OOS 1-4 - Combining machine learning and eco-evolutionary theory to incorporate CO2 fertilization in global gross primary productivity upscaling

Terrestrial ecosystems play a vital role in mitigating climate change by sequestering carbon dioxide (CO2) from the atmosphere through photosynthesis. Growing evidence suggests that elevated atmospheric CO2 concentration has increased photosynthesis, a phenomenon known as the CO2 fertilization effect. However, uncertainties in quantifying CO2 fertilization have led to significant discrepancies in existing satellite-based estimations and Terrestrial Biosphere Models (TBMs) simulations of photosynthesis which prevents a robust assessment of global carbon dynamics and prediction of future changes. To address this issue, we employed machine learning and eco-evolutionary theory to evaluate CO2 fertilization in ecosystem-level photosynthesis measurements, also called Gross Primary Productivity (GPP), from globally distributed flux tower sites. We then trained machine learning models at tower locations and upscaled the models using satellite images to generate global GPP maps from 1982 to 2020.

Our results demonstrated widespread positive effects of elevated atmospheric CO2 on global photosynthesis, while standard satellite-based data products underestimated these impacts, particularly in tropical areas, due to a neglect of the direct CO2 effects on photosynthetic light use efficiency. By incorporating the direct CO2 effects through a combination of data-driven and biophysical theory approaches, our datasets showed improved consistency in GPP long-term trend with TBM simulations from the TRENDY ensemble, compared to previous satellite-based estimations. Our datasets also demonstrated greater agreement in the spatial and temporal patterns of GPP interannual variability with TRENDY. Our results reconciled the discrepancy between satellite-derived and TBM-simulated photosynthesis estimations, providing important constraints to the global carbon cycle.