Modelling Carbon Sequestration Rates (AI for Good)

With a noticeable increase in anthropogenic carbon dioxide emissions, climate change is one of the most influential problems facing society. The release of greenhouse gases, such as carbon dioxide, into our atmosphere results from a variety of different human activities, including the burning of fossil fuels and the production of electricity, with one of the leading causes being deforestation. The recovery of previously-deforested land and the expansion of forests into new areas provide an incredibly meaningful mitigation strategy, with the potential to offset a significant portion of emissions.
Starting with the temperate forests in the Pacific Northwest, a team from UC Berkeley’s Data X class was able to create and validate a model, using field-collected training data, that predicted that region’s carbon sequestration rates with an r2 of about 80 percent, in order to help quantify the exact carbon benefit of recovering forests – a complex problem, considering the great geographical variability of sequestration rates and the difficulty of field-based measurements. An Extremely Randomized Trees (Extra Trees Regression) on the climate variables, soil data, and existing carbon was the basis for the model, which uses data that can be collected for any location with relative ease. This project will be built upon by introducing training data from other regions and incorporating other publically-available predictor variables to create an more accurate predictive tool, helping to inform decisions about and justify forest conservation by governments and private landowners.