Spatio-temporal images are a class of complex dynamical systems that evolve over both space and time. With increased interest in the investigation of nonlinear complex phenomena, especially spatio-temporal data analysis governed by evolutionary laws that are dependent on both spatial and temporal dimensions, there has been an increased need to investigate data-driven modeling methods for this class of complex systems. Compared with pure temporal processes, the identification of spatio-temporal models from observed images is much more difficult and quite challenging. Starting with an assumption that there is no a priori information about the true model but only observed data are available, this work introduces a new type of wavelet network that utilizes the easy tractability and exploits the good properties of multiscale wavelet decompositions to represent the rules of the associated spatio-temporal evolutionary system. An application to a chemical reaction exhibiting a spatio-temporal evolutionary behaviour, is investigated to demonstrate the application of the proposed modeling and learning approaches.