Stochastic Modeling to Allocate and Assess
Disaster Response Capacity in Logistics Networks
MASSACHUSETTS INSTITI ITF
OF rECHNOLOLGY
by
Lauren A. Seelbach
B.S. Civil Engineering
Syracuse University, 2010
JUN 0 4 2015
LIBRARIES
SUBMITTED TO THE ENGINEERING SYSTEMS DIVISION IN PARTIAL
FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE IN TECHNOLOGY AND POLICY
AT THE
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
JUNE 2015
C 2015 Lauren A. Seelbach. All Rights Reserved.
The author hereby grants MIT permission to reproduce and to distribute publicly paper
and electronic copies of this thesis document in whole or in part
in any medium now known or hereafter created.
Signature redacted
Signature of Author:
Technology and Policy Program, Engineering Systems Division
May 8, 2015
Signature redacted
Certified by
Jarrod Goentzel
Director, MIT Humanitarian Response Laboratory
Thesis Supervisor
Accepted b
y:
Signature redacted
Dava J. Newman
Professor of eronautics and Astronautics
and Engineering Systems
and Policy Program
Technology
Director,
Stochastic Modeling to Allocate and Assess
Disaster Response Capacity in Logistics Networks
by
Lauren A. Seelbach
Submitted to the Engineering Systems Division
on May 8, 2015 in partial fulfillment of the
requirements for the degree of
Master of Science in Technology and Policy
Abstract
When a disaster occurs in the United States, individuals in the impacted areas look to the
local, state and federal government agencies to provide aid in the form of food, water,
shelter, and other essential commodities. The responding agencies have the task of
preparing their logistics networks in advance of a disaster to enable them to meet the
demand for these essential commodities shortly after an event occurs. This thesis uses
stochastic modeling and related metrics to answer three central questions that inform the
development of a framework for applying these metrics for evaluating the disaster
response capacity of a logistics network.
The thesis answers the questions of (1) where should inventory be placed, considering
both normal, steady-state allocation of inventory as well as allocation of inventory in
advance of a notice event such as a hurricane, (2) how does private sector involvement
change the response capacity of a logistics network, and (3) what is the impact of reduced
demand for critical commodities on the logistics network's ability to respond?
Two case studies are conducted on the logistics networks of the Federal Emergency
Management Agency (FEMA) and the Florida Division of Emergency Management
(FDEM). Results from these case studies indicate that applying stochastic modeling and
the associated metrics to inform the allocation of critical commodities in a logistics
network demonstrate measurable benefits in terms of fraction of overall demand met,
time to serve a disaster affected population, and related metrics. For prepositioning
decisions in FDEM's logistics network the benefits were particularly convincing. By
locating stock closer to the areas predicted to suffer the greatest losses, FDEM is able to
decrease the time per unit served of two critical commodities by 10-15%. For FEMA's
logistics network, results indicate that restructuring the terms of contract stock has the
potential to increase fraction of overall demand served by 14% within 24 hours after a
disaster and 16% within 36 hours after a disaster.
Thesis supervisor: Jarrod Goentzel
Title: Director, MIT Humanitarian Response Laboratory
3
Acknowledgements
I am incredibly grateful to everyone who has challenged, inspired, and cheered me on
throughout my graduate studies and the writing of this thesis. In particular, I would like
to thank:
Jarrod Goentzel, for consistently challenging me to think outside the box and push the
limits of what I thought was attainable, and for the opportunity to work for the past two
years in the Humanitarian Response Lab.
Jason Acimovic, for your patience, advice, and lightning speed responses to my questions
throughout the writing of this thesis.
Marianne Jahre, for all you have taught me about approaching an academic paper, and
especially for introducing me to the wonders of Norwegian cuisine.
Adam Norrige and Gregg Hogan at Lincoln Laboratory, for supporting this research
endeavor and providing thoughtful, inspiring insight throughout the process.
The analytics team at DataCo and the logistics team at FEMA and in Louisiana, for
providing data to make answering these questions possible.
Michael, for lots of things, but most of all for your constant, unwavering support and for
reminding me that there is life outside of MIT and making sure that I live it to the fullest.
My family: Mom, Dad, Nikki and Leeza, for being awesome. I truly would not be writing
these words if it weren't for your consistent love and encouragement (and personal
moving services, programming tutoring, graphic genius and motivational messages)
every step of the way.
My Baltimore family, for laughter, inspiration, and a second home when I re-discovered
what a northeastern winter entailed. A special thank you to Cassie, for keeping me
grounded and laughing these past two years.
My coworkers-and-friends, past and present in the Humanitarian Response Lab. Corinne,
Julia, Emily, Mark, Tim, Jaime, and Erin. You have made some of the most challenging
times in my time at MIT seem not so challenging after all. And, you guys make a mean
lunch.
My TPP family, the forty or so classmates turned friends who have inspired me,
introduced me to more new experiences, foods, and ideas than I can count, and made
these past two years so much fun.
To Barb, Ed and Frank, for helping me navigate the maze of MIT requirements and for
providing advice, listening, and coming through with the occasional piece of chocolate.
4
Contents
A cknow ledgem ents......................................................................................................
4
List of Figures.....................................................................................................................
8
List of Tables ....................................................................................................................
10
A cronym List ....................................................................................................................
11
I
Introduction.................................................................................................................
12
2
Literature Review ....................................................................................................
15
2.1
3
Humanitarian Logistics: Preparing the Response Infrastructure ...................... 16
2.1.1 Needs assessm ent......................................................................................
2.1.2 Prepositioning .............................................................................................
2.1.3 Vendor-managed inventory and framework agreements...........................
2.2 Public-Private Partnerships...............................................................................
17
18
19
20
2.3
21
Indices and M etrics for Evaluation.................................................................
2.3.1 Logistics indices.........................................................................................
2.3.2 D isaster response and resilience indices....................................................
2.3.3 Metrics for evaluating performance in disaster response ..........................
M ethods.......................................................................................................................
22
22
24
26
3.1
Research Question .............................................................................................
26
3.2
Research D esign................................................................................................
26
3.3
The M odel............................................................................................................
27
3.3.1 M odel form ulation ....................................................................................
3.4 M etrics D erived from the M odel .........................................................................
28
31
3.4.1 The balance m etric....................................................................................
3.4.2 Fraction served and fraction of disasters covered......................................
3.4.3 Tim e per unit delivered.............................................................................
3.4.4 Response Capacity Index...........................................................................
3.5 D ata......................................................................................................................
32
33
33
33
35
3.5.1 D isaster data................................................................................................
3.5.2 Stockpile data.............................................................................................
3.5.3 Tim e and cost data....................................................................................
3.6 Case Study Scenarios........................................................................................
35
42
42
44
3.6.1 A llocation of inventory .............................................................................
3.6.2 Impact of private sector commitments or contractual agreements ............
3.6.3 Im pact of resilience goals ..........................................................................
3.7 Lim itations ...........................................................................................................
44
45
46
48
5
4
Case Study: Federal Emergency Management Agency...........................................
49
4.1
Background......................................................................................................
49
4.2
A nalysis................................................................................................................
50
4.2.1 EM-DAT disaster data...............................................................................
4.2.2 FEMA inventory data ...............................................................................
4.2.3 Analysis scenarios......................................................................................
4.3 Scenario 1: FEMA Physical Inventory Allocation ..........................................
50
53
54
55
4.3.1 Ability to meet demand.............................................................................
4.3.2 Network inventory allocation strategies ...................................................
4.3.3 Delivery deadline cutoff times..................................................................
4.3.4 Response Capacity Index...........................................................................
4.4 Scenario 2: FEMA Physical Inventory and Contract Inventory Allocation ........
56
56
60
61
65
4.4.1 Delivery deadline cutoff times..................................................................
65
4.5 Scenario 3: FEMA Physical Inventory Allocation with Reduced Demand......... 70
5
4.5.1 Ability to meet demand.........................................o
..................................
4.5.2 Network inventory allocation strategies ...................................................
Case Study: Florida Division of Emergency Management......................................
5.1
Background................................................
74
5.2
A nalysis................................................................................................................
75
5.2.1 Insurance risk data ....................................................................................
5.2.2 Florida inventory and demand data.............................
5.2.3 Analysis scenarios.....................................................................................
5.3 Scenario 1: Current Allocation of Inventory ...................................................
76
79
80
81
5.3.1 Ability to meet demand.............................................................................
5.3.2 Network inventory allocation strategies ....................................................
5.3.3 Delivery deadline cutoff times..................................................................
5.3.4 Response Capacity Index...........................................................................
5.4 Scenario 2: Prepositioning of Inventory ...........................................................
81
82
87
89
91
5.5
6
70
72
74
Scenario 3: FDEM Physical Inventory Allocation with Reduced Demand......... 95
5.5.1 Ability to meet demand.............................................................................
5.5.2 Network inventory allocation strategies ....................................................
D iscussion ..........................................................
...................................................
95
96
98
6.1
Federal Emergency Management Agency ..............................
6.2
Florida Division of Emergency Management....................................................
100
6.3
Cross-Case Findings .........................................................................................
101
6.3.1
Allocation of critical commodities............................................................
6
98
102
7
6.3.2 Impact of partnerships in m eeting needs ....................................................
6.3.3 Resilience goals ..........................................................................................
6.4 Fram ework .........................................................................................................
104
104
105
6.5
109
Future Research .................................................................................................
6.5.1 Impact of model and data limitations .9..........................
6.5.2 Com plimentary future research...................................................................
Bibliography .............................................................................................................
7
110
113
List of Figures
Figure 2-1 Conceptual framework of the literature review ..........................................
16
Figure 3-1 National Hurricane Center cone and warnings for Hurricane Wilma on
October 21, 2005 at 4PM CDT (5PM EST) (National Hurricane Center, 2005)..... 41
Figure 3-2 National Hurricane Center cone and warnings for Hurricane Wilma on
October 23, 2005 at 10AM CDT (1 1AM EST) (National Hurricane Center, 2005) 41
Figure 4-1 EM-DAT disaster locations and affected population, continental United States
1990-2013 (Esri, 2015)..........................................................................................
51
Figure 4-2 FEMA permanent stock warehouse locations (Esri, 2015).......................... 53
Figure 4-3 FEMA contract stock warehouse locations (Esri, 2015)............................. 54
Figure 4-4 Actual and optimal allocation of MREs (minimizing time or cost)............ 57
Figure 4-5 Actual and optimal allocation of water bottles (minimizing cost or time) ..... 58
Figure 4-6 Optimal allocation of MREs (minimize time) with increasing total system
inventory, current inventory level indicated with dashed line.............................. 59
Figure 4-7 Optimal allocation of water bottles (minimize time) with increasing total
system inventory, current inventory level indicated with dashed line................... 60
Figure 4-8 Fraction of demand served (minimize time) actual vs. optimal allocation of
physical inventory only........................................................................................
61
Figure 4-9 Locations of potential FEMA warehouses used in the sensitivity analysis for
the balance m etric (Esri, 2015)............................................................................
64
Figure 4-10 Fraction of demand served (minimize time) actual vs. optimal allocation of
physical inventory only........................................................................................
66
Figure 4-11 Fraction of demand served (minimize time) actual vs. optimal allocation of
contract inventory only ..........................................................................................
67
Figure 4-12 Fraction of demand served (minimize time) actual vs. optimal, physical and
contract stock ............................................................................................................
68
Figure 4-13 Actual vs. optimal allocation of FEMA contract inventory only (minimize
tim e)..........................................................................................................................
69
Figure 4-14 Actual vs. optimal allocation of inventory with original and reduced demand
for M REs (m inimize time)....................................................................................
72
Figure 4-15 Actual vs. optimal allocation of inventory with original and reduced demand
for water bottles (minimize time) ..........................................................................
73
Figure 5-1 Count of disasters by county, disaggregated disaster dataset, for counties with
TAP>100 (Esri, 2015)...........................................................................................
77
Figure 5-2 Maximum total affected population by county, disaggregated disaster dataset,
for counties with TAP>100 (Esri, 2015) ..............................................................
77
Figure 5-3 Map of Counties in Florida, noting the county seats (Geology.com, n.d.)..... 78
Figure 5-4 Florida Logistics Staging Areas (Including Orlando, the location of permanent
warehouse) (Florida Division of Emergency Management, 2013)....................... 79
Figure 5-5 Actual and optimal allocation of MREs (minimizing time or cost)............ 83
Figure 5-6 Actual and optimal allocation of water bottles (minimizing cost or time)..... 84
Figure 5-7 Optimal allocation of MREs (minimize time) with increasing total system
inventory, with current inventory indicated by dashed line................................. 85
Figure 5-8 Optimal allocation of water bottles (minimize time) with increasing total
system inventory, with current inventory indicated by dashed line...................... 86
8
Figure 5-9 Fraction of demand served, actual allocation of Florida MRE inventory,
m inim ize time .......................................................................................................
87
Figure 5-10 Fraction of demand served, actual allocation of Florida water bottle
inventory, minimize tim e......................................................................................
88
Figure 5-11 Optimal allocation of MREs, minimize time, for two prepositioning time
intervals.....................................................................................................................
91
Figure 5-12 Optimal allocation of water bottles, minimize time, for two prepositioning
tim e intervals.............................................................................................................
92
Figure 5-13 Fraction of demand served, minimize time, for water bottles in Report 1
(10/22/2005 at 7PM CDT).......................................
94
Figure 5-14 Fraction of demand served, minimize time, for water bottles in Report 2
(10/23/2005 at 1PM CD T)........................................................................................
94
Figure 5-15 Actual vs. optimal allocation of inventory with original and reduced demand
for MREs (minimize time).............................................
97
Figure 5-16 Actual vs. optimal allocation of inventory with original and reduced demand
for water bottles (minimize time) ...................................
98
Figure 6-1 Framework for assessing logistics and other preparedness decisions. Graphic
designed and illustrated by Nicole Seelbach (Seelbach & Seelbach, 2015)........... 107
9
List of Tables
Table 3-1 Structure of disaster risk data (DataCo, 2015) ............................................
Table 3-2 Summary of case study analysis process......................................................
Table 4-1 FEMA case study scenarios...........................................................................
Table 4-2 Ability of FEMA physical inventory to meet demand .................................
Table 4-3 Response Capacity Index for FEMA physical inventory .............................
Table 4-4 Balance metric sensitivity analysis results ...................................................
Table 4-5 Ability of FEMA physical inventory to meet demand with 10% reduction in
Florida dem and .....................................................................................................
Table 5-1 Florida Case study scenarios ........................................................................
Table 5-2 Ability of Florida physical inventory to meet demand..................................
Table 5-3 Number of counties within one to four hours driving distance from Orlando,
F L ..............................................................................................................................
Table 5-4 Response Capacity Index for Florida ..........................................................
38
47
55
56
62
65
71
81
82
89
89
Table 5-5 Average time to serve per unit of commodity for prepositioning scenarios .... 93
Table 5-6 Ability of Florida inventory to meet demand with 10% reduction in demand in
M iam i-Dade County ..............................................................................................
