Improving Disruption Management With Multimodal Collaborative

IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016 2699

Improving Disruption Management With Multimodal

Collaborative Decision-Making: A Case Study

of the Asiana Crash and Lessons Learned

Aude Marzuoli, Emmanuel Boidot, Pablo Colomar, Mathieu Guerpillon,

Eric Feron, Alexandre Bayen, and Mark Hansen

Abstract—Transportation networks constitute a critical

infrastructure enabling the transfers of passengers and goods,

with a signiﬁcant impact on the economy at different scales.

Transportation modes are coupled and interdependent. The

frequent occurrence of perturbations on one or several modes

disrupts passengers’ entire journeys, directly and through ripple

effects. Collaborative decision-making has shown signiﬁcant

beneﬁts at the airport level, both in the U.S. and in Europe. This

paper examines how it could be extended to the multimodal

network level, discusses the supporting evidence, and provides

recommendations for implementation. A case study on the

disruption management following the Asiana Crash at San

Francisco International Airport is presented. The crash led

to a large number of ﬂight diversions to many airports,

such as Oakland, Los Angeles, but also Seattle for instance,

disrupting the journeys of thousands of passengers. Passenger

reaccommodation varied greatly from airline to airline and airport

to airport. First, a passenger-centric reaccommodation scheme is

developed to balance costs and delays, for each diversion airport.

Second, assuming better information sharing and collaborative

decision-making, we show that there was enough capacity at the

neighboring airports, Oakland and San Jose, to accommodate

most of the diverted ﬂights and reoptimize the allocation of ﬂight

diversions to the Bay Area airports. Based on this case study,

recommendations for the adoption of multimodal CDM are

elaborated. This paper paves the way for further data-driven re-

search for increased resilience of passenger door-to-door journeys.

Index Terms—Air trafﬁc control, air transportation, rail trans-

portation, road transportation, robustness.

I. INTRODUCTION

N 2012, 2.9 billion passengers boarded an airplane, whether

for business or leisure, across the world [1]. Yet, air transport

is only a portion of the passenger door-to-door journey, which

also relies on other modes of transportation, such as rail, road

and water. Transportation modes are usually studied separately

as if not interacting, although they are intrinsically coupled

Manuscript received June 21, 2015; revised December 28, 2015; accepted

February 11, 2016. Date of publication April 20, 2016; date of current version

September 30, 2016. The Associate Editor for this paper was F.-Y. Wang.

A. Marzuoli, E. Boidot, P. Colomar, M. Guerpillon, and E. Feron are with

the Department of Aerospace Engineering, Georgia Institute of Technology,

Atlanta, GA 30332 USA (e-mail: am[email protected]).

A. Bayen and M. Hansen are with the Department of Civil and Environmental

Engineering, University of California at Berkeley, Berkeley, CA 94720 USA.

Color versions of one or more of the ﬁgures in this paper are available online

at http://ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/TITS.2016.2536733

through passenger transfers. The failure of one mode disrupts

the entire passenger journey. Over the past few years, many

disruptions have highlighted the rigid structure of transport

infrastructures and the potential for perturbations to snowball

across multimodal infrastructures. The failures and inefﬁcien-

cies of the air transportation system not only have a signiﬁcant

economic impact but they also stress the importance of putting

the passenger at the core of the system [2]–[5]. The objective of

making each passenger or cargo’s door-to-door journey seam-

less cannot be achieved without a better understanding of the

multi-modal transportatio n network. The regular occurrence of

perturbations that propagate through the system and sometimes

even paralyze it highlights the need for further research on

its resilience and agility and for adequate coordination a t the

network level. As the number of passenger keeps growing [1],

congestion and snowball effects threaten the resilience of the

whole multimodal transp ort infrastructure.

On the transportation side, there has been extensive research

on disturbance propagation in the airspace [6]–[9], the impact

on airline scheduling of aircraft and crew [10] and the best

recovery optimization schemes [11], [12]. When a disruption

occurs, airline schedule recovery tries to maintain operations

and get back to schedule as quickly as possible while mini-

mizing additional costs. The different recovery mechanisms are

aircraft swaps, ﬂight cancellations, crew swaps, reserve crews

and passenger rebooking. Usually airlines react by solving

the problem in a sequential manner. First, infeasibility of the

aircraft schedule is examined, then crewing problems, then

ground problems and the impact on passengers.

In the coming decades, air trafﬁc demand is expected to

increase signiﬁcantly [13]. The present airspace capacity limits

are predicted to be exceeded. Delays caused by congestion or

weather perturbations are increasing on the ground and in the

air. The cost of congestion in such a tightly interconnected

network of airports and aircraft reached $41 billion in the US

in 2008 [14]. In 2012, 18.22% of ﬂights were delayed in the

United States [6].

Most of the trafﬁc demand g rowth is expected to take place in

major metropolitan areas. Metropolitan areas with high demand

are often served by a system of two or more airports whose

arrival and departure operations are highly interdependent,

referred to as a metroplex [15]. Atkins [16] examined the

San Francisco Bay Area metroplex, providing a deﬁnition

and an initial framework to measure metroplex performance.

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

2700 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

Clarke et al. [17], [18] identiﬁed six types of interdependencies

between trafﬁc ﬂows in a metroplex based on observations.

Li et al. [19] studied the metroplex operational interdepen-

dencies, resulting from sharing limited c ommon resources in

airspace, such as common ﬁxes, ﬂight paths, airspace volumes

and downstream restrictions. DeLaurentis [20] evaluated a con-

cept of ﬂexible operations at a metroplex to optimize the use of

common resources.

In 1991, the FAA’s Air Trafﬁc Management Ofﬁce com-

missioned an analysis to measure the effects of the airlines’

ﬂight-substitution process on the efﬁcacy of ground-delay pro-

grams (GDPs) [21]. The FADE (FAA Airlines Data Exchange)

project aimed at the development of operational procedures and

decision support tools for implementing and managing GDPs.

However, the CDM philosophy and principles can and should

be applied to a much broader class of problems in air trafﬁc

management. Prototype GDP operations started in 1998 at San

Francisco and Newark airports [22]. The collaboration between

government and industry was born out of the FAA’s need

for real-time operational information from the airlines and the

airlines’ desire to gain more control over their operations during

a GDP [23]. Burgain et al. [24] developed a Collaborative

Virtual Queue (CVQ), which uses virtual queuing to keep

aircraft away from runway queues and enable last-minute ﬂight

swapping. Gupta [25] built an integrated system, SARDA-

CDM, for improving surface operations by metering departing

aircraft. Three need s led to the creation of CDM and are still

at the heart of the concept today: the need for a shared global

picture of predicted capacity and demand for various airspace

resources, leading to common situational awareness and sup-

ported by appropriate information sharing; the need for real-

time models that predict the impact of potential control actions

and user decisions, supported by data from all stakeholders; the

need for collaborative resource allocation tools, mechanisms

and procedures.

Over the past few years, severe weather p erturbations have

paralyzed the transportation system. On the European side,

the eruption of the islandic volcano in 2010 had the longest

and biggest economic impact on aviation [26], with more than

100,000 ﬂights canceled. Bolic et al. offer recommendations

to better address such large disruptions, stressing the need

for better information exchanges between all the stakeholders.

Zhang [27] develops a framework to reduce passenger “disu-

tility,” to help airlines recover schedule more promptly, and to

assist trafﬁc ﬂow managers to utilize scarce resources m ore efﬁ-

ciently and equitably. When there is a signiﬁcant capacity short-

fall, airlines with hub-and-spoke networks could incorporate

ground transport modes into their operations. Such intermodal-

ism triggered by disruptions was reported by Evans regarding

Continental Airlines at Newark [28]. Real-time intermodalism

includes the substitution of ﬂights by ground transportation and,

when the hub is part of a regional airport system, the use of

inter-airport ground transport to enable diversion of ﬂights to

alternate airports, while limiting the impact on airlines.

Recently, a shift towards passenger-centric metrics in air

transportation, as opposed to ﬂight-centric, h as been promoted,

after the disproportionate impact of airside disruptions on

passenger door-to-door journeys was highlighted [29]–[32].

Flight delays do not accurately reﬂect the delays imposed upon

passengers’ full multi-modal itinerary. The growing interest to

measure ATM performance calls for m etrics that reﬂect the

passenger’s experience. Cook and al. [29] design propagation-

centric and passenger-centric performance metrics, and com-

pare them with existing ﬂight-centric metrics. In [30],

Bratu et al. calculate passenger delay using monthly d ata from

a major airline operating a hub-and-spoke network. They show

that disrupted passengers, whose journey was interrupted by

a capacity reduction, are only 3% of the total passengers, but

suffer 39% of the total passenger delay. Wang [31] showed that

high passenger trip delays are disproportionately generated by

canceled ﬂights and missed connections. 17% of routes, o r 9 of

the busiest 35 airports, cause 50% of total passenger trip d elays.

Congestion, ﬂight delay, load factor, ﬂight cancellation time

and airline cooperation policy are the most signiﬁcant factors

affecting total passenger trip delay.

The goal o f this paper is to examine how Multimodal Collab-

orative Decision Making can support better crisis management

at the network level, fr om passenger-centric and ﬂight-centric

perspectives. This paper tackles mitigation strategies fo llowing

the Asiana Crash, both for passengers and for ﬂights. Section II

brieﬂy describes the impact of the crash on operations and

on passengers. In Section III, ﬂight diversions are introduced,

from an airline and an Air Trafﬁc Management perspective.

Flight diversions are rare events, but they are harder to recover

from than cancellations for instance. The Asiana crash was a

striking example of m assive ﬂight diversions because of an

unexpected airport closure. In Section IV, a passenger-centric

optimization is proposed to analyze the reaccommodation, via

bus or aircraft, of passengers diverted to airports within rea-

sonable drive distance of San Francisco. In Section V, a ﬂight-

centric optimization model examines how remaining capacity

at Bay Area airports may have p layed a role in diversions of

SFO-bound ﬂights to far away airports. Section VI provides

recommendations for improved crisis management. Finally,

Section VII draws the conclusions of the p aper and suggests

future research paths.

