Impact of observation density in data assimilation: A study with simulated observations

The potential of high-density observations is studied in a practical context of the 4DVAR assimilation. A series of observing system simulation experiments (OSSEs) are carried out. Observations with both uncorrelated and correlated observation errors are simulated in sensitive areas. The results show that: for the observations with uncorrelated error, increasing the observation density generally improves the analysis and the forecast; for the observations whose error is correlated and by using a sub-optimal scheme (i.e., no modelling of the error correlation), the assimilation system can still extract useful information and one can determine an observation density leading to a minimum error of analysis and forecast. A risk of using horizontal high-density observations is that it could produce unrealistic increments and degrade the analysis on the levels without observations in the case of inappropriate background error correlations.