This work introduces a bilevel, stochastic optimization problem aimed at robust, regional evacuation network design and shelter location under uncertain hazards. A regional planner, acting as a Stackelberg leader, chooses among evacuation-route contraflow operation and shelter location to minimize the expected risk exposure to evacuees. Evacuees

then seek an equilibrium with respect to risk exposure in the lower level. An example network is solved exactly with a strategy that takes advantage of a fast, low-memory, equilibrium algorithm and general purpose computing on graphical processing units.