II. T

HE IMPACT OF THE ASIANA CRASH

A. Impact on Operations

First let us brieﬂy summarize the events leading to the Asiana

crash at San Francisco International airport (SFO). SFO is

the seventh busiest airport in the United States, with about

400,000 movements and 45 million passengers per year. On

July 6th, 2013, the weather was good, the winds were light. The

instrument landing system vertical guidance (glide slope) on

runway 28L was, as scheduled, out of service. At 11:28

A.M.,

Asiana Airlines Flight 214, a Boeing 777-200 ER aircraft,

crashed just short of runway 28L’s threshold at SFO. Of the

307 people aboard, 3 died, 181 others were injured. The ac-

cident investigation submission [33] states that “the probable

cause of this accident was the ﬂight crew’s failure to monitor

and maintain a minimum airspeed during a ﬁnal approach,

resulting in a deviation below th e intended glide path and an

impact with terrain.”

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2701

The crash resulted in a ﬁve hour total closure of the runways

at the airport. By 3:30

P.M. PDT, the two runways perpendicular

to 28L were reopened; runway 10L/28R (parallel to the runway

of the accident) remained closed for more than 24 hours. The

accident runway, 10R/28L, reopened on July 12. This crash is

a powerful example of node failure leading to ripple effects

on several networks. Indeed, an airport is a node for the air

transportation network, for the road network because of easy

highway access and for the transit network, with a BART (Bay

Area Rapid Transit) station in the Bay Area.

The work presented is based upon publicly available data

from the Bur eau of Transportation Statistics (BTS) that are

primarily used to evaluate airline on -time performance and

ETMS data, that provides aircraft radar latitude and longitude

every one minute.

The crash led to the closure of SFO and, even after the

airport reopened, its capacity was reduced signiﬁcantly. The

crash led to cancellations, diversions and d elays at SFO, and

impacted the r est o f the airspace with ripple effects. Over four

days, more than 660 ﬂights to SFO and 580 ﬂights from SFO

airport had either been canceled or diverted. Diversions mostly

occurred on Saturday as well as on Sunday. The proportion of

domestic diversions is high: 74 arrival ﬂights, that is 17% of

arrival ﬂights to SFO, were diverted on Saturday. There were

also 180 cancellations of arrival ﬂights, and 231 cancellations

of departure ﬂights from SFO on Saturday. On Sunday, the

situation improved a little, with still 30 arrival ﬂights to SFO

and 3 0 departure ﬂights from SFO diverted. According to the

BTS, 0.2% of domestic ﬂights were diverted in 2013, and it is a

steady number since 2004. Operations were worse on Tuesday,

July 9th than on Monday, July 8th. Moreover, due to the closure

of the crash runway, runway capacity was still signiﬁcantly

reduced, leading to many cancellations. There are very few

diversions after Sunday. This is to be expected since diversions

are usually tactical operations. Cancellations and delays due

to the crash at SFO propagated through the airspace and the

ripple effect lasted several days. To analyze this propagation

phenomenon, the tail numbers of all aircraft involved with

ﬂights canceled at departure or arrival to SFO airport from

July 6th to July 9th were tracked. On the day of the crash, the

propagation of cancellations due to the Asiana crash accounts

for more than 85% of all cancellations in the entire US airspace,

more than 50% on Sunday and more than 25% on Monday and

Tuesday. Over the four days, the Asiana crash led to more than

49% of all cancellations in the US.

The major carrier ﬂights were diverted to a number of

airports. The other Bay Area airports, Oakland (OAK) and

San Jose (SJC) accommodated most ﬂights, from Saturday to

Tuesday. Several other airports, as far as Denver (DEN), Los

Angeles (LAX) and Las Vegas (LAS), received many diverted

ﬂights on the crash day. Fig. 1 displays the estimated number of

diverted passengers. 9,770 domestic passengers were diverted

the 6th of July, 4,260 on the 7th and approximately 1,470

on the 8th and 9th of July. Only 21% of these passengers

could reach SFO, their ﬁnal destination, with the same ﬂight.

The news showed how the disruption left most of the diverted

passengers unattended and uninformed, waiting for the airline

representatives to ﬁgure out how to reaccomodate them on the

Fig. 1. Percentages of domestic ﬂights diverted to different airports from

July 6th to July 9th 2013.

spot. Twitter quickly started to be the most updated channel of

information for passengers. In the ETMS data, we found 25 ad-

ditional diversions, by analyzing the trajectories of international

ﬂights scheduled to arrive at SFO but landing elsewhere. These

diverted ﬂights landed in Vancouver (YVR), Calgary (YYC),

SEA, LAX, SMF, LAS, OAK and SJC.

Most stakeholders only have access to a partial v iew of

the crisis and, in most cases, for only one mode of transport.

Following the Asiana crash, if the main stakeholders had had

access to real-time data feeds of reliable trafﬁc data via collab-

orative decision making, it is likely that the recovery process

could have been more efﬁcient. Our work therefore focuses

on optimization of aircraft operations and diverted passenger

reaccommodation. At the present stage, only hypotheses can

be drawn when it comes to how diverted passengers who

landed sometimes several hundred miles from their destination

airport actually made it to their ﬁnal d estination. Social media,

such as Twitter, provides pieces of information suggesting that

the treatment of passengers and the crisis management varied

greatly from airport to airport and from airline to airline.

B. Impact on Passengers

Twitter proved p articularly usefu l to access infor mation on

diversions and their management depending on the airports

the diverted ﬂights landed at. It also provides a passenger-

centric view of the crisis, which is rarely taken into account

while analyzing the air transportation performance under crises.

When it comes to timing, tweets provide a means to access spe-

ciﬁc information about diverted passengers. Such information

was otherwise very unlikely to resurface with usual internet

searches because the large news coverage ﬂooded the internet

with similar summaries of events but little precision on the

timing of events.

Many passengers were diverted to airports where their airline

operates at low frequency. For instance, at the Virgin America

counter of Seattle Tacoma International Airport on the 6th of

July, [34] customer-service representatives collected travelers’

2702 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

names and phone numbers so the airline could rebook or cancel

ﬂights without the people standing in line. Passengers were

advised that the quickest option to get home would be to

rebook through another carrier and obtain a refund, as “the

soonest ﬂights on Virgin America will be Monday or Tuesday.”

Although notiﬁed that waiting times could reach the two days,

passengers argued they had no extra money to purchase new

ﬂights and be refunded later.

CBS news reports [35] how in Sacramento, located less than

100 miles from SFO, people waited for hours uninformed,

queueing around help desks and waiting for airline represen-

tatives to inform about the rerouting options. CBS News inter-

viewed passengers who said: “We were not even aware of what

had happened until someone on the ﬂight was able to turn on the

cell phone,” “Our carrier had no information whatsoever. We

were booted off the plane and with no direction whatsoever.”

Such witness accounts support the fact that airlines had n o

systematic rerouting scheme for such disruptive situation.

On July 6th, Salt Lake City International Airport (SLC)

absorbed most of the international ﬂights, instead of SJC or

OAK. SLC was the closest airport with an international custom

capable of processing international ﬂights diverted, as OAK and

SJC had no such facilities ready. However, some international

ﬂights could not be diverted to international hubs such as Seattle

or Phoenix, due to low fuel reserves. Therefore, some of them

were forced to land in Sacramento, which is not equipped

in terms of customs, forcing the custom ofﬁcers to enter the

aircraft to proceed with the security checks in situ. Other

customs issues were reported in Oakland.

The LAX Airport Operations Center stated on July 6th:

“Three international ﬂights that diverted to LAX and deplaned

their passengers have all left LAX and busing their passengers

to SFO. Airlines that cancelled ﬂights between LAX and SFO

are making arrangements for their passengers, including: re-

booking passengers on future ﬂights, adding special ﬂights if

aircraft are available, busing p assengers to SFO, putting u p

passengers in airport-area hotels, asking passengers to return

to LAX tomorrow, etc. Tomorrow’s (Sunday’s) ﬂights between

LAX and SFO are heavily booked due to the combination of

holiday weekend and peak summer travel, so it is expected the

airlines will require one to two days to catch up on the backlog

of cancelled ﬂights” [35]. “Transportation to San Francisco

for passengers diverted to Sacramento (SMF) depends on the

airline. Delta Airlines arranged taxi and shuttle services for

passengers to get to San Francisco after their planes were

forced to land in SMF. US Airways passengers were loaded

on to shuttle buses at SMF to be taken to San Francisco. (...)

United Airlines did not have a deﬁnitive plan in place to help

passengers who were diverted to SMF. SMF has a Customs

area at the airport, but has limited space” [36]. “Ad ditional

staff was brought in to help accommodate the more than 1,000

passengers that were diverted to SMF after the plane crash in

San Francisco. Ofﬁcials say they had to bring in extra staff

to accommodate all those passengers that were landing at the

same time. It was a mad rush as staff scrambled to get everyone

to where they needed to go” [37]. Many more issues arose

when ﬂights were diverted to airports in wh ich their carrier doe s

not operate. For instance, a SFO-bound United Airlines ﬂight

from Seattle was diverted to Oakland. Local news reporters

[38] interviewed passengers, who reported that “United has no

support here. They sent a dislocation team, but basically what

they keep saying is: “You’re dislocated.” “Many passengers

were diverted to airports where their airline op erates at low

frequency.

Regarding how airports handled ﬂight diversions, San Jose

airport stated [39]: “SJC handled 25 ﬂight diversions including

two international ﬂights while SFO was closed to all operations.

Airport Operations staff handled the majority of duties directly

associated with the diversions—the complex logistics of locat-

ing arriving aircraft at the terminal or at remote locations, and

closing taxiways to allow fo r aircraft parking (...) Thirteen of

the diverted ﬂights were accommodated at the terminals; the

remaining eight deplaned at the North Cargo Area where more

than 500 passengers were bused from the airﬁeld to the termi-

nal. (...) In addition to Airpor t staff, many airport contractors

assisted with the additional trafﬁc Saturday including First

Alarm, Shuttleport and Ta xi San Jose (which saw a 100 percent

increase in business). (...) The airlines serving SFO contin-

ued with “planned” diversions throughout Sunday (15 ﬂights)

and Monday (5 ﬂights). The advance notice makes the planned

diversions much easier to handle from an operational perspec-

tive and provides better service to airline customers.”

The lack or the little amount of information delivered by

airlines, added to the highly booked situation of the holiday

week-end, created additional crisis situations in the diverted

airports. At that point, a few airlines decided to implement

inter-modal operational measures by placing buses and taxis

to reroute their passengers. However this solution was imple-

mented in a non-collaborative, case-to-case fashion and resulted

in wide discrepancies in the treatment of passengers affected by

the same situation.

