Leveraging synthetic data to improve regional sea level predictions

Area of Science:

• Climate Science
• Geophysics
• Artificial Intelligence

Background:

• Rising sea levels due to climate change pose significant threats to global coastal communities.
• Existing sea level prediction models often overlook developing regions lacking extensive tide gauge data.
• Satellite altimetry data offers a more globally available resource for sea level monitoring.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for enhanced regional sea level predictions.
• To address the data scarcity issue in developing countries by leveraging synthetic data generation.
• To improve the accuracy of sea level forecasts using satellite altimetry data.

Main Methods:

• A hybrid deep learning model combining Times Generative Adversarial Network (TimesGAN) with Convolutional Long Short-Term Memory (ConvLSTM) was developed.
• TimesGAN was employed to generate synthetic training data, augmenting sparse real-world datasets.
• The enhanced ConvLSTM model was trained and validated using satellite altimetry data across diverse global regions.

Main Results:

• The integrated TimesGAN-ConvLSTM model significantly improved sea level prediction accuracy across all tested regions.
• Average Mean Squared Error (MSE) reductions ranged from 64.5% (Lisbon) to 85.1% (Somalia).
• The method demonstrated particular effectiveness in improving predictions for developing regions with limited tide gauge data.

Conclusions:

• Synthetic data generation using TimesGAN is a viable strategy to enhance deep learning-based sea level prediction models.
• The proposed approach effectively bridges the data gap in under-resourced regions, improving climate change adaptation capabilities.
• This study highlights the potential of advanced AI techniques to provide more reliable sea level rise forecasts globally.