With excellent air-ground multi-mode movements, the vertical take-off and landing (VTOL) air-ground integrated vehicle can easily traverse complex terrains and maintain high energy efficiency. During movement, path planning plays an important role in achieving the autonomous operation of the vehicle, which faces the following difficulty. A short air-ground multi-mode path requires efficient planning, with proper judgment of the timing and position for mode switching. To address this difficulty, we propose a new path planning algorithm, named Dynamically Directed Graph Algorithm (DDGA). It can realize short path search in limited search nodes via dynamically extracting key search nodes in maps and forming a dynamically directed graph. To be specific, adjacent nodes of the first obstacle traversed by the connection line from the current search node to the destination are defined as key search nodes. As the current node changes, key search nodes are dynamically updated. The above key search nodes and the directed paths between them form a dynamically directed graph. Considering the air-ground movement capability, obstacle areas below maximum flight altitudes in maps are defined as pending flight areas. The directed paths traversing these areas are considered in the above graph. Besides the two-dimensional distance cost, flight altitude cost is added to the cost values of different directed paths. This cost contributes to judging the proper switching timing and position. Compared to other algorithms, DDGA can find short paths with fewer search nodes in multiple obstacle maps. It efficiently plans a short air-ground multi-mode path for the VTOL air-ground integrated vehicle.