95
Table 6-1 Logistics decisions and associated metrics evaluated and data required for
analysis....................................................................................................................
108
10
Acronym List
API
Application programming interface
FA
Framework agreement
FDEM
Florida Division of Emergency Management
FEMA
Federal Emergency Management Agency
LP
Linear programming
LSA
Logistics Staging Area
MRE
Meal ready to eat
NFIP
National Flood Insurance Program
NIMS
National Incident Management System
NRF
National Response Framework
PPP
Public-private Partnership
RCI
Response Capacity Index
VMI
Vendor-managed Inventory
11
1 Introduction
When a disaster occurs in the United States, individuals are displaced from their homes
or are faced with prolonged interruptions in essential services such as water and power.
This interruption in service or displacement means that disaster survivors look to their
local and state government emergency response organizations and, in the case of large
disasters that exhaust local supply, the Federal Emergency Management Agency (FEMA)
to fill the gap and provide supplies to meet basic needs. One of the most important
functions of FEMA, as well as state and local disaster response agencies is to ensure that
the basic needs of a disaster-affected population are met. However, it is currently
difficult to evaluate how well these agencies are prepared to meet the needs just
described. How do the decisions that response organizations make in the planning and
preparedness phases before a disaster strikes impact their ability to respond?
The response to a disaster reflects the continuous efforts of communities led, in
many cases, by an emergency management organization or other government entity and
supported by private, non-profit and faith-based organizations, as well as individuals.
FEMA has coined the term "whole community" to refer to this broader collective of
organizations, government entities, and individuals that is essential to a successful
response to a natural disaster or other emergency (Federal Emergency Management
Agency, 2015b). The response to a disaster not only reflects the efforts of this group of
organizations and individuals in the moments immediately after a disaster, but also their
efforts throughout the preparedness cycle. The preparedness cycle is defined by the
National Incident Management System (NIMS) as "a continuous cycle of planning,
organizing, training, equipping, exercising, evaluating and taking corrective action in an
effort to ensure effective coordination during incident response" (Federal Emergency
Management Agency, 2015a). FEMA emphasizes the importance of preparedness
because it is, as the NIMS definition suggests, a cycle, and decisions made long before a
disaster strikes can have a huge impact on the response.
An important preparedness decision made long before a disaster occurs is where
resources should be located throughout a federal, state, local or tribal government's
logistics network. Logistics Management is defined as "that part of supply chain
management that plans, implements, and controls the efficient, effective forward and
12
reverse flow and storage of goods, services, and related information between the point of
origin and the point of consumption in order to meet customers' requirements (Council of
Supply Chain Management Professionals, 2013). In the disaster response context,
logistics management means ensuring that survivors whose homes and livelihoods have
been damaged or destroyed have food, water, and shelter. Humanitarian logistics is the
term that has developed to encompass this aspect of logistics management, and similarly
to logistics management in general, includes considerations for effective storage
(allocation) of goods in a logistics network, highlighting the importance of this decision
before a disaster occurs.
This thesis aims to answer three central questions that will inform the
development of a framework for evaluating the capacity of disaster response
organizations to meet the needs of a disaster-affected population after an event using a set
of metrics and an index recently developed by Acimovic and Goentzel (2015). The
questions inform first where critical stock should be allocated long before a disaster as
well as shortly before a disaster where an emergency management organization would
have notice, such as a hurricane second, how the private sector's involvement changes
the response capacity of a logistics network, and third, what the impact of reduced
demand for critical commodities is on the logistics network and the ability of the network
to meet demand. The framework will outline how these considerations relate to the types
of organizations involved in a response, and metrics and an index can be applied in the
United States context for evaluating performance of a logistics network.
Two case studies are conducted on the logistics networks of the Federal
Emergency Management Agency and the Florida Division of Emergency Management to
answer the research questions. The case studies also explore the implications of varying
quality of disaster risk data, assumptions about how to determine affected population, and
stockpile data. They present the existing frameworks in place governing disaster response
at the organizations studied, with a focus on the logistics guidance and/or requirements.
The case study analyses rely on forecast disaster impact data to evaluate strategic
logistics decisions made in the preparedness cycle. Decisions made in the preparedness
cycle are evaluated here with forecast disaster data to facilitate an evaluation of the
ability of these decisions to hold up against reality in a response.
13
The goal of this thesis is to provide a framework for assessing domestic disaster
response capacity based on analytical metrics of how well equipped the supply chain is to
respond with critical commodities to meet the needs of disaster survivors. This
framework and the associated models and metrics are proposed to further the discussion
about a broader index for assessing disaster resilience, and to aid decision makers at
local, state and federal emergency response organizations as well as in the private and
non-profit sectors in making decisions about allocation of critical commodities before,
during, and after a disaster.
The Literature Review section presents the relevant literature in humanitarian logistics,
with a focus on needs assessment, prepositioning, vendor managed inventory, framework
agreements, as well as the concept of public-private partnerships, and also includes a
discussion on indices and metrics as they relate to this research.
The Methods section describes the overall research design, model and data used in
conducting the case study analyses.
The FEMA Case Study section describes the assumptions and data unique to FEMA in
greater detail and presents the findings of the case study on the FEMA logistics network.
The FDEM Case Study section describes the assumptions and data unique to FDEM in
greater detail and presents the findings of the case study on the FDEM logistics network.
The Discussion section summarizes the key findings from each of the two case studies. It
also includes a discussion of relevant cross-case findings as they relate to the literature,
and finally it presents the framework for assessing disaster response capacity under the
broader umbrella of preparedness decisions.
14
2 Literature Review
Logistics is a critical aspect of any disaster response because it ensures that supplies get
to those in need. An understanding of logistics in the humanitarian context is important
because it differs from traditional supply chain and logistics theory and because this
understanding will facilitate the more detailed evaluation of certain aspects of
humanitarian logistics here. The aspects of humanitarian logistics that covered here are
elements of preparing the response infrastructure - how are needs determined, how are
goods stored and moved to reach the end user, and what policies and procedures support
that. Within the topic of preparing the response infrastructure the focus is on
prepositioning resources, vendor managed inventory and framework agreements. Next,
public-private partnerships (PPP) will be covered as an overall approach to engaging the
private sector in emergency response, presenting the general concept of PPP as well as
their application to logistics is discussed. Mechanisms that govern the overall disaster
response, such as federal or private organization policies and procedures, are not included
here but are presented with the individual case studies as they constitute the context for
these studies. Following the discussion of humanitarian logistics, literature related to
logistics and disaster response and resilience indices is reviewed. An index often
represents the quantification of a framework using benchmarks or ratings, and takes a
framework from conceptual to something that can be measured. Similarly, literature and
examples of performance metrics and evaluation were reviewed as a compliment to the
literature on indices for their application to the development of a framework.
The literature was identified through the review of several reports and articles
related to the central themes of this research, including humanitarian logistics, publicprivate partnerships, prepositioning, and indices (Acimovic & Goentzel, 2015; Gustetic,
2007; L N Van Wassenhove, 2005). Subsequently a backward look at papers that had
been cited in these initial papers was conducted to identify and supplement the review.
Then, a search looking forward at papers that had cited these original papers was
conducted (Webster & Watson, 2002). Some papers reviewed were newer and had not
yet been cited.
15
Figure 2-1 provides a visualization of the conceptual framework for this literature
review. The headings represent the broad topic, and the bullets beneath each heading
represent the main subtopics covered.
Preparing
Response
Infrastructure
-prepoddonWng
Public-Private
Partnerships
-
in giesnd
aer"SOeY
ru~ponsn
Indices and
Metrics
_nratrutpoure
-
Me'
tndkoe
id aMetrics In the
Hmiralits
meve
V man~d
-Disaster Response
framewok of the liertue
Figure* -1 Cocpta
Context
I
Pramawrk
Figure 2-1 Conceptual framework of the literature review
2.1
Humanitarian Logistics: Preparing the Response Infrastructure
Humanitarian logistics is defined as 'the process of planning, implementing and
controlling the efficient, cost-effective flow and storage of goods and materials, as well
as related information, from point of origin to point of consumption for the purpose of
meeting the end beneficiary's requirements' (Thomas & Mizushima, 2005).
'Requirements' in this definition is similar to needs, which drive the demand for goods
and services in a disaster. The planning, implementation and control processes are akin to
the activities for preparing the response infrastructure covered here. This section will
discuss both the needs as well as four strategies for preparing the response infrastructure
in advance of a disaster: prepositioning, vendor-managed inventory and framework
an
agreements. These strategies are presented because they are examples of methods that
emergency management organization might employ to streamline the response and
reduce the time to serve the disaster-affected population.
16
2.1.1 Needs assessment
Needs is a broad term that encompasses many things in daily life as well as in a disaster.
Darcy and Hofhiann (2003) present need as having several applications in the
humanitarian context including to describe basic needs (e.g. food), to describe a lack of
something considered essential, and to describe the need for assistance of some sort. The
focus in this study is most closely aligned with the first application outlined by Darcy and
Hofmann (2003), where needs are made up of items or supplies that fall into the 'basic
needs' category.
In a disaster response, an individual's needs are not satisfied by just one
organization. Needs are satisfied through a patchwork of resources provided by federal,
state and local government as well as non-governmental and faith-based organizations.
FEMA, through its Ready campaign, provides an 'Emergency Supply List' of items to
keep on hand in case of a disaster. At the top of this list are water, food, a means of
battery-operated communication, and a flashlight (Federal Emergency Management
Agency, 2014). Similarly, in Florida's Emergency Management Handbook, the top
needs listed are medical, water, food, shelter, and fuel (Fugate, 2009). These basic needs
drive the demand for goods that the humanitarian logistics network must respond to.
In order to ensure an effective response to meet the needs of disaster survivors,
organizations must make decisions on how to structure their logistics network. These
decisions are for the most part made in the preparedness phase, before a disaster occurs.
The goal of preparedness is to ready the whole community, including emergency
management organizations, state, local, tribal, and federal governments, faith based
organizations, as well as the non-profit and private sectors and individuals in the
community to more effectively respond in the event of a disaster. The compliment to the
preparedness mission is mitigation, reducing the risk of catastrophic damage in the event
of a disaster. Mitigation can be defined as "a sustained action to reduce or eliminate risks
&
to people and property from [such] hazards and their effects" (Haddow, Bullock,
Coppola, 2011). Sustained action is the operative phrase in the definition of mitigation, as
it implies a goal of long-term reduction of risk, which has real potential to drive reduction
in demand for critical commodities. This reduction in demand can in turn impact the
planning and allocation of inventory within the logistics network.
17
In order to prepared for and respond to these needs quickly and at minimal cost,
humanitarian logisticians rely on a number of strategies that are an important part of the
overall infrastructure of a response. As Kovics and Spens note, one key difference
between humanitarian and industry supply chains is that in the humanitarian supply
chain, 'relationships are built "just in case" and suppliers are needed with a capacity at
the time of need' (Kovacs & Spens, 2011). It is these "just in case" relationships,
enabling physical strategies such as prepositioning, how they are accomplished (e.g.
public-private partnerships, vendor-managed inventory), and the impact that these
strategies have on the overall response that is emphasized in this thesis.
2.1.2
Prepositioning
Prepositioning is defined as 'the storage of inventory at or near the disaster location for
seamless delivery of critical goods' (Ukkusuri & Yushimito, 2008). Because the goods
are located closer to the disaster affected population, the time that it takes responding
organizations to provide these life-saving goods to people in need is far less than if
prepositioning had not been initiated (Duran, Gutierrez, & Keskinocak, 2011). Because
the goods must be in place prior to an event occurring in order to be the most effective,
prepositioning requires additional coordination by the responding organization to ensure
that enough of each of the essential commodities are placed in the appropriate locations.
The literature on prepositioning in the humanitarian context relates to either where
the optimal location of prepositioning sites should be (e.g. G6rmez, K6ksalan, & Salman,
2010; Ukkusuri & Yushimito, 2008), or what quantity of goods should be placed in
&
designated prepositioning locations (e.g. Guo, Wang, & Liu, 2014; Lodree Jr., Ballard,
Song, 2012; Ozbay & Ozguven, 2008), and sometimes both (Duran et al., 2011).
Specifically focusing on the allocation of goods in designated prepositioning locations
here, Guo, Wang, and Liu have proposed a methodology that takes into account risks and
dynamics in the humanitarian supply chain on both the supply and demand sides to
optimally allocate supplies throughout a logistics network. The stockpile capacity model
developed by Acimovic and Goentzel (2015) and adapted for use here suggests a method
for answering the question of: given locations for prepositioning supplies, what quantity
of each item should be stocked in each location within a network to minimize the
response time or cost. Similarly, Duran et al. (2011), work with the relief organization
18
CARE International to determine the optimal warehouses of a finite set of nine
warehouses to open, and how much to stock in each warehouse. The key differences
between the work of Acimovic & Goentzel (2015) and Duran et al. (2011) are that the
former considers cost as well as time in decisions about where to locate stock, and also
includes truck travel. Acimovic & Goentzel (2015) also develop a set of metrics through
their analysis in order to facilitate future evaluation of a logistics network. Also similar to
these is the method used by Lodree Jr. et al. (2012), where a two stage stochastic linear
programming model is used to evaluate the optimal allocation of supplies throughout the
logistics network of a commercial retailer to meet the pre-event demand for commodities.
2.1.3 Vendor-managed inventory and framework agreements
In addition to the tool of prepositioning resources closer to the disaster-affected
population, emergency response organizations have options for how they procure and
store essential goods like water and food. In Spens and Goentzel's case study on the
Florida Division of Emergency Management they present the concept of vendor-managed
inventory (VMI), where the vendor of the good rather than the emergency response
organization, in this case the State of Florida Division of Emergency Management
(FDEM) owns the good (Goentzel & Spens, 2011). The FDEM purchases the good only
when it leaves the warehouse in a disaster response. This benefits both the FDEM as well
as the vendor. The FDEM does not have to stockpile large quantities of goods in case of
an event and risk these goods expiring, and the vendor gains additional space to store
goods to supplement their own supply chain in the event of a disruption. VMI is a
concept that has been used in commercial supply chains (Waller, Johnson, & Davis,
1999), but is relatively new to humanitarian logistics.
Similar to VMI, Framework Agreements (FAs) are another tool used by
humanitarian agencies to ensure that they have the critical commodities to meet the needs
of disaster survivors immediately following an event. In FAs, suppliers "reserve
inventory for the relief organization and promise to deliver supplies according to prespecified terms (such as pricing, packaging, etc.) once an order is made" (Balcik & Ak,
2014). The response organization then decides, after a disaster, whether or not to activate
the agreement. Similar to prepositioning, depending on the locations of the supplier's
warehouses, FAs could place critical commodities closer to the disaster-affected
19
population. Compared to VMI, FAs mean that the stock is stored in the suppliers'
warehouses rather than a response agency's warehouse. Another key mechanism that
disaster response organizations rely on is public-private partnerships, which can and often
do go hand-in-hand with prepositioning, VMI, and framework agreements.
