The increasing threat of unpredictable and no-notice disasters due to climate change, such as tsunamis, wildfires, and dam-failure floods, should call for more attention to disaster management. These disasters leave limited time for the affected population to respond. Hence, Shelter-In-Place (SIP) strategies have been suggested to complement large-scale evacuation, aiming to secure individuals within local communities immediately. In the problem context of dam-failure floods, the priority of SIP is to elevate the individuals above the flooded ground, and Vertical Evacuation Structures/Shelters (VESs) are designed to contain these SIP evacuees. However, a portion of the flood-affected population cannot be accommodated because of the limited capacities of VESs. Accordingly, these evacuees must evacuate outside the flooded area, known as Horizontal Evacuation (HE). This study formulates the VES location problem considering VE and HE: the established VESs will cover as many evacuees as possible and reduce the system evacuation time for HE. The VES locations and the HE evacuee assignment are interdependent, as reflected by factoring individual evacuation time. Therefore, this study develops a bi-level programming model to address the interdependence. Furthermore, a heuristic algorithm is proposed to analyze the evacuation time for evacuees at each node. The bi-level model is verified using the network of Tucheng District, New Taipei City, Taiwan, impacted by dam-failure floods. Results show that the total evacuation time can be reduced significantly after integrating HE into VES planning.