Flooding presents a major hazard to urban citizens, particularly during the initial phase of an event when evacuation decisions must be made under limited spatial information. Despite recent advances in hydrological modelling and early warning systems, these approaches often do not provide route-level information required for operational evacuation planning. This study presents a graph-based spatial framework for computing safe evacuation routes under flood conditions. The approach constructs a weighted graph from open geospatial data, integrating buildings, street network, and flood extent. Buildings are classified according to their extent of flood-exposure, while road segments intersecting flooded areas are assigned a prohibitively large cost and are therefore excluded from the routing process, whereas unaffected segments retain a finite cost proportional to their length. Safe routes are calculated using a shortest-path algorithm on the resulting flood-constrained weighted graph. The methodology is applied to a case study in Kremmen and Oranienburg, Germany, a flood-prone region in Brandenburg, demonstrating how inundation alters network connectivity and constrains evacuation options. The proposed framework supports the integration of semantic information and optimisation in evacuation planning under flood scenarios. Whilst the current framework operates on the basis of static flood extents, its graph-based modular structure can potentially incorporate dynamic flood data, which would enable its applicability to be extended to provide operational support for evacuation during the course of an event.