III. F

LIGHT DIVERSIONS

According to an FAA ofﬁcial [40], “Thunderstorms are re-

sponsible for the majority of aircraft diversions each year. (..)

Diversion ﬂights are a rare occurrence. But when this does hap-

pen, we need to make the information available to help airlines,

controllers and airport operators decide the optimal airport for

a diversion.” Jenkins et al. [41] describe disruptions as follows:

“Diversions are an expensive, chronic, and disruptive element

of ﬂight operations, costing at least $300 MM annually for US

carriers for domestic ﬂights alone. (..) A diversion is not a

single, discrete event but rather a set of cascading actions that

cause severe disruptions to airline schedules, major costs, and

signiﬁcant passenger frustration and ill will. Diversion costs can

range from $15,000 for a narrow-body domestic ﬂight, to over

$100,000 for a wide-body international ﬂight.”

A. Airline Policy

Most airlines indicate on their websites how diversions are

handled and what the consequences may be for passengers.

Even for non-refundable tickets, airlines usually are supposed

to refund ancillary f ees. Delta Airlines stipulates in its contract

that “in the event of ﬂight cancellation, diversion, delays of

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2703

greater than 90 minutes, or delays that will cause a passenger to

miss connections, Delta will (at passenger’s request) cancel the

remaining ticket and refund the unused portion of the ticket.”

US Airways indicates on its website that “When a ﬂight

is diverted to an alternate airport and cancelled, the pilots or

ﬂight attendants will advise the customers of the reason for

the diversion. The customers may need to remain onboard.

(...) Some irregular operations may require landing at alternate

airports, with bus service to the ﬁnal destination. It is acceptable

to allow a customer to leave directly from an alternate airport

without requiring him/her to travel to the ﬁnal destination. (..)

When a ﬂigh t (aircraft) is diverted to a city served by US

Airways or codeshare partner, and canceled (meaning it will

not eventually reach its ﬁnal destination), the customer service

representatives in that city will reaccommodate customers on

either the next available US Airways ﬂight or the next avail-

able ﬂight via another carrier. (...) When a ﬂight (aircraft) is

diverted and then canceled in a city not served by US Airways

or a codeshare partner, the customer service manager in US

Airways’ Operations Control Center will make arrangements

with other carriers and/or hotel accommodations. Once the

ﬂight attendants receive word from the ﬂight deck, they will

communicate to the customers the reason for the diversion,

the estimated time of departure and/or accommodations. I f

the ﬂight is canceled, subject to availability, passengers will

be reaccommodated via another airline. The ﬂight attendants

and ﬂight crew will be the US Airways representatives for the

customers. (...) When alternate tran sportation is unavailable

until the following day and overnight accommodations are

required, the ﬂight attendants and ﬂight crew will communicate

to the passengers which expenses US Airways will pay. The

following is a list of what US Airways will pay for providing

the cancellation is due to anything other than weather: hotel

room; ground transportation; passengers without baggage will

be reimbursed upon presentation of receipts for reasonable

incidentals such as toiletries needed until they are reunited with

their baggage.”

B. FAA Rules

The FAA diversion recovery plan provides details on the

chain of command in such events: “A diversion is a ﬂight that is

required to land at other than its original destination for reasons

beyond the control of the pilot/company, e.g., periods of signiﬁ-

cant weather. Diversion r ecovery is an initiative orch estrated by

the ATCSCC (Air Trafﬁc Control System Command Center)

and system users to minimize the impact of system disruption.

Diversion recovery will be utilized during a nd after periods

of signiﬁcant weather or other phenomena that have adversely

impacted the system resulting in ﬂight diversions. The goal of

the d iversion recovery initiative is to ensure that ﬂights, which

have already been penalized by having to divert to another

airport, do not receive add itional penalties or delays. Flights

identiﬁed for diversion recovery must receive priority handling

over other ﬂights from their point of departure.

Diversion ﬂights are identiﬁed by having “DVRSN” in the

Remarks section of the ﬂight plan, or the user inputs the

information into the Diversion Recovery Tool (DRT).

The ATCSCC must: implement diversion recovery; transmit

an advisory to inform both ﬁeld facilities and users that a

diversion recovery initiative has been implemented and the

DRT has been activated; adjust the initiative as necessary to

meet changing conditions; transmit an advisory when the DRT

has been deactivated.

The ARTCCs (Air Route Trafﬁc Control Center) must: im-

plement diversion recovery as directed by the ATCSCC; notify

the ATCSCC if they do not intend to use the DRT. In such

cases, the ATCSCC must send the Center a general message

with the information, every 60 minutes until diversion recovery

is no longer in effect; provide expeditious handling in returning

to the system those ﬂights identiﬁed by the ATCSCC/DRT as

diversion ﬂights; forward user diversion recovery requests to

towers and TRACONs.

Towers and TRACONs must: provide expeditious handling

in returning to the system those ﬂights identiﬁed by the

ARTCC/DRT as diversion ﬂights; notify the overlying ARTCC

TMU if they will utilize the DRT.”

C. IATA Diversions Management

A representative of IATA summarized the criteria for se-

lecting airports where diverted ﬂights land as follows: Safety

of Flight, Airspace or Airport Restrictions, Overﬂight autho-

rization, Landing authorization, Immigration, Customs, Air-

port Services, Crew considerations, Service recovery options,

Schedule recovery options. More precisely, the Safety of Flight

includes choosing the emergency airport (nearest available and

nearest suitable), evaluating the fuel remaining and getting

thealternateapproved.Theprimary objectives are to safely

land and support the aircraft. The diversion airport is selected

based on the following criteria: approved alternate; weather

at diversion airport; airport services company, maintenance,

fuel; aircraft servicing—tow bar, air stairs, main deck loader,

ground power, air conditioning, parking; passenger handling

facilities, Customs and Immigration, food, accommodations;

other scheduled service.

It should be best p repared to handle the aircraft, service the

customer, return the ﬂight to original destination. The crew

aspects are also taken into account, such as on-duty times, the

legal to ﬁnish limitations, the accommodations, the replacement

crew availability and the crew pairing disruptions. Regarding

service recovery, passenger disruption is examined regarding

the delay to ﬁnal destination, the 3 hour tarmac rules and the

Customs and Immigration requirements, onward connections,

company re-schedule options, and re-booking schedule options.

IV. P

ASSENGER MULTIMODAL REROUTING FROM

AIRPORT S WHERE THE DIVERTED FLIGHTS LANDED

This section studies how multimodal rerouting of passengers

could have helped in the recovery process.

Hansen and Zhang [27] conducted a study on bus charter

companies’ response to service inter-modal service requests.

To evaluate how promptly charter companies could respond

to service requests, they conducted a telephone survey for

ten randomly picked charter companies for six regions in the

2704 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

Fig. 2. Intermodal service times reported by Hansen and Zhang.

US: San Francisco, Los Angeles, New York, Chicago, Miami

and Texas. All of the regions are supposed to have a charter

companies’ offer comparable, if not bigger, to San Francisco.

They constructed a scenario motivating an urgent request for

a motor coach service at an airport, and asked for a motor

coach that could accommodate at least 30 passengers and their

personal belongings and be available for at least 6 hours. Their

results are shown in Fig. 2 and show that buses can be made

available at most airports in less than four hours.

After the Asiana crash, and considering the airports that wel-

comed most ﬂight diversions, the airports from which a complete

inter-modal substitu tion is feasible are: Sacramento Interna-

tional Airport (SMF) (100 miles from SFO); Reno Tahoe In-

ternational Airport (RNO) (230 miles from SFO); Los Angeles

International Airport (LAX) (390 miles from SFO); Las Vegas

McCarran International Airport (LAS) (565 miles from SFO).

The Department of Transportation proposes an approach to

measure the hourly values of travel time for aviation passen-

gers. These values are used by the FAA, and are not to be

updated for changes in price levels. The present analysis study

only examines what could have been the best case scenario, in

hindsight, based on the data recorded for this disruption.

A. Model Formulation

1) Nomenclature Used in the Model: Let us deﬁne the input

sets as follows:

• A = a

,...,a

be the set of diverted airports (RNO,

SMF, LAS, LAX).

• F = f

,...,f

be the set of departure ﬂights from the

diverted airports to the Bay Area (SFO, OAK, SJC).

• Γ=t

,...,t

the set of discrete time periods.

• P = p

,...,p

the set of diverted passengers.

• A = g

,...,g

MaxAircraft

the set of aircraft available

to charter in the diverted airport.

• B = b

,...,b

MaxBuses

the set of available buses for

inter-modal substitution in the d iverted airport.

Input time and delay variables:

• ADT

: Actual departure time of ﬂight f from diverted

airport a.

• ADivAT

: Actual arrival time of passenger p at the di-

verted airport a.

•BDT

OAK

, BDT

SJC

, BDT

: Bus driving time from OAK

to SFO, from SJC to SFO and from the diverted airport a

to SFO.

• FlightTime

: Flight time from diverted airport a to SFO.

Input capacity variables:

• Seats

is the number of seats left on maintained ﬂight f.

• CapAircraft

is the passenger capacity of chartered

ﬂight g.

• CapBus is the passenger capacity of any bus.

Input cost coefﬁcients:

• CostCharter: Cost of chartering a new aircraft [$/ hour ·

passenger].

• CostBV: Cost of bus reaccommodation [$/passenger].

• CostP: Passenger delay cost per time unit [$/ hour ·

passenger].

Other input coefﬁcients:

• β

Wait

: Weight coefﬁcient f or passen ger waiting time.

• β

Transp

: Weight coefﬁcient for passenger reaccommoda-

tion time.

• MinloadBus: Minimum passenger load (percentage) to

allow a bus to depart.

• MinloadCharter: Minimum passenger load (percentage)

to allow an aircraft to depart.

• TimeFactor: Conversion factor used to convert time peri-

ods into minutes. 15 minutes time intervals are chosen.

•MaxBuses

: Maximum number of buses available at di-

verted airport a.

• MaxAircraft

: Maximum number of aircraft available to

be chartered at diverted airport a.

Input binary variables:OAK

= 1 if the destination of

departure ﬂight f from diverted airport a is OAK, OA K

= 0

otherwise. SJC

= 1 if the destination of departure ﬂight f

from diverted airport a is SJC, SJC

= 0otherwise.

