Combination of long time series of tropospheric parameters observed by VLBI R. Heinkelmann, J. Boehm, H. Schuh Institute of Geodesy and Geophysics, TU Vienna 4th IVS General Meeting 2006, Concepción, Chile
Introduction 24th IVS General Meeting Aim of this study Combined long time series of tropospheric parameters for the assessment of climatological trends Approach Direct combination on result level with scaling factors for the individual AC solutions Estimation of linear trends using common arrays scaled by variance components (VC)
BKG GSFC IAA IGG MAO Data description 3 Long time series of tropospheric parameters from Analysis Centers (ACs) 4th IVS General Meeting
Data description 4 Different parameter models Least-squares (LS) estimates of piecewise linear function model (PWLF) Kalman forward running filter (FRF) estimates of random walk stochastic model Square root information filter (SRIF) (forward + backward) estimates of random walk stochastic model * BKG GSFC IGG IAA MAO AC Model * Bierman G.J th IVS General Meeting
Data description 5 Different constraints on the parameters Rates of piecewise linear function are constrained using different weights A-priori values for the estimated parameters and the covariance matrices, dynamic model BKG GSFC IGG IAA MAO n.a. loose constraint [mm/h] AC Model 4th IVS General Meeting
Data description 6 Different epoch and interpolation Linear interpolation of parameters and standard deviations integer hours Linear integer hours, stdv are from tropospheric offset Average of 1 hour interval time reference, stdv forwarded by error propagation BKG GSFC IGG IAA MAO AC Model 4th IVS General Meeting
Data description 7 Different solution strategies NMF VMF NMF BKG GSFC IGG IAA MAO AC mf cutoff 5° 3° 5° 0° w.r.t. TRF VTRF2003 ITRF2000 VTRF2003 ITRF2000 datum NNT/NNR fixed coord. NNT/NNR 4th IVS General Meeting
Data description 8 Summary different functional and stochastic models different a-priori information / constraints different analysis options, geodetic datum different relation of time of reference, interpolations and stdv treatment Common ground Results characterize the same physical phenomenon Averaging analysts’ noise 4th IVS General Meeting
Combination on result level 9 Strategy Independent analysis of each station and each parameter (wet, hydrostatic zenith delay, gradients) Elimination of outliers of individual time series Determination of linear trends using weight factors obtained by variance component (VC) estimation a.with a-priori variances b.without a-priori variances 4th IVS General Meeting
Elimination of outliers 10 Strategy Decomposition of time series by frequency analysis until residuals follow a white noise process, i.e. normal distribution Detection of outliers w.r.t. the functional model using the BIBER algorithm Minimal modification of observations to fulfill normal distribution 4th IVS General Meeting
Decomposition of time series 11 Characteristics: Begin: 1993/04/21 End: 2004/12/ data points Irregular sampling Big data gap Begin: 1997/05/27 End: 1998/02/12 4th IVS General Meeting Example: Fortaleza, Brazil, IGG - wet zenith delays
: systematic part: offset, trend, seasonal component : vector of observations : vector of residuals: 14 Gauss-Markov model * Functional model: p1, p2 annual and semiannual periods Functional model of outlier elimination * Koch K.R th IVS General Meeting
Characteristics of BIBER outlier elimination: Only one modification per iteration step Correlations are considered Observations are minimally modified 1)compute 2)if 3)where 4)modify observation 15 BIBER algorithm * * Wicki F th IVS General Meeting
16 Outlier cleaned time series # observations linear trend modified observations IGG: mm/year BKG: mm/year GSFC: mm/year IAA: mm/year MAO: mm/year 4th IVS General Meeting Example: Fortaleza, Brazil - wet zenith delays
Method: Global best invariant quadratic unbiased estimation (global BIQUE) * applied iteratively ** Minimal computational costs At convergence point independent of approximate values i 17 Gauss Markov model with unknown VC Variance component estimation * Förstner W ** Koch K.R th IVS General Meeting
18 Relative variance components VC considering a-priori stdvneglecting a-priori stdv VC strongly depend on a-priori variance information Example: Fortaleza, Brazil 4th IVS General Meeting
19 Linear trend of common data Example: Fortaleza, Brazil ALL: without VC linear trend: mm/year VC neglecting a-priori stdv mm/year VC considering a-priori stdv mm/year 4th IVS General Meeting
20 Conclusions Seasonal signal - must be included in functional model of both outlier elimination and trend determination, trend and sin/cos functions are not orthogonal A-priori variance information - significantly influence the variance component estimation - stdv from different stochastic models have different level Linear trends of tropospheric parameters - strongly depend on models, analysis options, combination strategy - from combined time series average the Analyst noise 4th IVS General Meeting
21 Outlook: Combination on NEQ level Within VLBI: One model for tropospheric parameters Same constraints Same geophysical models and and analysis options Homogeneous meteorological input data Tropospheric parameters estimated at epoch, i.e. no interpolation Output: SINEX files including tropospheric parameters With other space geodetic techniques Local ties Same meteorological data, models, and height reference Observations at same epoch Output: SINEX files including tropospheric parameters 4th IVS General Meeting
contacts: R. Heinkelmann Thank you for your attention end Acknowledgements: All IVS ACs which contribute to this study are greatly acknowledged. project 16992
Reference Bierman G.J Factorization Methods for Discrete Sequential Estimation, Mathematics in Science and Engineering 128, edited by R. Bellman Foster G Wavelets for period analysis of unevenly sampled time series, The Astronomical Journal 112 (4), Förstner W Ein Verfahren zur Schätzung von Varianz- und Kovarianzkomponenten, AVN 11-12, Koch K.R Parameterschätzung und Hypothesentests, 3rd edition, Dümmler, Bonn Lomb N.R Least-Squares Frequency Analysis of unequally spaced data, Astrophysics and Space Science 39, Roberts D.H. et al Time series analysis with CLEAN. I. Derivation of a spectrum, The Astronomical Journal 93 (4), th IVS General Meeting