Decision Support for Planning Maritime Search and Rescue

Operations in Canada

Irène Abi-Zeid, Michael Morin and Oscar Nilo

Operations and Decision Systems, Université Laval, Québec, Canada

Keywords: Search and Rescue, Decision Support System, Search Theory, Design Science Research, Optimal Search

Planning.

Abstract: In this project we constructed and evaluated research artifacts to support Search and Rescue (SAR) mission

coordinators in planning searches for missing persons or objects at sea. An iterative heuristic based

optimization model was formulated and implemented in a prototype that is integrated in a Decision Support

System. Using representative examples, we show that the new planning method can help coordinators with

the complex task of allocating search resources to search areas in a way that maximizes the chances of finding

survivors quickly. Although developed for the Canadian Coast Guard, our method can be used in other

countries. We followed Design Science Research guidelines and our design process was according to the

Design Science Research Methodology. The research entry point was client and context initiated and beta

testing with users is planned in the spring of 2019. It is expected that our innovative artifacts will contribute

to improving the SAR system and saving more lives.

1 INTRODUCTION

Search and Rescue (SAR) comprises “the search for,

and provision of aid to, persons who are, or who are

feared to be, in distress” (Canadian Coast Guard,

2014). The Canadian government is responsible for

providing SAR in an area covering over 18 million

square kilometers of land and water and the Canadian

Coast Guard is involved in an average of 6000

incidents per year (Quadrennial SAR review, 2013).

Maritime SAR operations are under the control of

three joint rescue coordination centres and two

marine rescue sub-centres where SAR mission

coordinators (SMC) are responsible for planning,

coordinating, controlling and directing the response

to incidents. They are decision-makers who must

make timely decisions in situations where lives are at

risk. Search planning is a complex task where time is

a crucial factor for survivors who must be found

quickly. SAR operations are among the most critical

responsibilities of the Canadian Coast Guard and can

be difficult to carry out. Each situation is unique:

Particular constraints limit the choice of the search

resources and their deployment, difficult climatic and

weather conditions may be present and operations are

often carried out in remote and unfamiliar areas. It is

therefore of the utmost importance to plan searches

that ensure the best use of available search resources

in order to maximize the chances of finding survivors.

The Canadian Coast Guard (CCG) is currently

working on developing the Advanced Search

Planning Tool (ASPT), the next generation decision

support system (DSS) to replace CANSARP

(Canadian Coast Guard, 1998), its current SAR

planning system. During the requirements

specification phase, the need for an intelligent search

planning module, that can recommend search plans

designed to ensure the optimal use of available search

resources, was identified and confirmed. This led our

research team to formulate the following research

question:

RQ: How can optimal or near-optimal search

operations be planned in reasonable time and

decision support provided to assist maritime search

mission coordinators?

The research presented in the paper was

conducted within the Design Science Research (DSR)

paradigm defined by Hevner and Chatterjee (2010),

as“a research paradigm in which a designer answers

questions relevant to human problems via the

creation of innovative artifacts, thereby contributing

new knowledge to the body of scientific evidence. The

designed artifacts are both useful and fundamental in

understanding that problem.”

328

Abi-Zeid, I., Morin, M. and Nilo, O.

Decision Support for Planning Maritime Search and Rescue Operations in Canada.

DOI: 10.5220/0007730303280339

In Proceedings of the 21st International Conference on Enterprise Information Systems (ICEIS 2019), pages 328-339

ISBN: 978-989-758-372-8

In order to answer the research question relevant

to the human problem of searching for and quickly

finding missing persons or objects at sea, we created

a method consisting of innovative artifacts as follows:

we first formulated a search planning model

involving simulation and optimization based on

search theory (Stone, 2004). Search theory can be

seen as the application of Bayesian statistics to the

question of where to search for a missing object. We

refer the reader interested in learning about search

theory to the seminal work of L. D. Stone (Stone,

2004).

Our model was then translated into an algorithm

and further implemented in Search Planner, a

prototype that provides optimal or satisficing (Simon,

1956) feasible search plans to the SMC. The figures

of merit, probability of success, associated with the

resulting search plans are obtained via simulation of

moving search objects and available search and

rescue units. Although a standalone application,

Search Planner has been integrated within the ASPT

DSS and beta testing with the users is planned in the

spring of 2019. The work presented here was carried

out over a span of three years, from 2016 to 2018.

The rest of this paper is structured as follows: In

Section 2, we address the research background along

with related work and research design. In Section 3,

we describe and explain our search planning model,

algorithm and present the implemented prototype.

Using representative application examples in Section

4, we show some results. In Section 5 we provide a

discussion as well as limitations. Finally, we conclude

in Section 6.

