A recent study in the journal BiocharBiochar is a carbon-rich material created from biomass decomposition in low-oxygen conditions. It has important applications in environmental remediation, soil improvement, agriculture, carbon sequestration, energy storage, and sustainable materials, promoting efficiency and reducing waste in various contexts while addressing climate change challenges. More explores how machine learning can be used to optimize the hydrothermal treatment (HT) of biowastes, with a specific focus on swine manure. The authors, Xiaofei Ge, Tao Zhang, and their colleagues, investigated the fate of phosphorus (P) in both the solid hydrochar and the liquid phase after treatment. Their work highlights a novel application of artificial intelligence (AI) to address the environmental challenges of livestock biowaste management and promote resource recycling.

The rapid growth of intensive farming has created a significant environmental issue: managing large quantities of livestock biowastes. Swine manure is rich in valuable nutrients like carbon, nitrogen, and phosphorus, but it can also contain toxic elements that pose a risk to the environment. Hydrothermal treatment (HT) is an effective technology for processing these biowastes without pre-drying, offering high conversion efficiency. The distribution of P within the solid hydrochar and the liquid phase is influenced by factors like feedstockFeedstock refers to the raw organic material used to produce biochar. This can include a wide range of materials, such as wood chips, agricultural residues, and animal manure. More composition, pHpH is a measure of how acidic or alkaline a substance is. A pH of 7 is neutral, while lower pH values indicate acidity and higher values indicate alkalinity. Biochars are normally alkaline and can influence soil pH, often increasing it, which can be beneficial More, and processing conditions. However, traditional experimental methods are often insufficient to fully understand these complex interactions, which is where machine learning (ML) models become valuable.

The study used three different ML models—XGBoost, Decision Tree, and Random Forest—to predict the outcomes of HT. These models were trained on a dataset of 423 experiments, including 32 original experiments conducted for the study. The models predicted the amount of total P in the hydrochar (TPS) and inorganic P in the liquid phase (IPL). The research found that the XGBoost model consistently outperformed the other models, with a strong predictive accuracy for TPS (R² = 0.77) and an almost perfect fit for IPL (R² = 1.0) on the training set. This means the model’s predictions closely matched the actual experimental results, confirming its reliability.

The research also identified key factors influencing P distribution. A feature importance analysis from the XGBoost model showed that feedstock characteristics, particularly oxygen content, had the greatest overall impact on the output parameters. Operational conditions were also important, with reaction time having a more significant effect than temperature. The presence of calcium (Ca2+) and iron (Fe2+) ions also had a significant impact on P distribution, ranking second only to reaction time in importance. The study found that adding these ions promoted the transfer of P into the hydrochar, as P formed stable complexes with the added metals, significantly reducing its solubility in the liquid phase.

The research validated the model’s predictions with a case study on swine manure. The results confirmed that the XGBoost model accurately predicted the distribution of P. As the reaction severity increased, the P in the hydrochar became more uniform in its organic form, and the crystallinity of the hydrochar decreased, indicating crystal degradation under intense conditions. This confirms that precise control over parameters like temperature, time, and the addition of Ca and Fe ions can effectively regulate P distribution and improve hydrochar quality for use in soil and as fertilizer.

In conclusion, this study demonstrates that machine learning models, especially the XGBoost model, can be a powerful tool for optimizing hydrothermal treatment processes for agricultural waste. By accurately predicting how P and other elements behave under various conditions, this approach provides a scientific basis for better waste management and resource recovery. This innovative method not only offers a solution for reducing environmental pollution but also contributes to the circular economy by producing valuable resources from waste.

Source: Ge, X., Zhang, T., Mukherjee, S., Chen, Y., Wang, X., Chen, X., … & Shaheen, S. M. (2025). Optimizing the conditions of biowastes hydrothermal treatment and predicting phosphorus fate in the hydrochar and liquid phase using machine learning. Biochar, 7(96).