Self-organized pattern formation is vital for many biological processes. Mathematical modeling using reaction-diffusion models has advanced our understanding of how biological systems develop spatial structures, starting from homogeneity. However, biological processes inherently involve multiple spatial and temporal scales and transition from one pattern to another over time, rather than progressing from homogeneity to a pattern. One possibility to deal with multiscale systems is to use coarse-graining methods that allow the dynamics to be reduced to the relevant degrees of freedom at large scales. Unfortunately, these approaches have the major disadvantage that the eliminated scales cannot be reconstructed from the large-scale dynamics and thus one loses the information about the patterns. Here, we present an approach for mass-conserving reaction-diffusion systems that overcomes this issue and allows one to reconstruct information about patterns from the large-scale dynamics. We illustrate our approach by studying the Min protein system, a paradigmatic model for protein pattern formation. By performing simulations, we first show that the Min system produces multiscale patterns in a spatially heterogeneous geometry. This prediction is confirmed experimentally by in vitro reconstitution of the Min system. On the basis of a recently developed theoretical framework for mass-conserving reaction-diffusion systems, we show that the spatiotemporal evolution of the total protein densities on large scales reliably predicts the pattern-forming dynamics. Since conservation laws are inherent in many different physical systems, we believe that our approach can be generalized and contribute to uncover underlying physical principles in multiscale pattern-forming systems.