Improvement of global and regional mean sea level estimation from satellite altimetry

Abstract

The accurate measurement of the global mean sea level (GMSL) record and of the estimation of GMSL’s trend and acceleration are key goals of high precision satellite altimetry. These measurements are indeed crucial to tackle important scientific questions related to the closure of the sea level budget and to the assessment of the Earth Energy Imbalance (EEI) in the context of climate change. Quantifying precisely observational sea level uncertainties is also required because uncertainties inform on the reliability of sea level observations and prevent from misinterpretations of artifacts arising from the limitations of the observing system. Great efforts have been put in the last decade to better understand and improve the uncertainties associated to the GMSL measurements from radar altimetry, leading to the design of an error variance-covariance matrix describing the temporal correlations of the GMSL uncertainty budget at global (Ablain et al 2009, Ablain et al 2019, Guerou et al 2023) and local (Prandi et al., 2021) scales. The matrix quantifies the uncertainties in the GMSL record at different time scales, from months to decades, allowing for a robust and accurate propagation of the errors to the estimation of metrics like the trend of sea level rise and acceleration. Thanks to these efforts, the 28-year GMSL trend and acceleration uncertainties are currently down to 0.3mm/yr-1 and 0.05 mm/year-2 (90%CL), respectively (Guerou et al 2023). Yet, further improvements are still required to address three main scientific questions 1) closure of the sea level budget, 2) detection, and attribution of the signal in sea level that is forced by greenhouse gases emissions (GHG) and 3) estimate of the current EEI (Meyssignac et al 2023). Meeting such requirements will require improving the accuracy and precision of satellite altimetry data but also improving the error description and analysis. Indeed, in the current analysis an Ordinary Least Square (OLS) estimator is used which is not the most adapted when the errors show strong correlations and heterogeneous variance, as in the GMSL error variance-covariance matrix, leading to sub-optimal estimation of the trend and acceleration uncertainties. In this talk we will show that using a General Least Square (GLS) estimator allows to significantly improve the estimation of both trend and acceleration uncertainties, reducing the uncertainty by 15% and 20%, respectively, with respect to current constraints. We will also present a Bayesian based statistical analysis of GMSL data, compare the results to the ones obtained with the GLS and OLS and discuss the benefit of this approach for future analysis. Finally, we will show how these methods can be applied to improve the MSL analysis at regional scales.