Abstract: The possibility of extracting the mesoscale information from large-scale observations is evaluated with numerical simulation experiments by using the shallow water model and the method of four-dimensional variational data assimilation. The initial field for the simulation is set to a circular vortex with radius of 5 times mesh scale added to a straight flow. The distances between the observation stations are 5,7 or 9 times mesh scale separately. The observation is available every 1,2 or 3 hours. The data assimilation period is defined to be 6 or 12 hours. The initial fields resulting from the assimilation are evaluated The conclusions are: When the distance between the observation stations is 5 times mesh scale, the basic properties of the vortex can not be represented by the analysis fields from the observations; The basic structure of the vortex is well defined in the initial fields from the assimilation, but the intensity is slightly weakened; The assimilation results can be improved by increasing the observation frequency or lengt hening assimilation time; The assimilation results are better for the fast moving system; When the distance between the observation stations is 9 times mesh scale, there is noobvious improvement from the assimilation results compared with analysis fields.