Optimal experimental design for bio-accelerated mineral weathering
Conference
65th ISI World Statistics Congress
Format: IPS Abstract - WSC 2025
Keywords: experimental-design
Session: IPS 925 - Machine Learning improved Time Series Analysis
Monday 6 October 10:50 a.m. - 12:30 p.m. (Europe/Amsterdam)
Abstract
Silicate weathering, a natural carbon dioxide sequestration process, can be enhanced by identifying the optimal combination of its abiotic and biotic drivers. Due to the multitude of potential drivers and the limited number of experiments that can be conducted, making the experiments as informative as possible is paramount. We propose a customized methodological approach centered on designing efficient experiments within a batch-based framework. Leveraging Bayesian optimization, the project iteratively refines combinations based on insights from previous trials. Given the diverse dataset with many categorical variables, an ensemble of gradient-boosted decision trees, specifically the CatBoost algorithm, is employed to address this challenge. This ensemble approach not only estimates expected outcomes but also quantifies knowledge uncertainty associated with different combinations. A genetic algorithm is used to maximize the upper confidence bound, aiding in the identification of the most promising combinations for accelerating weathering rates. The methodology incorporates different exploration/exploitation trade-offs to determine batches of combinations to determine which combinations of potential drivers would yield the highest information gain in the next batch of experiments.