Output binary variables:

The ﬁrst type of output binary variables are Squeeze

p,f,a

Subst

p,b,a

and Charter

p,g,a

. These three variables assign

passengers to one of the three possible rerouting options:

Squeeze

p,f,a

= 1 if passenger p is squeezed into ﬂight f depart-

ing from d iverted airport a in time interval t; Subst

p,b,a

= 1if

passenger p is rerouted with motor coach b from diverted airport

a in time interval t;ACharter

p,g,a

= 1 if passenger p is rerouted

with chartered ﬂight g from diverted airport a in time interval t.

The second type of output binary variables are DTBus

b,a

and

DTCharter

g,a

, indicating when a bus o r an chartered aircraft

leaves diverted airport a.DTBus

b,a

= 1ifbusb departs in time

period t;DTCharter

g,a

= 1 if chartered aircraft g departs in

time period t.

2) Model Input Data: The input data for the mathematical

programming is the following:

• Set of diverted passengers to the airport of study.

• Set of departure ﬂights F from the diverted airport to

SFO, OAK or SJC.

• Number of passengers booked and capacity of each

ﬂight f.

• Scheduled and actual departure and arrival times of each

ﬂight f.

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2705

Airlines’ operations are difﬁcult to optimize as a whole due

to the interaction of m any factors and feasibility constraints of

different resources. Four main constraints a ffect the feasibility

of airline planning and disruption management: aircraft main-

tenance checks, pilot work rules, ﬂeet assignment and passen-

ger accommodation. Therefore, the following assumptions are

made to ensure an admissible problem com plexity: connecting

passengers will connect to their ﬁnal destination from the Bay

Area; there are o nly a limited number of aircraft available to

be chartered; when the rerouting is done through the alternative

airports in the Bay Area, only 80% of the passengers will be

rerouted to SFO; the model does not take into account aircraft

maintenance checks and pilot work rules, nor that pilots and

crew are eligible to continue their scheduled tasks for 135

maximum hours of service; it is assumed there is enough arrival

capacity for the chartered ﬂights to land in the Bay Area.

3) Objective Function: The objective of the mathematical

model is to minimize the cost of reaccommodation of di-

verted passengers. The input data is the actual schedule on

July 6th, 2013 (e.g. what ﬂights were diverted, which ﬂights

were cancelled and which ones could reach SFO), and the

model computes the a cost-effective way to bring passengers

from the diversion airport to their ﬁnal destination, SFO. The

reaccommodation takes into account the following costs: pas-

sengers delay cost while remaining at the diverted airport, the

cost of squeezing passengers into remaining seats on ﬂights to

the Bay Area, the cost of chartering an aircraft to ferry back

diverted passengers, the cost of transporting passengers with

motor-coaches, either from the diverted airport, or just within

the Bay Area. At the end of the chosen time horizon, no diverted

passengers must remain in the diverted airport. The optimiza-

tion minimizes the value of the following objective function:



[CSqueeze

+ CSubst

+ CCharter

]. (1)

Cost of squeezing passengers into departure ﬂights

CSqueeze



Squeeze

(p,f,a)

× [FlightTime

transp

CostP



ADT

− ADivAT



CostP β

wait

+ CostBV



SJC

+ OAK





BDT

OAK

· OAK

+ BDT

SJC

· SJC



× β

transp

CostP] . (2)

The ﬁrst term computes the passengers waiting time before

being r eaccommodated, translated to economic terms with the

passenger value of time (CostP) and weighted with the variable

Wait

. The second term adds the operational costs of using

motor-coaches, in case a passenger is reaccommodated with

remaining seats on ﬂights to Bay Area airports. The third

term evaluates the economic value of the ground transportation

times, weighted by the variable β

Transp

Cost of reaccommodation via ground transportation sub-

stitution

CSubst



Subst

(p,b,a)

[BDT

wait

CostP



t · TimeFactor − ADivAT



CostPβ

wait

+ CostBV



. (3)

The ﬁrst term of the equation computes the cost of passengers

waiting time. The time o f arrival to the diverted airport DivAT

is substracted from the departure time of the motor coach. The

second term computes the cost of passenger transportation time,

by multiplying the bus driving time BDT

Div

by the passenger

value of time CostP, weighted by β

transp

. The third term

computes the cost per passenger o f contracting the motor coach

service CostBV.

Cost of chartering an aircraft

CCharter



Charter

(p,g,a)



CostCharter +



t · TimeFactor − ADivAT



× CostP β

wait

+(FlightTime

transp

CostP)] . (4)

The ﬁrst term computes the cost of passengers waiting

time. The time of arrival to the diverted airport DivAT

substracted from the departure time of aircraft a. The second

term computes the cost of passenger transportation time, by

multiplying the ﬂight transportation time FlightTime

DivAirp

the passenger value of time CostP, weighted by β

transp

.The

third term computes the cost per passenger of chartering an

aircraft CostCharter. Additionally, it has been assumed in this

particular rerouting option there is a limited amount of aircraft

available to charter.

4) Constraints: Constraints of squeezing passengers into

scheduled ﬂights The number of passengers squeezed into

ﬂight f in time p eriod t + 1, should be less or equal to the

number of remaining seats:



Squeeze

(p,f,a)

≤ Seats

∀ f, ∀ a. (5)

A passenger at diverted airport a can not be squeezed into

ﬂight f, if the ﬂight has already departed:



t · TimeFactor − ADT



× Squeeze

(p,f,a)

≤ 0 ∀ t, ∀ f, ∀ a, ∀ p. (6)

a) Constraints of the complete inter-modal substitution

option: The number of diverted passengers in airport a as-

signed to each motor coach must be less than or equal to the

motor-coach capacity, at every time interval:



Subst

(p,b,a)

≤ CapBus ∀ b, ∀ a, ∀ t. (7)

2706 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

The bus b and the passengers leaving with this bus leave at

the same time:

DTBus

b,a

≥ Subst

(p,b,a)

∀ p, ∀ b, ∀ a, ∀ t. (8)

The motor coach b contracted for inter-modal substitution can

only depart if it is ﬁlled up to a minimum bus load

If DTBus

b,a

≥ 1, then



Subst

(p,b,a)

≥ MinloadBus

· CapBus ∀ p, ∀ b, ∀ a, ∀ t. (9)

The motor coach b can only depart once from airport a:



DTBus

b,a

≤ 1 ∀ b, ∀ a, ∀ t. (10)

Constraints corresponding to chartering a new aircraft to

ﬂy to one of the Bay Area airports

The number of passengers assigned to a new chartered air-

craft g must be less than the aircraft remaining capacity, at every

time slot:



Charter

(p,g,a)

≤ CapAircraft

∀ g, ∀ a, ∀ t. (11)

The chartered aircraft g and the passengers leaving with this

aircraft leave at the same time:

DTCharter

g,a

≥ Charter

(p,g,a)

∀ p, ∀ g, ∀ a, ∀ t. (12)

An chartered aircraft g can only depart if it is ﬁlled up to a

minimum aircraft load

If DTCharter

g,a

≥ 1, then



Charter

(p,g,a)

≥ MinloadAc · CapAc ∀ p, ∀ g, ∀ a, ∀ t. (13)

This conditional constraint is transformed into a pair of linear

constraints with auxiliar y variables.

An chartered aircraft g can only depart once:



DTCharter

g,a

≤ 1 ∀ g, ∀ a. (14)

Passenger conservation constraints

Each passenger must be assigned to one of three rerouting

options during the time horizon considered



⎡

⎣



Squeeze

(p,f,a)



Subst

(p,b,a)



Charter

(p,g,a)

⎤

⎦

= 1 ∀ p, ∀ a. (15)

Fig. 3. Reaccommodation of diverted passengers on remaining seats in ﬂights

that reached the Bay Area, regardless of their original carrier.

Passengers can only b e assigned to a rerouting option 30

minutes after landing in the d iverted airport:

∀ a, ∀ p, ∀ t, ∀ f,∀ b, ∀ g

[t · TimeFactor − (30 + ActualDivAT

p,a

)]



Squeezed

(p,f,a)

+ Subst

(p,b,a)

+ Charter

(p,g,a)



≥ 0.

(16)

B. Optimization Results

1) Baseline for Study—Reaccommodation of Passengers

Under a Unimodal Scenario (Flying Only): The best case real-

life scenario for most diverted p assengers is to be rebooked

on ﬂights to the Bay Area (SFO, OAK, SJC) that were not

cancelled, on July 6th (crash day) and the following days,

regardless of their original carrier. From passenger tweets, we

do know that some airlines provided shuttles to the Bay Area in

a few diverted airports, but most of them did not. Moreover, if

we consider that a passenger can only be reaccommodated on

later ﬂights operated by his or her original carrier, passengers

who landed in airports where the carrier does not operate would

not have been rebooked. American Airlines policy indicates

that, under special circumstances, passengers may be rebooked

on another carrier. Fig. 3 shows how long it would have taken

for all passengers to be rebooked depending on the airport

they were diverted to: RNO welcomed about 230 passengers,

who would have been rebooked by Monday morning, July 8th;

SMF welcomed about 600 passengers, who would have been

rebooked by Monday evening, July 8th; LAX welcomed about

1030 passengers, who would have been rebooked by Tuesday

morning, July 9th; LAS welcomed about 670 passengers, who

would have been rebooked by Tuesday morning, July 9th.

Moreover, fewer than 300 passengers would have reached the

Bay Area b efore Monday, July 8th at noon, that is, less than

48 hours after the crash. This means most passengers’ expenses

probably included two hotel nights, and two full days of meals.

This cost incurred is not included in the optimization model

proposed, but adds to the beneﬁts of multimodal rerouting

described below.

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2707

Fig. 4. Number of chartered aircraft available and used in each scenario.

Fig. 5. Number of buses used in each scenario.

2) Study of Different Scenarios Involving Various Numbers

of Aircraft Available for Chartering: The input coefﬁcients are

set to the following values:

• β

Wait

= 0.5

• β

Transp

= 1 − Beta

Wait

• MinloadBus = 0.75

• MinloadCharter = 0.9

To understand the role o f available multimodal substitution,

four scenarios are studied. In each of them, the number of

buses available in each airport remains the same, to ensure

that all diverted passengers can be rerouted by bus. However,

the number of aircraft available for chartering varies from zero

aircraft in scenario 1, to one aircraft in RNO, two in SMF,

two in LAX and 2 in LAS. The details for each scenario are

shown in Fig. 4. The maximum number of chartered aircraft

was chosen small, because ﬁrst, capacity at OAK and SJC

would have been limited, even in the evening of the crash, and

second, few aircraft might have been available at the diverted

airports, except the diverted aircraft themselves.

