Approximate Dynamic Programming for MEDEVAC Dispatch Optimizationunder Catastrophic Flood Conditions: A Case Study of the Gharb Region, Morocco

Abstract

When floods inundate Morocco’s Gharb plain, roads vanish and communities are isolated, turning medical evacuation into a desperate race against time. In these mass-casualty events, the deployment of aerial vectors—helicopters and other aircraft—becomes the only viable lifeline. However, coordinating limited air assets amidst shifting floodwaters, stranded populations with varying injury severities, and overwhelmed hospitals presents an immense logistical puzzle.This study introduces an adaptive optimization model designed to solve this puzzle in real time. Framed as a sequential decision problem, our approach dynamically manages the dispatch, rerouting, and repositioning of air ambulances across the disaster zone. The system continuously accounts for triage-prioritized casualty lists, aircraft positions and flight times, and the fluctuating availability of medical treatment facilities.We trained the model using two machine learning techniques—Random Forest and Gradient Boosting—and tested it against three catastrophic flood scenarios in the Gharb region, including a simulated once-in-a-century event generating over eight thousand evacuation requests. Results show that our approach consistently outperforms standard decisionmaking rules, improving the speed and priority of casualty transport. By learning which decisions save the most lives, the model consistently prioritizes the most critically injured. This work provides Moroccan emergency services with a practical, data-driven tool to enhance disaster response, ensuring that when the waters rise, the chain of survival remains unbroken.

BJMS, Vol. 25 No. 02 April’26 Page: 473-486