Mechanistic and Machine Learning–Guided Design of Biomimetic Biochar–Mineral Sorbents for Aquatic Anion Capture

Biochar has emerged as a versatile platform material in environmental engineering, offering significant potential in nutrient retention, water remediation, soil regeneration, and long-term carbon sequestration (Lehmann & Joseph, 2015). Its functional performance, defined by properties such as surface area, porosity, surface charge, and functional group density, is highly dependent on both feedstock characteristics and pyrolysis conditions. These properties directly influence its interactions with critical and emerging pollutants (Abbas et al., 2018). While many studies have sought to optimise these variables empirically, less attention has been given to the underlying structural drivers of sorption mechanisms, particularly how biomass cellular architecture and pyrolytic transformation govern material reactivity. The evolution of carbon structure during pyrolysis, from amorphous to turbostratic to graphitic, influences a biochar’s surface chemistry, particularly its carbon stability, defect density and retention of oxygen-containing functional groups (Petersen et al., 2023). Amorphous and turbostratic forms tend to retain more surface functionality and edge-plane defects, enhancing their potential for electrostatic attraction (when surface charge is favourable) and ligand exchange with anions (Farsad et al., 2023). These structures, especially when combined with metal oxides via co-pyrolysis impregnation, may also support surface precipitation processes by promoting local supersaturation at the solid–liquid interface. Emerging evidence suggests that feedstocks with high parenchyma content (e.g., hemp hurd) may promote the formation of reactive turbostratic carbon domains and strong electrostatically charged properties due to differential collapse and devolatilisation of tissue types (Marrot et al., 2022; Yang et al., 2022). However, this relationship between feedstock anatomy, carbon domain development, and sorption mechanism expression remains underexplored. In parallel, whilst the utilisation of biochar for a swathe of environmental management applications is increasingly emerging, the integration of machine learning and artificial intelligence remains largely underexplored. Most optimisation efforts still emphasise relationships between feedstock, pyrolysis conditions, and production technologies, with outcomes reported as yields and bulk properties (e.g., ash, BET). However, the field remains fragmented when it comes to predicting functional performance, particularly in the contexts of anion removal. A critical challenge being that datasets are commonly heterogeneous with varying protocols and key variables regarding aquatic compositional data (e.g. pH, ionic strength, competitors etc), and key biochar-mechanistic data (e.g., pHPZC, mineralogy, etc) are inconsistently reported. With controlled inputs (feedstock architecture, mineral donors, pyrolysis temperature/time) and controlled testing (pH, ionic strength, competing anions, contact time), ML can prioritise candidates and reveal which properties, such as surface charge (pHPZC), Fe/Al/Ca/Mg phases, functional groups, and microstructure (to name a few), most strongly drive performance toward new anions based upon their chemical descriptors (e.g. Charge at pH, pK?, logD/logP, headgroup etc) Accordingly, the overarching aim of this PhD proposal is to synthesise evidence on biochars and their mineral–hybrid counterparts for anion capture and to identify the minimal set of mechanistic descriptors needed to enable robust, interpretable ML. The goal is to define a reproducible testing and reporting framework that supports predictive optimisation. ML algorithms, when informed with a