Discover more from Sean’s Research Newsletter
[Journal Article] Using tornado-related weather data to route unmanned aerial vehicles to locate damage and victims
OR Spectrum, Quantitative Approaches in Management
Citation : Grogan, S., Pellerin, R. & Gamache, M. Using tornado-related weather data to route unmanned aerial vehicles to locate damage and victims. OR Spectrum (2021). https://doi.org/10.1007/s00291-021-00640-1
Read more here : https://rdcu.be/cnwnr
In this paper we propose a framework for the use of Unmanned Aerial Vehicles (UAVs) equipped with cameras and wireless sensors to search an area after the occurrence of a tornado. We attempt to demonstrate how tornado weather data can be incorporated into search and rescue procedures to allocate and route the UAVs. Traditionally, the time to assess and search an area after a tornado strikes is on the order of several days. Incorporating UAVs into a search and rescue team's available tools can reduce this time-span to the order of hours. We apply the methods and model in this project to three real-world cases. Several methods for creating "waypoints", points of interest for the UAVs to inspect, to route the UAVs were tested. We analyzed the time it took to generate the waypoints and the resulting objective function value. We observed that while there is an opportunity to use exact methods to generate waypoints, our proposed heuristic is sufficient for the rapid response needed in post disaster relief.
What does this mean for practitioners?
For OR practitioners, we often discuss about the the tradeoffs of using exact and approximate solution methods (and finding faster solution methods) when solving optimization models. I believe we, as OR practitioners, have an intuitive sense of this tradeoff, but we don’t necessarily quantify this tradeoff.
Search and Rescue operations provide a strong base to conduct this research. Time spent planning and time spent executing the plan is important. Specifically, in our paper the objective function of the model proposed is in units of time, and the time value is on the order of several hours, which means there is a quantifiable and obvious relationship between time spent planning (solving the mathematical model) and the time executing the plan. In the paper, we can see solution methods that can be resolved in a “reasonable” time (less than a hour), but potentially are not “worth it” because a solution that solves in minutes yields a solution that is about 20 minutes longer. Explicitly from one test :
Heuristic solution, takes 15s to solve, solution is of 3h28m.
Metaheuristic solution, takes 6m8s to solve, solution is of 3h28m (does not find a better solution than the heuristic to the precision of a minute)
Exact solution, takes 40m17s to solve, yields a solution of 3h09m
For disaster response practitioners, we hope this poses some creative use of well known and recorded data and novel equipment to respond to disasters. As we discuss in our conclusion, this paper is an American-centric and probably not widely applicable as-is to everywhere in the world. However, we encourage the reader to reflect and explore how to use expert-generated data and the flexibility wireless sniffers can provide members of the search and rescue community.
Off the top of our heads, we pose that an example to get data of where people may be is to use light pollution maps and imagery from artificial satellites. One such idea to identify bounding locations for search areas is via observation towers (such as those tracking forest fires) and triangulating a search area from observers.