Artificial intelligence convolutional neural networks map giant kelp forests from satellite imagery

Area of Science:

• Marine ecology
• Remote sensing
• Artificial intelligence

Background:

• Climate change is altering marine species distribution, impacting vital kelp forest ecosystems.
• Accurate, long-term kelp forest mapping is crucial for understanding ecosystem dynamics and predicting climate change responses.
• Traditional mapping methods are labor-intensive and costly, hindering efficient ecological monitoring.

Purpose of the Study:

• To develop and validate an automated system for mapping kelp forest canopy cover using Mask R-CNN and satellite imagery.
• To create a cost-efficient tool for long-term marine ecological monitoring and biodiversity conservation.
• To reconstruct a 32-year time series of kelp forest distribution in Baja California.

Main Methods:

• Utilized Mask R-CNN deep learning model for automated kelp forest detection.
• Integrated open-source Landsat Thematic Mapper satellite imagery for data assimilation.
• Focused on giant kelp (Macrocystis pyrifera) along California and Baja California coastlines.
• Optimized model hyperparameters through cross-validation, including data augmentation and learning rates.

Main Results:

• The Mask R-CNN model achieved high performance in detecting kelp forests (Jaccard's index: 0.87, Dice index: 0.93) with minimal overprediction (0.06).
• Successfully reconstructed a 32-year time series of kelp forest distribution in Baja California, capturing high variability.
• Demonstrated the framework's cost-efficiency and effectiveness for large-scale, long-term ecological monitoring.

Conclusions:

• Automated kelp forest mapping using Mask R-CNN offers a significant advancement over traditional methods.
• This cost-efficient tool facilitates enhanced marine biodiversity conservation and management decisions.
• The developed framework provides vital data for understanding and predicting kelp forest responses to climate change.