2 BACKGROUND

Following the receipt of an alert pertaining to a

maritime incident, SAR mission coordinator must

gather information in order to establish whether a

search is to be conducted, in which case he/she must

begin the search planning process by verifying the

search resources available, choosing the resources,

determining the area to be searched and developing a

search plan. A search plan is a distribution of the

search resources over a search area, also called effort

allocation. Optimal search planning may be defined

as the allocation of the available search resources in

such a way to maximize the chances of locating and

rescuing survivors, subject to operational constraints.

Over the years, manual methods and procedures

have been devised to develop search plans. However,

these methods do not in general take advantage of

today’s computer power and advances in search

theory and simulation, and may not be specifically

tailored to the search equipment on hand, which may

result in plans that do not have the highest success

probabilities.

Nonetheless, it has been known since the Second

World War that search theory-based planning can

result in significant gains in search effectiveness. It is

now recognized that the use of search theory and

organized planning results in higher success rates and

a significant increase in the number of lives preserved

(Frost and Stone, 2001; Abi-Zeid and Frost, 2005;

Ferguson, 2008; Abi-Zeid et al., 2011). Furthermore,

Stone et al. (2016) give examples of some high profile

cases including the response to the submarine threat

in the Atlantic, the search for a lost H-bomb in the

Mediterranean, the search for the US nuclear

submarine Scorpion, the clearing of unexploded

ordnance in the Suez Canal, the search for the sunken

treasure ship, the SS Central America, and more

recently the locating of the wreckage of AF 447. The

authors further identify two unsuccessful searches

that might have benefited from better planning. In

more recent years, search theory has also been applied

in the area of autonomous searching by robots in

structured environments, and by unmanned air

vehicles for outdoor searching of large areas

(Ablavsky and Snorrason, 2000; Lau et al., 2008; Sato

and Royset, 2010; Kriheli et al., 2016; Venkatesan,

2016; Bernardini et al., 2017).

The need for specific decision support systems

that can assist a SMC has long been identified in the

scientific literature (Abi-Zeid and Frost 2005; Hillier,

2008; Aronica et al., 2010; Kratzke et al., 2010; Stone

et al., 2014; Małyszko and Wielgosz, 2016;

Bellantuono et al., 2016). Various SAR information

systems are currently available in various countries

(Vidan et al., 2016), including the widely used SARIS

(sold by BMT ships & Coastal Dynamics product),

SARMAP (sold by RPS ASA), SAR (sold by

TRANSAS), SARGIS (Guoxiang and Maofeng,

2010), and CANSARP (Canadian Coast Guard,

1998). However, at the present time, only SAROPS,

a maritime SAR DSS developed for the US Coast

Guard, provides capabilities for search theory-based

search planning (Kratzke et al. 2010). Our research

project was therefore created as an answer to an

expressed requirement of the Canadian Coast Guard

to fill an existing gap. Nonetheless, the knowledge

created and our designed artifacts can benefit the

whole international SAR community, since there is

willingness on the part of the Canadian Coast Guard

to share knowledge with other countries.

In the next subsection, we describe our research

design and provide methodological context.

Decision Support for Planning Maritime Search and Rescue Operations in Canada

329

Figure 1: The three design research cycles (adapted from Hevner and Chatterjee, 2010, Fig. 2.2).

2.1 Research Design

Our project was conducted following

recommendations from Hevner and Chatterjee (2010)

who proposed three design research cycles, where the

aim is to ensure that the research is both rigorous and

relevant, and provided design science research

guidelines. Figure 1 shows the three research cycle

while providing additional context-specific

information. Table 1 provides context-specific

information to design science research guidelines.

Based on our research question, our objective and

design requirement was to develop and implement

methods and algorithms for optimal search planning

that would provide results in reasonable time. Our

artifacts are therefore, the search planning method,

the algorithms and the resulting prototype. These are

viable since they are compatible with existing

technical systems, were transferred to an operational

system in the organizational environment, and will be

supported by training and technical teams.

The relevance of our research is established since

the requirements were expressed by the organization

responsible for search planning, namely the Canadian

Coast Guard and by future users of our artifacts.

Operations and management had identified problems

with the current methods and foreseen opportunities

to remedy this situation by taking advantage of

advances in computing power and in simulation. As

for rigor, our approach was based on search theory, a

well-established theory for search and detection

(Stone et al., 2016). We conducted a thorough

literature review and had exchanges with other search

theory experts. Furthermore, we have extensive

experience with search theory for SAR, since two of

the authors have been conducting research on this

topic for over 20 years, and the third for over 10 years.

In fact, we had previously developed SARPlan for

overland search planning, a DSS that had won awards

for innovative technology on the national level.