2.2
Public-Private Partnerships
Humanitarian logistics research often cites the benefits of pulling strategies from and
working with the private sector (Van Wassenhove, 2005; Van Wassenhove & Pedraza
Martinez, 2012). This makes intuitive sense because private organizations are moving
goods through their supply chains daily, whereas humanitarian supply chains are
activated relatively infrequently. Additionally, in these infrequent activations,
humanitarian supply chains are moving essential and often life-saving goods in less-thanoptimal conditions and under immense pressure to shorten delivery time. Van
Wassenhove and Pedraza Martinez (2012) stress that a "successful response implies
quickly building a supply chain," and that "responding is a less difficult task if the
response system is well prepared". Partnerships between private sector organizations that
are moving goods on a daily basis and emergency response organizations are an
important avenue through which expertise and resources can be shared to ensure that the
response system is well prepared.
Public private partnerships (PPP), or public private collaboration is defined by
Donahue and Zeckhauser as 'the pursuit of authoritatively chosen public goals by means
that include engaging the efforts of, and sharing discretion with, producers outside of
government' (Donahue & Zeckhauser, 2006). This definition is similar to those in other
literature and reports on public private collaboration or partnerships (Gustetic, 2007;
National Council for Public-Private Partnerships, n.d.; Osborne, 2000). A few have
looked at applying public-private partnerships to disaster or emergency response or to
logistics in a broad sense. Gustecic (2007) suggests a separation of responsibilities
between public and private that is divided between emergency preparedness and
emergency response, with the private sector handling the response and recovery side.
The Council on Foreign Relations report titled 'Neglected Defense' states that the private
sector is "adept at providing material, logistics, and know-how to provide relief in the
aftermath of a disaster" and echoes the sentiment expressed in many papers on public-
20
private partnership in disaster response: that the private sector should not be an
afterthought in planning (Flynn & Prieto, 2006).
How then, should these observations be enacted? How should the private sector
be included in the disaster response process? Several make the observation that private
organizations, like individuals, feel some sense of responsibility to their communities and
have a sense of civic duty, and that public organizations charged with disaster response
should take note of this and engage with the private sector (Business Response Task
Force, 2007; Flynn & Prieto, 2006; Tomasini & Van Wassenhove, 2009). In a review of
several case studies on public private partnerships undertaken at each stage of the disaster
response life cycle, Chen et. al found that "trust, reciprocity, and commitment to the
collective", as well as alignment of incentives were key factors of success for a PPP
(Chen, Chen, Vertinsky, Yumagulova, & Park, 2013). The Business Response Task force
suggested systematic integration of the private sector into the nation's response to
disasters, including removing legal hurdles that make working with the government less
predictable and efficient (2007). This is especially important, they note, for the
purchasing and vendor relationship regulations. In summary, the interests and incentives
of public and private entities must be aligned, and legal and procedural hurdles to
partnerships between the two should be removed for public-private partnerships to be
successful.
Once partnership can be achieved, the information and resource sharing can benefit
both the public and private sectors (Bergqvist & Pruth, 2006). The public sector gains the
expertise from companies moving goods on a daily basis, and the private sector benefits
from a business perspective with the opportunity for learning and growth as a result of
their involvement (Tomasini & Van Wassenhove, 2009).
2.3
Indices and Metrics for Evaluation
Having sound strategies such as prepositioning and vendor managed inventory, as well as
strategic relationships - including public private partnerships - are all essential in
facilitating disaster response. An understanding of each of these mechanisms is essential
in framing this research. As part of the development of a framework for assessing a
community's ability to respond to a disaster, it was also important to understand the body
of research around indices. The review of these indices will inform what is needed in a
21
framework to get towards an index. An index combines quantitative benchmarks for each
aspect of a framework, and in doing so allows for a normalized means of comparison
across sectors, organizations, states, etc. Much of the literature on indices focuses on
economic indices, but for the purposes of this research, only logistics, disaster response,
and resilience indices are reviewed.
2.3.1
Logistics indices
Many of the logistics indices that exist were developed by international organizations like
the World Bank, and not unsurprisingly focus on metrics related to international trade.
For example, the air connectivity index was developed as a new measure of connectivity
in the global air transport network. The intent of the index is to show how connected a
given country is via air as an important indicator of the overall level of service provided
by the global air transport system (Arvis & Shepherd, 2011). Hoffmann (2012)
developed the Liner Shipping Connectivity Index (LSCI) in an effort to measure the
containerization of trade and access to containerized transport services because these are
important factors in determining a country's competitiveness in trade. It is calculated as
the normalized average of five components. Next, the Logistics Performance Index
(Arvis et al., 2014) "measures the on-the-ground efficiency of trade supply chains, or
logistics performance." The index is developed based on results from a survey of freight
forwarders, and uses Principal Component Analysis to weigh the responses to the survey
questions into one number.
2.3.2
Disaster response and resilience indices
As with other indices, the disaster response and resilience indices are aimed at
quantifying the state of some service or entity, usually with the intent of providing a
decision-maker or policymaker with more data and evidence to support their decisions.
Berkeley's Institute of Governmental Studies BuildingResilient Regions Report presents
the Resilience Capacity Index. The index is accessed through a web interface, similar to
the National Health Security Preparedness Index, which is described next. The index,
from the webpage is a "single statistic summarizing a region's score on 12 equally
weighted indicators-four indicators in each of three dimensions encompassing Regional
Economic, Socio-Demographic, and Community Connectivity attributes" (Foster,
22
201 1).The purpose of this is to gauge economic resilience and to give decision makers a
tool for making more informed decisions on shoring up economic resilience. It is a
simple tool to use, and in that is powerful in its messaging.
The National Health Security Preparedness Index is an index that aims to quantify
health security preparedness in the United States by looking at health security
preparedness in each individual state (Robert Wood Johnson Foundation, 2013). It is
updated annually, and the target scores for each portion of the index were determined by
an expert panel, by regulatory requirements, by the leading literature, or were statistically
determined. The purpose of this index is to "Strengthen preparedness, inform decision
making, guide quality improvement, and advance the science behind community
resilience" (Robert Wood Johnson Foundation, 2013).
While not explicitly focused on developing an index, Cutter et. al (2010)
formulate a set of indicators and baseline characteristics to measure community
resilience. Indicators for social, economic, institutional, and infrastructure resilience as
well as community capital are included in what Cutter et. al (2010) term the disaster
resilience index. The main focus of the study is to develop baseline metrics from which to
measure improvement within a community or compare across communities on similar
metrics that would inform an index (referred to as a disaster resilience index), rather than
on an index as a whole. Similarly to the Resilience Capacity Index (Foster, 2011),
measures of socioeconomic resilience are included, but in addition to the Resilience
Capacity Index, the indicators proposed also look at the role of the physical and
community infrastructure in disaster response (Cutter et al., 2010). The indicators and
sub-indicators included by Cutter et. al are fairly comprehensive and were developed to
use readily available data. They do not include indicators to measure resilience of a
logistics network or a community's stockpiles of critical commodities, which might be
informed by the set of indicators and index that were developed recently by Acimovic
and Goentzel (2015). Acimovic and Goentzel (2015) have developed the Response
Capacity Index (RCI), which "measures the quality of stock deployment in the global
disaster response network." This index, which is formulated using the results from a twostage stochastic linear programming (LP) model that analyzes stockpile capacity, will be
applied in this thesis, along with several related metrics. The metrics and index
23
specifically evaluate the ability of a logistics network to meet the needs of affected
individuals after a disaster, and could be used in complement to a broader set of disaster
resilience indicators or index, like that outlined by Cutter et al. (2010) to inform
community decision-making.
2.3.3
Metrics for evaluating performance in disaster response
A number of the topics covered thus far have related to or been relevant for an evaluation
of a logistics network's ability to meet the needs of a disaster-affected population.
Prepositioning, if implemented effectively, has the potential to greatly reduce the time to
serve a disaster-affected population (Duran et al., 2011). Indices are tools that can be
used on a continuous basis to update and evaluate the allocation of supply throughout the
logistics network. In addition to indices, literature on developing metrics in the
humanitarian context in general is reviewed here to provide additional background on
techniques to evaluate a logistics network.
FEMA's National Disaster Recovery Framework (2011) stresses the importance
of metrics for tracking and evaluating the progress of disaster recovery, and is one of the
few frameworks reviewed that includes detail on how this might be achieved for disaster
recovery. It does not suggest specific benchmarks or metrics but it does walk the reader
through a number of activities that should be done to facilitate tracking progress in
disaster recovery, and developing metrics for assessing how well a community is doing.
The three overarching activities suggested include conducting a baseline impact
assessment after a disaster to understand the extent of the disaster's impact on the
community, identifying the desired outcome as holistic results rather than simply
numerical targets for units delivered or infrastructure constructed, and conducting a
cross-sector assessment that includes sectors like environmental, housing, health,
business, and infrastructure in the picture for what it means to conduct a successful
recovery (Federal Emergency Management Agency, 2011). This is a good point of
reference, but does not provide clear benchmarks or metrics for evaluating progress.
Developing clear, effective metrics in logistics is challenging. Davidson (2006)
cites some of the challenges of implementing metrics in the non-profit sector, specifically
focusing on the humanitarian sector. She notes that the culture of the non-profit sector
may make the implementation of a set of performance indicators or metrics difficult,
24
citing overall organizational culture as well as communication within the organization,
across departments. Caplice and Sheffi (1994) develop a method for evaluating logistics
metrics in general, suggesting eight criteria for assessing the overall quality of a metric,
including the metric's validity, robustness, usefulness, integration, economy,
compatibility, level of detail, and behavioral soundness. While developed for application
to traditional logistics, these criteria can be used to evaluate metrics in humanitarian
logistics as well, and to inform the use of metrics to assess the logistics network in
humanitarian response organizations.
The background literature in humanitarian logistics, including needs assessment,
prepositioning, VMI, FAs, as well as PPPs, indices and metrics were used to frame the
case studies and discussion around how this research can be applied. An understanding of
these concepts informed the methods used and the analyses conducted in the case studies.
Finally, the Discussion section is used to describe how the concepts presented here relate
to the findings of the research.
25
3 Methods
This section describes the methods used to answer the three central research questions
First, the research questions and the overall research design are presented, including an
outline of the case study analysis, the formulation of the model used in the study and the
metrics derived from the model. Next, the data collected and the data collection and
cleaning process is described. Finally, the methods by which the three central research
questions will be analyzed using the model and metrics derived from the model are
described.
3.1
Research Question
This thesis answers three central questions that informed the development of a framework
for evaluating the capacity of disaster response organizations to meet the needs of a
disaster-affected population after an event. These are:
1. Where should inventory be placed, considering both normal allocation of inventory as
well as allocation of inventory in advance of a notice event such as a hurricane?
2. How does private sector involvement change the response capacity of a logistics
network?
3. What is the impact of reduced demand for critical commodities on the logistics
network's ability to respond?
These questions can inform important decisions regarding the logistics network in an
emergency management organization. The research relied on metrics developed by
Acimovic and Goentzel (2015) to evaluate these questions, including an index combining
several evaluation metrics. The sections that follow describe the specific research design
and the model, as well as the metrics used to assess these three research questions.
3.2
Research Design
An exploratory case study method was used here for two reasons. First, because this
thesis aims to describe natural phenomena in depth and in the context of realistic
scenarios and assumptions, and second, because the study incorporates multiple data
sources with the understanding that there may be items of interest for which no data are
available (Yin, 2009). Many emphasize the connection of theory to reality as an
important reason for applying case study research in general, and in particular emphasize
the use of exploratory case studies to advance theory (e.g. Edmonson & McManus, 2007;
26
Pedraza Martinez, Stapleton, & Van Wassenhove, 2009; Stuart, McCutcheon, Handfield,
McLachlin, & Samson, 2002). Furthermore, the case study method has been used some in
humanitarian logistics literature but could and should be used more often, Kunz and
Reiner (2012) assert in their meta-analysis of the literature.
The research was conducted using a two-stage stochastic linear programming
(LP) model (Acimovic & Goentzel, 2015), actual supply data from emergency response
organizations, and disaster risk data generated for use in the insurance industry. The LP
model is used to further advance theory around the application of this type of model to
inform decision-making in the humanitarian context. Actual data from emergency
response organizations are used to simulate the intended context as accurately as
possible.
Two exploratory case studies were conducted on the disaster response logistics
networks for the Federal Emergency Management Agency (FEMA) and the Florida
Division of Emergency Management (FDEM). More specifically, an embedded multiplecase study design was used in this study. This design was chosen as the results from the
analysis embedded within each case will not be pooled across all cases, rather it will be
applied to the specific case in which it was conducted (Yin, 2009). The cases were
selected on the basis of available data as well as presence of natural hazards. Florida
experiences frequent natural hazards, and logistics data such as location of warehouses
and stock levels could be obtained either through agency contacts or publicly available
sources. In the case of FEMA, the Agency responds to disasters of all types across the
country, and inventory data from their warehouses were provided in support of this study
(FEMA Staff 1, 2015). The unit of analysis in each of the case studies is the logistics
network in the emergency management organizations studied (FEMA and FDEM), with
embedded units of analysis on specific decisions that would be made by these emergency
management agencies in preparing for a disaster. The analysis was conducted using the
stockpile capacity model as well as relevant metrics and the associated Response
Capacity Index (RCI) developed by Acimovic and Goentzel (2015).
3.3
The Model
In order to conduct the evaluation in the case studies, the stockpile capacity model, a twostage stochastic LP model ("the model") (Acimovic & Goentzel, 2015) was used. The
27
model takes in the locations of and stock in existing warehouse facilities as well as the
locations and quantities of post-disaster needs and seeks to minimize the time or cost to
meet the needs. It uses the existing locations of the warehouses and the existing stock
levels to determine both the optimal allocation of stock across warehouses to minimize
time or cost, as well as the time and cost of keeping the commodities in their original
locations. It is important to note that the model is intended to inform the positioning of
emergency response capacity in advance of a disaster's impact and it is not intended to be
precisely predictive of when supplies will arrive in a given response effort. The model
was originally developed to assess the stock allocation in the six United Nations
Humanitarian Response Depots located across the world to serve disasters that are also
spread across the world. Despite its development for this international context, the model
is flexible to analyze smaller geographic areas depending on the data that are input
(Acimovic & Goentzel, 2015). In this study, the model will be applied to the United
States as a whole and to the individual state of Florida.
3.3.1
Model formulation
The model has two stages, the first of which records the allocation of inventory (if
utilizing actual allocations) or determines the optimal allocation (if utilizing optimal
allocations) and the second of which allocates the resulting supply to demand. The
demand can occur at multiple disaster nodes for each disaster scenario. The model
minimizes either total expected time or cost to deliver items from the supply nodes to the
disaster locations. A dummy supply node is used to satisfy disasters where the demand
exceeds the available inventory in the system. Additionally, in certain cases a delivery
deadline can be imposed to include only solutions able to meet the needs within a given
time window.
