Surface observations cut errors in a Southwest Atlantic ocean model
A peer-reviewed simulation study adapted the RODAS observation-assimilation system to a regional ocean model. Surface-temperature error fell in four years of hindcasts, and heat-content estimates also improved. Deeper waters showed mixed results, including poorer temperature estimates in parts of the thermocline. Researchers tested one regional configuration and propose adding further kinds of observations.
Science··Morning
Observations are added to the regional simulation
A peer-reviewed computational study adapted the RODAS data-assimilation system to the Regional Ocean Modeling System, known as ROMS, in the Southwest Atlantic. Data assimilation combines observations with model output to adjust the simulated ocean state. The system had originally been developed for a different ocean model. Researchers compared four years of hindcasts with observations and with a run without assimilation. The work tests an economical regional modelling approach rather than an intervention in the ocean.[1]
Surface error falls against several baselines
The root-mean-square deviation in sea-surface temperature fell by 70 per cent relative to the run without assimilation. Errors were 55 per cent lower than in 72-hour hindcasts started from earlier observation updates. Against the GLORYS ocean reanalysis, the reduction reached 64 per cent. These comparisons use different baselines. For ocean heat content, the seasonal decrease in mean absolute error reached 15 per cent in summer and 25 per cent in winter. The results describe this regional experiment and its particular evaluation measures.[1]
Deeper temperatures remain harder to constrain
Improvements did not extend uniformly down the water column. Argo floats provided temperature profiles for checking deeper waters. Some results worsened in the thermocline, the layer marked by a steep vertical temperature gradient. The surface observations therefore did not resolve the problem of estimating temperature and salinity at depth. The authors propose adding Argo profiles, sea-level anomalies and further satellite observations. Those additions were not demonstrated here. Model output is shared, while public release of the code remains a planned step.[1]
