electronic Engineers IEEE
Pages1757-1760
ISBN (Electronic)978-1-5090-3332-4, 978-1-5090-3331-7
ISBN (Print)978-1-5090-3333-1
DOIs
Publication statusPublished - 3 Nov 2016
MoE publication typeA4 Article in a conference publication
Event36th IEEE International Geoscience and Remote Sensing Symposium: Advancing the understanding of our living planet - Beijing, China
Duration: 10 Jul 201615 Jul 2016
Conference number: 36

Publication series

Name
ISSN (Print)2153-7003

Conference

Conference36th IEEE International Geoscience and Remote Sensing Symposium
Abbreviated titleIGARSS
CountryChina
CityBeijing
Period10/07/1615/07/16

Fingerprint

boreal forest
coniferous forest
climate modeling
pixel
carbon
prediction
modeling
parameter
method

Keywords

  • boreal forest
  • Gaussian process
  • pixel-level uncertainty
  • regression
  • tree height

Cite this

Mutanen, T., Sirro, L., & Rauste, Y. (2016). Tree height estimates in boreal forest using Gaussian process regression. In Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International (pp. 1757-1760). Institute of Electrical and Electronic Engineers IEEE. https://doi.org/10.1109/IGARSS.2016.7729450
Mutanen, Teemu ; Sirro, Laura ; Rauste, Yrjö. / Tree height estimates in boreal forest using Gaussian process regression. Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International . Institute of Electrical and Electronic Engineers IEEE, 2016. pp. 1757-1760
@inproceedings{b03e9de189a947799922429c2bb461bc,
title = "Tree height estimates in boreal forest using Gaussian process regression",
abstract = "Tree height is one of the continuous essential climate model variables. This work applies Gaussian process regression to tree height estimation. Estimates are produced in a coniferous boreal forest area in Southern Finland. GP regression produces both estimate and predictive variance for each pixel. Results show that Gaussian process regression produces estimates which are as good as the ones produced by existing methods. The root-mean-square error from GP regression is lower, estimates have less bias and maximum value is closer to the actual maximum. Predictions can be applied and ingested to carbon modelling where estimates are either direct input parameters or they serve for validation the parameters computed by the models.",
keywords = "boreal forest, Gaussian process, pixel-level uncertainty, regression, tree height",
author = "Teemu Mutanen and Laura Sirro and Yrj{\"o} Rauste",
year = "2016",
month = "11",
day = "3",
doi = "10.1109/IGARSS.2016.7729450",
language = "English",
isbn = "978-1-5090-3333-1",
publisher = "Institute of Electrical and Electronic Engineers IEEE",
pages = "1757--1760",
booktitle = "Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International",
address = "United States",

}

Mutanen, T, Sirro, L & Rauste, Y 2016, Tree height estimates in boreal forest using Gaussian process regression. in Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International . Institute of Electrical and Electronic Engineers IEEE, pp. 1757-1760, 36th IEEE International Geoscience and Remote Sensing Symposium, Beijing, China, 10/07/16. https://doi.org/10.1109/IGARSS.2016.7729450

Tree height estimates in boreal forest using Gaussian process regression. / Mutanen, Teemu; Sirro, Laura; Rauste, Yrjö.

Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International . Institute of Electrical and Electronic Engineers IEEE, 2016. p. 1757-1760.

Research output: Chapter in Book/Report/Conference proceedingConference article in proceedingsScientificpeer-review

TY - GEN

T1 - Tree height estimates in boreal forest using Gaussian process regression

AU - Mutanen, Teemu

AU - Sirro, Laura

AU - Rauste, Yrjö

PY - 2016/11/3

Y1 - 2016/11/3

N2 - Tree height is one of the continuous essential climate model variables. This work applies Gaussian process regression to tree height estimation. Estimates are produced in a coniferous boreal forest area in Southern Finland. GP regression produces both estimate and predictive variance for each pixel. Results show that Gaussian process regression produces estimates which are as good as the ones produced by existing methods. The root-mean-square error from GP regression is lower, estimates have less bias and maximum value is closer to the actual maximum. Predictions can be applied and ingested to carbon modelling where estimates are either direct input parameters or they serve for validation the parameters computed by the models.

AB - Tree height is one of the continuous essential climate model variables. This work applies Gaussian process regression to tree height estimation. Estimates are produced in a coniferous boreal forest area in Southern Finland. GP regression produces both estimate and predictive variance for each pixel. Results show that Gaussian process regression produces estimates which are as good as the ones produced by existing methods. The root-mean-square error from GP regression is lower, estimates have less bias and maximum value is closer to the actual maximum. Predictions can be applied and ingested to carbon modelling where estimates are either direct input parameters or they serve for validation the parameters computed by the models.

KW - boreal forest

KW - Gaussian process

KW - pixel-level uncertainty

KW - regression

KW - tree height

U2 - 10.1109/IGARSS.2016.7729450

DO - 10.1109/IGARSS.2016.7729450

M3 - Conference article in proceedings

SN - 978-1-5090-3333-1

SP - 1757

EP - 1760

BT - Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International

PB - Institute of Electrical and Electronic Engineers IEEE

ER -

Mutanen T, Sirro L, Rauste Y. Tree height estimates in boreal forest using Gaussian process regression. In Geoscience and Remote Sensing Symposium (IGARSS), 2016 IEEE International . Institute of Electrical and Electronic Engineers IEEE. 2016. p. 1757-1760 https://doi.org/10.1109/IGARSS.2016.7729450