28
The model formulation is as follows (adapted from the model developed by Acimovic
and Goentzel (2015) to include multiple locations per disaster event):
I 3 i - Set of all depots and warehouses except the dummy node
iw
Iw
=I
U Lw
- the dummy supply node
- Set of all depots including dummy node
K 3k
- Set of possible disaster scenarios
J 3j
- Set of possible disaster locations
M 3 m
- Set of disaster types
N 3 n - Set of line items
R 3 r - Set of transportation modes
Cijnr
Tijr
cwTw
- Cost in dollars to transport a single item n from i toj via mode r
- Time to ship a single item from i to j via mode r (item independent)
- The cost (time) from the dummy supply node to a disaster
)
Pk - Probability of scenario k occurring (1 pk = 1
SJmnt
- The domestic/local capacity to respond to a disaster in locationj of type m
at period t for item n
X - The inventory vector dictating how many items are stored at each depot i.
X's are the elements of this vector.
XE
N - Starting inventory in the system as a whole, not including the dummy node
TAP - The population vector dictating the Total Affected Population at each
locationj for disaster scenario k. TAPj,k's are the elements of this vector
fljmnt
dk,;
- Factor converting number of people affected into the demand for item n at
locationj at time t for disaster type m
max (TAP X/3j,m,n,t - Sjm,n,to 0)
- Actual demand for item n for disaster k
ykjnr
- Decision variable for how much of n to send from i to the disaster in
scenario k, which has locationsj, via mode r. (Note n is in in the subscript,
but for now the problem is decomposed by item, so it is redundant)
X
- The III dimensional vector of starting inventory in each supply node. Its
elements are X. (Depending on the formulation, this may be an input by the
29
user or a decision variable)
0 - Delivery deadline: If 0 < oo, arcs whose Tyr > 0 are removed for both cost
and time objectives
The stochastic LP that minimizes time is (note: the LP that minimizes cost is analogous):
Vw(X,n)
min
pk
k
I
Ti,j,rkYnr
iElW,j,r
Vk
yy.dk
S. t.
i EIW,r
I
Y~nr
yunr -
niIJ
XVi
E I, k
r
Yynr
0
Vi, k,r
The formulation that optimizes inventory allocation is very similar, with X as a decision
variable, and with an additional constraint to ensure the sum of the supply is equal to X.
VoPT,W x
=n)min
k
k
k
iEiW,j,r
S. t.
yL
= dnk
i EIW,J,r
y nr
X
Vi E I, k
r
Xi =,x
tel
>0nr
y
X1
30
0
ViE Iw,k,r
ViEI
The objective values with the dummy costs subtracted are defined as:
Vw(X, n) -
V(X,n) E
p
w, 1 r Yw,n,r
r
k
VOPT (X, n)
j
VOPTW(X, n) -
pk
k
TLwjr
Ywjk
r
r
&
where the yiw,n,r's are fixed as the respective solutions for the above LPs (Acimovic
Goentzel, 2015).
3.4
Metrics Derived from the Model
The model solutions are used to derive several operational metrics, some of which were
applied in the analysis. The most relevant metrics for this study and how they are derived
are explained here. The dummy value TMw can be replaced by cw if cost if being
minimized or measured instead of time.
V(Xn)
VOPT (, n)
I
y
Balance Metric
pk dk
Weighted average of demand
pkmin (dkX)
Average demand met
k
I
I'
k
Y'
>Lpk
Fraction of disasters served completely
k:dk X
V(X)
It I
Average time (cost) per unit delivered
,
S8
Fraction of demand served
31
To calculate A requires solving both LPs. The metrics y, it', y, and 6 can easily be
calculated without solving any LPs. The following sections describe the metrics selected
for this analysis from those described in Acimovic & Goentzel (2015) in detail and how
they were used to evaluate inventory decisions.
3.4.1
The balance metric
The balance metric is intended to measure whether a given amount of inventory is
generally in the correct place or not. More specifically, it estimates how far out of
balance the actual allocation of inventory is relative to the optimal, and is based on the
metric set forth by Acimovic and Graves (2015).
Properties of the balance metric, as outlined in Acimovic and Goentzel (2015) include:
1. It is an approximation of the fractional increase in cost (time) to serve
beneficiaries given that one's inventory is allocated as it actually is as opposed to
being allocated optimally. As such, if inventory should be in Mountain View, CA
but it is actually across the country in Frederick, MD, the metric will suffer.
However, if the inventory should be in Frederick, MD, and it is actually nearby in
Cumberland, MD, the balance metric value will not change substantially from
optimal.
2. The optimal value is 1. Anything greater than 1 is considered out-of-balance.
3. It is not strongly affected by abnormally large disasters in the dataset, i.e., it is
relatively robust to outliers. Often, the largest few disasters in a set of scenarios
far exceed the on-hand supply for any item. The objective functions and solutions
of the model depend only on those people who can be served by the on-hand
inventory. People affected by a disaster who cannot be served by the current
inventory due to a lack of supply have no bearing on the optimal allocation. Thus,
if an item can serve only 1,000 people, then the optimal allocation, the time-toserve those who can be served, and other output metrics will be identical whether
the largest disaster across the scenarios affects 1,000 people or 10,000,000 people.
4. The metric is affected by which depots are considered. If a new depot is opened in
a disaster hotspot and no inventory is actually moved there, the balance metric
will increase (because V"PT(X, n)will decrease). In this sense, operational
32
managers can be alerted to the fact that the inventory is out-of-balance given the
new depot.
3.4.2
Fraction served and fraction of disasters covered
Fraction served (X) represents the fraction of the weighted average of demand met, where
the weights are derived from the disaster scenario probabilities. This value gives a sense
as to whether the inventory of items stored in the depots in total is appropriate. It does not
depend on how the items are allocated among the depots (unless the demand deadline
0 < oo). This metric can be influenced by very large disasters. The inclusion of a disaster
scenario affecting tens or hundreds of millions of people will have a significant impact on
g, the denominator in the fraction that defines X. Thus, these values in general may
appear low, and must be interpreted with this in mind.
Fraction of disasters covered (6), on the other hand, is relatively robust to outliers.
If there are only 1000 items in stock of an item, then whether an unserved disaster affects
1001 or 10,000,000 people is irrelevant in the calculation of S. This robustness comes at
the cost of not conveying the magnitude of the disasters that go unserved. 8 provides
different information from X, and the two together can help operational managers
understand the adequacy of the total inventory level (Acimovic & Goentzel, 2015).
3.4.3
Time per unit delivered
The value <p represents the average time (cost) to deliver one unit to a beneficiary from a
depot. This will, of course, not be equal to the actual time to deliver an item (which
would be stochastic itself and would depend on specific factors such as traffic and
weather), but rather will be the time from the depot to the location of the disaster as
calculated using the assumptions and data described in Section 3.5.3. Even though this
number should not be used to estimate how long it will take for supplies to arrive at a
disaster site, it can provide valuable and objective information when comparing scenarios
and strategies with each other (Acimovic & Goentzel, 2015).
3.4.4
Response Capacity Index
The RCI is an index to evaluate the quality of stock allocation for an organization or set
of organizations providing humanitarian aid (Acimovic & Goentzel, 2015). It is meant to
33
be a tool to allow decision-makers to evaluate and compare options for positioning of
inventory, because it provides a common standard on which to compare these decisions.
Each of the five metrics included in the RCI is scaled from 1-10 according to the
best and worst possible values for that metric. In the case study sections where RCI is
used, the specific values used as the best and worst for each of the metrics are explained,
as well as where the actual value fell between those values and how it was scaled.
1. Optimal time to serve
(OPT)
refers to the average time to serve if the inventory is
optimally allocated. This evaluates not only the total inventory level, but also the
network of warehouses because it takes into account the optimal allocation of
inventory. In essence, it is a way to evaluate how well positioned the warehouses
in a logistics network are because the stock allocation cannot be improved further
from the optimal.
2. Balance metric, time (Ar) refers to the balance metric when minimizing time. The
balance metric is calculated as shown earlier to describe how well allocated the
inventory is currently as compared to the optimal allocation. The best and worst
possible values may differ depending on the range of potential scenarios present
in the logistics network being evaluated.
3. Balance metric, cost (Ac) refers to the balance metric when minimizing cost. It is
calculated similarly to the balance metric for time.
4. Fraction of disasters covered (6) refers to the fraction of disasters that are served
completely given the current inventory level. This number is robust to outliers.
The worst value is 0, and the best possible value is 1 for all cases, so the value is
simply multiplied by 10 to scale it for the RC.
5. Fraction of disasters covered in 12 hours (612) refers to the fraction of disasters
that are served completely in 12 hours time. This is included in the RCI to serve
as another way of evaluating the overall positioning of warehouses, independent
of total inventory in the system. Similar to the fraction of disasters covered, the
worst value is 0 and the best possible value is 1 for all cases, so the value is
simply multiplied by 10 to scale it for the RCI.
34
It is important to note that the RCI represents an initial proposal of a methodology
for evaluating the overall capacity of a logistics network to meet the needs of disaster
survivors. It could benefit from additional input from experts in terms of what metrics
should be included as part of the RCI and how each of the metrics should be weighted.
The intent of the application of the RCI in this thesis was to apply it to a new context and
evaluate how well it transfers to this new context, as well as to suggest methods for
evaluating components of the index.
3.5
Data
This section describes the data that are used in this thesis, as well as key assumptions
made regarding these data. First, the disaster risk data are described, followed by
stockpile and finally time and cost data are described.
3.5.1
Disaster data
The model used in this thesis is not a model focused on forecasting potential future
disasters and their impact. Instead, it assumes that the future is similar to the past, and
uses a catalog of past events to estimate future damage, assuming each has an equal
probability of occurring in the future (Acimovic & Goentzel, 2015). The model is flexible
to any proposed disaster forecast, and two types of disaster data are used in the analysis
conducted for this thesis. They are described in detail below:
3.5.1.1
EM-DA T data
The Centre for Research on the Epidemiology of Disasters manages the Emergency
Events Database (EM-DAT). This database includes records on the impacts of over
18,000 disasters across the world from 1900 to present (Centre for Research on the
Epidemiology of Disasters, 2015). For the purposes of this thesis, EM-DAT data were
used for disasters impacting the continental United States from 1990 until the summer of
2013 in the assessment of FEMA stockpile data, and are only considering sudden onset
natural disasters. The key adjustments to these disaster data for analysis include:
1. Each disaster event is associated with only one location. While the model as
adapted from Acimovic & Goentzel (2015) can accommodate greater than one
location per disaster event, this assumption is a function of the EM-DAT dataset.
Most disasters in the dataset only have one location and where events have
35
multiple locations the affected population is not separated by location, which
makes it difficult to discern the distribution of affected population throughout the
locations given. There were a total of 317 events in the database impacting the
continental United States, 24 of which had locations that were difficult to
reconcile (e.g. impacted locations that were far apart geographically, or included a
county name with no state name, or were blank). These 24 records were removed
from the total set of 317 events. For events where multiple locations were
recorded in the EM-DAT record, a central location was estimated (e.g. for
Caroline County, VA the town of Milford, VA).
2. In line with Acimovic & Goentzel (2015) the "Total Affected" field in the EMDAT database was used to measure the number of people affected by a disaster.
The EM-DAT database description of "Total Affected" is the sum of "people
suffering from physical injuries, trauma or an illness requiring medical treatment
as a direct result of a disaster", "people needing immediate assistance for shelter"
and "people requiring immediate assistance during a period of emergency; it can
also include displaced or evacuated people" (Centre for Research on the
Epidemiology of Disasters, 2015). This number was used instead of "Number
Killed" because it provides a better estimate of the number of people requiring
assistance in the form of the critical commodities which are evaluated here: water
and food. Records with null values for "Total Affected" were excluded from this
study.
3. The estimate for affected population is translated into units of an item demanded
by multiplying the per-person units of need for the initial push of supplies that a
logistics network would need to support following a disaster by the number of
individuals affected. In line with Acimovic and Goentzel (2015), the initial push
of supplies is assumed to be enough supplies for 72 hours (e.g. 2
MREs/person/day, over 3 days equals 6 MREs/person).
4. Only sudden onset disasters and epidemics affecting the continental United States
were included in the study of the FEMA disaster response logistics network. This
includes: earthquakes, epidemics, floods, mass movement dry, mass movement
wet, storm, volcano, and wildfire. Not included were: complex disasters,
36
droughts, extreme temperature disasters, industrial accidents, insect infestations,
miscellaneous accidents, and transport accidents.
&
5. Post-1990 data were used for completeness and homogeneity (Acimovic
Goentzel, 2015).
3.5.1.2
Insurance risk data
Data were obtained from DataCo, a firm specializing in catastrophe risk quantification
and decision analytics firm serving the risk and insurance industry, among others, to
enhance the model that estimate the residential and commercial damage as a fraction of
total insurable value in each county in Florida. DataCo combines a catalog of natural and
man made catastrophes with exposure data, vulnerability functions and mitigation
measures to develop estimates of property damage. For this project, the perils considered
included only hurricane wind, storm surge and associated precipitation caused damage
impacting the State of Florida, and did not take into account the man made events in
DataCo's catalog of events (DataCo Staff, 2015a; DataCo, 2015).
The DataCo Hurricane Model for the United States contains a 10,000-year
stochastic catalog of potential hurricane events, which provides 10,000 iterations of next
year's potential hurricane activity under current climate conditions. The Hurricane
Model used to generate the damage data for this project begins with the historical
hurricane events that have occurred since 1900. For each historical event, a number of
characteristics for each type of event, such as radius of maximum winds, minimum
central pressure, wind speed, and depth of storm surge in hurricanes, for example, to
generate the stochastic catalog of events. Each unique combination of these
characteristics defines a simulated event. For the State of Florida the catalog included
thousands of disaster event scenarios. The characteristics of the disasters combined with
physical characteristic data and vulnerability functions for each property allow DataCo to
provide an estimate of expected damage. The data used for this study are the aggregated
expected damage levels of all insurable commercial or residential properties for all 67
counties in Florida. It is important to note that these are out of all insurableproperty,
rather than all insuredproperty. This gives a more accurate representation of the
damages experienced in a disaster, as disasters do not solely impact properties that are
insured. The dataset used included both historic and stochastic disaster events.
37
The disaster risk data are structured in the format shown in Table 3-1:
Table 3-1 Structure of disaster risk data (DataCo, 2015)
Event ID
County FIPS
Industrial Damage
Commercial Damage
Residential Damage
X1,C%
X1,1,R%
1
1X1,1,%
1
3
XI,3,I%
XI,3,C%
X1,3,R%
1
5
X 1,5,1%
XI, 5,c%
X1,5,R%
The Event ID field is a unique identifier for each disaster event in the dataset. The
County FIP field is the field for the County Federal Information Processing Standard
(FIPS) code, a unique identifier given to counties and county equivalents in the United
States. The industrial, commercial and residential damage fields represent the fraction of
damage or loss over total insurable value. For each Event ID, industrial, commercial and
residential damages were reported for each county in Florida.
