Machine learning-assisted life cycle assessment of biochar soil application

Pyrolysis of waste biomass to produce biochar for soil application is receiving great attention for its potential to achieve negative carbon emissions. This study presents an environmental impact assessment framework combining machine learning modelling and life cycle assessment to evaluate the carbon footprints of biochar production from agricultural waste for soil application. Five machine learning models were compared for predicting biochar yields and properties, with multi-layer perceptron neural network and Gaussian process regression models showing excellent performance for the prediction of yield, and carbon and nitrogen contents of biochar ( R 2 = 0.97, RMSE = 3.5; R 2 = 0.92, RMSE = 3.2; R 2 = 0.94, RMSE = 0.36, respectively). The multi-layer perceptron neural network model predicted a maximum GWP saving associated condition is PT = 400 °C, HR = 15 °C/min, and RT = 40 min. The environmental impact assessment was carried out considering carbon sequestration and two fertiliser substitution scenarios. It was shown that the highest carbon saving potentials were −1323 and −1355 kg CO 2 -eq/t feedstock achieved by the scenarios of urea ammonium nitrate and calcium ammonium nitrate fertiliser substitutions, respectively. This framework is capable of simulating the influences of various operating conditions of pyrolysis towards the environmental impacts of its biochar soil application. It offers a useful tool for maximizing the environmental benefits of pyrolysis while accounting for the complex interdependencies between process parameters. The results highlight the importance of optimizing biochar production parameters while assessing the life cycle environmental impacts of biochar soil application to minimize trial and error and facilitate process up-scaling. • Five different Machine Learning models are developed to predict biochar production. • Biochar yield, carbon and nitrogen content are predicted simultaneously. • The framework combines Machine Learning modelling and Life Cycle Assessment. • Exploration of multiple process parameter combinations associated GWP, CED, and EP. • Sensitivity analysis reveals influence of different parameters effects on total GWP.