Improved estimates and understanding of interannual trends of CO₂ fluxes in the Southern Ocean
The Southern Ocean plays an important role in mitigating the effects of anthropogenically driven climate change. The region accounts for 43% of oceanic uptake of anthropogenic carbon dioxide (CO₂). This is foreseen to change with increasing greenhouse gas emissions due to ocean chemistry and climate...
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | Doctoral or Postdoctoral Thesis |
| Language: | English |
| Published: |
University of Cape Town
2017
|
| Subjects: | |
| Online Access: | http://hdl.handle.net/11427/25320 https://open.uct.ac.za/bitstream/11427/25320/1/thesis_sci_2017_gregor_luke.pdf |
| Summary: | The Southern Ocean plays an important role in mitigating the effects of anthropogenically driven climate change. The region accounts for 43% of oceanic uptake of anthropogenic carbon dioxide (CO₂). This is foreseen to change with increasing greenhouse gas emissions due to ocean chemistry and climate feedbacks that regulate the carbon cycle in the Southern Ocean. Studies have already shown that Southern Ocean CO₂ is subject to interannual variability. Measuring and understanding this change has been difficult due to sparse observational data that is biased toward summer. This leaves a crucial gap in our understanding of the Southern Ocean CO₂ seasonal cycle, which needs to be resolved to adequately monitor change and gain insight into the drivers of interannual variability. Machine learning has been successful in estimating CO₂ in may parts of the ocean by extrapolating existing data with satellite measurements of proxy variables of CO₂. However, in the Southern Ocean machine learning has proven less successful. Large differences between machine learning estimates stem from the paucity of data and complexity of the mechanisms that drive CO₂. In this study the aim is to reduce the uncertainty of estimates, advance our understanding of the interannual drivers, and optimise sampling of CO₂ in the Southern Ocean. Improving the estimates of CO₂ was achieved by investigating: the impact of increasing the gridding resolution of input data and proxy variables, and Support vector regression (SVR) and Random Forest Regression (RFR) as alternate machine learning methods. It was found that the improvement gained by increasing gridding resolution was minimal and only RFR was able to improve on existing error estimates. Yet, there was good agreement of the seasonal cycle and interannual trends between RFR, SVR and estimates from the literature. The ensemble mean of these methods was used to investigate the variability and interannual trends of CO₂ in the Southern Ocean. The interannual trends of the ensemble confirmed trends ... |
|---|