| In this contribution,several key techniques of the Terrestrial Reference Frame(TRF)establishment and maintenance using multi-source space geodetic data are studied in detail,such as datum constraints and their influence,fusion modes,combination models of multi-source data,data processing methods in regional reference frames,horizontal velocity field establishing,coordinate transformation between different reference frames,and so on.A complete set of multi-source data processing models and algorithms for the TRF is investigated.The main contents and innovations of this thesis are summarized as follows:1.Adaptive parameter estimation and inner and external precision are discussed.Adaptive adjustment of stochastic model includes three types: inner precision-based adaptation,external accuracy-based adaptation and semi-accuracy-based adaptation.The statistics correspond to the adaptive estimation developed these years are analyzed respectively at first.It is shown by analysis that the existing adaptive parameter estimations are related to the inner precision,external accuracy and semi-accuracy.Thus the corresponding adaptive estimations can be divided into inner precision-based,external accuracy-based and semi-accuracy-based adaptive estimation,respectively.The characteristics of the different adaptive estimations are analyzed.Some examples are performed to analyze their main properties.Finally,the premise of applications and some existed problems are pointed out.2.For TRF definition,different constraint methods and their effects on datum definition are analyzed and compared.The similarity transformation constraint is established based on the assumption that the coordinate transformation parameters are zero between origin frame and target frame,and the inner constraint of translations,rotations and scale for geodetic control networks is deduced,respectively.It is theoretically proved that the network barycenter datum constraint is a special minimum constraint(MC).The optimal MC is chosen based on the posteriori precision of estimated parameters.Under the MC,the influence of data noise,datum noise and abnormal errors on the reference frame establishing is analyzed.It shows that the datum noise does not affect the shape of the network,but it degrades the posteriori precision of the estimated coordinates.3.The parameter estimator and its posteriori precision for dual functional and stochastic information constraints are presented in the frame of least squares principle.Three special examples,which are the parameter estimators with only functional model constraints,stochastic model constraints or without any constraints,are derived respectively.The properties of the estimator with dual constraints are discussed theoretically.It is shown that any error in the functional model constraints will result in compulsive twist in estimated parameters which is called “hard twist”.On the other hand,the errors of the stochastic model constraints will also result in the bias of parameter estimation,which is called “soft bias”.However,the effects of errors in stochastic models can be reduced by posteriori estimation.4.Based on the model of coordinate transformation and time-variable reference frame,the combination model for establishing secular reference frame is deduced.The combination model of coordinate time series in inter-and intra-technique are proposed,and two different approaches which are two-step and one-step fusion mode are also derived.It is shown that when the input covariance matrices are the normal equation coefficient matrices,the two approaches are equivalent.A long-term solution has been established by combining the IGS repro2 daily solution from the year 2010.0 to 2015.0,the WRMS is less than 3 mm for X,Y and Z component,respectively.Compared with ITRF2014,for position,the standard derivations(std)are 3.45,4.04 and 2.84 mm for X,Y and Z component,respectively.For velocity,the std are 1.53,1.46 and 1.21 mm/a for X,Y and Z component,respectively.5.The alignment strategies between regional or national TRF and global TRF are studied,including the datum definition,the choice of reference stations and the effect of coordinates of reference station in different reference frames on the results.Using a MC by similarity transformation,different coordinates in ITRF97 and ITRF2008 for the same set of fiducial stations are used to calculate the network positions,and some systematic discrepancies exists between these two cases.The std of discrepancy are 2,17 and 6 mm for X,Y and Z component,respectively.After removing the systematic discrepancies,the std are still 1.1,0.8 and 0.5 mm for each component.However,for baseline,the difference of baseline length between ITRF97 and ITRF2008 is small;the largest one is smaller than 1 mm.When using coordinates in ITRF97 for datum definition,the std of differences between stochastic constraint and similarity constraint are 5.6,1.9 and 5.27 cm for X,Y and Z component,respectively.But when using coordinates in ITRF2008,the std are only 1.26,0.27 and 0.15 cm for each component.6.A