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, derived from biomassBiomass is a complex biological organic or non-organic solid product derived from living or recently living organism and available naturally. Various types of wastes such as animal manure, waste paper, sludge and many industrial wastes are also treated as biomass because like natural biomass these More, is gaining attention as a sustainable alternative to fossil fuels, but optimizing its production to minimize environmental risks remains a challenge. In the journal Biochar, Zhang et al., present a novel machine learning approach using a ResNet-based autoencoder to predict biochar yield and composition more accurately.

The newly developed model addresses the common problem of data uncertainty in training datasets. By incorporating previously discarded data with high missing rates and including three newly collected covariates, the model’s predictive performance was enhanced. The model achieved a mean R² of 0.985, outperforming other methods such as MLP-NN (mean R² = 0.907), Random Forest (mean R² = 0.798), XGBoost (mean R² = 0.826), and Gaussian Process (mean R² = 0.786). A significant advancement of this model is its ability to handle data uncertainty, a prevalent issue in biochar research. The model effectively manages missing values and deviations from true values, common in multi-source datasets with mismatched variables or experimental environments. This robustness was demonstrated through robust sensitivity analyses of the input covariates.

The model distills informative features using a convolutional neural network, enhancing the precision of predictions for multiple targets. This research provides a reliable method for determining optimal biochar production conditions, which is crucial for maximizing biochar yield and reducing toxic effects. The model’s efficiency is highlighted by its computational speed, taking only 5 minutes to train with 250 epochs on an NVIDIA RTX A6000 GPU and 0.15 seconds for inference on the testing set.

This demonstrates the feasibility of the method for industrial applications. The study not only advances the application of machine learning in biochar research but also offers valuable guidance for optimizing biochar treatment conditions, thereby expanding its application and potential value.  

SOURCE: Zhang, Y., Lei, B., Mahdaviarab, A., Wang, X., & Liu, Z. (2025). Robust biochar yield and composition prediction via uncertainty-aware ResNet-based autoencoder. Biochar, 7(61). https://doi.org/10.1007/s42773-025-00446-2   Sources and related content