Solving the foregoing optimization problem for the four

scenarios described, the results show different trends. First,

Fig. 5 indicates the number of buses used to reroute passen-

gers to the Bay Area. As the number of aircraf t available for

chartering increases, the number of buses used decreases. Even

though chartering an aircraft is more costly than using a bus,

the fact that aircraft have a larger passenger capacity and that

Fig. 6. Rerouting assignment of passengers in each scenario.

Fig. 7. Average passenger delay in each scenario.

ﬂying provides a much shorter travel time make ﬂying the most

cost-effective option overall. This is conﬁrmed by examining

the rerouting option provided to each passenger in Fig. 6,

showing that chartering a total of ﬁve aircraft in scenario 4

largely changes the proportion of passengers accommodated on

new ﬂights.

The goal of this multimodal o ptimization problem is to

ensure that passengers reach the Bay Area faster than they

would have if they had waited to ﬁll remaining seats on later

ﬂights operated out of the diverted airports to the Bay Area.

Fig. 7 presents the average passenger delay, that is either the

average difference between the arrival time of a passenger in

SFO and his/her scheduled arrival time in SFO or the average

difference between the arrival time of a passenger in SFO and

his/her actual arrival time in the diverted airport. The average

passenger delay is less than six hours, which is small consider-

ing the bus travel time from the diverted airports further away

from SFO. It also highlights that chartering aircraft reduces the

average delay by about an hour. Figs. 8 and 9 show that, across

all scenarios, the differences in passenger departure times is

very small. Moreover all passengers have lef t the diverted

airports before midnight. However, the more chartered aircraft

available, the earlier p assengers arrive in the Bay Area o n

average. It should be noted that, because of bus travel time,

some passengers reach the Bay Area at about 2 am on Sunday.

For these small portion of passengers, a bus departure early

the next day might be preferable and should be considered in

future work.

2708 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

Fig. 8. Cumulative number of passengers by departure time from the diverted

airport, for each scenario.

Fig. 9. Cumulati ve number of passengers by arrival time in the Bay Area, for

each scenario.

Fig. 10. Objective value for each value of β

wai t

3) Sensitivity Analysis on β

wait

: We want to understand the

weighting in the objective function depending on whether we

attach more importance to the delay suffered by passengers

or the cost incurred by reaccommodating them on aircraft or

buses. Therefore we consider scenario 4 from the previous

example, with one chartered aircraft available in RNO, two

in SMF, two in LAX and 2 in LAS, is tested for β

wait

= 0.3,

wait

= 0.4, β

wait

= 0.5, β

wait

= 0.6, β

wait

= 0.7.

Fig. 10 shows that the objective value decreases as β

wait

increases. This is intuitive since the optimization model forces

all passengers to be rerouted. Second, in Fig. 11, the number

Fig. 11. Number of buses used for each value of β

wai t

Fig. 12. Average passenger delay for each value of β

wai t

of buses used by airport increases with β

wait

, while the number

of chartered aircraft used remains constant and the number of

passengers reaccommodated on buses remains the same. This

means that the load in each bus decreases, the optimization is

no longer trying to ﬁll buses to their maximum capacity but

only to ﬁll them to their minimum capacity before letting them

depart. Finally, Fig. 12 indicates that the inﬂuence of β

wait

passenger delay is limited.

V. M

ETROPLEX GATE ASSIGNMENT

In the foregoing section, we studied how to better reac-

commodate diverted passengers.

We wonder if it could have

been possible to land more d iverted ﬂights in the Bay Area,

under metroplex operations, with real-time shared information

between all stakeholders and collaborative decision making. In

a metroplex environment, neighboring airports could dynami-

cally act as reliever airp orts to SFO. Therefore, we ﬁrst examine

the situation, the remaining capacities at OAK and SJC, the

domestic and international diverted ﬂights, and then propose a

reoptimization scheme.

Trajectories of diverted ﬂights arriving in the Bay Area in the

hours following the crash are displayed in Figs. 13 and 14, and

show the unusual number of holding patterns and reroutings in

the air, so that the ﬂights can land at SJC or OAK.

Part of this research was previously published by the authors at the AIAA

Infotech Conference [42].

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2709

Fig. 13. Trajectories of diverted ﬂights in the Bay Area following the Asiana

crash at SFO from 11:30 to 12:30.

A. Inputs

Consider the time period f rom 1 1:30

A.M. (local time at

SFO) to 4:30

P.M.. The horizon is broken down into 15-minutes

Fig. 14. Trajectories of diverted ﬂights in the Bay Area following the Asiana

crash at SFO from 13:00 to 14:00.

time intervals, denoted by t in T. Let each diverted ﬂight f be

in the set F. Let the two other main airpo rts in the Bay Area,

OAK and SJC, be denoted by a in A.

2710 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

TABLE I

ESULTING DELAY PER FLIGHT F OR EACH OPTIMIZATION SCENARIO

For each ﬂight f ∈ F, its scheduled arrival time at SFO is

denoted by arrSFO

, where arrSFO

= t corresponds to time

bin t. The ﬂight estimated arrival time (ETA) to the potential

arriva l airports at the time of the crash is denoted timetoa pt

f,a

corresponding to the arrival time bin. Assume that all ﬂights

stay gate_time time intervals at their gate, where gate_time

enforces constraints on the minimum turnaround time and the

maximum time at the gate allowed by the airports.

Each ﬂight en-route is assumed to arrive at SFO at the ETA

of the TZ table in the ETMS data, at the latest location message

before the crash, to avoid taking into account any change of

destination airport and by consequent an irrelevant ETA to SFO.

This ETA is not available for ﬂights that took off after the

crash (it is unclear if the ETA in the data is for an arrival at

SFO or elsewhere), in this case the ETA at SFO is determined

using the earliest TZ data available, corresponding to the ﬁrst

point after take-off. The result is similar to the scheduled

arrival time plus any actual departure delay. From the ETA to

SFO, an approximation of timetoapt

f,a

can be obtained. The

particularity of a metroplex is to have several airport close to

each other. For the bay area SFO, OAK, SJC are close enough to

have an approach path extremely restricted to avoid any hazard

due to the neighboring airports trafﬁc. The incoming ﬂux ﬁx for

OAK and SFO are only a few miles apart and therefore the ETA

to OAK would be less than 5 minutes apart from the ETA for

SFO. Knowing the precision of the ETA used for this study is

15 minutes, the two ETA can be estimated equal. For SJC the

incoming ﬂux ﬁx is about 30 miles south of the SFO incoming

ﬁxes. The estimated time to go from one ﬁx to the other is

about 10 to 15 minutes. Therefore depending of the origin of

the ﬂight, its ETA would be its ETA to SFO plus an additional

15 minutes if the ﬂight is arriving from the North, or minus

15 minutes when arriving from the South, or its original ETA

when arriving from the East or West.

For each metroplex airport, SJC and OAK, the remaining

runway capacity at each time bin is computed, while taking

into account the actual (not the scheduled) arrival and depar-

tures (including international ﬂights from ETMS), except the

diversions from SFO on that day. The corresponding variable

is rwycapa

a,t

. Moreover, the remaining gate capacity at each

time bin is calculated b y taking into account actual arrival and

departures (including international ﬂights), and referred to as

gatecapa

a,t

. For domestic ﬂights, the BTS data provides the

gate arrival time. For international ﬂights, the gate arrival time

is extrapolated from the runway arrival time (see Table I).

B. Problem Formulation

The objective is to minimize the arrival delay of the diverted

ﬂights in the Bay Area. In a disruptive situation, the goal is

to land aircraft safely as close as possible to their original

destination , i.e., SFO. The model optim izes the arrival times of

each diverted ﬂight and provides the following information. For

each ﬂight f , its runway arrival time is denoted by arr

f,t,a

, its

gate occupancy time is indicated by gateocc

f,t,a

. The objective

function can be formulated as follows:



f ∈F



t∈T



a∈A

t · arr

f,t,a

−



f ∈F

arrSFO

(17)

The following constraints are deﬁned:

Flight Duration Constraints:

The landing time at the actual arrival airport is bounded

above and b elow. A ﬂight f cannot land at the arrival airport

before it has at least ﬂown there. Aircraft are legally required to

be dispatched with enough fuel to allow 45 minutes of airborne

holding [43], therefore reserve time = 2 time intervals, so it has

to land before this time elapses

∀ f,



a∈A

timetoapt

f,a

+reserve time



=timetoapt

f,a

arr

f,t

= 1. (18)

A ﬂight can only land once and at one airport:

∀ f ∈ F,



a∈A



t∈T

arr

f,t,a

= 1. (19)

A ﬂight cannot arrive a gate earlier than 15 minutes after

landing and not later then 45 min [44]

∀ f ∈ F, ∀ a ∈ A, ∀ t ∈ T, if arr

f,t,a

= 1

Then ∀ t

∈ [t + 2,t+ gate_time], gateocc

f,t

= 1

and gateocc

f,t+1,a

+ gateocc

f,t+gate_time+1,a

= 1. (20)

An aircraft remains gate_time intervals time at one gate in one

airport

∀ f ∈ F,



a∈A



t∈T

gateocc

f,t,a

= gate_time. (21)

Airport Constraints

The number of landing ﬂights must be below the remaining

runway capacity:

∀ t ∈ T, ∀ a ∈ A,



f ∈F

arr

f,t,a

≤ rwycapa

a,t

. (22)

The number of ﬂights parked at the gate must be less than the

remaining gate capacity:

∀ t ∈ T, ∀ a ∈ A,



f ∈F

gateocc

f,t,a

≤ gatecapa

a,t

. (23)

C. Results

Each one of the 90 diverted ﬂights initially scheduled to land

at SFO between 11:30

A.M. and 4:30 P.M. could theoretically

have landed in the Bay area. The question remains whether

good communication and information sharing regarding the

remaining capacities at SJC and OAK co uld enable it. Even

in the worse case scenario where each ﬂight occupies its gate

for two hours, OAK and SJC could accommodate the incoming

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2711

Fig. 15. Runway Occupancy at SJC and OAK for each scenario. (a) Runway

occupancy for gate_time = 60 min. (b) Runway occupancy for gate_time =

90 min. (c) Runway occupancy for gate_time = 120 min.

trafﬁc, wh ile keeping airborne ﬂight delay relatively low and

within safety limits.

