SeADL: Self-Adaptive Deep Learning for Real-Time Marine Visibility Forecasting Using Multi-Source Sensor Data.

Area of Science:

• Oceanography
• Atmospheric Science
• Artificial Intelligence

Background:

• Marine visibility prediction is crucial for safe maritime operations, especially in data-sparse and dynamic ocean environments.
• Traditional deep learning methods face limitations in marine settings due to sparse fixed stations and high spatiotemporal variability.
• Short-term visibility forecasting improvements can significantly enhance navigational safety and operational planning.

Purpose of the Study:

• To introduce SeADL, a self-adaptive deep learning framework for real-time marine visibility forecasting.
• To address the challenges of data scarcity and dynamic conditions in marine visibility prediction.
• To enhance maritime situational awareness and operational safety through improved forecasting.

Main Methods:

• Developed SeADL, a self-adaptive deep learning framework.
• Integrated multi-source time-series data from onboard sensors and drone-borne atmospheric measurements.
• Implemented a continuous online learning mechanism for real-time model parameter updates.

Main Results:

• SeADL demonstrated high prediction accuracy in marine visibility forecasting.
• The framework maintained robust performance across diverse and extreme weather conditions, including storm simulations.
• Continuous online learning enabled adaptation to both short-term weather fluctuations and long-term environmental trends.

Conclusions:

• SeADL offers a robust solution for real-time marine visibility forecasting.
• Combining self-adaptive deep learning with real-time sensor data enhances marine situational awareness.
• The framework has significant potential to improve operational safety in dynamic ocean environments.