Application of the Multi-Adaptive Regression Splines to Integrate Sea Level Data from Altimetry and Tide Gauges for Monitoring Extreme Sea Level Events

Abstract

This article determines sea level fields with nonlinear components along the northern coast of Australia using a state-of-the-art approach of the Multi-Adaptive Regression Splines (MARS). The 20 years of data from multi-missions of satellite altimetry (e.g., Topex, Jason-1, Jason-2) and 14 tide gauges are combined to provide a consistent view of sea levels. The MARS is chosen because it is capable of dividing measured sea levels into distinct time intervals where different linear relationships can be identified, and of weighting individual tide gauge according to the importance of their contributions to predicted sea levels. In the study area, the mean R-squared (R2) of 0.62 and Root Mean Squared (RMS) error of 6.73 cm are obtained from modelling sea levels by MARS. The comparison of the MARS with the multiple-regression shows an improved sea level prediction, as MARS can explain 62% of sea level variance while multiple-regression only accounts for 44% of variance. The predicted sea levels during six tropical cyclones are validated against sea level observations at three independent tide-gauge sites. The comparison results show a strong correlation (˜99%) between modelled and observed sea levels, suggesting that the MARS can be used for efficiently monitoring sea level extremes.