The Moho depth can be determined using seismic and/or gravimetric methods. These methods will not yield the same result as they are based on different hypotheses as well as different types, qualities and distributions of data. Here we present a new global model for the Moho computed based on a stochastic combination of seismic and gravimetric Moho models. This method employs condition equations in the spectral domain for the seismic and gravimetric models as well as degree-order variance component estimation to optimally weight the corresponding harmonics in the combination. The preliminary data for the modelling are the seismic model CRUST2.0 and a new gravimetric Moho model based on the inverse solution of the Vening Meinez-Moritz isostatic hypothesis and the global Earth Gravitational Model EGM08. Numerical results show that this method of stochastic combination agrees better with the seismic Moho model (3.6 km rms difference) than the gravimetric one. The model should be a candidate for dandifying the frequently sparsely data CRUST2.0. We expect that this way of combining seismic and gravimetric data would be even more fruitful in a regional study.
The problem of handling outliers in a deformation monitoring network is of special importance, because the existence of outliers may lead to false deformation parameters. One of the approaches to detect the outliers is to use robust estimators. In this case the network points are computed by such a robust method, implying that the adjustment result is resisting systematic observation errors, and, in particular, it is insensitive to gross errors and even blunders. Since there are different approaches to robust estimation, the resulting estimated networks may differ. In this article, different robust estimation methods, such as the M-estimation of Huber, the “Danish”, and the
-norm estimation methods, are reviewed and compared with the standard least squares method to view their potentials to detect outliers in the Tehran Milad tower deformation network. The numerical studies show that the
-norm is able to detect and down-weight the outliers best, so it is selected as the favourable approach, but there is a lack of uniqueness. For comparison, Baarda’s method “data snooping” can achieve similar results when the outlier magnitude of an outlier is large enough to be detected; but robust methods are faster than the sequential data snooping process.