Separablity of deformations and measurement noises of GPS time series with modified Kalman filter for landslide monitoring in real-time

The separation of the deformations and measurement noise of GPS coordinate time series and accuracy improvement of GPS real-time coordinates are major aspects of the thesis. In order to reduce the influence of the colored noise in the GPS position time series, three different methods have been compared: the Finite Impulse Response (FIR) filter, the Kalman filter model, and the sequential algorithm. Among these three methods, the Kalman filter is investigated in detail. The GPS real-time series contains the colored noise, yet the Kalman filter model requires white noise. The state vector can be augmented by appending to the state vector components of the shaping filter which can describe the long term movement of the colored noise. Thus the deformation analysis based on the Kalman filter model with a shaping filter technique, has been applied in the different movement trends of GPS real-time series. From the results, the Kalman filter model with a shaping filter can be widely used to process the GPS short baseline time series in real-time. The precise position coordinate can be obtained and the deformation epoch can be detected in time and with high reliability. It can be applied in the early warning system of the natural hazards. The detection of a deformation with less time delay and the improvement of reliability of detecting deformation epoch is another key issue of the investigation. The proposed model makes use of the statistical criterion (MDL criterion) comparison instead of the hypothesis test. Considering the affection of colored noise in the GPS time series the multiple Kalman filters model was augmented by shaping filters which describe the long-term movement of the colored noise. By the GPS experiments, it has been verified that the proposed models have the ability to better capture the deformation epoch and to improve the reliability of detecting the deformation epoch. The proposed models can be used to detect stepwise changes of a variety of fields in real-time or near real-time.