Tampere University of Technology

TUTCRIS Research Portal

Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images

Research output: Contribution to journalArticleScientificpeer-review

Standard

Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images. / Schraik, Daniel; Varvia, Petri; Korhonen, Lauri; Rautiainen, Miina.

In: JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER, Vol. 233, 01.08.2019, p. 1-12.

Research output: Contribution to journalArticleScientificpeer-review

Harvard

Schraik, D, Varvia, P, Korhonen, L & Rautiainen, M 2019, 'Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images', JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER, vol. 233, pp. 1-12. https://doi.org/10.1016/j.jqsrt.2019.05.013

APA

Schraik, D., Varvia, P., Korhonen, L., & Rautiainen, M. (2019). Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images. JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER, 233, 1-12. https://doi.org/10.1016/j.jqsrt.2019.05.013

Vancouver

Schraik D, Varvia P, Korhonen L, Rautiainen M. Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images. JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER. 2019 Aug 1;233:1-12. https://doi.org/10.1016/j.jqsrt.2019.05.013

Author

Schraik, Daniel ; Varvia, Petri ; Korhonen, Lauri ; Rautiainen, Miina. / Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images. In: JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER. 2019 ; Vol. 233. pp. 1-12.

Bibtex - Download

@article{f650b8505fb74d2da1499367228ca5e4,
title = "Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images",
abstract = "The inversion of reflectance models is a generalizable tool to obtain estimates on forest biophysical parameters, such as leaf area index, with theoretically little information need from a study area, instead relying on the knowledge about physical processes in the forest radiation regime. The use of prior information can greatly improve the reflectance model inversion, however, the literature does not yet provide much information on the selection of priors and their influence on the inversion results. In this study, we used a Bayesian approach to invert the PARAS forest reflectance model and retrieve leaf area index from Sentinel-2 MSI and Landsat 8 OLI multispectral satellite images. The PARAS model is based on the theory of spectral invariants, which describes the influence of wavelength-independent parameters on forest radiative transfer. The Bayesian inversion approach is highly flexible, provides uncertainty quantification, and enables the explicit incorporation of prior knowledge into the inversion process. We found that the choice of prior information is crucial in inverting a forest reflectance model to predict leaf area index. Regularizing and informative priors for leaf area index strongly improved the predictions, relative to an uninformative prior, in that they counteracted the saturation effect of the optical signal occuring at high values for leaf area index. The predictions of leaf area index were more accurate for Landsat 8 than for Sentinel-2, due to potential inconsistencies in the visible bands of Sentinel-2 in our data, and the higher spectral resolution.",
keywords = "Bayesian inversion, Clumping, Forest reflectance, Landsat 8, Leaf area index, PARAS, Recollision probability, Sentinel-2, Spectral invariants",
author = "Daniel Schraik and Petri Varvia and Lauri Korhonen and Miina Rautiainen",
year = "2019",
month = "8",
day = "1",
doi = "10.1016/j.jqsrt.2019.05.013",
language = "English",
volume = "233",
pages = "1--12",
journal = "JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER",
issn = "0022-4073",
publisher = "Elsevier",

}

RIS (suitable for import to EndNote) - Download

TY - JOUR

T1 - Bayesian inversion of a forest reflectance model using Sentinel-2 and Landsat 8 satellite images

AU - Schraik, Daniel

AU - Varvia, Petri

AU - Korhonen, Lauri

AU - Rautiainen, Miina

PY - 2019/8/1

Y1 - 2019/8/1

N2 - The inversion of reflectance models is a generalizable tool to obtain estimates on forest biophysical parameters, such as leaf area index, with theoretically little information need from a study area, instead relying on the knowledge about physical processes in the forest radiation regime. The use of prior information can greatly improve the reflectance model inversion, however, the literature does not yet provide much information on the selection of priors and their influence on the inversion results. In this study, we used a Bayesian approach to invert the PARAS forest reflectance model and retrieve leaf area index from Sentinel-2 MSI and Landsat 8 OLI multispectral satellite images. The PARAS model is based on the theory of spectral invariants, which describes the influence of wavelength-independent parameters on forest radiative transfer. The Bayesian inversion approach is highly flexible, provides uncertainty quantification, and enables the explicit incorporation of prior knowledge into the inversion process. We found that the choice of prior information is crucial in inverting a forest reflectance model to predict leaf area index. Regularizing and informative priors for leaf area index strongly improved the predictions, relative to an uninformative prior, in that they counteracted the saturation effect of the optical signal occuring at high values for leaf area index. The predictions of leaf area index were more accurate for Landsat 8 than for Sentinel-2, due to potential inconsistencies in the visible bands of Sentinel-2 in our data, and the higher spectral resolution.

AB - The inversion of reflectance models is a generalizable tool to obtain estimates on forest biophysical parameters, such as leaf area index, with theoretically little information need from a study area, instead relying on the knowledge about physical processes in the forest radiation regime. The use of prior information can greatly improve the reflectance model inversion, however, the literature does not yet provide much information on the selection of priors and their influence on the inversion results. In this study, we used a Bayesian approach to invert the PARAS forest reflectance model and retrieve leaf area index from Sentinel-2 MSI and Landsat 8 OLI multispectral satellite images. The PARAS model is based on the theory of spectral invariants, which describes the influence of wavelength-independent parameters on forest radiative transfer. The Bayesian inversion approach is highly flexible, provides uncertainty quantification, and enables the explicit incorporation of prior knowledge into the inversion process. We found that the choice of prior information is crucial in inverting a forest reflectance model to predict leaf area index. Regularizing and informative priors for leaf area index strongly improved the predictions, relative to an uninformative prior, in that they counteracted the saturation effect of the optical signal occuring at high values for leaf area index. The predictions of leaf area index were more accurate for Landsat 8 than for Sentinel-2, due to potential inconsistencies in the visible bands of Sentinel-2 in our data, and the higher spectral resolution.

KW - Bayesian inversion

KW - Clumping

KW - Forest reflectance

KW - Landsat 8

KW - Leaf area index

KW - PARAS

KW - Recollision probability

KW - Sentinel-2

KW - Spectral invariants

U2 - 10.1016/j.jqsrt.2019.05.013

DO - 10.1016/j.jqsrt.2019.05.013

M3 - Article

VL - 233

SP - 1

EP - 12

JO - JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER

JF - JOURNAL OF QUANTITATIVE SPECTROSCOPY AND RADIATIVE TRANSFER

SN - 0022-4073

ER -