Our design process was iterative, as we defined

and refined our artifacts. All along the project, there

were regular meetings and exchanges with

practitioners, technical staff and intended users and

their representatives, who provided input, criticism,

requests for modifications, constraints, etc. Five

versions of the model and optimization algorithms

were developed, compared and tested before adopting

the current version. The aim of these iterations was to

increase the solution’s quality and reduce

computation time. The acceptance of our artifacts was

an ongoing process where we produced various

documents, demonstrated the prototype, and had

multiple meetings with the stakeholders.

Furthermore, external scientific experts have been

tasked to evaluate our methods and artifacts.

During the design process, we followed the

Design Science Research Methodology as shown in

Figure 2 (Peffers et al., 2007). Our research entry

point was client/context centered. The problem

identified was how to plan search operations for

moving objects simulated using Monte Carlo based

drift models. Our objective was to demonstrate a

search planning method, which led us to design and

develop search theory based artifacts that were

implemented and demonstrated in a prototype. The

client integrated the artifacts in their operational

system. Evaluation and communication are on-going.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

330

Table 1: Design research guidelines (adapted from Hevner and Chatterjee, 2010, Table 2.1).

Guideline

Description (Hevner and Chatterjee, 2010)

Specific context

1. Design as an

Artifact

Design science research must produce a viable

artifact in the form of a construct, a model, a

method, or an instantiation.

Search planner prototype for ASPT

2. Problem relevance

The objective of design science research is to

develop technology-based solutions to important

and relevant business problems.

Relevant problem: maritime search and

rescue operations planning

3. Design evaluation

The utility, quality, and efficacy of a design artifact

must be rigorously demonstrated via well-executed

evaluation methods.

Comparison with search mission

coordinators manual planning

(forthcoming)

4. Research

contributions

Effective design science research must provide clear

and verifiable contributions in the areas of the

design artifact, design foundations, and/or design

methodologies.

Proof of concept and prototype in

response to search mission

coordinators needs

5. Research rigor

Design science research relies upon the application

of rigorous methods in both the construction and

evaluation of design the artifact.

Agile development approach,

implementation of search theory

concepts, optimization approaches,

search mission coordinators expertize

6. Design as a search

process

The search for an effective artifact requires utilizing

available means to reach desired ends while

satisfying laws in the problem environment.

Development of a prototype

compatible with existing systems,

algorithms and simulator short

response time needed

7. Communication of

research

Design science research must be presented

effectively to both technology-oriented and

management-oriented audiences.

Open source, professional and

scientific publications

Figure 2: Our process iteration following the Design Science Research Methodology.

Decision Support for Planning Maritime Search and Rescue Operations in Canada

331

Figure 3: The SAR Planning process.

3 SEARCH PLANNER

When a SMC begins the process of planning a

maritime search mission, he/she starts by creating a

SAR case containing all the available information

concerning the emergency, the characteristics of the

vessel, the number of persons involved, the last

known point, possible sightings, relevant

communications, etc. The next step is to run, in

ASPT, a Monte Carlo (MC) based stochastic drift

simulation (particle filter) for computing probability

distributions of the search object location. The first

step in the simulation is to seed, in space, a certain

number of particles (typically 5000), equally likely to

be the search object, using a 2D Gaussian distribution

with a standard deviation specified by the user. The

locations where the particles are seeded in the

simulation represent plausible last known positions of

the search object. The particles are then moved, by

simulation, in time and space, according to a drift

model, as a function of surface currents and winds.

The drift model calculates, over a simulation horizon,

the positions of the particles at each time step. Each

set of particle’s positions in time represents a search

object’s likely trajectory (Breivik and Allen, 2008).

The simulation’s output, the MC drift file containing

the particles’ positions at each time step, is an input

to Search Planner.

Subsequently, the SMC must identify available

search resources that will be tasked to conduct the

search operations. This is also an input to Search

Planner. At this point, the SMC can either manually

produce a search operation and send it to Search

Planner for evaluation purposes, i.e. computing its

probability of success, or request that Search Planner

suggest a search operation in which case the

Optimizer module is invoked. Planning a search

operation (SO) consists of assigning search plans to

the available search resources (SRU). A search plan

(SP) is defined for a SRU by a search pattern over a

search rectangle. A search area (SA) is a rectangular

region where SRUs are tasked. It is defined over a

drift simulation time horizon and is based on the

simulated particles’ positions.

Our prototype Search Planner contains three

submodules, a Simulator (to simulate the positions of

the SRU), an Evaluator, and an Optimizer. The

Optimizer uses the Evaluator that in turns uses the

SRU Simulator. The maritime SAR planning process

using ASPT along with Search Planner is presented

in Figure 3.

