The experiment setting is similar to Twin Experiment 3, but with two extra algorithms: Dud-EnKF and EnKF-GS. Like in Twin Experiment 3, the noise parameters used in the filters are identical with the ones used in generating the truth. Hence, the filters should provide the most accurate results as possibly allowed by the ensemble size (8 members).
The Dud-EnKF and EnKF-GS took 76 and 69 hours respectively to finish the simulation over the period of 2015 (EnKF took about 59 hours; on v-wcp128 EnKF took 74 hours).
Results
At assimilation station
RMSE over 51 analyses
At all stations:
Discussion
It should be noted here that at the moment the smoothing of the Kalman gain takes place in OpenDA only when the simulation covers the previous analysis time. The results of EnKF-GS shown above are obtained by running the simulation over 1 year splitted into 3 days periods. Hence, no smoothing has actually taken place. The result can be seen simply as another result of the EnKF.
