Prepare or Wait? The Marine Forces Reserve Hurricane Decision Simulator

Imagine you are the commander of the U.S. Marine Forces Reserve (MFR) in New Orleans, and there is a major hurricane in the Gulf of Mexico. You are responsible for 3,000 people, plus families, as well as the mission to protect the U.S. homeland. Your decision matrix calls for deploying staff to set up an alternate headquarters at least 72 hours before gale-force winds, but this is a costly decision. The storm is currently tracking over Alabama, and the probability of hurricane-force winds in the next 72 hours is 10%. You know the track could change, but also that on average a hurricane will hit New Orleans only once every seven years. You decide to wait. The National Hurricane Center (NHC) issues its next forecast—now the storm is tracking over Mississippi, the probability of a hurricane reaching New Orleans is 25%, and the lead time is less than 60 hours. Should you pull the trigger now?

Preparing for a hurricane is the archetypal multistage decision under uncertainty. For the MFR commander and for local officials, it has very high stakes—even early preparation steps may incur $10 million in unrecoverable costs to prepare for a hurricane that may or may not affect their operations. But failing to prepare can be even worse. This challenge is compounded by another common analytics problem—a gap between the people who generate information and the people who use it, and a resulting mismatch between the information provided and the decision processes that use it. 

For the MFR headquarters in New Orleans, this mismatch is stark. Their decision process is based on a timeline and presumes the hurricane arrival time is known, when it is actually highly uncertain. A collection of NHC products forecast many of the important variables like storm track, size, and intensity, but none of them corresponds directly to the MFR decision matrix. Accuracy improves with each forecast update, but the changing forecasts make the decision environment that much more challenging. And while the NHC is conservative about modifying its forecast products, they do change season to season and their accuracy improves. This makes it hard for even long-term hurricane professionals to learn to use the forecasts optimally. Frequent rotations in leadership pose a special challenge for the U.S. Marines. At the MFR headquarters in New Orleans, there have been seven changes of command since 2000.

After Hurricane Ivan in 2012, the MFR asked the Naval Postgraduate School for support with this challenging decision. As described in Regnier and MacKenzie (2018), the Hurricane Decision Simulator (HDS) is an online simulation-based training tool that lets individual users experience four threatening storms per hour, instead of the natural rate of about one per year.

We worked with the Marines and other local officials to understand and model their decision and implementation process for hurricane operations, and the possible consequences of preparation scenarios as a function of storm scenarios. Realistic context includes follow-on implementation actions from their execution matrix. In addition, the HDS simulates other officials’ decisions that affect the Marines such as issuing evacuation orders or closing levees. Unlike other hurricane-preparation training, the HDS is a choose-your-own-adventure. It allows users to make decisions throughout a storm and receive outcome feedback that depends on their choices and the storm’s impacts.

Fig1

Screen shot from Hurricane Decision Simulator, showing simulated storm Andy threatening the Marine Forces Reserve Headquarters Training Center (HTC) in Hialeah, FL.

In order to give users the opportunity to experience decades’ worth of storms in a few hours, the HDS requires hundreds of distinct simulated storms. The simulated storms are based on 38 years of hurricane data. We use a discrete-time Markov model to generate the simulated storm’s center. The simulated storms span the diversity of possible storm and forecast behavior so that users encounter not just typical storms but also realistic storms with unexpected behavior. The forecast errors are consistent with NHC forecasts for the most recent five-year period (now 2012–2017), and forecast products first used in 2017 are included.

The MFR has been using the HDS successfully at the MFR headquarters for two seasons and are extending the HDS to reserve training centers on the Gulf and Atlantic Coast. Originally built for the headquarters office in New Orleans, version 2.0 includes expansion to more locations and a storm selector to support team-based training. In addition, a stylized scenario has been added to support experiments.

The MFR headquarters uses the HDS in its two tabletop hurricane exercises each year as well as in individual training for the crisis action team and emergency response staff. In team exercises, MFR reports that the HDS training provides a more realistic experience than hand-generated scenarios used previously. The HDS encourages the Marines to focus on the uncertainty in the storm forecast and difficult trade-offs inherent in the preparation decisions with this uncertainty.

In 2017, we worked with the Hialeah, Florida Marine Forces Reserve Headquarters Training Center to develop their decision matrix and represent their processes in the HDS. When the command was handed over in June 2017, the departing and arriving commanders used the HDS to communicate the knowledge the new commander needed to take over responsibility for hurricane preparation. They reported that the process proved very helpful in September, when Hurricane Irma threatened the entire Florida peninsula. 

You can try out the simulator at http://hurricane.mfr.nps.edu.

 

Reference

Regnier ED, MacKenzie CA (2018) Finalist—2017 M&SOM Practice-Based Research Competition—The hurricane decision simulator: A tool for Marine forces in New Orleans to practice operations management in advance of a hurricane. Manufacturing Service Operations Management, ePub ahead of print May 24, https://doi.org/10.1287/msom.2017.0694.

Comments