<p>Permeability is affected by complex factors such as the subsurface geological structure and porosity-permeability correlation. For highly heterogeneous reservoirs with complex pore structures, it is extremely challenging to spatially characterize (predict) permeability using seismic data. The conventional way of permeability prediction intends to convert underground reflection data into the elastic parameters sensitive to underground fluids, build a universal low-dimensional template via petrophysical modeling and ultimately deliver spatial prediction of permeability. However, this method is restrained by the actual subsurface condition, selected well-logging sensitive parameters and the accuracy of the computed elastic parameters and fails to simulate the petrophysical mechanisms of complex reservoir permeability, which reduces the permeability prediction accuracy. The method proposed in this paper combines petrophysics and artificial intelligence and integrates multiple types of information to build the high-dimensional petrophysical template for permeability, in an attempt to improve the spatial characterization and prediction accuracy of permeability. The field testing demonstrates the high application performance and effective improvement in prediction accuracy and fluvial channel characterization.</p>