Forecasting of Typhoon-Induced Wind-Wave by Using Convolutional Deep Learning on Fused Data of Remote Sensing and...

Area of Science:

• Marine Science and Engineering
• Meteorology
• Data Science and Artificial Intelligence

Background:

• Taiwan's maritime economy is vulnerable to typhoons due to its island geography and location in a typhoon-prone region.
• Typhoons pose significant threats to port operations through strong winds and large waves.
• Accurate real-time prediction of wind and wave conditions is crucial for maritime safety and economic stability.

Purpose of the Study:

• To develop and validate models for predicting wind speed and wave height near Taiwanese ports during typhoon events.
• To forecast conditions within a 1 to 6-hour horizon.
• To assess the efficacy of combined neural network approaches for typhoon impact prediction.

Main Methods:

• Developed hybrid models integrating gated recurrent unit (GRU) neural networks and convolutional neural networks (CNNs).
• Designed two wind speed prediction models (WIND-1, WIND-2) and four wave height prediction models (WAVE-1 to WAVE-4).
• Utilized ground station data, buoy observations (Longdong and Liuqiu), and radar reflectivity images for model training and validation.

Main Results:

• The WIND-2 model demonstrated superior wind speed prediction compared to WIND-1, effectively capturing spatio-temporal wind variations.
• The WAVE-4 model achieved optimal wave height prediction performance, outperforming WAVE-1, WAVE-2, and WAVE-3.
• Directly using in-situ and reflectivity data in the WAVE-4 model yielded the best wind-based wave height predictions.

Conclusions:

• Combined GRU and CNN models effectively extract spatial features and time-series information for typhoon impact prediction.
• The developed models show promising capabilities for real-time forecasting of hazardous wind and wave conditions.
• These advanced predictive models can enhance the safety and efficiency of maritime operations in typhoon-affected regions.