A comparison of observed and modelled relative sea-level change over the satellite observational period 1993–2013

Abstract

During the past two decades (1993–present) relative sea-level changes (RSL) have been monitored by satellite altimetry techniques. The European Space Agency Climate Change Initiative (ESA CCI) project on “Sea Level” has produced an improved set of sea-level products by applying a number of corrections—reduction of orbit errors, wet/dry atmospheric correction errors, reduction of instrumental drifts and bias, intercalibration biases, intercalibration between missions, combination of missions, and an improved reference mean sea-surface—into multi-mission (ERS-1&2, Envisat, TOPEX/Poseidon, Jason-1&2, and Geosat Follow-on) altimetry data over the period 1993–2013. We compared those data to RSL changes calculated by separation of mass (land-ice, terrestrial water storage) and volume (steric) changes of water, including the fingerprint due to mass changes. Based on CMIP5 (Coupled Model Intercomparison Project Phase 5) data and independent data sets for groundwater, reservoir storage, and glacial isostatic adjustment (GIA), local RSL changes can be calculated by summation of all the individual contributions at every grid point—taking into account gravitational and rotational aspects as well—over the satellite altimetry period at 1o×1o resolution with a land-ocean mask extending from ± 68.5o in latitude. Over the period 1993–2013, and after correcting for GIA, the observed global mean sea-level (GMSL) change is 3.2 ± 0.5 mm/yr and the modelled GMSL change from the CMIP5 multi-model ensemble is 3.1 ± 0.5 mm/yr, a result which suggests that on a global scale observations and model results are in good agreement within the ± 1σ uncertainty. However, differences among individual models and between observations and models are larger on a regional scale (in some locations > 200%), particularly in the eastern equatorial Pacific. For some other regions, like the Indian and South Atlantic Ocean, the agreement between models and observations is better (ranging within ±50%). The global RMS error of the ensemble of 8 CMIP5 models which contain all necessary fields to calculate RSL is 2.3 mm/yr, which is 73% of the observed global mean. The RMS error of the individual models ranges from 1.0–5.9 mm/yr. We conclude that while models perform well on a global scale against observations, they do not perform as well on a regional scale over the period 1993–2013.