Fig. 15 shows the runway capacity is stressed right after the

crash, between 11:30

A.M. and 11:45 A.M.. However, even at

its peak, the runway occupancy remains below 90% of capacity.

The second peak is visible for the scenario with 120 minutes

of gate time, and shows a short saturation of gate occupancy at

both SJC and OAK, which leads to a temporary and manageable

increase in airborne holding. The gate occupancy level reaches

its limit for two scenarios and for the last scenario, the peak

Fig. 16. Gate Occupancy at SJC and OAK for each scenario. (a) Gate Occu-

pancy for gate_time =60min; (b) Gate Occupancy for gate_time = 90 min;

occupancy reached is 90% (see Fig. 16). Summarizing it all,

most diversions, if not all, with some uncertainty margin, could

have landed in the Bay Area.

The repartition of diverted ﬂights at SJC and OAK is similar,

and balance each other, as illustrated in Figs. 17 and 18. The

larger the gate time enforced, the more balanced the repar-

tition is. A noticeable fact is that no ﬂights lands between

12:30

P.M. and 12:45 P.M. because all gates are full in the next

time interval. Since the limit in gate o ccupancy lasts more than

30 minutes in the second and third scenarios, the taxi-in time

2712 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

Fig. 17. Number of ﬂights landing at SJC and OAK, and associated delays for

each scenario. (a) Delays for gate_time = 60 min. (b) Delays for gate_time =

90 min. (c) Delays for gate_time = 120 min.

margins are not enough to mitigate the peak, and therefore air-

borne holding time is used to delay aircraft. However, as shown

in Fig. 19, the airborne holding times are below 15 minutes

and well below safety limits and reserve times. Because of

the objective function that minimizes arrival d elay, the aircraft

land as soon as possible, favoring taxi-in d elay over airborne

holding. Fig. 20 illustrates when the taxi-in modulations are

not sufﬁcient. The larger the gate occupancy, the more airborne

holding time is allocated, but it remains low. Also it should be

Fig. 18. Number of ﬂights landing at SJC and OAK, and associated delays

for each scenario. (a) Number of ﬂights landing for gate_time = 60 min.

(b) number of ﬂights landing for gate_time = 90 min. (c) number of ﬂights

landing for gate_time = 120 min.

noted that a few ﬂights arrive a few minutes earlier at OAK

or SJC than their scheduled time at SFO, but this comes from

the fact that their ﬂight time slightly diminishes, for the ﬂights

coming from the East for instance.

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2713

Fig. 19. Trade-offs between Airborne holding at SJC and OAK for each

scenario. (a) Average Taxi vs airborne holding time for gate_time = 60 min;

(b) Average Taxi vs airborne holding time for gate_time = 90 min; (c) Average

Taxi vs airborne holding time for gate_time = 120 min.

Overall, the optimization performed demonstrates that most

diverted ﬂights could have landed in the Bay Area, even with

uncertainty present. However, it is understandable that safety

was a priority and that grounding ﬂights wherever possible

when the situation was unclear was th e best solution. With fu-

ture concepts of operations, including better information shar-

ing throughout the system between airports, carriers, regarding

estimated arrival times and remaining capacities, it is expected

Fig. 20. Average Flight Delay, Airborne Holding and Taxi-in Times at SJC and

OAK for each scenario. (a) Delay, taxi and airborne holding time e volution for

gate_time = 60 min. (b) Delay, taxi and airborne holding time evolution for

gate_time = 90 min. (c) Delay, taxi and airborne holding time ev olution

for gate_time = 120 min.

that crisis management will be handled better and avoid the

diversions of thousands of passengers to far-away airports,

causing many issues related to reaccommodation, customs, and

carrier support to name a few.

2714 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

VI. RECOMMENDATIONS FOR T HE ADOPTION OF

MULTIMODAL COLLABORATIVE DECISION MAKING

The Asiana crash constitutes one of many adversarial events

that signiﬁcantly disrup t the passengers and airlines alike. In

this context, it contributes factual information supporting how

operations could be improved under what we call a “multimodal

CDM” concept of operations.

A. Stakeholders and Action Plans

1) Passengers: The pivotal role of the passenger, though

obvious, b ecomes clear when considering the role of the main

stakeholders involved in the door-to-door travel experience in

normal and disrupted conditions. Only passengers (or their

luggage) interact with all stakeholders. Other stakeholders have

one or a few connections up or down the line, but they do not

have an immediate operational reason to be aware of the needs,

priorities and issues facing the full run of stakeholders involved

in the journey process. Passengers can differ signiﬁcantly in

their travel behavior, requirements and preferences.

a) Timing of passenger information: The accuracy of

exchanged data is c ritical to enable informed decisions for im-

proved travel. In the case of a long journey some uncertainties

normally will cancel each other out which could be estimated

with error propagation. However, some sources of delay may

affect multiple stages, lead ing to greater-than-usual journey

times throughout the journey. The timeliness of data exchange

is very impor tant for empowering the traveler, and enables the

travel service provider to make a good prognosis of the progress

of the journey. The later information is exchanged, the more

limited will be the availability of alternatives and/or counter-

measures. For example, if the traveler received information on

congestion on his or her way to the airport too late and is already

within a tr a fﬁc jam on the highway, either this delay can be

absorbed through journey time buffers, or processes later in the

journey could be shortened to allow the traveler to board their

critical travel connection in time. If neither of these options

were possible, the journey would have to be re-planned. For

instance, when SFO closed after the Asiana crash, passengers

with later ﬂights were still trying to go to the airport.

A disruption corresponds to an episode that results in many

cancellations at one or more airports, for example, major snow

events, volcanic ash, aircraft accidents, strikes, technical fail-

ures, ﬁres or terrorism. Such situations differ from events that

lead primarily to delays.

The short term beneﬁts of multimodal CDM are the im-

provement of passenger satisfaction by reducing door-to-door

travel time, reducing uncertainty, and improving information

provision.

2) Airlines: Airlines bear a high cost during disruptive

events. Their objective is to recover from the disruption quickly

and efﬁciently, while supporting passengers’ needs. If airlines

were willing to share their connection matrices under disrupted

conditions, the diversions might be able to land in airports more

suited to facilitate the reaccommodation of diverted passengers.

For instan ce, they could land closer to the ﬁnal destination

airport. A ranking process, similar to CTOP (Collaborative

Trajectory Options Program) [45], could be envisioned. For

ﬂights still far away from the arrival airport that closes, but

already airborne, the airline could be asked to provide a ranking

of preferred d iverted airports. Whether airlines can provide

complete or partial preference information, the details of the

expression of such preferences remain to be explored.

The short term beneﬁts of multim odal CDM for airlines

include reducing congestion in airport terminals, particularly at

airline counters, both under normal conditions (as passengers

spend less unnecessary time in the terminal) and in disruptive

situations situations.

3) Airports, Ground Transportation, Infrastructures and

Transportation Authorities: The short term beneﬁts of mul-

timodal CDM consist in: helping airlines to better maintain

schedules by reducing the uncertainty associated with late

passenger arrival at the gate; allowing stakeholders to optimize

resource allocation (for example, improving prediction of how

many immigration desks will need to be open at a given time in

a g iven airport).

Multimodal issues and bottlenecks resulting from operational

processes, deﬁciencies in existing technology, and lack of

information exchange have to be examined using the avail-

able data sources. Given the level of complexity encountered

in the multimodal transportation system, advanced methods

could be used to explore uncertainty issues and networked

interdependencies in order to reveal both the current issues

and bottlenecks which have not yet occurred, but may do in

future due to currently-foreseen system changes (e.g. increases

in demand). Public Transport authorities need to be involved

to better manage the interconnections between transportation

modes. Private transport companies (taxi for instance) can be

locally contracted. For instance, the BART operates a reduced

scheduled on Saturdays. It may be possible to have more trains

operate a given origin-destination pair between airports in the

metroplex to facilitate passenger movements on the ground.

Regarding cooperation between airlines and ground transport,

Evans [28] reported that Continental Airlines had an agreement

at Newark to make buses available to air p assengers in case of

severe disruptions.

B. Support Tools for Multimodal Collaborative

Decision Making

An analysis of how the conclusions of research in the pre-

vious areas may change under disruptive conditions, would

help in setting recommendations for how data provision and

exchanges would need to be modiﬁed when the aviation system

is disrupted. A key research and development area identiﬁed

is the need to further foster solution s that enable a seamless

door-to-door journey for the passenger. Due to signiﬁcant bar-

riers related to incompatibilities between systems and data,

it is recommended to focus on the development of direct

communications between the passengers and each transport

provider. Technical and algorithmic aspects n eed to be ad-

dressed, they inclu de computing abilities of decentralized

solutions, ﬂeet needs, communication channels, human

decision-making support tools. Provision of door-to-door travel

support, such as alternatives in case of ﬂight cancellation, could

be implemented as a de-centralized service.

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2715

1) Information Sharing: According to SITA Air Transport

IT review [46], “the air transport industry is shifting towards a

new era of continuous engagement. (...) it is creating rising

demand for more relevant services and giving airlines and

airports opportunities to offer passengers enhanced personal-

ization.” This SITA survey indicates that “airlines” sights are

set on providing a real-time service experience, targeted at their

passengers’ journeys, via smartphone apps: 65% of airlines

plan to do this by the end of 2017, up from 13% today. High

on the airport agenda are updates on wait-times and local

trafﬁc issues: 18% offer them today, with another 55% making

plans for this service.” Moreover, in the recovery process,

“[passeng ers] are expecting airlines and airports to provide

a personal alert and response when ﬂights are canceled or

delayed.” Smartphones and other communication technologies

open up a myriad of opportunities to provide disruption man-

agement services to passengers. The use of new technologies is

becoming more and more commonplace in case of unexpected

events: “A third of airports are able to p rovide real-time infor-

mation to passenger mobiles in the event of disruption, with

a further third doing so by 2017.” The continuous provision

of information to the passenger is facilitated by the fact that

“almost every passenger (97%) carries a smartphone, tablet or

laptop.”

Recent events such as the Asiana Crash and the ensuing

multimodal ripple effects clearly illustrated the fragility of the

system, the costs associated with not reacting effectively and

hence the need of coordination. The FAA, the EC, Eurocontrol

and others have responded positively to mitigate d isruptive

events and spread the CDM concept but more could be done,

such as: delivery protocols that enable levels of ﬁltered alert in-

formation to be passed through the network; a web “dashboard”

of status information to which stakeholders could contribute,

which provides real-time information to all stakeholders re-

questing access. For instance, CDG website provides real-time

on-time performance reports to anyone in the air transportation

industry that previously requested an account. Airlines across

the world use it to monitor their ﬂights and evaluate if their

passengers are likely to make their connections; the establish-

ment of intelligence/alert units that can capture non-operational

features such as meteorological or security data and make them

available to the network.

