Using laboratory models of continental collisions, we study the mechanisms responsible for large-scale deformation and the nature of the penetrative deformation, localized or homogeneous, and its characteristic scales and structure. In order to focus the study, one of the most spectacular cases of continental collisions, namely the India--Asia collision, is considered Different models with varying rheologies are analyzed which attempt to respect the brittle--ductile stratification corresponding to the crust and mantle structure in the Earth. Each experiment is quantified by studying the strain field and the fault pattern as a function of the position P of the indenter within the system. The strain field is characterized by (1) the second invariant of the two-dimensional strain tensor and its variation with position for a fixed P, the evolution of its average as a function of P, (2) the evolution with P of the participation ratio S2, which quantifies the fraction of the system surface area which participates in the deformation, (3) the evolution of the ''escape ratio,'' calculated as the surface increase gained by the system during its eastward lateral escape divided by the surface covered by the indenter penetration. Qualitatively, we find that the deformation first spreads out but later on undergoes localization. Once the fault pattern is ''mature'' (the cumulative fault length does not increase anymore), we observe that the strain field is essentially controlled by the kinematics of the larger faults.

Each fault pattern is quantified by studying (1) the histograms of fault orientations, (2) the ''capacity'' and barycenter fractal dimensions, (3) the ''multifractal'' generalized dimensions, and (4) the distribution of fault lengths. The fractal dimension Df is found to almost constant within experimental uncertainty (Df=1.7¿0.1) and thus appears rather insensitive to the particular chosen rheology. We find a correlation between the generalized multifractal dimensions and two exponents, the barycenter fractal dimension ''b'' and the exponent ''a'' of the fault length distribution. This shows that the scaling properties of fault patterns can be characterized by the knowledge of only two exponents ''b'' and ''a'' of the spatial and length distributions of faults.

Our main results are (1) observation of the growth of self-similar fault patterns, (2) complete characterization of the fault patterns with two scaling exponents ''a'' and ''b'', (3) wide distribution of undeteriorated domains and large heterogeneity of the deformation field, (4) maturation of the fault structure corresponding to a localization of the deformation on a few large faults in the late stage of the deformation. ¿ American Geophysical Union 1993