Automated system for near-real time modelling and prediction of altimeter-derived sea level anomalies

• New system and service for near-real time prediction of sea level was developed.
• The system is based on automated gridwise modelling using time series methods.
• Every day, sea level anomaly predictions up to 14 days in the future are computed.
• Along with prognoses, errors of previous model outputs are provided.
• Inaccuracies of forecasts correspond to areas where main mesoscale eddies occur.

This paper serves as a presentation of a novel geoinformation system and a dedicated service, jointly named as Prognocean and based at the University of Wrocław (Poland), that aim to predict Sea Level Anomaly (SLA) maps and publish them online. The system works in near-real time and is updated daily. The data are provided by the Archiving, Validation and Interpretation of Satellite Oceanographic data (AVISO), and the time series processed by Prognocean is delivered by various altimetric satellites. The emphasis is put on gridded SLA maps, also known as MSLA, which are provided as Delayed Time (DT) and Near-Real Time (NRT) daily products. The daily sampling interval, however, does not coincide with typical repeat cycles of altimetric satellites and is obtained through reprocessing produced by AVISO. The two-module infrastructure forms the system. The first module is responsible for the near-real time communication with AVISO to download the most recent MSLA data and acquire the corrected data when the geophysical corrections have been available. The second module forms the main engine which does data processing, modelling, forecasting, statistical quality control and finally generates products as maps. The online service, however, publishes the products online every day. The above-mentioned components and infrastructure are described in detail. The performance of the system was evaluated using at least 150 predicted MSLA maps, available after half year of computations carried out in near-real time. We identified a few regions of imperfect performance of our prognoses and found that they spatially correspond to the mouth of the Amazon River and locations of key mesoscale eddies, the vast majority of which being nonlinear and hence unmodelled in our experiment.