The identiﬁcation o f existing data availability, technology

and data ﬂows is necessary. To accurately evaluate perfor-

mance, the available data from many data sources and reporting

methods needs to be understood as a whole. Unless given

incentives or provided with potential beneﬁts, stakeholders are

concerned that b y sharing their data they are submitting them-

selves to open comparison with co mpetitors. Data provisio n and

analysis could also be a way to enable multimodal ticketing,

which could help signiﬁcantly streamline multimodal journeys.

The duality between competition and cooperation can be an

obstacle to multimodal CDM. The proposed concept involves

information exchange between various stakeholders who may

be competing. The different data sources, their availability,

and aspects of conﬁdentiality have to be investigated. A trade-

off between the performance o f the solution of a multimodal

network optimization and constraints in data provision should

be established. Antitrust concerns should be addressed as well.

2) Resource Allocation: Several limits to passenger reac-

commodation need to be addressed. The capacity of other air

services to provide spare seats for passengers from canceled

ﬂights is a key factor affecting recovery from crisis events. In

normal oper ations, airlines try to maximize their load factors.

In crisis events, however, faster recovery is aided by lower load

factors on subsequent ﬂights. Putting passengers on ground

transport is feasible only if there are enough seats at suitable

times. The ability of ground transportation p roviders to support

stranded p assengers varies greatly by location. Several limita-

tions may arise, includ ing lack of spare rolling stock, staff avail-

ability and training for the routes needed, an d inf rastructure

limits. Multimodal CDM could be decomposed in to o ptions,

with several multimodal touch points, such as the BART station

at OAK airport. In the context of metroplex operations under

disruptions, playbooks could be developed, to provide structure

and guidance. Cases such as a sudden and unforeseeable airport

closure could be addressed.

Zhang et al. proposed a Regional Ground Delay Program

(Regional GDP) [47]. When a hub airport located in a regional

airport system encounters a severe airside capacity reduction,

air trafﬁc ﬂow managers could not only evaluate the imbalance

of trafﬁc demand and terminal capacity at the hub airport but

also excess capacity at other airports in the same region, as-

suming that airlines could incorporate ground modes into their

disruption management and use ground vehicles to transport

passengers and crew members between original scheduled and

diverted airports. The case study of the Asiana Crash suggests

that a regional GDP, including LAX, LAS, SMF and RNO air-

ports, may have helped mitigate the d isruption. The feasibility

of multimodal CDM could be facilitated through the existing

and successful initiatives constituted by GDPs and CDM.

Moreover, mathematical models, supported by the appro-

priate data sources and processes, and feeding into decision-

support tools, need to be developed, for different time

horizons, from strategic to tactical planning. Algorithms sup-

porting distributed multimodal optimization with reasonable

computational times and robust margins need to be studied and

implemented.

3) Decision Making: Multimodal collaborative Decision

Making necessarily involves multiple stakeholders. The sys-

temic nature of aviation means that those stakeholders are

international as well as national and local. In the case of the

Asiana crash, not only was SFO airport involved and all entities

dealing with airport closur e and passenger evacuation, but also

the California Highway Patrol, Taxi Companies operating out

of SJC and OAK, custom and border representatives at any

airport where ﬂight diversions landed, to name a few. Similarly,

working together, these organizations can enhance the passen-

ger experience in normal as well as disrupted conditions. The

key to delivering effective multim odal CDM is communication.

Communication is a means to an end: it will h elp improve

decision-making, and the data sharing will support the use of

decentralized optimization models. The structure o f the deci-

sion space for all stakeholders needs to be better understood.

There could be a pre-established metroplex playbook indicating

which stakeholder to involve at which stage of the decision

process. Real-time, tactical and strategic decision-making may

require the involvement of different stakeholders. Efﬁciency,

2716 IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, VOL. 17, NO. 10, OCTOBER 2016

whether cost efﬁciency or time efﬁciency for instance, of

the decision-making aims at making the passenger journey

seamless or recovering from a disruption as fast as possible.

Multimodal CDM could also be tied to SWIM (System Wide

Information Management), in the sense that it would provide a

broad base information management system including passen-

gers and local transportation networks.

The timing of decisions on the Air Trafﬁc Control side could

be investigated. By comparing the initial ﬂight plan and the

trajectories followed by diverted aircraft, the timing of the

diversions could be retrieved. More might be uncovered on

the tactical trafﬁc control aspects for the entire airspace.

VII. C

ONCLUSION

The present paper aimed at making the case for the extension

of Collaborative Decision Making to the Multimodal Network

level. It tackled, in hindsight, how the disrup tion caused by the

Asiana Crash could have been better managed, at the system

level. The consequences of the crash may have been better

mitigated, for both the stakeholders and passenger s, had Multi-

modal Network CDM been in place. Two optimization models

were developed to improve the crisis management following the

crash. The passenger-centric optimization aimed at balancing

cost and delays with a multimodal reaccommodation scheme

from each diversion airpor t. It showed that multimodal col-

laboration to reroute passengers could have helped passengers

within an 8-hour bus drive radius reach the Bay Area on the

crash day, instead of waiting up to several days for ﬂights

in diverted airports. The ﬂight-centric optimization aimed at

allocating ﬂight diversions to SJC and OAK while balancing

runway and gate capacity, and minimizing ﬂight delays. It

showed that there was potentially more capacity at SJC and

OAK to accommodate more diverted ﬂights on the crash day,

which could have mitigated many of the ripple effects for both

passengers, airlines and airports. One of the m ain obstacles to

optimal capacity utilization in crises is information sharing and

collaborative decision making between all stakeholders. This

would improve the performance of the air transportation system

both from a ﬂight-centric and a passenger-centric perspective.

Then recommendations were elaborated to expand CDM to the

multimodal network level and hig hlighted the expected beneﬁts

for all stakeholders and passengers.

The higher-level goal of this paper is to foster a better under-

standing of multimodal transportation to increase its resilience

and facilitate the passenger door-to-door journey. This research

can provide the ﬁrst experimental basis upon which several

system engineering methods could be applied to improve the

entire passenger journey.

EFERENCES

[1] Annual Passenger Total Approaches 3 Billion, International Civil Aviation

Organization (ICAO), Montreal, QC, Canada, 2012.

[2] Flightpath 2050—Europe’s Vision for Aviation (Report of the High Level

Group on Aviation Research), European Commission, Brussels, Belgium,

2011. Accessed on: Jul. 18, 2014. [Online]. Available: http://ec.europa.eu/

transport/modes/air/doc/ﬂightpath2050.pdf

[3] White Paper: Roadmap to a Single European Transport Area—Towards

a Competitive and Resource Efﬁcient Transport System, European Com-

mission, Brussels, Belgium, 2011. [Online]. Available: http://ec.europa.

eu/transport/themes/strategies/2011_white_paper_en.htm

[4] World Economic Forum, “Connected world: Transforming travel,

transportation and supply chains,” Cologny, Switzerland, 2013. [Online].

Available: http://www3.weforum.org/docs/WEF_MO_ConnectedWorld

_Report_2013.pdf

[5] Smart Travel: Unlocking Economic Growth and Development Through

Tra vel Facilitation, World Economic Forum, Cologny, Switzerland,

2014. [Online]. Available: http://www3.weforum.org/docs/GAC/2014/

WEF_GAC_TravelTourism_SmartTravel_WhitePaper_2014.pdf

[6] Airline On-Time Statistics and Delay Causes, U.S. Department of Trans-

portation, Washington, DC, USA, 2009. [Online]. Available: http://www.

transtats.bts.gov/ot_delay/ot_delaycause1.asp

[7] A. Mukherjee, M. Ball, and B. Subramanian, “Models for estimating

monthly delays and cancellations in the NAS,” presented at the NEXTOR

National Airspace System Performance Metrics Conf., Paciﬁc Grove, CA,

USA, 2006.

[8] N. Pyrgiotis, K. Malone, and A. Odoni, “Modelling delay propagation

within an airport network,” Transp. Res. C, Emerging Technol., vol. 27,

pp. 60–75, Feb. 2011.

[9] N. Nayak, “Estimation of the impact of single airport and multi-airport

system delay on the National Airspace System using multi variate simulta-

neous models,” Ph.D. dissertation, Dept. Civil Environ. Eng., Uni v. South

Florida, Tampa, FL, USA, 2012.

[10] S. AhmadBeygi, A. Cohn, Y. Guan, and P. Belobaba, “Analysis of the

potential for delay propagation in passenger airline networks,” J. Air

Transp. Manage., vol. 14, no. 5, pp. 221–236, Sep. 2008.

[11] N. Jafari and S. Hessameddin Zegordi, “Simultaneous recovery model for

aircraft and passengers,” J. Franklin Inst., vol. 348, no. 7, pp. 1638–1655,

Sep. 2011.

[12] J. D. Petersen, G. Sölveling, J.-P. Clarke, E. L. Johnson, and S. Shebalov,

“An optimization approach to airline integrated recovery,” Transp. Sci.,

vol. 46, no. 4, pp. 482–500, Nov. 2012.

[13] NextGen Implementation Plan 2015, Federal Aviation Administration,

Washington, DC, USA. Accessed on: Apr. 6, 2016. [Online].

Available: https://www.faa.gov/nextgen/media/NextGen_Implementation

_Plan-2015.pdf

[14] C. Schumer and C. Maloney, Your Flight has Been Delayed Again: Flight

Delays Cost Passengers, Airlines, and the US Economy Billions,” The

U.S. Senate Joint Economic Committee, Washington, DC, USA, 2008.

[15] Concept of Operations for the Next Generation Air Transportation Sys-

tem, vol. 3, Joint Planning and Development Ofﬁce, Washington, DC,

USA, 2007.

[16] S. Atkins, “Observation and measurement of metroplex phenomena,” in

Proc. IEEE/AIAA 27th DASC, 2008, pp. 3.E.2-1–3.E.2-15.

[17] L. Ren et al., “Contrast and comparison of metroplex operations, pre-

sented at the 9th AIAA Aviation T echnology, Integration, Operations

Conf., Hilton Head, SC, USA, 2009, Paper AIAA 2009-7134.

