In the heart of California, researchers at Sandia National Laboratory and the Joint BioEnergy Institute are harnessing the power of machine learning to revolutionize how we predict the future of bioenergy crops. Led by Sagar Gautam, a scientist at both institutions, a groundbreaking study has combined machine learning with traditional agroecosystem modeling to project biomass yields and soil organic carbon changes for three key bioenergy crops: Miscanthus, sorghum, and switchgrass. The findings, published in Carbon Capture Science & Technology, could significantly impact the energy sector’s approach to bioenergy production and land use.
The research addresses a critical challenge in agroecosystem modeling: the high computational and data demands that limit scalability. Traditional models, while detailed, struggle to efficiently predict outcomes across multiple scenarios and timescales. Enter machine learning, which offers a streamlined approach, reducing computational resources and enhancing predictive accuracy.
Gautam and his team integrated a machine learning model with the Daily Century (DAYCENT) agroecosystem model to project baseline (2009–2018) and future (2021–2100) biomass yields and soil organic carbon changes. The projections were made under the Shared Socio-economic Pathway 8.5 (SSP5 8.5), a scenario that assumes high greenhouse gas emissions and significant socio-economic changes.
The results are promising. The machine learning-based reduced-order model accurately predicted baseline biomass yields with R2 values ranging from 0.96 to 0.98, and soil organic carbon changes with R2 values between 0.93 and 0.98 across the three bioenergy crops. This high level of accuracy suggests that the model could be a game-changer for the energy sector.
Under the SSP5 8.5 scenario, Miscanthus and sorghum exhibited lower sensitivity to precipitation and temperature impacts in terms of biomass yield and soil organic carbon changes compared to switchgrass. This resilience could make them more attractive options for bioenergy production in a changing climate. “The model’s ability to predict these dynamics with such accuracy is a significant step forward,” said Gautam. “It allows us to make more informed decisions about where and how to cultivate these crops.”
The study also projected changes in economically viable land area for each crop. Sorghum and Miscanthus are expected to see increases of 29% and 10%, respectively, with the most significant gains for sorghum at higher latitudes. In contrast, switchgrass is projected to see a decline of 11% by 2100. These projections could influence investment decisions and land use policies in the energy sector.
The implications of this research are far-reaching. By providing accurate predictions of biomass yields and soil organic carbon changes, the machine learning-based model could help energy companies make more informed decisions about bioenergy crop selection and land use. It could also aid in developing user-friendly agroecosystem analysis tools, making advanced modeling more accessible to a broader range of stakeholders.
As the energy sector continues to explore sustainable bioenergy options, tools like this could play a crucial role in shaping the future of bioenergy production. The research, published in Carbon Capture Science & Technology, translates to English as ‘Carbon Capture Science and Technology’, underscores the potential of machine learning to drive innovation in the field. As Gautam puts it, “This is just the beginning. The possibilities for applying machine learning in agroecosystem modeling are vast, and we’re excited to see where this journey takes us.”