Comparing object- and pixel-based random forest landcover classification techniques for better pre-flooding carbon balance estimation of a boreal reservoir
(2026) NGEK11 20261Department of Earth and Environmental Sciences (MGeo)
- Abstract
- The impoundment of hydroelectric reservoirs is associated with significant emissions of green house gases. In addition, the loss of the pre-flooded landscape is a substantial factor in the long-term carbon footprint of hydroelectric reservoirs. The Paix-des-Braves reservoir in Québec was one of the first reservoirs to be studied in regards of its total carbon footprint, combining pre- and post-impoundment carbon balance estimates. However, carbon balance estimates for the pre-impoundment landscape of the Paix-des-Braves reservoir currently rely on landcover information which has not been assessed for its accuracy or otherwise validated. This leads to uncertainties in the accuracy of the carbon balance estimates. Therefore, this study aimed... (More)
- The impoundment of hydroelectric reservoirs is associated with significant emissions of green house gases. In addition, the loss of the pre-flooded landscape is a substantial factor in the long-term carbon footprint of hydroelectric reservoirs. The Paix-des-Braves reservoir in Québec was one of the first reservoirs to be studied in regards of its total carbon footprint, combining pre- and post-impoundment carbon balance estimates. However, carbon balance estimates for the pre-impoundment landscape of the Paix-des-Braves reservoir currently rely on landcover information which has not been assessed for its accuracy or otherwise validated. This leads to uncertainties in the accuracy of the carbon balance estimates. Therefore, this study aimed to improve the landcover classification for the pre-impoundment landscape of the Paix-des-Braves reservoir in Québec by using Landsat 5TM imagery and auxiliary data for object- and pixel-based random forest landcover classification. The study was able to produce a map accurate enough for carbon balance estimates and proposes two approaches for accuracy assessment without field validation. Both the out-of-bag error per class and the correlation matrices using a subset of the training data as validation yielded very high accuracies for both modelling approaches. Based on the updated landcover information, the pre-impoundment PDB reservoir landscape was found to be a sink of CO2 and a small source of CH4. Furthermore, differences in landcover classification were found to clearly impact carbon balance results with diffusive flux values for CH4 being outside the range of previous studies. The methodology proposed in this study can be transferred to other boreal hydroelectric reservoirs to improve our understanding of the carbon footprint of hydroelectric energy. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9241113
- author
- Hamel, Lia Sophie LU
- supervisor
- organization
- course
- NGEK11 20261
- year
- 2026
- type
- M2 - Bachelor Degree
- subject
- keywords
- Hydroelectric reservoir, carbon balance, greenhouse gases, random forest, boreal, Québec, PDB, LULC, remote sensing
- language
- English
- id
- 9241113
- date added to LUP
- 2026-06-18 12:40:57
- date last changed
- 2026-06-18 12:40:57
@misc{9241113,
abstract = {{The impoundment of hydroelectric reservoirs is associated with significant emissions of green house gases. In addition, the loss of the pre-flooded landscape is a substantial factor in the long-term carbon footprint of hydroelectric reservoirs. The Paix-des-Braves reservoir in Québec was one of the first reservoirs to be studied in regards of its total carbon footprint, combining pre- and post-impoundment carbon balance estimates. However, carbon balance estimates for the pre-impoundment landscape of the Paix-des-Braves reservoir currently rely on landcover information which has not been assessed for its accuracy or otherwise validated. This leads to uncertainties in the accuracy of the carbon balance estimates. Therefore, this study aimed to improve the landcover classification for the pre-impoundment landscape of the Paix-des-Braves reservoir in Québec by using Landsat 5TM imagery and auxiliary data for object- and pixel-based random forest landcover classification. The study was able to produce a map accurate enough for carbon balance estimates and proposes two approaches for accuracy assessment without field validation. Both the out-of-bag error per class and the correlation matrices using a subset of the training data as validation yielded very high accuracies for both modelling approaches. Based on the updated landcover information, the pre-impoundment PDB reservoir landscape was found to be a sink of CO2 and a small source of CH4. Furthermore, differences in landcover classification were found to clearly impact carbon balance results with diffusive flux values for CH4 being outside the range of previous studies. The methodology proposed in this study can be transferred to other boreal hydroelectric reservoirs to improve our understanding of the carbon footprint of hydroelectric energy.}},
author = {{Hamel, Lia Sophie}},
language = {{eng}},
note = {{Student Paper}},
title = {{Comparing object- and pixel-based random forest landcover classification techniques for better pre-flooding carbon balance estimation of a boreal reservoir}},
year = {{2026}},
}