Predicting Peatland Net Ecosystem Exchange of CO2 during the Non-Growing Season using Machine Learning

The world’s cold regions are experiencing some of the fastest warming, especially during the winter and shoulder seasons. Recent studies have highlighted the significance of carbon dioxide (CO2) emissions during the non-growing season (NGS) to the annual carbon budgets of northern peatlands. Because of the positive feedback of soil microbial respiration to warming, a warmer NGS may be expected to alter the carbon balance of peatlands, which are estimated to store about one-third of global terrestrial organic carbon stocks. However, estimates of NGS net ecosystem CO2 exchange (NEE) remain highly uncertain. In this study, we determine key environmental variables affecting the NGS-NEE from a temperate peatland (Mer Bleue Bog; Ottawa, Canada) and predict future NGS-NEE under three climate scenarios (RCP2.6, RCP4.5, and RCP8.5) using a variable selection methodology, global sensitivity analysis, and data-driven model. The model successfully reproduces the observed NGS-NEE fluxes using only 7 variables, with NGS-NEE being most sensitive to changes in net radiation. Our projections estimate that mean NEE during the NGS could increase by up to 103% by the end of the 21st century; thus, reinforcing the urgent need for a comprehensive understanding of peatlands as evolving sources of atmospheric CO2 in a warming world.