Performance evaluation of machine learning methods in determining sea level trends based on tide gauge and satellite altimetry data: A case study of Australian coasts

Abstract

This research focused on the performance evaluation of machine learning techniques employed in identifying sea level changes from 1993 to 2023 utilizing tide gauge and satellite altimetry data from 43 stations along the Australian coastline.In that respect, in addition to classical linear regression, machine learning methods such as Random Forest (RF), Decision Tree (DT), Support Vector Machines (SVM), and Gaussian Process Regression (GPR) were applied, while the models were analyzed based on the R2, MAE, and RMSE criteria. DT explained at least 76% of the variance in tide gauge data and 70% in satellite altimetry data, thus giving the best results with lower error metrics, MAE and RMSE compared to other approaches. Regional sea level trends were estimated based on the best performing approaches in the range of 3.55-4.06 mm/yr for the tide gauge data and 2.90-3.19 mm/yr for satellite altimetry data. These findings demonstrate that machine learning techniques, particularly the DT algorithm, offer significant advantages in modeling sea level trends compared to traditional methods. The results provide valuable insights for long-term coastal management and for understanding and developing strategies to address the impacts of sea level rise on communities and ecosystems.