Back to search

Carleton University

Random Forest Modelling and Monitoring of Surface-Atmosphere Carbon Dioxide Flux in the Hudson Bay Lowlands

Abstract

dc:description.abstract

Peatlands are a critical component of the global carbon cycle despite covering only 3% of the earth’s surface. Within Canada the largest contiguous peatland is the Hudson Bay Lowland (HBL), storing an estimated 33 Gt of carbon as peat as a result of a small but persistent difference between gross primary productivity (GPP) and ecosystem respiration (ER) over millennia. The vastness and remote nature of the HBL makes in-situ monitoring of its carbon cycle difficult. To address this, random forest (RF) regression models driven solely by remote sensing imagery (either 500 m MODIS or 30 m Harmonized Landsat Sentinel) are trained to predict GPP based on eddy covariance CO2 flux data from 2012-2019 for five sites spanning a climatic gradient in the HBL. There is little difference between daily GPP simulations for spatial resolutions between 30 m and 500 m (R2 = 0.65-0.78), but better temporal resolution improves model results at the annual time step. Additional RF models are trained to predict ER and the net difference between ER and GPP (net ecosystem exchange; NEE) using MODIS data. While ER remains accurately predicted (R2 ~0.82) by the models, NEE models are weaker (R2 ~0.48) but generally do better than published semi-empirical methods based on light use efficiency and temperature-respiration relationships. These RF models are then applied to three 48 km x 48 km regions around the field sites and compared with established global products from other machine learning algorithms, process-based, atmospheric inversion, and remote sensing-based models for the period 2000-2020. These products predict greater GPP compared to the RF models given that few include peatland-specific parameterizations. However, all models (RF and global products) present a common latitudinal trend with the greatest GPP in the south and decreasing northward. Most of the models agree that the HBL is a net sink of CO2 for this period but there is less agreement on the magnitude and latitudinal gradient. Remote sensing-based RF models of CO2 flux show promise for monitoring changes in the HBL’s carbon flux, particularly when there is land use change or other impactful disturbances on a large scale.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy (Ph.D.)
Level thesis:degree_level
Doctoral
Discipline thesis:degree_discipline
Geography
Grantor dc:publisher
Carleton University
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Beaver, Jason Glen

Subjects

dc:subject × 2

Rights

dc:rights
Statement dc:rights
  • Copyright © 2023 the author(s). Theses may be used for non-commercial research, educational, or related academic purposes only. Such uses include personal study, distribution to students, research and scholarship. Theses may only be shared by linking to the Carleton University Institutional Repository and no part may be copied without proper attribution to the author; no part may be used for commercial purposes directly or indirectly via a for-profit platform; no adaptation or derivative works are permitted without consent from the copyright owner.
Language dc:language.iso
en

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Carleton University
Base URL
carleton.scholaris.ca/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Beaver, Jason Glen. Random Forest Modelling and Monitoring of Surface-Atmosphere Carbon Dioxide Flux in the Hudson Bay Lowlands. Doctoral thesis, Carleton University, 2024. https://hdl.handle.net/20.500.14718/41359