Our objective in designing and implementing the

Optimizer is to provide a tool that recommends

feasible search operations with the highest probability

of success (POS). A POS is a figure of merit,

associated with a search operation, loosely defined as

the probability of finding the search object.

In order to illustrate the concepts used, Figure 4

presents an example of a search area containing a 12-

hour simulated drift.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

332

Figure 4: An example of a Search Operation.

The points are positions of drifting particles in time.

The green polygon is the possibility area, a convex

hull containing all search particles positions over the

12-hour period. The Search operation is composed of

three SRUs each having their own search pattern over

their search rectangle. A helicopter (SRU

) is present

on-scene between hours 1 and 3 of the simulation. A

fixed wing aircraft (SRU

) is present between hours 4

and 8 and a vessel (SRU

) is present between hours 4

and 10. The patterns shown are parallel search

patterns. Our artifacts are designed to recommend the

search plans of a search operation, namely identify

the best combination of search rectangles (in red) and

the enclosed patterns.

We describe below the algorithm, based on search

theory, implemented in the Optimizer module.

3.1 The Optimizer

We developed a heuristic based optimization

algorithm that provides feasible optimal or near-

optimal search plans, given appropriate input

parameters, operational constraints and data. It takes

into account the various characteristics of the SRUs

namely their endurance, speed, altitude, and detection

capability as a function of the environment and of the

search object type. The activity diagram depicting the

Search Planner process for suggesting a search

operation is presented in Figure 5.

Given a SAR case, the optimal search planning

problem or SRU allocation problem can be viewed in

terms of the global problem of identifying a search

area, and then locally defining a search plan for each

SRU by assigning it to its optimal or satisfactory

feasible search rectangle, yielding a feasible optimal

or near-optimal search operation. This division of the

optimization task is similar to the approach used by

the SMCs. They tend to define the general SA first

and then position the search patterns of various SRUs

inside that large SA. The optimization procedure we

propose uses a simulation-based evaluation of the

POS of the candidate SO at each iteration.

Decision Support for Planning Maritime Search and Rescue Operations in Canada

333

3.1.1 Evaluating a Search Operation

Optimizing search operations requires that we find

the best combination of search plans for the available

SRUs. The figure of merit used to compare candidate

search operations is the probability of success

associated with the SRUs and a search object

(Kratzke et al., 2010). This is the probability of

detecting the object It is obtained via the Simulator

and the Evaluator.

Suppose that we have P particles. Let O(p) be the

probability that particle p is the search object (prior to

searching) and let pfail(p) be the probability that

particle p has not been detected. Before any search is

conducted, this probability is 1. Consider a SRU u

searching along K straight-line legs where d

is the

distance at the closest point of approach between the

SRU u on leg k and the particle p. Let f

) be the

probability that SRU u detects particle p when the

distance, at the closest point of approach, between the

particle and the SRU is d

. This is computed from

lateral range curves developed following extensive

experiments under various weather and search

conditions (Frost, 2002). It is a function of the search

object and of the SRU. The probability the SRU u

fails to detect the particle on the k

leg is: 1−f

Detections are assumed to be independent along

different search legs. The probability of SRU u not

detecting particle p is therefore:





 









  

















(1)

Suppose now that U SRUs are searching, then the

probability of particle p not being detected is:













  



 











(2)

and the probability of a particle being detected is









  







 (3)

Figure 5: The Activity Diagram for optimization in Search Planner.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

334

The probability of success of a search operation S is

then defined as:



















 (4)

After an unsuccessful search, the prior probability

O(p) is updated in a Bayesian fashion and the

posterior probability (after the search) O

(p), used as

an a priori probability for the subsequent search, is

computed as follows:











 















(5)

3.1.2 The Optimization Algorithm

We begin by receiving, from ASPT, drift data

containing the particle’s position in space and time.

Next, we create a convex hull (called a possibility

area or PA) containing all the particles during the

whole MC simulation. There are many algorithms to

generate convex hulls from a set of 2D points. We use

Andrew's monotone chain algorithm (Andrew, 1979)

that has a O(n log n) complexity. We then compute a

search area, a minimum spanning rectangle around

the convex hull. The SA can be oriented according to

the mean drift vector, or following the convex hull’s

longest side. Using the information about the

available resources such as time on-scene, duration

on-scene, detection capability as a function of the

search object, operational constraints on track spacing

in search patterns, etc., the next step is to generate, for

each SRU, a region of interest (ROI), a rectangle

enclosing the convex hull of particles present during

a given time period, normally its on-scene duration.

