Machine learning-enhanced biomass forecasting for spatial bioenergy planning under uncertainty

Bioenergy is a key energy source for achieving climate targets and enabling a low-carbon economy, making long-term biomass availability assessment essential. This study develops a spatially explicit biomass forecasting framework to assess future biomass supply and bioenergy potential for 2030, 2040, and 2050. The model integrates observed data for 24 biomass feedstocks across agricultural, forestry, and marine sectors in Ireland as a case study with a hybrid modelling approach combining machine-learning growth estimation, economic mobilization modelling, and Monte-Carlo uncertainty analysis. National growth multipliers are estimated using Gradient Boosting regression and spatially distributed using county-level shares, while economic and trade indicators are incorporated through feedstock mobilization factors under low, baseline, and high scenarios. Results indicate that agricultural biomass remains dominant: Perennial pasture grass and cattle slurry are identified as the most robust and predictable feedstocks, offering high reliability for long-term planning. Spatial analysis reveals strong regional concentration of biomass, with counties such as Cork, Tipperary, and Galway emerging as major supply hubs. In these regions, the projected energy potential significantly exceeds local residential demand, identifying them as ideal candidates for regional biogas clusters and centralized processing facilities. In the 2050 Baseline scenario, national biomass availability reaches approximately 48.7 million tonnes from agriculture, 5.36 million tonnes from forestry, and 16.8 thousand tonnes from marine biomass resources. Furthermore, uncertainty analysis indicates that while core feedstocks maintain tight prediction intervals, emerging resources exhibit higher variability, requiring risk-informed strategies. The findings of this study can support better understanding of spatial biomass availability, regional bioenergy planning, feedstock prioritization, and policy design. The framework is also transferable to other countries and regions facing similar data limitations and long-term planning needs. To support transparency and reproducibility, the full dataset, model outputs (including observed and predicted values), and source code are made publicly available.