Reconstruction of Extreme Sea Levels in coastal China using Multiple Deep Learning models

Area of Science:

• Oceanography
• Coastal Engineering
• Data Science

Background:

• Accurate coastal water level data is crucial for understanding and mitigating coastal hazards.
• Existing datasets may have limitations in spatial coverage or temporal resolution, particularly in regions with sparse observations.

Purpose of the Study:

• To create a comprehensive dataset of daily maximum coastal water levels for China's coast spanning 1970-2020.
• To enable spatiotemporal analyses of extreme coastal water levels and support coastal hazard mitigation efforts.

Main Methods:

• Developed an Informer-based deep learning model to predict storm-surge residuals, benchmarked against LSTM, CNN-LSTM, and ConvLSTM.
• Combined predicted storm-surge residuals with astronomical tides estimated by UTide.
• Utilized ERA5 reanalysis data and multi-source tide-gauge records for model training and validation.

Main Results:

• The reconstructed daily maximum coastal water levels achieved a mean correlation coefficient of 0.81 and RMSE of 11.7 cm.
• For extreme events (95th percentile), the mean correlation was 0.68 with RMSE typically below 20 cm.
• A mean non-coincidence bias of 14.9 cm (14.8%) was estimated between the peaks of tide and surge.

Conclusions:

• The released dataset provides a valuable resource for coastal research and management in China.
• The deep learning approach offers a robust method for reconstructing historical coastal water levels.
• The dataset facilitates improved coastal hazard assessment and mitigation strategies, especially in data-scarce areas.