This uses the same procedure as computing a SA but

with a subset of the particles. We then apply a sub-

optimization filtering procedure: For each SRU, we

modify its ROI by shrinking it or enlarging it and by

moving its centre, within the operational constraints

related to the SRU’s capabilities. Each modified

version of the ROI is a possible search plan (rectangle

and search pattern). We only retain feasible search

plans. We then evaluate all combinations of feasible

search plans for all SRUs and we obtain a POS

evaluation of the SO (Equation 4). The algorithm is

iterative, we continue until we have evaluated all the

combinations or until we reach a given stopping

criterion such as a time limit or minimum global POS

improvement. The pseudo-algorithm is summarized

in Algorithm 1.

Algorithm 1: The optimization pseudo-algorithm.

Search operations are successive in time (Alpha,

Bravo, Charlie, etc.) and are planned as a function of

the previous search operations. Consider for example

an Alpha search that was conducted over N hours

starting at time T. If it is unsuccessful, a subsequent

search operation Bravo, must be planned over M

hours, starting at time T+N. This is done by redrifting

Input:

U SRUs, search object, the

drifted particles positions in

time

Output:

A best POS Search Operation

containing a search rectangle

and pattern for each of the U

SRUs

Steps:

1 - Let z(u) be the total amount of

available search effort for SRU

u (hours)

2 - For each SRU u = 1 to U:

i. Construct the convex hull

containing the particles

during the on-scene

period of the SRU u

ii. Construct its ROI: A(u)

iii. Run the sub-optimization

procedure to generate

SR(u), a set of feasible

search rectangles. This

set is constructed by

varying the location, the

size and the orientation

of A(u).

3 - Add the sets SR(u) to a

candidate SO

4 - Simulate (Simulator) and

evaluate (Evaluator) to obtain

the SO’s POS

5 - Let S be the POS value of the

current candidate SO. Let S* be

the POS value of the best so

far incumbent SO. If S is

strictly better than S* then

the candidate SO becomes the

best so far incumbent SO and S*

is updated

6 - If the stopping criteria are

not met, return to step 4.

Otherwise recommend the best so

far incumbent SO.

Decision Support for Planning Maritime Search and Rescue Operations in Canada

335

the particles starting at simulation time T for a

duration of M+N hours and by re-evaluating the

Alpha search as a function of the real environmental

conditions observed between T and T+N. Re-

evaluating the Alpha search produces more realistic

figures of merit to update the probability that a given

search particle is the actual search object, depending

on whether it came within detection range of a SRU

during the Alpha search. The optimized Bravo search

is then obtained by maximizing Equation 4, between

T+N and T+N+M, where O (p) are replaced by the

updated particles probabilities, O

(p). Particles with

lower O

(p), ones that came within detection range,

are less likely to be the search object and their

importance in planning subsequent searches are

thereby reduced. Furthermore, the cumulative

probability of success is computed for successive

search operations. This information is very useful for

the search reduction process. At some point, the

commander will have to decide to reduce and stop the

search. This happens when he/she is reasonably

convinced that the chances of finding survivors are

practically nil, reflected by many hours of

unsuccessful search operations that had a very high

theoretical cumulative POS. The rationale is, if

survivors were to be found, they would have been

found by the end of these well-planned searches.

Other factors to take into account in the reduction

decision include the average survival times of persons

in water under the given environmental conditions.

4 APPLICATION EXAMPLES

In order to illustrate the applicability of our method,

we present two representative examples of realistic

maritime SAR cases. The first incident, described in

Section 4.1, pertains to a person in water (PIW) and

the second one, in Section 4.2, to a fishing vessel

(FV). The drift model in ASPT is used to estimate the

particles positions at time steps of 15 minutes.

Currents and winds and last known points are entered

in the MC module of ASPT. A total of 5000 particles

are seeded. The corresponding lateral range curves

for computing detection probabilities are used. The

results are presented using the geographic

information system QGIS (QGIS development team,

2019). It is important to note that both the search

object and the SRUs are moving. Detection

opportunities depend on both positions being

synchronized. The heat maps presented correspond to

the position of all particles during the whole search.

They do not necessarily convey a good idea of the

movement of the particles. The ability of our search

planning method to take into account the particles’

movements is one of its main advantages over the

current manual planning method using deterministic

drift. In fact, a person manually developing a plan will

be tempted to position the search pattern over the heat

map. Although a good starting heuristic, it does not

automatically produce the best probabilities of

success since it is not possible to synchronize in one’s

head the positions of the particles and of the SRU.

This can only be done by simulation, as in our

Simulator component. The white triangle shows the

direction of the drift. The solution for both examples

was obtained in under one minute. The area searched

is much larger for the FV because its search duration

is longer.

