Physics
Scientific paper
Dec 2002
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2002agufmsa22a..04m&link_type=abstract
American Geophysical Union, Fall Meeting 2002, abstract #SA22A-04
Physics
0355 Thermosphere: Composition And Chemistry, 2447 Modeling And Forecasting, 3230 Numerical Solutions, 3337 Numerical Modeling And Data Assimilation
Scientific paper
Most data assimilation techniques, including Kalman filtering, consider errors in the model and use this information, along with estimated measurement errors, to provide a `best' state estimate. As a down side, however, these techniques assume that the model error is random and has a zero mean, which is often not the case. Typically, errors in the model are not random but change slowly with the dynamics of the system. Additionally, the actual amount of error in the model may be unknown - especially in the early design stages of the data assimilation system. This lack of knowledge about the model error decreases the accuracy of the state estimate when the data assimilation technique is applied. To demonstrate the effects this problem, two electric field models are used to provide the neutral composition in the thermosphere. A Weimer electric field is used to create a simulated `truth' data set and a Foster electric field is used in a Kalman filter to reproduce, as best as possible, the original truth data set. The vice versa application of these models can also be used since their differences are the main point to consider. The two electric fields provide two slightly different composition patterns to mimic a slowly changing, non-random error as may occur in a realistic situation. As expected, the traditional application of the Kalman filter shows a bias in the state estimate when trying to reproduce the truth set. In this research, it is proposed that the Kalman filter can be supplemented by an auxiliary method to account for the unknown error. The auxiliary method is applied based on the assumption that the errors in the measurements are better known in comparison to the model errors. The measurement errors alone are used in an auxiliary method to estimate the bias, and this calculated bias is then used to correct the model in the main Kalman filter solution. Methods for calculating the model error by the auxiliary method are varying. These methods, including neural nets, two-point boundary value problems, and minimum model error estimators, are compared. The results are evaluated by comparing the estimated state to the original simulated truth file.
Codrescu Mihail V.
Fuller-Rowell Tim J.
Minter C. F.
No associations
LandOfFree
Considering Model Error when Applying Data Assimilation to New Systems does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Considering Model Error when Applying Data Assimilation to New Systems, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Considering Model Error when Applying Data Assimilation to New Systems will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1434223