Data assimilation setting
The experiment setting is similar to Twin Experiment 3, but with three more different ensemble sizes: 16, 32 and 64 members.
Computational time are 127, 290, and 475 hours respectively (i.e. about 1, 2.3, and 4 hour per 3 day simulation or 2.4, 5.6 and 9.1 hours per 7 day simulation).
Results
At assimilation station
At a validation station
Over all locations
Discussions
Generally, it is expected that the accuracy of an EnKF increases with ensemble size. In this experiment, however, there is no clear indication about the trend. It is not always true that EnKF with a larger ensemble size is more accurate than EnKF with a smaller ensemble. It is likely due sample variation; here for each ensemble size the EnKF is run only once. To reduce the effect of sample variation, it is necessary to run the experiments a number of times and to average the results out.