4.1 Case 1: Person in Water

A person has gone overboard in water and is assumed

to have a survival suit. Two SRUs are available: One

helicopter with endurance (time on-scene) of 2 hours

and one fixed wing with endurance of 4 hours. The

visibility is of 10 nautical miles (NM). The waves are

5 m high. The Search Planner results are presented in

Figure 6 where we see the search rectangles and

corresponding search patterns (expanding square in

blue and parallel in purple) assigned to the two SRUs.

The helicopter, flying at 50 knots at an altitude of 750

ft, has a recommended search rectangle of 6 by 9

(NM) (smaller rectangle). The fixed wing has a

recommended search rectangle of 11 by 14 (NM)

(larger rectangle). Its search speed is 100 knots and

altitude is 1000 feet. The combined probability of

success is 31%. This may seem not very high but the

probability of detecting a person in water is very

small because of the size of the person. The total area

searched is 154 NM

and represents a 4 hour drift.

Deconfliction between SRUs is based on altitudes.

Figure 6: The proposed search operation for the PIW.

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

336

4.2 Case 2: Fishing Vessel

A fishing vessel is missing. Two SRUs are available:

One helicopter with endurance (time on-scene) of 2

hours and one fixed wing with endurance (time on-

scene) of 6 hours. The visibility is of 10 nautical miles

(NM). The results are presented in Figure 7 where we

see both search rectangles (in red) and corresponding

search patterns (parallel patterns, dotted blue and

purple lines). The helicopter, flying at 50 knots at an

altitude of 100 ft, has a recommended search

rectangle of 17 by 20 (NM). The fixed wing, flying at

120 knots and an altitude of 1500 feet, has a search

rectangle of 38 by 45 (NM). All particles were

covered (came within detection range). Under these

conditions, the probability of success is very high:

99%. This can be explained by the fact that the object

has high detectability. The total search area is 1710

and represents a 6 hour drift. Deconfliction

between SRUs is based on altitudes.

Figure 7: The proposed search operation for the FV.

5 DISCUSSION

The two examples presented above illustrate how, by

designing Search Planner, we were able to answer our

research question. Our new method, implemented in

a prototype and integrated in a DSS, can assist search

mission coordinators in planning optimal or near-

optimal search operations in reasonable time. As

required, Search Planner produces search plans for

each available SRU that have the best combined POS

within a limited computation time defined by the user.

The software has been verified in the sense that it

meets the specifications of the designed method. Its

validation, i.e. ensuring that the software meets the

requirements of the users is planned in the spring of

2019. However, the evaluation of the results produced

(search plans quality) is an ongoing process that

requires some months, and is planned as follows:

First, “beat the DSS” sessions where experienced

SMCs are asked to provide manual plans will be

conducted. The POS of the manually produced plans

will be compared with the automatically produced

plans. It is expected that the DSS will compete

advantageously with the human operator. In all cases,

this experiment will contribute to improving the

algorithm’s heuristics based on practical human

experience and knowledge. Second, past solved SAR

incidents will be used to validate the artifacts. They

will be defined as new cases, and the search plans

produced by the DSS will be evaluated as a function

of the locations where the search objects were

actually found in the past incidents. DSS-produced

search plans will be considered valid if they contain

the locations of the found search objects. Third, to

help validate their drift simulation module,

experiments are planned by the Canadian Coast

Guard where buoys will be released in water and

tracked. Their actual physical trajectories will be

compared to the simulated particles positions.

Comparing the output of our DSS with existing

similar systems could further validate our artifacts.

However, at the current time, we do not have access

to the only other DSS, developed in the US, that has

similar optimal search planning functionalities

(Kratzke et al., 2010).

Our search planning method has limitations. The

main one is related to the model itself. Any model is

a simplified assumption of reality based on

underlying hypotheses. In our case, in order to use

theoretical search theory and lateral range curves to

compute probabilities of detection and of success, we

make the assumption of independent detections along

infinitely long parallel search tracks. This a

reasonable assumption when the tracks are longer

than the detection range. Furthermore, the lateral

range curves, constructed and validated in

experiments (US unpublished reports, 1998; 2005),

are simplified detection models. However, both the

underlying hypotheses and the lateral range curves

are behind a DSS that has been operational and is

successfully used for over 10 years in the US (Stone

et al., 2016).

Another limitation is related to scaling up. Drift

simulation may imply the use of thousands of

particles that drift for many hours. Their positions can

be computed at various time steps ranging from 1

minute to 30 minutes. This has great implications on

the size of the problem. Computing, using step by step

search pattern simulation, probabilities of detections

Decision Support for Planning Maritime Search and Rescue Operations in Canada

337

of millions of particles, in order to evaluate a possible

search operation, can be very time consuming.

