UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Using machine learning to identify and map controls of growing-season carbon dioxide and methane fluxes in the Mackenzie Delta region Skeeter, June


With climate change, this permafrost carbon pool is at risk of destabilization and emission to the atmosphere. Two field campaigns were conducted in the Mackenzie Delta Region to measure carbon dioxide and methane fluxes with eddy covariance and chamber methods. Illisarvik, a drained thermokarst lake, was studied during the peak growing season in 2016. Fish Island, a low center polygonal peatland, was studied over the 2017 growing season. Half-hourly fluxes were calculated and filtered with quality control tests. Flux footprints were calculated and overlaid on landscape classification maps to estimate the relative flux contributions from different vegetation types and microtopographic features. For both data sets, neural networks (NN) were trained to map flux responses to large sets of potential soil and weather drivers and flux contributions by landscape classification. The NN were pruned to identify the strongest drivers, which is a novel approach that has not been applied to eddy covariance data elsewhere. The pruned NN were used to gap-fill the flux time series to calculate net ecosystem exchange (NEE) and net methane exchange (NME). The functional relationships identified by the NNs were visualized by plotting model derivatives and projecting estimates across the observed ranges of drivers. Radiative input and soil temperature were the primary determinants of NEE, and the NN were able to closely map the functional relationships. Spatial heterogeneity had a significant influence on NME and the NN estimates could not map the relationships as accurately. Further analysis was conducted with the Fish Island data. A set of surface observations from a weather station at Fish Island was paired with reanalysis and satellite data. Regression models were trained on the 2017 field data to estimate a time series of flux drivers and NN were used to project NEE and NME over the 2009 to 2019 snow-free seasons. These estimates contextualize the 2017 flux observations. They indicate Fish Island was a net growing season CO2 sink and CH4 source from 2009 to 2019, but shoulder season CO2 emissions may have offset growing season uptake in some years. They also indicated climate warming may reduce carbon uptake at Fish Island.

Item Media

Item Citations and Data


Attribution-NonCommercial-ShareAlike 4.0 International