The increasing need for continuously available software systems has raised two key-issues: self-adaptation and design evolution. The former one requires software systems to monitor their execution platform and automatically adapt their configuration and/or architecture to adjust their quality of service (optimization, fault-handling). The later one requires new design decisions to be reflected on the fly on the running system to ensure the needed high availability (new requirements, corrective and preventive maintenance). However, design evolution and selfadaptation are not independent and reflecting a design evolution on a running self-adaptative system is not always safe. We propose to unify run-time adaptation and run-time evolution by monitoring both the run-time platform and the design models. Thus, it becomes possible to correlate those heterogeneous events and to use pattern matching on events to elaborate a pertinent decision for run-time adaptation. A flood prediction system deployed along the Ribble river (Yorkshire, England) is used to illustrate how to unify design evolution and run-time adaptation and to safely perform runtime evolution on adaptive systems.