Modeling regional mean sea level based on climate measurements using a stacked ensemble approach

Area of Science:

• Environmental Science
• Climate Science
• Data Science

Background:

• Assessing mean sea level (MSL) changes is critical due to climate change impacts.
• Soft computing offers efficient alternatives to traditional MSL estimation methods.
• Limited research exists on applying soft computing to analyze climate change effects on MSL.

Purpose of the Study:

• To develop and compare soft computing techniques for modeling MSL fluctuations.
• To utilize meteorological data for predicting MSL changes.
• To assess model effectiveness at Damietta station, Egypt.

Main Methods:

• Employed Random Forest (RF), Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Deep Neural Network (DNN), Gaussian Process Regression (GPR), and stacked ensemble methods.
• Utilized environmental variables including surface water temperature, air temperature, humidity, and wind attributes.
• Statistically assessed model performance using correlation coefficient (R) and normalized root mean square error (RMSE).

Main Results:

• RF, KNN, and GPR models demonstrated superior performance in MSL modeling during training and testing.
• A weighted stacked ensemble model integrating RF, KNN, and GPR achieved a correlation coefficient (R) of 0.88 and RMSE of 0.056 m.
• MSL modeling sensitivity was highest for water temperature, wind speed/direction, and atmospheric pressure.

Conclusions:

• The developed soft computing models provide a robust framework for MSL forecasting.
• This methodology is valuable for coastal regions with limited tide records, supporting coastal resilience.
• The study contributes to UNESCO's Ocean Decade Challenge 5 by enhancing coastal adaptation strategies.