The key assumptions and adjustments made to the disaster risk dataset for Florida
include:
1. Each disaster event is associated with multiple locations. This is different than the
EM-DAT disaster database, where each disaster event is associated with only one
location. Only disaster-county pairs where the impact was greater than 100
persons were considered, as it is assumed that the county has the ability to
respond to events affecting less than 100 persons. A disaster-county pair
represents the unique combination of the disaster event ID and the county. That is
to say, no disaster impacted the same county twice in the dataset.
2. Total Affected Population for each event-county pair was determined by
multiplying Residential Damage (%)x 2014 County Population(State of
Florida Office of Economic & Demographic Research, 2015). Additional methods
for determining Total Affected Population were considered, but ultimately this
simple and intuitive method was used. This method was chosen over others
considered as it did not risk double-counting affected individuals by including the
effects of commercial and industrial losses (e.g. an individual whose home is
damaged and whose local grocery store is also damaged), or risk misinterpreting
38
the demand for critical commodities by estimating this figure based on the typical
demand for commodities in non-disaster times.
3. The estimate for affected population is translated into units of an item demanded
by multiplying the per-person units of need for the initial push of supplies that a
logistics network would need to support following a disaster by the number of
individuals affected. In line with Acimovic and Goentzel (2015), the initial push
of supplies is assumed to be enough supplies for 72 hours (e.g. 2
MREs/person/day, over 3 days equals 6 MREs/person).
4. Each disaster event, whether historic or stochastic, was assumed to have an equal
probability of occurring in the future.
5. Only hurricane wind, storm surge and associated precipitation damages were
included in the disaster risk dataset used in this project (DataCo Staff, 2015a).
6. The damage level was modeled at the property location level, then aggregated for
each county.
7. The location of disaster events was assumed to be the county seat for the affected
county identified in the disaster risk dataset, because this is most likely to
represent a central location within the county where individuals might go to
receive aid, or where local emergency management or response organizations
might set up distribution points.
3.5.1.3
Simulated propositioningdata
Data from a disaster response training exercise were also obtained from DataCo to run
the prepositioning scenario analysis. The data provided were based on an actual hurricane
event: Hurricane Wilma in 2005, and consisted of disaster impact forecasts at several
time intervals prior to the time when the hurricane made landfall on October 24, 2005 at
approximately 6:00 a.m. CDT. The data from these forecast reports that were used for
the prepositioning analysis was the population at risk reported in two state level reports
(DataCo, 2014). The first report and associated analysis were generated for the conditions
and forecasts that existed at October 21, 2005 at 7:00 p.m. CDT, and the second report
and associated analysis were generated for the conditions and forecasts that existed at
October 23, 2005 at 1:00 p.m. CDT. Each analysis used DataCo's stochastic modeling
39
techniques to expand the operational forecast model ensemble (40-50 discrete simulation
tracks) to create an enhanced ensemble (500 discrete simulation tracks).
The population at risk is reported as a statewide average displaced population, and
is broken into population at risk from just storm surge, just wind, and both storm surge
and wind. The combined figure for displaced population from wind and surge is used
here. For each member of the enhanced ensemble considered in each report (forecast
period), the displaced population was considered to be100% of the census block
population if estimated winds in a census block were over 74 mph or storm surge was
over 4 feet. Summing that figure across the state for each scenario and taking the simple
average of the 500 scenarios gives the average population at risk estimate for that forecast
period (DataCo Staff, 2015b).
In addition, for each report, every county in the state is classified as having a high,
medium, or low risk of being displaced by wind only, surge only, or wind and surge. To
arrive at a county-by-county distribution of the displaced population for each of the two
forecast periods considered these two pieces of information were combined. First,
counties were coded with a (1) if designated as low risk, (3) if medium, and (5) if high,
the sum of scores for the state was used to normalize each county as a fraction of the
overall state score. This normalized fraction was then multiplied with the average
displaced population to estimate the total affected population in each county for the two
forecast periods considered in this analysis.
Hurricane Wilma made landfall at on October 24, 2005 at approximately 6:00
a.m. CDT (NOAA, 2005). The two forecast periods considered for the prepositioning
analysis were Module 1 on October 21, 2005 at 7:00PM CDT, and Module 3 on October
23, 2005 at 1:00 p.m. CDT. Traditionally, a key piece of data that would be used by
decision makers in the United States to make prepositioning decisions is information on
the storm's path from the National Hurricane Center. Figure 3-1 and Figure 3-2 show the
National Hurricane Center's cone forecasts of the disaster impact for the closest time
approximations to the reporting times available from DataCo, for comparison of the data
that would have been available and used in decision making in Hurricane Wilma.
40
Figure 3-1 National Hurricane Center cone and warnings for Hurricane Wilma on October 21, 2005 at 4PM
CDT (5PM EST) (National Hurricane Center, 2005)
)
:ox. Distance Scale ( Statute Miles
125. tSO 3S 500
4urfcane Wma
:ctobor 23, 2005
10 AM CDT Sunday
WS TPCiatlonaI Hurricane Center
kdvlsory 33
:urrent Center Location 22.7 N 5.8 W
Wac Sustained Wind 100 mph
iiiM Movement NE at 8 mph
Cunent Center Location
Forecast Center Positions
H Sustained wind , 73 mph
S Sustained wind 39-73 mph
Potential Day 1-S Track Area
Hurricane Warning
'
Hurricane Watch
-
Tropical Storm Warning
Figure 3-2 National Hurricane Center cone and warnings for Hurricane Wilma on October 23, 2005 at 10AM
CDT (11AM EST) (National Hurricane Center, 2005)
41
The intent of the analysis for prepositioning is to answer the question of whether the
allocation of goods to prepositioned locations made, in this case, 57 hours (Report 1)
before an event differs significantly from prepositioned locations made 17 hours (Report
3) before landfall of a hurricane. While these intervals do represent relevant planning
intervals, ideally a third time window even earlier than 57 hours before the event would
have been provided, as that would more likely be when decisions about prepositioning
would first be made.
3.5.2
Stockpile data
Local, state and federal emergency management organizations keep stockpiles of critical
commodities in preparation for disasters. These stockpiles allow the organizations to
meet the immediate needs of disaster survivors faster than would be possible if supplies
had to be procured from a third party following an event. This section describes the
stockpile data that was obtained from the two organizations studied.
Stockpile data, including quantity in each warehouse and location of warehouses
were obtained from FEMA Logistics for both water bottles (liters) and meals ready to eat
(MREs). The stockpile data used for water bottles is current as of March 2014, and the
MRE stockpile data is current as of March 2015. In the State of Florida, data were
obtained from publicly accessible sources. Warehouse location data were estimated based
on the State of Florida Unified Logistics Section Logistics Operations and Management
training course and stockpile data are assumed to be similar to those cited by Goentzel
and Spens in their case study on FDEM (Florida Division of Emergency Management,
2013; Goentzel & Spens, 2011).
3.5.3
Time and cost data
This section describes how the distance, time, and cost from each warehouse to each
disaster site were calculated. These figures are calculated for truck transportation only,
and air and sea transportation are not considered as the study focuses on immediate
deployment of the stockpile. In the case of disasters impacting the continental United
States, it is assumed that stockpiles will be moved via truck. In line with Acimovic and
Goentzel (2015), it is also assumed that the cost of transporting goods is linear in the
number of units being transported. Thus, for each warehouse-disaster-mode triplet, the
42
time and cost-per-metric-ton-km are calculated for the route. This, paired with
information on the weights of each item, allows the cost of shipping to be calculated for a
single unit on a specific route (Acimovic & Goentzel, 2015).
In order to calculate the distance and time between a warehouse and a disaster,
Google's "Distance Matrix" application programming interface (API) (Google, 2014) is
used. This provides the duration and distance between two points based on driving.
Although one can prefer to not use ferries with this API, if a ferry route exists, it will
return the distance using a ferry. In line with Acimovic and Goentzel (2015), this is
assumed to be reasonable because if a ferry route exists, organizations may use this route
(or a similar one) to transfer goods via boat. With the distance for truck routes, the time
and cost are then calculated assuming there is a fixed and variable component. For cost,
moving a truckload of goods is assumed to incur 237USD fixed cost and 0.59USD per
metric ton per km variable cost via truck. In line with Acimovic and Goentzel (2015), the
time and cost parameters were estimated based on data spanning numerous commercial
and humanitarian projects conducted at the MIT Center for Transportation & Logistics.
As such, the parameters may vary among different organizations working in different
contexts in different locations.
To calculate the times associated with truck travel along the routes, the Google
API travel times were used for driving, rounding up to the nearest whole hour for all
drive times output using the Google API (Google, 2014). Driving is possible for all
scenarios studied in this thesis, including for FEMA and the State of Florida. In addition,
a fixed time was added to truck travel in the United States for loading and unloading
goods. This time varies for different organizations, but the standard time used for all
organizations was 6 hours. For inventory that is owned by the emergency response
organization being evaluated (e.g. FEMA physical inventory) this fixed time represents
the sum of the time that it takes for the truck to arrive at the distribution center where
supplies are located which was assumed to be 3 hours, the time to load the truck which
was also assumed to be 3 hours. For inventory that is not owned by the emergency
response organization but that is obtained through existing contract agreements, this fixed
time of 6 hours to acquire and load a truck was assumed to hold, and that there is an
additional fixed time of 6 hours added to the time to place and process the order. For the
43
purposes of this study it is assumed that driver rest will not be an issue as team drivers
will be hired to meet the need in the fastest time possible.
3.6
Case Study Scenarios
For each of the case studies an embedded analysis was conducted using the model output
and relevant metrics developed by Acimovic and Goentzel (2015) to evaluate a set of key
resource allocation decisions, representing the central research questions. The decisions
represent the types of issues that an emergency manager or logistician would face in
preparing for a disaster, and they are (1) Where should inventory be placed, considering
both normal allocation of inventory as well as allocation of inventory in advance of a
notice event such as a hurricane? (2) How does private sector involvement change the
response capacity of a logistics network? (3) What is the impact of reduced demand for
critical commodities on the logistics network's ability to respond? The following
sections describe how the metrics will be applied in the case studies to answer these
research questions.
3.6.1
Allocation of inventory
Allocation of inventory refers to decisions about where to locate stock of critical
commodities. These decisions are made during non-disaster times as well as in advance
of a notice event (e.g. a hurricane), and both applications are studied here. The analysis of
physical inventory allocation in non-disaster times will be done in both case studies and
will aim to assess the current allocation using several of the metrics derived from the
model output, including the RC. It will compare the actual and optimal distribution of
commodities across warehouses; show the allocation of critical commodities by inventory
level (as a unit of inventory is added, where should it be placed?), and an idea of the
demand served by hour with the current allocation as compared with the optimal
allocation.
These metrics will be evaluated for FEMA's current allocation of inventory, as
well as its contract (private sector) inventory, and FDEM's current allocation of
inventory in one warehouse, and prepositioning of inventory in staging areas throughout
the state.
44
In addition to evaluating metrics for the current allocation, the model is extended
from Acimovic and Goentzel (2015)'s application to evaluate prepositioning strategies
for a notice event in the State of Florida, such as a hurricane. In this case, emergency
managers can move supplies closer to locations most likely to be impacted by the event.
To conduct this analysis, the catalog of disaster risk scenarios considered in the model is
limited to only those that may potentially occur given the current trajectory of the storm
and date of prediction, as is described in Section 3.5.1.3. In general, the geographic scope
of these scenarios is narrowed as the storm approaches land and the forecasted track
becomes more accurate.
To evaluate prepositioning decisions using the model, the metrics for the actual
allocation of stock in the system in one central warehouse will be compared with metrics
for the optimal solution when sites that have been pre-identified as locations for
prepositioning supplies in a notice vent are included in the model. The metrics that will
be most important in evaluating a prepositioning strategy include the allocation of
inventory, the change in demand met by hour when adding a prepositioning location, and
the change in fraction of overall demand served when adding prepositioned locations.
Prepositioning makes sense when the disaster-affected population can be served faster,
but the emergency manager must also consider costs in decision-making about early
deployment of supplies.
3.6.2
Impact of private sector commitments or contractual agreements
Private sector organizations become involved in disaster response through a few
mechanisms. The details of the mechanisms (contracts, donations, etc.) by which they are
involved in the supply chain are not explored. Instead, this study is concerned with the
amount of goods that private sector organizations have committed to the response, when
they will be available, and where those goods are located.
The impact of private sector commitments will be evaluated here for contracted
private sector commitments to FEMA's logistics network in order to provide a method
for conducting future analyses of other forms of private sector commitments. The key
metrics that will be evaluated include the RCI, the actual and optimal times and costs to
serve for the inventory owned by the organization as compared with adding capacity
from private sector commitments, and the change in overall demand served with the
45
addition of the private sector capacity. Using FEMA as the test case, the metrics for the
current inventory levels will be compared with those when outside suppliers stock
becomes available after 24, 48, and 72 hours, according to the contract structure set up by
FEMA Logistics. This assumes the same demand, in aggregate, as is used to assess
physical inventory allocation.
3.6.3
Impact of resilience goals
To evaluate the impact of resilience goals put in place by emergency response
organizations or state governments, a change in demand is used as one way to quantify
the effect of resilience goals. Methods for estimating how much demand changes as a
result of resilience goals are not covered in this thesis, rather a demand reduction of 10%
is modeled to evaluate its impact on response metrics. Any value could have been chosen,
but 10% was selected as it is high enough to be expected to have some impact, but low
enough to be a plausible reduction in demand that could be achieved with increased
resilience.
The metrics evaluated to measure the impact of resilience goals include the
overall demand met with reduced demand as a result of resilience goals, as well as the
cost and time to serve the original demand as compared to the decreased demand.
Table 3-2 summarizes the case study analyses just described, including the
decisions informed by the analyses, the scenarios and metrics evaluated, and the data
required to conduct a given analysis. This provides a general guide for how the case
studies will be structured.
46
Table 3-2 Summary of case study analysis process
Decision
Scenarios
_________Considered
FEMA: Current
Allocation of
Inventory
Where to allocate
inventory?
Where to
preposition
inventory in a
notice event (e.g.
hurricane?)
Florida: Current
allocation of
inventory (all in one
warehouse) vs.
Allocation evenly
throughout the state
Metrics Evaluated
MtisEaued
*
Response
Capacity Index
e
Ability to meet
*
Current allocation of
inventory with
option to stock
closer to disaster
events in LSAs for
two time intervals
before a notice event
*
(e.g. hurricane)
*
*
demand
Comparison of
current vs.
optimal
inventory
allocation
Comparison of
current vs.
optimal
inventory
allocation
Ability to meet
demand
Time to serve
Data Required
DaRqird
*
Case Applied
To
Warehouse
locations
*
*
Warehouse
stock
Disaster data
(historic and
forecast)
*
*
FEMA
FDEM
All from base
allocation, plus:
prepositioning
*
locations
Cone of
expected
isaster
impact several
* FDEM
storm
Should I enter
into a contract
with the private
sector?
