How have terrestrial ecosystems responded to global environmental change?

In the past century, we have witnessed the most substantial rise in global surface temperatures for millennia. A considerable increase in the emission of carbon dioxide (CO2) originating from human activities has been identified as the leading cause of this rapid climate change. The land and ocean sinks take up roughly one half of anthropogenic CO2. In response to rising atmospheric CO2 levels and global surface temperatures, an increasing quantity of CO2 has been absorbed by land vegetation each year. This land sink is, however, highly variable and the precise mechanisms behind the recent positive trend in carbon uptake are poorly understood. Overall, this thesis aims to answer the question of to what extent has land vegetation functioning adapted in response to climate change over the past decades. This work focuses largely on northern boreal ecosystems, which have accounted for up to one half of the total land sink over recent years. First, atmospheric CO2 records are compared with land surface temperature data, which was combined with an analysis of model simulations to determine the extent to which northern high-latitude vegetation uptake is still controlled by temperature. From this analysis, it is determined that high latitude spring carbon uptake remained strongly controlled by temperature during the 1979-2016 period, contrary to previous findings. Following on from this, the thesis analyses the carbon-13 (13C) isotope record in the atmosphere, which is a key indicator of land vegetation functioning. A suite of simulations is then produced to determine the key driving factors behind the variation of 13C in the atmosphere. Uncertainty in the oceans is determined to be the dominating factor over atmospheric 13C, with vegetation productivity and soil turnover times also emerging as important players. Finally, shifts in the coverage of Alaskan forests are examined using remote sensing to detect changes over time. A deep learning model is trained with the aim of enhancing the consistency of the satellite record and hence improving the robustness of the estimated long-term trends of tree cover. This deep learning model is demonstrated to be more effective at enhancing the consistency of satellite data than classical means and allowing more accurate reconstructions of tree cover change over time.