# Machine learning-based identification of key biotic and abiotic drivers of mineral weathering rate in a complex enhanced weathering experiment

## TL;DR

This study uses machine learning to identify which abiotic and biotic factors most influence mineral weathering, a process that can help remove carbon dioxide from the atmosphere.

## Contribution

The novel contribution is the use of machine learning and explainability methods to disentangle the complex interactions between abiotic and biotic drivers of mineral weathering in a large-scale experiment.

## Key findings

- Abiotic factors were consistently key drivers of mineral weathering indicators.
- Earthworms and microbes significantly influenced carbon dioxide removal metrics.
- Machine learning models successfully predicted and explained the impact of multiple drivers on weathering rates.

## Abstract

The optimization of enhanced mineral weathering as a carbon dioxide removal technology requires a comprehensive understanding of what drives mineral weathering. These drivers can be abiotic and biotic and can interact with each other. Therefore, in this study, an extensive 8-week column experiment was set up to investigate 29 potential drivers of mineral weathering simultaneously.

The setup included various combinations of mineral types and surface areas, irrigation settings, biochar and organic amendments, along with various biota and biotic products such as earthworms, fungi, bacteria and enzymes; each varying in type or species and quantity. The resulting changes in dissolved, solid, and total inorganic carbon (∆TIC), and total alkalinity were calculated as indicators of carbon dioxide removal through mineral weathering. Three machine learning models, Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest and eXtreme Gradient Boosting (XGB) regression, were used to predict these indicators. Dominant drivers of the best performing model were investigated using SHapley Additive exPlanations (SHAP).

SHAP analysis revealed that each CDR indicator was influenced by different factors. However, key drivers were consistently abiotic, though biota also made a significant contribution to the predictions. The most representative CDR indicator, ∆TIC, was predominantly driven by steel slag addition and mixed mineral grain sizes but was also substantially impacted by earthworms and microbes.

These findings provide valuable insights into the complex interplay of numerous abiotic and biotic factors that affect mineral weathering, highlighting the potential of machine learning to unravel complex relationships in biogeochemical systems.

The increasing concentration of carbon dioxide (CO
2) in the atmosphere, driven by human activities, is the primary cause of global warming. To limit global temperature rise to below 2°C, CO
2 emissions need to decline to net zero, and excess CO
2 must be removed using carbon dioxide removal (CDR) technologies. Many existing carbon capture technologies are energy-intensive and compete for resources such as land, making them challenging to scale. As a result, alternative methods like enhanced weathering are gaining attention due to their potential for CO
2 removal without these drawbacks.

Natural mineral weathering is a process where carbonic acid, formed from CO
2 and water, reacts with minerals, breaking them down and capturing carbon in the form of dissolved bicarbonate ions or carbonate minerals. This natural process slowly reduces atmospheric CO
2 over long time scales. The process can be enhanced, amongst others by grinding minerals to increase their surface area, which leads to an increase in CO
2 removal, but potentially also yields co-benefits, such as improving soil health. Besides grinding minerals, biota could possibly further enhance weathering, for example by producing organic acids, enzymes and siderophores, or by physically mixing soil and rocks. However, the impact of these biota on mineral weathering and the interactions between multiple drivers of mineral weathering are complex and not yet fully understood.

To explore these drivers, an extensive column experiment was conducted with 200 columns per round, running across 10 rounds. Each column contained unique combinations of minerals, organic amendments, and biota, including bacteria, fungi and earthworms, and was irrigated with different flow rates and frequencies. The impact on CDR was assessed using indicators such as changes in dissolved inorganic carbon, solid inorganic carbon, total inorganic carbon, and total alkalinity.

Given the complexity of the dataset, machine learning (ML) was used to predict the CDR indicators. To address the "black box" nature of the used ML models, they were combined with explainability methods, which identified the most influential drivers of mineral weathering. These drivers were mainly abiotic, but several biota also had a substantial impact on the predictions.

This study demonstrates how combining large-scale experiments with ML and explainability techniques can provide valuable insights into optimizing enhanced weathering strategies for CDR.