Compare current
inventory levels
with consideration
Current
allocation
strategy for
Private sector
perspective: How
much am I
improvng the
of ude t ader
capacity added after
24, 48, 72 hours,
according to the
contract structure
contract stock
vs. optimal
allocation
Demand served
by hour
*
All from base
allocation, plus:
e
Contractual
*
FEMA
private sector
agreements
Locations of
private sector
warehouses
response?
What is the
benefit, in terms
of cost and time
to serve the
affected
population, if
resilience goals
reduce overall
demand?
*
Reduction in
demand of a given
state or county of a
set percentage
e
Ability to meet
demand
Comparison of
current vs.
optimal
allocation of
inventory
47
All from base
allocation, plus
* Resilience
goals for a
county or state
*
e
FEMA
FDEM
3.7
Limitations
The limitations of this study are largely related to two factors. First, limitations exist
related to the data available for analysis. This section discusses the limitations to the
study, and areas where the results would be strengthened with better data. First, the
DataCo disaster risk data provided for Florida (DataCo, 2015) to represent the hazards
present include only hurricane hazards. This means that they exclude riverine flooding
hazards as well as other natural disaster risk that is present in these states (e.g. tornado,
severe thunderstorm).
Another limitation of this study related to data availability was the availability of
accurate stockpile data. For example, in the state of Florida, while data were obtained
from publicly available sources, these data are outdated. As a result, the FDEM analysis
findings are more useful in understanding general concepts and cannot be used to inform
current decision-making. Also, data on stock committed through partnerships and
contract agreements with the private sector were not obtained for Florida, which meant
that analysis on the impact of these commitments could not be conducted as was done for
FEMA's logistics network. This study would benefit from a comparison of the impact of
private sector commitments on state-level response to the impact of these commitments
on federal emergency response.
The second set of limitations relate to the model as currently formulated. First, the
model is not formulated to consider time-phased availability of stock in the optimal
solution (it can accommodate time-phased availability in the actual solution). That is to
say that if stock was not made available through contract terms until 72 hours after the
disaster, the model cannot, as currently formulated, delay the availability of that quantity
of stock for the optimal solution. Second, additional costs or time constraints associated
with decisions such as hiring team drivers or running into traffic slowdowns in transit are
not included. This is something that could likely be incorporated into the model relatively
easily. Third, the model does not consider the cost that would be incurred if the demand
for a disaster were overestimated and the goods had to be sent back to warehouse
facilities. These model limitations as well as limitations on data availability are important
to note, and if remedied would further enhance the results of the study.
48
4 Case Study: Federal Emergency Management Agency
This section presents the results of the case study on the disaster response logistics
network at the United States Federal Emergency Management Agency (FEMA). First, the
framework used within FEMA to guide disaster response is presented with a focus on the
parts of this framework that address logistics concerns. Next, the analysis assumptions
and process are described, followed by the results of the analysis and finally a discussion
of the implications of these results for FEMA.
4.1
Background
The National Response Framework (NRF) is a framework for response to natural and
man-made disasters and emergencies in the United States. The framework serves as a
guide for all responding agencies, including FEMA, and aims to coordinate the efforts of
the multiple stakeholders involved in a response operation. It "describes the principles,
roles and responsibilities, and coordinating structures for delivering the core capabilities
required to respond to an incident and further describes how response efforts integrate
with those of the other mission areas" (United States Department of Homeland Security,
2013).
The response mission area, according to the NRF (2013), is focused on ensuring
that the Nation is able to respond to all types of incidents, and the priorities of the mission
area are to "save lives, protect property and the environment, stabilize the incident and
provide for basic human needs". The FEMA logistics network is essential to the response
mission area, as it facilitates the movement of goods to a disaster-affected population.
When a disaster exceeds or, in certain cases, is expected to exceed the resources of a
state, the state may request Federal assistance under the Robert T. Stafford Disaster
Relief and Emergency Assistance Act (Stafford Act). It is in these cases where the FEMA
logistics network may be used to deliver essential resources.
The NRF describes the coordinating structures that serve as a framework for
organizing resources from state and tribal governments, the federal government, and
other actors in a response (e.g. private organizations). The federal government and many
states organize their coordinating structures into Emergency Support Functions (ESF),
which combine resources and actors into key functional areas. ESF #7 is logistics, and
includes a number of core capabilities that are to be carried out by the ESF coordinators,
49
one of whom is DHS/FEMA. Relevant to this study are the response core capabilities of
providing the broad categories of public and private services and resources as well as
mass care services. ESF #7 is responsible for "comprehensive national incident logistics
planning, management, and sustainment capability" (United States Department of
Homeland Security, 2013). While the NRF sets up a framework for coordinating
response, and together with the Response Federal Interagency Operations Plan (FIOP)
(U.S. Department of Homeland Security, 2014) provides guidance to state and federal
agencies on critical tasks and responsibilities in a response, what is lacking in either of
these documents is a set of target performance metrics or measures of a successful
response.
Many of the decisions that impact response are made in the preparedness phase,
before a disaster strikes. This case study evaluated FEMA's logistics network using
forecast disaster data to assess stock level and warehouse decisions, and it suggests
metrics on which the ability of the logistics network to meet some of the goals outlined in
the NRF and Response FIOP can be evaluated. These metrics were selected from the
Stockpile Capacity Model ("the model") (Acimovic & Goentzel, 2015) presented in the
Methods section, and here those metrics that can be used readily in decision-making for a
logistics network are presented.
4.2
Analysis
This section will describe the analysis that was conducted using the model, associated
metrics and the Response Capacity Index (RCI) (Acimovic & Goentzel, 2015).
Supplementing the description of disaster and inventory data presented in the Methods
section, this section will describe the dataset for FEMA's logistics network in detail,
including any assumptions made for the purposes of this study.
4.2.1
EM-DAT disaster data
For the evaluation of FEMA's logistics network, only sudden-onset natural disasters
impacting the continental United States from the EM-DAT database were used. A total of
285 disasters are included, with total affected populations ranging from 11,000,048 to 1.
The geographic distribution and sizes of the disasters included in this analysis are shown
in Figure 4-1. The top 10 disasters range in total affected population from 11,000,048 to
50
-250,000, and are represented by the largest symbol on the map. The next 17% of
disasters represented by the medium-sized symbol range from -250,000 to -14,000 total
affected population, and the remaining 80% of disasters have a total affected population
below 14,000 and are represented by the smallest symbol. As described in the Methods
section, the EM-DAT field 'Total Affected' is used as 'Total Affected Population' in the
model calculations. According to conversations with FEMA Logistics staff, FEMA
estimates the affected population using a combination of modeling using FEMA's Hazus
model and assumptions based on the population in the impacted geographic area.
Specifically, the impacted population is made up of the "at-risk population" who would
come to a shelter to receive food, but would not stay overnight and the "shelter-seeking
population" who would seek overnight shelter following a disaster. If utility outages
continue longer than 72 hours after an event, the shelter-seeking population is expected to
increase significantly (FEMA Staff 3, 2015).
Vancouver
Sa
Montreal
@
OR
*
0fVc-c
die
L s
II
~Havana
Mexico
Santo
Cty
Domingo
44
8
,
LU,,
Guatemala
> 14,000 to 250,000
Caraca5
1 to 14,000
Figure 4-1 EM-DAT disaster locations and affected population, continental United States 1990-2013 (Esri, 2015)
51
The disaster events in the EM-DAT dataset of 285 disasters shown here are fairly evenly
distributed across the continental United States, accurately reflecting the different types
of hazards that FEMA must respond to.
A basic comparison of the data from EM-DAT with the disaster risk data
provided by DataCo for the state of Florida was conducted to assess the extent of disaster
events covered in both datasets. Of the 14 events in the EM-DAT database that were
recorded as having occurred in the state of Florida, only 6 were identified by a
comparison of the dates provided in the EM-DAT database with the dates in the DataCo
dataset to also be in the DataCo dataset. This may be a function of differences in how
historic disaster data are collected in each organization, or on how assumptions are made
to arrive at total affected population. In comparing the total affected population as
estimated by the methods used for the DataCo data to the total affected population as
estimated by EM-DAT, estimates by EM-DAT were higher than those for the DataCo
data for half of the common events (three events), and lower for the other half. This split
is likely a result of several factors, the first being that EM-DAT does not break down
affected population by location, even if multiple locations for one event are recorded in
the database. This means that for each disaster there is only one loss figure, regardless of
the number of locations listed. As described in the Methods section, if multiple locations
were recorded, the geographic center of these locations was used for the EM-DAT data,
because there was no good way to break down the affected population between the
multiple locations identified. A second reason for the discrepancy between the two
datasets might be the method used here for determining total affected population from the
DataCo disaster risk data. As outlined in the Methods section, this assumption is used
here for simplicity and intuitiveness. A final reason for the discrepancy may be that the
EM-DAT total affected number includes injured, affected, and homeless. This may
represent potential double- or triple- counting of individuals who fall into multiple
categories (e.g. injured and homeless). The method used for interpreting the DataCo
disaster risk data here was intended to eliminate the possibility of double-counting
individuals. In summary, improved historic or predicted disaster data will improve the
accuracy of the model in determining where critical commodities should be placed to
minimize time and/or cost to respond. In the absence of a common methodology for
52
estimating the impacts of disasters in terms of total affected population, the best available
data and assumptions are used here.
4.2.2
FEMA inventory data
Inventory data were provided by FEMA for this study for water bottles and shelf stable
meals or meals ready to eat (MREs). Water bottle data used here is current as of March
2014, and MRE data is current as of March 2015. In addition to data on the current
physical stock of inventory for these two items, data on contracted stock of MREs were
also provided. Physical inventory was used here to mean the organic inventory owned by
FEMA and kept in FEMA warehouses. This is different from contract inventory, which is
procured from outside vendors through pre-arranged contract vehicles.
FEMA's physical inventory is located in five warehouses throughout the
continental United States. Contracted inventory is located in three locations in the
southeast and midwest. Figure 4-2 and Figure 4-3 show the locations of these
warehouses.
tinrtivJ
Montreali
Chicago Dewrri
Was ngton
At M a
Los Angeles
Havana
Esri, HERE, DeLorme, NGA, USGS
Mexco
Figure 4-2 FEMA permanent stock warehouse locations (Esri, 2015)
53
I Esri, HERE, DeLorme
Seattle
Monteal
Washington
Los Angde
A
at
Figre
cntrct
-3FEM
d fdE
rcatis (Esi,
tok
Hrwowre7us
015
Havana
2
2M ant
Es , HERE, DeLorme, NGA, USPr I Es , HERE, Del
me
Figure 4-3 FEMA contract stock warehouse locations (Esri, 2015)
To estimate demand by item, 3L of water per person per day and two MREs per person
per day were assumed based on conversations with FEMA Logistics staff and the
international Sphere Standards for water needs for drinking and food (FEMA Staff 2,
2015; Florida Division of Emergency Management, 2013; The Sphere Project, 2011).
Finally, it is important to remember how affected population is translated into units of
demand. The estimate for affected population is translated into units of an item
demanded by multiplying the per-person units of need for the initial push of supplies that
a logistics network would need to support following a disaster by the number of
individuals affected. In line with Acimovic and Goentzel (2015), the initial push of
supplies is assumed to be enough supplies for 72 hours (e.g. 2 MREs/person/day, over 3
days equals 6 MREs/person).
4.2.3
Analysis scenarios
Three scenarios were run to evaluate the FEMA logistics network, and are summarized in
Table 4-1. First, the current physical inventory allocation of water bottles and MREs
were run to evaluate the base FEMA logistics network. Next, the allocation of the
contract inventory of MREs was run to evaluate the additional benefit that adding this
stock to the physical inventory provides. Finally, the current physical inventory allocation
54
was run against reduced demand in Florida to evaluate the impact of reduced demand in
one area of the country. The sections that follow describe the results of each scenario
analysis.
Table 4-1 FEMA case study scenarios
Decision
cnsdered
Metrics Evaluated
e
Where to allocate
inventory?
Should I enter into a
contract with the private
sector?
Private sector
perspective:How much
Current Allocation of
Inventory
Compare current
inventory levels with
consideration of
current and outside
am I improving the
response?
supplier capacity
added after 24, 48, 72
hours, according to
Where would an
additional unit of stock
be the most valuable?
the contract structure
What is the benefit, in
terms of cost and time
to serve the affected
population, if resilience
Reduction in demand
in the state of Florida
by 10%
*
*
0
*
Current allocation
strategy of contract stock
vs. optimal allocation
Demand served by hour
0
*
*
Warehouse
locations
Warehouse stock
Disaster data
(historic or
forecast)
All from base
allocation, plus:
0 Contractual
private sector
e
agreements
Locations of
private sector
warehouses
*
o
goals reduce overall
demand?
4.3
Response Capacity Index
RespitsetCameettyeIade
Ability to meet demand
Comparison of current
vs. optimal inventory
allocation
Data Required
Ability to meet demand
All from base
Comparison of current
vs. optimal inventory
allocation
allocation, plus
* Resilience goals
for a county or
state
Scenario 1: FEMA Physical Inventory Allocation
This scenario analyzed FEMA's current allocation of physical stock of water bottles and
MREs. The key metrics assessed were the ability of the physical stock to meet demand,
the quality of the inventory allocation, a comparison of current and optimal allocation
strategies and finally an assessment of the physical inventory using the RC.
55
4.3.1
Ability to meet demand
The first metric assessed for physical inventory was the ability of this inventory to meet
demand. Demand is represented as the weighted average of demand, as defined in the
Methods section, where each disaster has an equal weight or probability of occurring.
With this, large disasters carry no more weight than small disasters. It is important to note
that units of demand are reported here in terms of number of items, rather than in terms of
people demanding the items. Table 4-2 includes several metrics that are used to evaluate
the ability of FEMA's physical inventory to meet demand.
Table 4-2 Ability of FEMA physical inventory to meet demand
Demand
Demand Met
Fraction of
demand served
Fraction of
disasters served
Item
Units
MRE
7,319,000
1,281,604
477,673
0.37
completely
(6)
0.97
Water Bottle
9,144,876
1,922,983
656,000
0.34
0.97
(t)
(i )mn
Units here refer to the number of units of each item in stock in the entire network of
FEMA warehouses. Demand and demand met refer to weighted average demand and
average of demand met, respectively, as defined in the Methods section. Fraction of
demand served is also defined in the Methods section. In reviewing these metrics, it
seems contradictory that the fraction of overall demand served (r) would be low while
the fraction of disasters served completely (6) is high. This is the case because y is more
susceptible to outliers than 8. Many small disasters are covered completely by the
physical inventory, but there are a few very large disasters that cannot be served by the
physical inventory. This results in the high fraction of disasters served completely (0.97),
but a lower fraction of total demand is served (0.37 for MREs and 0.34 for water)
because there are a few large disasters that skew this result with high levels of demand.