total of 260 reference stations spanning 4 years and 2000 campaign stations in 2009 and 2011 come from Crustal movement observation network of china(abbreviated CMONOC)have been processed.A long-term final TRF has been established using the method of stacking normal equations based on the weekly input data of reference stations.The internal precisions are 0.3,0.5 and 0.4 mm for X,Y and Z component,respectively.Using MC,the internal precisions of 2000 campaign stations in year 2009 and 2001 are better than 2 cm for each component.7 transformation parameters between ITRF2008 and long-term solutions are calculated,the translations are 0.03,17.25 and 4.16 mm and the rotation angle are 0.278,-0.150 and 0.229 mars for X,Y and Z component,respectively.7.Using 1070 station velocities obtained from the observations from the year 1998 to 2009,two combined collocation models based on Euler vector(trend parameters)and the local distortions(signals)are proposed in this paper.An adaptive collocation estimator is applied by which the contributions of the signals and the observations to the estimated crustal movements are balanced.The calculation of Chinese horizontal crustal movements shows that the accuracy of the estimated horizontal crustal movement velocities is improved by the adaptive collocation model.Compared with the result from the classical collocation,the precisions of the estimated velocities are from 1.81 to 1.76 mm/a for east component and from 1.78 to 1.35 mm/a for north component.Based on collocation and multi-quadric equation interpolation,another combined estimation method for establishing velocity field is presented,in which a kernel function of multi-quadric fitting model substitutes for the covariance function of collocation signal.The result obtained by the new method shows that the precisions for east and north component are 1.25 and 0.89 mm/a respectively,while the precisions of Hardy Multi-quadric Fitting Model are 2.51 mm/a for east component and 1.77 mm/a for north component.8.Generally speaking,the similarity transformation method applied for coordinate transformation are not so accurate to unify different coordinate systems,an alternative collocation model is presented for coordinate unification,which treats the similarity transformation as trend part of transformation functional model and the residual errors of transformed models as stochastic part.The observation errors of common points do not only exist in the initial frame but also in the target frame,a new collocation method considering error in both frames is proposed.In collocation,it is required that the prior covariance matrices between the signals and the observations should be consistent with their uncertainties.Otherwise,the coordinate of the stations provided by the collocation transformation will be unreasonable.An adaptive collocation estimator is applied by which the contributions of the signals and the observations to coordinate transformation are balanced.Reasonable and consistent covariance matrices of the signals and the observations are achieved through the adjustment of the adaptive factor.The actual calculation from Xi'an Geodetic Coordinate System 1980(XAS80)to CGCS200 shows that the accuracy of coordinate transformation is improved by using the adaptive collocation model.9.The coordinates are obtained from observation by using least-squares method,so their precision should be contaminated by observation errors and the covariance also exists between common points and non-common points.The coordinate errors do not only exist in the initial frame but also in the target frame.But the classical step-wise approach for coordinate frame transformation usually takes the coordinate errors of the initial frame into account and overlooks the stochastic correlation between common points and non-common points in the initial frame.A new rigorous unified model for coordinate frame transformation which takes into account both the errors of all coordinates in both frame and inter-frame coordinate covariance between common points and non-common points,and the corresponding estimator for the transformed coordinates are derived and involved appropriate corrections to the standard approach,in which the transformation parameters and the transformed coordinates for all points are computed using least squares approach simultaneously.The inter frame coordinate covariance should be consistent to their uncertainties,but in practice they are usually not.To balance the covariance matrices of both frames,a new adaptive estimator for the unified model is thus derived in which the corresponding adaptive factor is constructed by the ratio computed by Helmert variance-covariance component estimation,reasonable and consistent covariance matrices are arrived through the adjustment of the adaptive factor.Finally,a real experiment with 2000 points from CMONOC is carried out to verify the implement of the new model;the results show that the proposed model can improve the precision of the coordinate transformation significantly. |
PDF Full Text Request