Alternatives including particle random sampling and

evaluating a subset of particles can reduce

computation time. Working with larger simulation

time steps is another option. Intensive sensitivity

analyses are necessary to arrive to a compromise in

terms of solution quality and computation time.

One of the challenges we face is the acceptance of

the DSS by the users as this implies a new way of

working and a new way of thinking. Furthermore, this

will imply extensive training (already planned) before

we can be confident that the tool is used to its full

capability. Moreover, the real ability of the proposed

method to increase the number of lives saved can be

assessed only after it has been in operation for a few

years when it is established that the average number

of lives saved has actually increased. Finally, search

operations with very high success probabilities do not

guarantee that the search objects will be found. There

have been many examples of SRUs flying over a

missing object and not seeing it. Although the POS is

accepted as a figure of merit for a search plan, it

remains only a probability.

6 CONCLUSION

We have designed research artifacts to support SAR

mission coordinators in planning searches for missing

persons or objects at sea. An iterative heuristic based

optimization model was formulated and implemented

in a prototype that is integrated in a DSS. Following

the identified limitations in the discussion section,

further research is needed to improve the quality and

performance of the heuristic optimization algorithm,

and to measure the real gains obtained in an

operational setting. For example, in order to try to

reduce computation time related to evaluating the

POS of each candidate search rectangle by

simulation, we are currently exploring machine-

learning techniques from Artificial Intelligence to

automatically “learn” then estimate, without having

to simulate the whole search pattern, the POS of a

search rectangle from a set of previously evaluated

similar rectangles in a similar area. This could result

in a significant decrease in computation time.

Future planned research includes the development

and evaluation of clustering algorithms that divide the

drifted particles set in clusters, which kernels can be

used as a starting centre points for the candidate

search rectangles. Another possible avenue is to

explore the influence of the search pattern type and its

starting point as a function of the drift’s direction.

Further algorithm enhancements could also be

achieved by adding some degrees of freedom in

designing the search plans: the initial convex hull

defining the possibility area could be constructed

using the rotating callipers algorithm to obtain the

most promising orientation of the search area

(Toussaint, 1983). In addition, the candidate search

rectangles could be rotated within the search area in

an attempt to improve the POS. Most importantly,

data on the DSS use in an operational context must be

gathered to continuously improve its acceptability

and performance over the next years.

In the future, as the users become more

comfortable with the new DSS, it is expected that

they will require additional functionalities, such as for

example, simultaneous planning for multiple search

objects, or planning with concurrent unequally likely

scenarios related to what might have happened in the

SAR incident and where.

ACKNOWLEDGEMENTS

The authors wish to thank the ASPT Working group

and the SMCs for their support during the project.

Special thanks to the project manager and the

technical team from the Canadian Coast Guard

College in Sydney. This project was funded by the

Canadian Government, contract number FP802-

150046 to Neosoft Technologies. Research and any

resulting information/theory gathered or compiled

during employment of the author of this published

work by the Government of Canada is licensed for

publication by the Department of Fisheries and

Oceans, and Canadian Coast Guard. No transfer of

Intellectual Property (IP) is implied by the granting of

said license.

REFERENCES

Abi-Zeid, I. and Frost, J.R., 2005. SARPlan: A decision

support system for Canadian Search and Rescue

Operations. European Journal of Operational Research,

162(3), pp.630-653.

Abi-Zeid, I., Nilo, O. and Lamontagne, L., 2011. A

constraint optimization approach for the allocation of

multiple search units in search and rescue operations.

INFOR: Information Systems and Operational

Research, 49(1), pp.15-30.

Ablavsky, V. and Snorrason, M., 2000, August. Optimal

search for a moving target-A geometric approach. In

AIAA Guidance, Navigation, and Control Conference

and Exhibit (p. 4060).

ICEIS 2019 - 21st International Conference on Enterprise Information Systems

338

Andrew, A.M., 1979. Another efficient algorithm for

convex hulls in two dimensions. Information

Processing Letters, 9(5), pp.216-219.

Aronica, S., Benvegna, F., Cossentino, M., Gaglio, S.,

Langiu, A., Lodato, C., Lopes, S., Maniscalco, U. and

Sangiorgi, P., 2010. An Agent-based System for

Maritime Search and Rescue Operations. In WOA.

Bellantuono, N., Camarda, P., Lisi, S., Caneva, P.,

Pontrandolfo, P., Romano, V., Striccoli, D., Scozzi, B.,

SpA, C. and Selex, E.S., 2016. Emergency management

at sea: A decision support system for Search and Rescue

operations. Journal of Software & Systems

Development.

Bernardini, S., Fox, M. and Long, D., 2017. Combining

temporal planning with probabilistic reasoning for

autonomous surveillance missions. Autonomous

Robots, 41(1), pp.181-203.

