Deep Sankar Banerjee, Indian National Centre for Ocean Information Services

A simpler approach of evaluating the Representative Error for Data Assimilation by regriding a high resolution ocean data into a coarser resolution grid structure


The purpose of this investigation is to find out the Representative Error (RE) which is incorporated due to regridding a high resolution ocean data into a coarser resolution grid structure. RE is the component of observational error that creeps into model due to the subgrid scale unresolved processes. For instance, if a large-scale circulation model is considered it does not incorporate the subgrid scale processes which causes the inclusion of the RE. It can be included in Data Assimilation by incorporating an estimate of the variance of RE and blending it with the observational error covariance matrix which will be used in further calculation regarding the assimilation of observational data. So it is very important to evaluate the RE with which one can get a model forecast that will be closer to the reality. In order to evaluate it in a simpler way a climatological output of ROMS (Regional Ocean Modeling Software) of 1/12 degree resolution is assumed as the best possible model estimate. Our goal is to find out how much physical information is lost due to regridding it into a coarser resolution. In this regard an arithmetic mean is produced from the given ROMS output for a single time step which is known as the 'redefined truth'. Then every grid points from the given high resolution data are subtracted from the redefined truth and the variance is calculated at each grid point. This experiment is done by taking the potential temperature of ocean and the sea water salinity as variables in the Arabian Sea, Bay of Bengal and Indian Ocean. It is found that in the vicinity of persian gulf and red sea within the range of thermocline depth the variance, rather the RE is a bit higher which can be anticipated by the fact that in these regions a certain number of mesoscale eddies exists which is underestimated after regridding the data into coarser resolution.