[18] J.-P. Clarke et al., “Ev aluating concepts for metroplex operations,” pre-

sented at the AIAA Aviation, Technology, Integration, Operations Conf.,

Fort Worth, TX, USA, 2010, Paper AIAA 2010-9249.

[19] L. Li, J. W. Park, and J.-P. Clarke, “A simulation-based method for es-

timating metroplex efﬁciency,” in Proc. IEEE/AIAA 30th DASC, 2011,

pp. 2E4-1–2E4-10.

[20] D. DeLaurentis, S. Landry, D. Sun, F. Wieland, and A. Tyagi, “A concept

for ﬂexible operations and optimized trafﬁc into metroplex regions,” Nat.

Aeron. Space Admin., Langley Res. Center, Hampton, VA, USA, NASA

CR-2011-217302, 2011.

[21] M. Wambsganss, “Collaborative decision making through dynamic infor-

mation transfer,” Air Trafﬁc Control Quart., vol. 4, no. 2, pp. 107–123,

1996.

[22] K. Chang et al., “Enhancements to the FAA ground-delay program under

collaborati ve decision making,” Interfaces

, vol. 31, no. 1, pp. 57–76,

Jan./Feb . 2001.

[23] M. O. Ball, C.-Y. Chen, R. Hoffman, and T . Vossen, “Collaborative

decision making in air trafﬁc management: current and future research

directions,” in New Concepts and Methods in Air Trafﬁc Management.

Berlin, Germany: Springer-Verlag, 2001, pp. 17–30.

[24] P. Burgain, E. Feron, and J.-P. Clarke, “Collaborative virtual queue:

Beneﬁt analysis of a collaborative decision making concept applied

to congested airport departure operations,” Air Trafﬁc Control Quart.,

vol. 17, no. 2, pp. 195–222, 2009.

[25] G. Gupta, W. Malik, and Y. C. Jung, “An integrated collaborative de-

cision making and tactical advisory concept for airport surface opera-

tions management,” presented at the 12th AIAA Aviation Technology,

Integration, Operations Conf., Indianapolis, IN, USA, 2012, Paper AIAA

2012-5651.

[26] T. Bolic and Z. Sivcev, “Eruption of Eyjafjallajökull in Iceland: Experi-

ence of European air trafﬁc management,” Transp. Res. Rec., no. 2214,

pp. 136–143, 2011.

[27] Y. Zhang and M. Hansen, “Real-time intermodal substitution: Strategy

for airline recovery from schedule perturbation and for mitigation of

airport congestion,” Transp. Res. Rec., J. Transp. Res. Board, no. 2052,

pp. 90–99, 2008.

MARZUOLI et al.: DISRUPTION MANAGE MENT WITH MULTIMODAL COLLABORATIVE DECISION-MAKING 2717

[28] A. D. Evans, “Responses to airport delays: A system study of new a rk inter-

national airport,” Ph.D. dissertation, Dept. Aerospace Eng. MIT,

Cambridge, MA, USA, 2002.

[29] A. Cook, G. Tanner, S. Cristóbal, and M. Zanin, “Passenger-oriented

enhanced metrics,” presented at the 2nd SESAR Innovation Days,

Braunschweig, Germany, Nov. 27–29, 2012.

[30] S. Bratu and C. Barnhart, “Flight operations recovery: New approaches

considering passenger recovery,” J. Sched., vol. 9, no. 3, pp. 279–298,

Jun. 2006.

[31] D. Wang, “Methods for analysis of passenger trip performance in a com-

plex networked transportation system,” Ph.D. dissertation, Dept. Syst.

Eng. Oper . Res., George Mason Univ., Fairfax, VA, USA, 2007.

[32] M. Ball, “Analysis of passenger delays: Developing a passenger delay

metric,” presented at the NEXTOR National Airspace System Perfor-

mance Metrics Conf., Paciﬁc Grove, CA, USA, 2006.

[33] National Transportation Safety Board, “Public hearing ﬁnal transcript,

investigation of the Asiana Airlines Flight 214 landing accident at

San Francisco International Airport, California, July 6, 2013,” Washington,

DC, USA, 2014. [Online]. Available: http://dms.ntsb.gov/pubdms/search/

document.cfm?docID=413233&docketID=55433&mkey=87395

[34] San Francisco Crash Sends Ripple Effect to Seattle, King 5 News, Seattle,

WA, USA, 2013. [Online]. Available: http://www.king5.com/story/tech/

science/aerospace/2014/08/04/13272742/

[35] Some Flights Scheduled to Land at SFO Diverted to LAX. [On-

line]. Available: http://losangeles.cbslocal.com/2013/07/06/some-ﬂights-

scheduled-to-land-at-sfo-diverted-to-lax/.UdilzcAoKEM.twitter

[36] Flights Forced to Land at Sacramento Airport After SFO Shuts Down.

[Online]. Available: http://www.news10.net/story/local/2013/07/06/

4755695/

[37] “Domestic and international ﬂights diverted to Sacramento following San

Francisco plane crash,” CBS Sacramento, Sacramento, CA, USA, 2013.

[38] E. K. Dan Nakaso and B. Newman, “SFO crash causes chaos in

California skies,” Willits News, Willits, CA, USA, 2013. [Online].

Available: http://www.willitsnews.com/ci_23610546/sfo-crash-causes-

chaos-california-skies

[39] Aircraft Diversions at SJC Due to Asiana Accident at SFO. [Online].

Available: http://www.piersystem.com/external/content/document/1914/

1841923/1/07-09-13AIRPORT.PDF

[40] T. D. Larsen, “A behavioral framework for managing massive airline ﬂight

disruptions through crisis management, organization development, and or-

ganization learning,” M.S. thesis, Dept. Manage., Eng., Aerosp., Transp.,

Uni v. Maryland Univ. College, Upper Marlboro, MD, USA, 2013.

[41] Jenkins and Cotton, “Improving airline proﬁtability through better es-

timated times of arrival and terminal area ﬂight information: A better

analysis of PASSUR,” PASSUR Aerospace, Stamford, CT, USA, 2002.

[Online]. Available: http://www.passur.com

[42] A. Marzuoli et al., “Multimodality in a metroplex environment: A case

study in the San Francisco Bay area,” presented at the AIAA Infotech

@ Aerospace, AIAA SciTech, Kissimmee, FL, USA, 2015, Paper AIAA

2015-1571.

[43] C. Quinn and R. E. Zelenka, “ATC/air carrier collaborativ e arrival plan-

ning,” in Proc. 2nd USA/Europe Air Trafﬁc Manage. Semin., 1998,

pp. 1–4.

[44] S. H. Kim and E. Feron, “Impact of gate assignment on departure

metering,” IEEE Trans. Intell. Transp. Syst., vol. 15, no. 2, pp. 699–709,

Apr. 2014.

[45] A. Kim and M. Hansen, “A framework for the assessment of collabo-

rative en route resource allocation strategies,” Transp. Res. C, Emerging

Technol., vol. 33, pp. 324–339, Aug. 2013.

[46] SITA, “Ready for the connected traveler, Air Transp. IT Rev., no. 1,

pp. 1–40, 2015. [Online]. Av a ilable: http://www.sita.aero/content/Air-

Transport-IT-Review-Issue-1-2015

[47] Y. Zhang and M. Hansen, “Regional gdp—extending ground delay pro-

grams to regional airport systems,” in Proc. 8th USA/Europe Air Trafﬁc

Manage. Res. Develop. Semin., Napa, CA, USA, 2009, p. 6.

Aude Marzuoli received French Engineering

Diploma from Supelec in 2012 and the Master’s

and Ph.D. degrees in aerospace engineering

from Georgia Institute of Technology, Atlanta,

GA, USA, in 2012 and 2015, respectively. She

previously worked with NASA Ames, SESAR Joint

Undertaking, and the French National School of

Aviation.

Emmanuel Boidot received the Master’s degrees in

computer science and in aerospace engineering from

Georgia Institute of Technology, and the Master’s

degree in electrical engineering and computer sci-

ence from Supelec in 2011. He is currently working

toward the Ph.D. degree in aerospace engineering

with Georgia Institute of Technology. His research

interests include robotics, game theory, networks,

and air transportation.

Pablo Colomar received the Master’s degree in

industrial engineering from Polytechnic University

of Catalonia in 2014 and the Master’s degree in

mechanical engineering and management from Tech-

nische Universitat Munchen in 2015.

Mathieu Guerpillon received the Master’s degree

in aerospace engineering from Georgia Institute of

Technology and the French Engineering Diploma

from Supelec in 2014.

Eric Feron receive d the B.S., M.S. and Ph.D. de-

grees from Ecole Polytechnique, France, Ecole Nor-

male Superieure, France and Stanford University,

U.S. Since 2005, he has been the Dutton-Ducoffe

Professor of Aerospace Software Engineering,

Georgia Institute of Technology. From 1993 to 2005,

he was with the Department of Aeronautics and

Astronautics, Massachusetts Institute of Technol-

ogy. His interests include the fundamental concepts

of control systems, optimization and computer sci-

ence to address important problems in aerospace

engineering, such as aerobatic control of unmanned aerial vehicles, multi-

agent operations, air trafﬁc control systems, and aerospace software system

certiﬁcation.

Alexandre Bayen received the Engineering degree

in applied mathematics from Ecole Polytechnique,

France, in 1998; the M.S. degree in aeronautics

and astronautics from Stanford University in 1999;

and the Ph.D. degree in aeronautics and astronautics

from Stanford University in 2003. He was a Visit-

ing Researcher with NASA Ames Research Center

from 2000 to 2003. In 2004, he worked as a Re-

search Director of the Autonomous Navigation Lab-

oratory at Laboratoire de Recherches Balistiques et

Aerodynamiques, (Ministere de la Defense, Vernon,

France), where he holds the rank of Major. He is currently an Associate Chan-

cellor Professor, and has been the Director of the Institute for Transportation

Studies (ITS) since 2014.

Mark Hansen receive d the Bachelor of Arts degree

in physics and philosophy from Yale Unive rsity in

1980 and the Master’s degree in city planning and

the Ph.D. degree in transportation engineering from

UC Berkele y in 1984 and 1988, respectively. He

is the codirector of National Center for Excellence

in Aviation Operations Research (NEXTOR). His

research interests include transportation economics,

policy and planning, air transportation, and public

transportation.