4.3.2
Network inventory allocation strategies
In order to evaluate the allocation strategies in FEMA's logistics network, it is useful to
compare the current allocation to the optimal allocation when minimizing time and/or
cost. The optimal solutions when minimizing time are very similar to those when
minimizing cost here because truck is the sole mode of transport being used in the model,
so the model cannot choose between air and truck when optimally allocating goods to
56
minimize either time or cost. This is because disasters in the continental United States
can all be served by truck. Figure 4-4 compares the actual allocation of MREs to the
optimal allocation of MREs when minimizing time and the optimal allocation of MREs
when minimizing cost, and Figure 4-5 does the same for water bottles.
I
Frederick,
fD
Actual
*Mountain View, CA
j Cumberland, MD
Minimize Cost
*Frederick, MD
U
Atlanta, GA
Fort Worth, TX
Minimize Time
2,000,000
4,000,000
6,000,000
8,000,000
Units in Inventory
Figure 4-4 Actual and optimal allocation of MREs (minimizing time or cost)
57
Fre
rick, MD
AmtG
Actual
mberlan
otWrh
MD
Mountain View, CA
Cumberland,
MD
Cumerlnd MD
0 Frederick, MD
Minimize Cost
N Atlanta, GA
'Fort Worth, TX
A
Minimize Time
-
G
2,000,000 4,000,000 6,000,000 8,000,000 10,000,000
Units in Inventory
Figure 4-5 Actual and optimal allocation of water bottles (minimizing cost or time)
The actual allocation of both MREs and water bottles is fairly evenly distributed across
all five depots, with greater inventory in Atlanta, GA and Fort Worth, TX. While the
stock in Maryland is in two separate warehouses (Cumberland and Frederick), they are
approximately two hours driving distance apart from one another so are likely serve the
same disasters and have a similar impact on the model that having only one location in
Maryland would have. When stock is optimally allocated (minimizing either time or cost)
the majority of the inventory is placed in Atlanta and Fort Worth. Referring back to
Figure 4-1 which shows the geographic distribution of disasters this is likely because a
greater number of disasters occur in the Southeast, South, and Midwest than the MidAtlantic and Northeast.
In addition to evaluating inventory allocation strategies for existing and optimal
inventory, decision-makers may be increasing or decreasing the system inventory levels
across multiple warehouses and need to determine where inventory should be allocated.
Figure 4-6 and Figure 4-7 show the optimal allocation of inventory in each warehouse as
the total inventory in the system increases for MREs and water, respectively, minimizing
58
time to serve. The current system inventories of MREs and water for FEMA are 7.3
million and 9.1 million respectively, and are indicated by dashed lines in the figures.
100%
90%
80%
PC
B
S4
70%
60%
U
Mountain View, CA
50%
U
Frederick, MD, USA
40%
U
Fort Worth, TX, USA
30%
*Cumberland, MD, USA
20%
N Atlanta, GA, USA
10%
0%
, S~~~~~~~~~N
ej
P0 q
R P51S 11 IIil
Total Inventory in the system
Figure 4-6 Optimal allocation of MREs (minimize time) with increasing total system inventory, current
inventory level indicated with dashed line
Similarly to the optimal allocation of the current level of physical inventory, when
progressively increasing the total inventory of MREs in the system, the optimal solution
(minimizing time) is to place stock primarily in Fort Worth, TX and Atlanta, GA. With
up to 2,000 units of stock, the optimal location to place 100% of stock is Fort Worth, and
then Atlanta is introduced. With approximately 200,000 units of total stock in inventory,
Mountain View, CA is introduced. Frederick and Cumberland, MD remain a low
percentage of the total inventory even up to 10,000,000 units of inventory.
59
100%
90%
I-
80%
Mountain View, CA
70%
"
ca
60%
" Frederick, MD, USA
PC
50%
* Fort Worth, TX, USA
40%
" Cumberland, MD, USA
30%
" Atlanta, GA, USA
20%
10%
0%
pQ
'ep
v'4> y~)
Total Inventory in System
Figure 4-7 Optimal allocation of water bottles (minimize time) with increasing total system inventory, current
inventory level indicated with dashed line
Like MREs, the optimal allocation of water bottles as the system inventory increases is
mostly placed in Fort Worth, TX and Atlanta, GA. Atlanta is not utilized as a second
warehouse location until there are 5,000 units of inventory in the system, and Mountain
View, CA is introduced when total inventory in the system is at approximately 200,000
units. At 2,000,000 units and after 5,000,000 units it appears as though the inventory in
Atlanta decreases, but it is important to remember that this graph shows the percent of
total inventory in each of the warehouses. This is showing that Atlanta decreased in
percent of total inventory held, but since the total inventory in the system simultaneously
increased this does not mean that the quantity of inventory in Atlanta decreased.
4.3.3
Delivery deadline cutoff times
To understand the performance of the current allocation of inventory over time, the
fraction of total demand served was evaluated for intervals of time from 0 to 72 hours,
the maximum time to respond considered by the model. Figure 4-8 compares the fraction
of demand served by the physical inventory as actually allocated and in the optimal
60
allocation. This figure is also presented later when contract inventory is added to the
analysis. Time to respond here refers to the time to deliver supplies to those in need and it
is based on a combination of the fixed and variable time associated with obtaining,
loading, and transporting goods as explained in the Methods section.
0.4
0.35
0.3
-o 0.25
0.2
-+--
Actual
Optimal
0.15
0.1
I
0.05
ii
0
0
10
20
30
40
50
60
70
80
Time to Respond (hours)
only
Figure 4-8 Fraction of demand served (minimize time) actual vs. optimal allocation of physical inventory
For FEMA's physical inventory, actual and optimal fraction served follow roughly the
same trend, except at 18-24 hours, the optimal allocation would increase the fraction
served above the optimal allocation by approximately 5%.
4.3.4
Response Capacity Index
for an
The RCI, as explained in the methods section is an index to evaluate the quality of stock allocation
was
organization or set of organizations providing humanitarianaid (Acimovic & Goentzel, 2015). The RCI
calculated for FEMA's current allocation of physical inventory. The results are shown in
Table 4-3 and calculations for each metric in the index are described in the paragraph that
follows.
61
Table 4-3 Response Capacity Index for FEMA physical inventory
Item
Optimal
tmto
serve
OPT)
Balance
metric time
Balance
metric cost
(Ar)
((pOP)
(AC)
Fraction of
dstes
covered
cvr
(6)(612)
Fraction of
disasters
covered in
12 hours
RCI
(simple
average)
Water Bottle
7.64
9.79
9.84
9.67
5.04
8.40
MRE
7.68
9.81
9.83
9.67
5.12
8.42
Each of the metrics included in the RCI is scaled from 1-10 according to the best and
worst possible values for that metric. The descriptions below describe what values were
used as best and worst for each of the metrics, where the actual value fell between those
values, and include a sample calculation showing how the scaling is done. Descriptions
of what each metric means can be found in the Methods section.
1. Optimal time to serve ($ OPT) The best time is assumed to be 6 hours, because that
is the fixed time associated with acquisition and loading of a truck. The worst
time is assumed to be the longest travel time that might potentially exist between
a warehouse and a disaster when traveling by truck within the continental United
States. This distance is from Point Arena, CA to West Quoddy Head, ME (United
States Geological Survey, 2001), and according to Google's driving distance
function represents a travel time of 52 hours. The values for optimal time to serve
per unit shipped for water bottles and MREs are 16.8 hours and 16.7 hours
respectively. Scaling is done by calculating the ratio of the difference between the
optimal time to serve per unit shipped for both commodities and the best possible
time to the difference between the best and worst possible travel times,
subtracting this ratio from 1, and multiplying by 10.
2. Balance metric, time (AT) The best possible value here is 1, meaning that the
current and optimal allocations are perfectly in balance. The worst possible value
is considered here to be 1.6 for cost, and 2.6 for time, based on a sensitivity
analysis that is described in this Section 4.3.4.1. The balance metric values for
water bottles and MREs (minimize time) were 1.032 and 1.030, respectively.
62
Intuitively, this means that the allocation of water bottles is 3.2% off from the
optimal allocation, and the allocation of MREs is 3.0% off from the optimal
allocation. As described in the Methods section, the balance metric increases if
warehouses are added closer to the locations of disasters but no stock is added to
them, or if new warehouses are added and all stock was moved to one warehouse
while leaving the others empty.
3. Balance metric, cost (Ac) refers to the balance metric when minimizing cost. It is
calculated similarly to the balance metric for time, with the same high and low
bounds. The balance metric values for water bottles and MREs (minimize cost)
were both 1.010.
4. Fraction of disasters covered (&) refers to the fraction of disasters that are served
completely given the current inventory level. This number is robust to outliers.
The worst value is 0, and the best possible value is 1, so the value is simply
multiplied by 10 to scale it for the RCI.
5. Fraction of disasters covered in 12 hours (612) refers to the fraction of disasters
that are served completely in 12 hours time. This is included in the RCI to serve
as another way of evaluating the overall positioning of warehouses, independent
of total inventory in the system. Similar to the fraction of disasters covered, the
worst value is 0 and the best possible value is 1, so the value is simply multiplied
by 10 to scale it for the RCI.
For FEMA, the RCI can be used as a tool to evaluate the tradeoffs between, for example,
adding and stocking another warehouse in a location that is closer to where frequent
disaster activity occurs (e.g. Miami). This might decrease the balance metric because the
optimal allocation across all disasters would not place inventory in Miami over more
centrally located warehouses, but it would also likely improve the fraction of disasters
served completely in 12 hours. Looking forward, the RCI will benefit from additional
expert input on the metrics to include and how each of the individual metrics should be
evaluated.
63
4.3.4.1 Balance metric sensitivity analysis
In order to evaluate the upper bound of the balance metric more accurately for FEMA
(the lower bound will always be 1), four scenarios were run in which warehouses were
added in all locations of FEMA regional offices (except in Atlanta, GA, where there is
already a warehouse) as well as the cities of Miami, FL, Minneapolis, MN, Billings, MT,
Salt Lake City, UT, Reno, NV and Memphis, TN to increase the geographic diversity of
the warehouse locations from what currently exist. Figure 4-9 shows where the additional
warehouses were located. The current warehouse locations are also included in this
analysis.
Vancouver
SA9h"I.WA
MNMontreal
W
wiNng,merMT
eMinnmapolis, MN
Ne )Yrr
I0
Salt Lake CI
0
SanF
Ph
Remno,CANV
Dn
s
co
%A",
S
Los Angeles
a
s
tonT
0
3
H1avana
600mI
Esri, HERE, DeLorme, NGA, USGS I Esri, HERE. DeLorme
metric (Esri,
Figure 4-9 Locations of potential FEMA warehouses used in the sensitivity analysis for the balance
2015)
Once the additional warehouses were added to the logistics network in the model, all
FEMA physical stock was assumed to be located in Bothell, WA, Oakland, CA, Miami,
FL, Boston, MA, and Billings, MT while the remaining warehouses remained empty to
represent five extreme cases of where stock could reasonably be located within the
continental United States. The balance metrics for these scenarios are used to conduct a
sensitivity analysis of the balance metric for FEMA and appropriately scale the balance
metric scores in the RCI for FEMA. This process should be repeated in other cases in
order to appropriately scale the balance metric for the purposes of the RCI. The resulting
balance metrics for cost for and time for water bottles and MREs are shown in Table 4-4.
64
Table 4-4 Balance metric sensitivity analysis results
MRE
Location of
Stock
Balance
Metric Cost
Water
Balance
Metric Time
Balance
Metric Cost
Balance
Metric Time
Bothell, WA
1.53
2.54
1.52
2.50
Oakland, CA
1.46
2.27
1.45
2.23
Miami, FL
1.14
1.38
1.14
1.39
Boston, MA
1.28
1.81
1.28
1.81
Billings, MT
1.32
1.90
1.31
1.88
As a result, the upper bound for the balance metric for cost is scaled to be 1.6 and the
upper bound for the balance metric for time is scaled to be 2.6. The lower bound remains
at 1.
4.4
Scenario 2: FEMA Physical Inventory and Contract Inventory Allocation
In the second scenario analyzed for FEMA, both physical and contract inventory were
considered. Physical inventory is the inventory owned by FEMA and stored in FEMA
owned warehouses. Contract inventory is inventory that FEMA has pre-emptively set up
contractual agreements to purchase in the event of a disaster. In this analysis, only
contract inventory from private suppliers was considered. Contract inventory from
private suppliers will likely be up for re-negotiation, and decision makers could use the
results of this analysis to restructure these contracts.
4.4.1
Delivery deadline cutoff times
To understand the performance of the current allocation of inventory over time, the
fraction of total demand served is evaluated for intervals of time from 0 to 72 hours, the
maximum time to respond considered by the optimization. Figure 4-10, which is the same
as Figure 4-8 compares the fraction of demand served by the physical inventory as
actually allocated and in the optimal allocation, minimizing time. Time to respond here
refers to the time to deliver supplies to those in need and it is based on a combination of
the fixed and variable time associated with obtaining, loading, and transporting goods as
explained in the Methods section.
65
0.4
0.3
/
PC
/
0.35
0.25
0.2
-
0.15
Actual
-
Optimal
0.1
0.05
n
0
10
20
30
40
50
60
70
80
Time to Respond (hours)
Figure 4-10 Fraction of demand served (minimize time) actual vs. optimal allocation of physical inventory only
For FEMA's physical inventory, actual and optimal fraction served follow roughly the
same trend, except that at 18-24 hours, the optimal allocation would increase the fraction
served above the optimal allocation by approximately 5%.
Next, the optimal and actual allocations are compared for the contract inventory
only, minimizing time. In order to conduct this evaluation, the locations of the contract
inventory were constrained to not be available to contribute to the network inventory
until the time specified in the contractual agreements with FEMA. Based on
conversations with FEMA staff, the times specified in the contract terms represent the
time range within which the stock is expected to arrive at the disaster location, rather than
when it is available at the warehouses. As the model does not predict arrival time,
assumptions were made to estimate time constraints for when inventory would need to
become available to arrive at disaster sites within the contract terms of 24, 48, and 72
hours. For contract terms where the stock was expected to arrive within 24 hours, stock
was assumed to become available according to the base fixed time constraint of 12 hours
as described in the Methods section. If stock was expected to arrive within 48 hours, 36
66
hours was used as the fixed time for the stock to become available (24 hours after the first
round of stock became available at 12 hours), and if stock was expected to arrive within
72 hours, 60 hours was used as the fixed time for stock to become available (48 hours
after the initial round of stock became available at 12 hours).