Breivik, Ø. and Allen, A.A., 2008. An operational search

and rescue model for the Norwegian Sea and the North

Sea. Journal of Marine Systems, 69(1-2), pp.99-113.

Canadian Coast Guard/Department of Fisheries and Oceans

Canada, 2014. Canadian Aeronautical and Maritime

Search and Rescue Manual, Combined Edition –

Volumes I, II and III, B-GA-209-001/FP-001.

Canadian Coast Guard College, 1998. A brief guide to

CANSARP, TP 11718E.

Ferguson, D., 2008, February. GIS for wilderness search

and rescue. In Proceedings of ESRI Federal User

Conference.

Frost, J.R. and Stone, L.D., 2001. Review of search theory:

advances and applications to search and rescue decision

support (No. CG-D-15-01). SOZA AND COMPANY

LTD FAIRFAX VA.

Frost, J.R., 2002. Elements, Modules and Algorithms for

Planning Optimal Searches - A High Level Overview.

Prepared for. U. S. Coast Guard Office of Search and

Rescue. Potomac Management group Inc. Alexandria,

Virginia.

Hevner, A. and Chatterjee, S., 2010. Introduction to design

science research. In Design Research in Information

Systems (pp. 1-8). Springer, Boston, MA.

Hillier, L.E., 2008. Validating and improving the Canadian

coast guard search and rescue planning program

(CANSARP) ocean drift theory (Masters dissertation,

Memorial University of Newfoundland).

Guoxiang, L. and Maofeng, L., 2010, May. Sargis: A gis-

based decision-making support system for maritime

search and rescue. In E-Business and E-Government

(ICEE), 2010 International Conference on (pp. 1571-

1574). IEEE.

Kratzke, T.M., Stone, L.D. and Frost, J.R., 2010, July.

Search and rescue optimal planning system. In

Information Fusion (FUSION), 2010 13th Conference

on (pp. 1-8). IEEE.

Kriheli, B., Levner, E. and Spivak, A., 2016. Optimal

search for hidden targets by unmanned aerial vehicles

under imperfect inspections. American Journal of

Operations Research, 6(02), p.153.

Lau, H., Huang, S. and Dissanayake, G., 2008. Discounted

mean bound for the optimal searcher path problem with

non-uniform travel times. European journal of

operational research, 190(2), pp.383-397.

Małyszko, M. and Wielgosz, M., 2016. Decision support

systems in search, rescue and salvage operations at sea.

Zeszyty Naukowe/Akademia Morska w Szczecinie, (45

(117)), pp.191-195.

Peffers, K., Tuunanen, T., Rothenberger, M.A. and

Chatterjee, S., 2007. A design science research

methodology for information systems research. Journal

of management information systems, 24(3), pp.45-77.

QGIS Development Team (2019). QGIS Geographic

Information System. Open Source Geospatial

Foundation Project. http://qgis.osgeo.org. Visited on

21/2/2019.

Quadrennial SAR Review, available at:

https://www.publicsafety.gc.ca/cnt/rsrcs/pblctns/archiv

e-nss-qdrnnl-rvw/index-en.aspx. Visited on

30/12/2018.

Simon, H.A., 1956. Rational choice and the structure of the

environment. Psychological review, 63(2), p.129.

Sato, H. and Royset, J.O., 2010. Path optimization for the

resource constrained searcher. Naval Research

Logistics (NRL), 57(5), pp.422-440.

Stone, L. D. ,2004.Theory of Optimal Search. Academic

Press, New York, 2

edition.

Stone, L.D., Keller, C.M., Kratzke, T.M. and Strumpfer,

J.P., 2014. Search for the wreckage of Air France Flight

AF 447. Statistical Science, pp.69-80.

Stone, L.D., Royset, J.O. and Washburn, A.R., 2016.

Optimal Search for Moving Targets, International

Series in Operations Research & Management Science.

Reston and Monterey, CA. doi, 10, pp.978-3.

Toussaint, G.T., 1983, May. Solving geometric problems

with the rotating calipers. In Proc. IEEE Melecon (Vol.

83, p. A10).

Venkatesan, S., 2016, November. AUV for Search &

Rescue at sea-an innovative approach. In Autonomous

Underwater Vehicles (AUV), 2016 IEEE/OES (pp. 1-

9). IEEE.

Vidan, P., Hasanspahić, N. and Grbić, T., 2016.

Comparative Analysis of Renowned Softwares for

Search and Rescue Operations. NAŠE MORE:

znanstveno-stručni časopis za more i pomorstvo, 63(2),

pp.73-80.

Decision Support for Planning Maritime Search and Rescue Operations in Canada

339