A Deep Learning-Based Method for Non-Destructive Estimation of Carbonate Carbon Storage in Biogenic Shells on Marine...

Area of Science:

• Marine biology
• Biogeochemistry
• Artificial intelligence

Background:

• Hard-shelled marine organisms sequester significant inorganic carbon in their shells.
• Current quantification methods involve destructive sampling, limiting longitudinal studies.

Purpose of the Study:

• To develop a non-destructive deep learning framework for estimating shell carbonate carbon storage.
• To enable long-term monitoring of carbon sequestration by marine biofouling communities.

Main Methods:

• Deployment of diverse surface material panels in nearshore waters for five months.
• Utilized an improved Mask R-CNN for identifying and measuring barnacles and bivalves from in situ images.
• Integrated image-derived dimensions with allometric models and measured carbonate fractions for carbon storage estimation.

Main Results:

• Achieved high model performance (recall/precision: 0.86/0.89) in complex nearshore environments.
• Image-derived shell dimensions showed strong agreement with manual measurements (R² = 0.95).
• Panel-scale carbon estimation errors were consistently below 15%.

Conclusions:

• The developed framework provides a non-destructive, quantitative method for assessing shell carbonate carbon storage.
• This approach facilitates comparative studies across different materials and environmental conditions.
• Enables robust long-term monitoring of carbon sequestration in marine biofouling communities.