- Home
- Scientific program
- Proceedings
- List of participants
- Photos
- Venue/Lodging
- Links
- Contact

Robert Heinkelmann
Type of contribution: poster

CONT14 analyzed by a Kalman filter: a test case

Maria Karbon, GeoForschungsZentrum Potsdam Benedikt Soja, GeoForschungsZentrum Potsdam Tobias Nilsson, GeoForschungsZentrum Potsdam Robert Heinkelmann, GeoForschungsZentrum Potsdam Harald Schuh, GeoForschungsZentrum Potsdam

n this work we investigate the performance of a Kalman filter within very long baseline interferometry (VLBI) data analysis. Our test case is the continuous VLBI campaign CONT14, observed from May 6th until May 20th 2014. We compare daily and hourly Earth orientation parameter (EOP) estimated by using the least squares method (LSM) with the results provided by various specifications of the Kalman filter. For external validation, hourly GNSS data are used. Our filter offers two main processing modes, one for near real-time processing, where the filter is run only forwards, and one for post-processing, where the data is filtered twice, forward and backward, followed by a smoothing algorithm. The latter is more precise than the first, since all the information contained in the data is used to improve the prediction.The drawback is the longer processing time, making it inapplicable for near real-time applications. Additionally, the filter includes a forecast tool, where EOP are predicted on the basis of angular momentum functions (AMF) of the atmosphere and the oceans. This tool not only forecasts EOP but also helps bridging gaps in the VLBI data and thus stabilizes the results. We assess the performance of the filter settings concerning internal precision (i.e., relative to each other and to the LSM solution), external accuracy (e.g., GNSS), and near real-time applicability (processing speed, memory consumption, and availability of the necessary external data like AMF).The performance of an adaptive Kalman filter will be tested as well because various authors suggest its use especially for EOP determination. So far our tests have not shown significant impact on the results.