top of page
Hemstitching is the embodiment of two concepts: The path tracked by the parameter estimation process in its search for the best iterative solution; And the shape of the available solution space where this parameter exploration takes place, literally.
The liberality of the modelling calibration relies in the relationship among three measures belonging to the Singular Value Decomposition (SVD) analysis, the directional vectors of orthonormal V and U matrixes and the scalar influence the diagonal S matrix.
Bellow, an expedite analysis took the licence to compare, as physical distances on the cardinal system, the different shapes formed by plotting the original vectors (columns of a given matrix A) and related eigenvalues (λ), after the cadential multiplication of its elements by themselves.
The SVD and hemstitching concepts should play together a basal rule for any numerical modelling proposition. "I personally bet if many of you MODFLOW users were or not bothered sometimes by the error: The matrix is severely non-diagonally dominant - stop execution." [?]
Thus, in considering mainly the earlier phases of manual regularization, in choosing reasonable sets of contour conditions and initial parameters, it may be important to keep in mind the liberty afforded by the numerical designing for the further search of a suitable, final combination of parameters for the fine tuning of this model.
To not say that “the decision-support modelling should be designed from the ground up" visualizing ahead, in advance, the burden to be overcome by PEST or other mechanism of calibration.
bottom of page