Figure 4-11 shows the comparison between the actual and optimal allocation of
the contract stock, minimizing time. It is important to note that the optimal solution in
this case factors in the best-case scenario of the ability to have all contracted inventory
available immediately, rather than the current delay imposed by the contract terms and/or
limitations on production ramp up for individual suppliers. Based on discussions with
FEMA staff, it seems that this is a scenario that may indeed be feasible, which is why it is
included here. The actual results here factors in the delay of availability of supplies from
certain vendors.
0.3
0.25
Ar
0.2
S0.15
-e-
Actual
Optimal
S0.1
0.05
0
0
10
20
30
40
50
60
70
80
Time to Respond (hours)
Figure 4-11 Fraction of demand served (minimize time) actual vs. optimal allocation of contract inventory only
The largest benefit for re-allocating contract inventory and/or changing when this
inventory is available comes between 24 and 48 hours, as Figure 4-11 shows. Later in
this section the optimal allocation of contract stock across vendors is presented, and
67
shows how the allocation of inventory in the warehouses throughout FEMA's logistics
network change at these different delivery deadline cutoff times.
0.7
0.6
V
0.5
V
Cu
'aC
0.4
-+-- Actual
0.3
-
C
Cu
I.E
-
Optimal
0.2
0.1
0
0
10
20
30
40
50
60
70
80
Time to Respond (hours)
Figure 4-12 Fraction of demand served (minimize time) actual vs. optimal, physical and contract stock
Figure 4-12 combines the fraction served by the physical inventory and the contract
inventory over a 0 to 72 hour period to respond. It incorporates the fraction served by the
physical inventory only beginning at the 6 hour response time as well as the benefit to reallocation of contract inventory as the time to respond moves towards 72 hours. The
overall fraction of demand served also increases with the combined contribution of
physical and contract stock to meet demand, as can be seen by comparing Figure
4-12with Figure 4-10 and Figure 4-11.
68
I
I
Optimal-72
E Mullins, SC, USA-12
N Mullins, SC, USA-36
Optimal-48
0 Mullins, SC, USA-60
* Norcross, GA, USA-12
Optimal-36
SA-1
E Norcross, GA, USA-36
Norcross, GA, USA-60
U Evansville,
Optimal-24
IN, USA-12
Evansville, IN, USA-36
'Evansville,
IN, USA-60
Actual Inventory
-
1,000,000
2,000,000
3,000,000
4,000,000
Figure 4-13 Actual vs. optimal allocation of FEMA contract inventory only (minimize time)
Figure 4-13 compares the current allocation of contract inventory across the three main
suppliers that are located in Mullins, South Carolina, Norcross, Georgia, and Evansville,
Indiana. The "-12", "-36", and "-60" appended onto each of these warehouse locations
represent the time-constrained availability of the supplies. For example, after 12 hours,
the warehouse in Mullins, South Carolina has contracted capacity to provide 150,000
MREs which are expected, according to conversations with FEMA officials, to arrive
within 24 hours (FEMA Staff 2, 2015). This is reflected in the model by adding a fixed
time constraint of 12 hours to the transportation parameters for the Mullins, South
Carolina warehouse. The model, however, is limited in its allocation of optimal inventory
- at present it will optimally allocate all contract inventory, regardless of when it
becomes available, to the warehouses with only a 12-hour fixed time constraint. Despite
this limitation, what the optimal solution does provide is a better idea of what would be
needed from a contract or set of contracts to serve a greater fraction of the demand
overall. At 36 hours, having stock distributed through Norcross, GA and Evansville, IN
with some in Mullins, DC is optimal, but at 48 and 72 hours, the optimal solution is to
69
place most of the stock in only Norcross, GA and Evansville, IN with a very minimal
contribution of stock from Mullins, SC. This is insight that may be used to structure
future contracts or to restructure existing contracts.
4.5
Scenario 3: FEMA Physical Inventory Allocation with Reduced Demand
This section evaluates FEMA's allocation of physical inventory with a 10% reduction in
demand in the state of Florida. This reduction in demand was chosen according to the
reasoning specified in the Methods section. Furthermore, in the case of the FEMA study,
it seems like a realistic estimate based on a presentation given by a representative of
Florida Emergency Management at a conference stating that the state emergency
management leadership has set a goal of increasing participation in the National Flood
Insurance Program (NFIP)'s Community Rating System (CRS) from approximately 40%
to 100% over the course of 18 months (Kay Roberts, 2015). Participation in the CRS
makes NFIP participants living in the community eligible for varying levels of insurance
benefits, depending on the community's CRS rating. The CRS rating is improved when
the community takes steps to promote reduction of flood vulnerability through a variety
of activities. Ten percent would likely represent a conservative estimate of the impacts of
this relatively aggressive goal of 100% CRS participation statewide in Florida. This is
because not all of the activities that qualify as CRS participation directly reduce
vulnerability on an individual household basis that would decrease the number of people
requiring aid after a disaster. This scenario is focused on assessing the impacts of reduced
demand on assessment metrics, but these resilience goals (CRS participation) represent
the type of activities that might cause a real reduction in demand in the long run.
4.5.1
Ability to meet demand
Similar to the evaluation of FEMA's physical inventory with the original disaster risk
data, the ability of the current physical inventory to meet the new reduced demand for
disasters in Florida was evaluated.
70
Table 4-5 Ability of FEMA physical inventory to meet demand with 10% reduction in Florida demand
Item
MRE
Units
Demand
Original
Item
Demand
Demand
Met
Reduced
Us OOriginal
O(')
Demand
Met
Reduced
(V')
Fraction
of
demand
served
Original
Fraction
of
demand
served
Reduced
(y)
(y)
Fraction
of
disasters
served
Original
and
7,319,000
1,281,604
1,239,351
477,673
475,082
0.37
0.38
Reduced
(8)
0.97
9,144,876
1,922,983
1,859,584
656,000
652,112
0.34
0.35
0.97
Water
Bottle
As is observed here, the fraction of demand served with the reduced demand is only
slightly higher, with the fraction of disasters served remaining exactly the same.
Additionally, the demand met with the reduced demand is less. To understand how these
three variables have changed (or not) with reduced demand in the FDEM logistics
network, one must consider how disasters are generally distributed. In the case of
disasters in the continental United States that were considered for the FEMA analysis,
there are a few very large disasters, but the total affected population for the remaining
disasters drops off significantly beyond these large events. In reducing demand in one
state weighted average demand (y) is reduced because some of the disasters have become
smaller in terms of total affected population. This information, combined with the fact
that the fraction of demand served (p')/(i) increased, means that with the current stock
FDEM's logistics network is able to meet a greater proportion of the demand given the
current allocation. As the demand overall has decreased, the average demand met still (y')
decreases, just less than the decrease in (p).
In summary, the increase in fraction of demand served means that the FDEM's
logistics network is serving proportionally more of the demand given the current
allocation of inventory, but the network is not serving any more disasters completely,
which is why fraction of disasters served remains the same. To serve more disasters
completely, demand would have needed to be reduced such that the current allocation of
stock that was unable to serve some of the larger disasters originally could now serve
those disasters. While this is only a slight benefit, an increase in fraction served might
still be used as justification for promoting resilience planning that is intended to reduce
overall demand for critical commodities in a disaster.
71
4.5.2
Network inventory allocation strategies
In addition to evaluating the figures described above, the actual allocation of physical
inventory was compared to the optimal allocation with original demand and the optimal
allocation with reduced demand in Florida. It was hypothesized that a reduction in
demand in one geographic location would change the optimal allocation of inventory
across warehouses. That was not the case for the reduction in demand in Florida, as is
shown in Figure 4-14 and Figure 4-15. The allocation for both water bottles and MREs
remains unchanged despite a reduction in demand in Florida. This is likely because the
reduction in the weighted average of demand was only 3% for both commodities being
considered. A larger reduction in the weighted average of demand in the state might have
pushed the optimal allocation to a warehouse farther from Florida.
T
G
FW
e
Actual
* Mountain View, CA
Cumberland, MD
1 1t (
OptTime-RedDmd
E Frederick, MD
N Atlanta, GA
* Fort Worth, TX
A i
OptTime-Orig
-
F
(
2,000,000
4,000,000
6,000,000
8,000,000
Units in Inventory
Figure 4-14 Actual vs. optimal allocation of inventory with original and reduced demand for MREs (minimize
time)
72
Actual
E Mountain View, CA
' Cumberland, MD
OptTime-RedDmd
0
Frederick, MD
N Atlanta, GA
E
Fort Worth, TX
OptTime-Orig
-
2,000,000 4,000,000 6,000,000 8,000,00010,000,000
Units in Inventory
Figure 4-15 Actual vs. optimal allocation of inventory with original and reduced demand for water bottles
(minimize time)
Despite not having observed significant change of the metrics considered thus far aside
from fraction of demand served, this method might still be used as a preliminary way of
putting a value on resilience. For example, the change in total cost of delivering the initial
push of supplies in a response can be estimated with this method. In this case, the
resulting change in total cost given the current allocation of inventory and a reduced
demand in the state of Florida of 10%, a savings of $493 is realized for MREs and $1160
for water bottles. These figures are small, but they show a savings and begin to answer
the question of how much money will be saved by reducing the number of people who
will require aid in a disaster.
73
5 Case Study: Florida Division of Emergency Management
This section presents the case study on the disaster response logistics network for the
Florida Division of Emergency Management (FDEM). First, the procedural framework
used by FDEM to guide disaster response is presented, with a focus on the parts of this
framework that address logistics concerns. Next, the analysis assumptions and process
unique to FDEM are described, followed by the results of the analysis and a discussion of
the implications of these results for FDEM.
5.1
Background
The FDEM has a robust logistics section and is often used as a best practices model of
state emergency response in the United States (e.g. Goentzel & Spens, 2011). The state
has also made publicly available much of their logistics documentation, which is
reviewed for this study. The state of Florida Unified Logistics Plan, which is an annex to
the State Comprehensive Emergency Management Plan, "includes plans, procedures, and
supporting documentation needed to ensure the state of Florida maintains a strong and
viable logistics capability" (State of Florida Division of Emergency Management Unified
Logistics Section, 2013) is primarily reviewed. This plan is authorized by Section 252.35
(1) of the Florida Statutes, which state that the "division is responsible for maintaining a
comprehensive statewide program of emergency management" which includes
coordination with the federal government as well as other agencies in the state
government, the private sector, and county and municipal governments. The statute also
requires that the state prepare a "comprehensive emergency management plan", which
the state has done, and which this logistics plan is part of (The Florida Legislature, 2014).
The majority of the Unified Logistics Plan focuses on operational activities and
coordination mechanisms, including the risk assessment, hazard analysis, roles and
responsibilities, the locations of key response, mobilization and distribution sites, and
other operational considerations. In addition, the state has outlined a resource
management system for supply chain management, which specifies the order in which
various resources will be called upon in a disaster, beginning with the state's physical
inventory. The state has also conducted a gap analysis to identify, for three levels of
disasters (10,000, 100,000 and 1,000,000), the resources that it has and the gaps that
exist, as well as plans for filling those gaps. The analysis conducted in this case study
74
compliments the plans and analyses already conducted by FDEM by expanding the
possible disaster scenarios and suggesting where the inventory of resources should be
allocated to best serve the disaster affected population.
In conclusion, the Unified Logistics Plan states the logistics goals, objectives, and
actions. The response goal cited is "to be able to effectively respond in an emergency
with Logistical Support to all emergency operations and field operations factions", with a
key objective to "ensure a timely and effective response to the many consequences that
may be generated by an emergency/disaster situation" (State of Florida Division of
Emergency Management Unified Logistics Section, 2013). This is one of four key
objectives for the response goal and most closely relates to the goals of this study. The
objectives outlined are qualitative in nature, which make the attainment (or not) of them
difficult to measure.
Also reviewed for the purposes of this case study was the County Logistics
Planning Standard Operating Guideline (State of Florida Division of Emergency
Management, 2006). This document is meant to serve as a guideline to communities in
developing their own individual logistics plans. It outlines what logistics finctions
counties are responsible for, such as designating points of distribution (POD) for critical
commodities, and how counties can request assistance from the state for events that it is
not able to respond to on its own. The document is primarily operational in nature and
does not include frameworks for evaluating performance, but does provide useful insight
into how communities operate their logistics response, which feeds into the overall state
logistics response.
This case study will aim to evaluate FDEM's logistics network and suggest
metrics on which the ability of the logistics network to meet the goals outlined in the
Unified Logistics Plan can be evaluated. These metrics were selected from the output of
the Stockpile Capacity Model ("the model") (Acimovic & Goentzel, 2015) presented in
the Methods section, and here those metrics that can be used readily in decision-making
for a logistics network are presented.
5.2
Analysis
This section will describe the analysis that was conducted using the model, associated
metrics, and the Response Capacity Index (RCI) (Acimovic & Goentzel, 2015). It is
75
presented as an addition to the description of disaster and inventory data presented in the
Methods section, and will describe the dataset for FDEM's logistics network in detail,
including any assumptions made for the purposes of this case.
5.2.1
Insurance risk data
For the evaluation of FDEM's logistics network, the disaster risk data were obtained from
DataCo (DataCo, 2015). These data are described in detail in the Methods section and as
noted, include only hurricane disaster data (storm surge, precipitation, and high winds).
This represents a portion of the disasters experienced in the state of Florida, but not all. A
more robust dataset would also include damages from riverine and severe storm flood
events, as well as other hazards that Florida is susceptible to. The original dataset from
DataCo for Florida included 9,842 disasters, 103 historic and 9,739 stochastic and their
impacts on the 67 counties in Florida.
Figure 5-1 and Figure 5-2 show the geographic distribution of the count of
disasters by county and the maximum total affected population by county for the
disaggregated disaster dataset. In this dataset, each disaster-county pair with a total
affected population of greater than 100 was included, because this is assumed to be the
maximum number of individuals affected in an event that a county within the state would
be able to respond to on its own. The figures separate the disaster data into the top 10%
(largest symbol), the next 20% (medium-sized symbol), and the bottom 70% (smallest
symbol) for the count of disasters per county and the maximum total affected population
by county. Disaster-county pairs with less than 100 TAP are not shown on the map. This
means that the top 7 counties for both count of disasters and maximum total affected
population are represented by the largest symbol, the next highest 13 are represented by
the medium sized symbol, and the rest are represented by the smallest symbol. As
anticipated given that this dataset includes losses from only hurricane wind and storm
surge, the counties along the coast and in the southern portion of the state experience the
most frequent and the largest disasters.
76
> 2,000 to 3,421
> 1,040 to 2,000
1 to 1,040
Esr,
HERE, DeLorme, NGA, USGS
I Esn HERE, DeUrme
2015)
Figure 5-1 Count of disasters by county, disaggregated disaster dataset, for counties with TAP>100 (Esri,
* 100,000 to 1,213,072
>
15,000 to 100,000
142
to
15,000
Havana
Esn, HERE. DeLorme, NGA, USGS I Esr., HERE, DeLorm
Figure 5-2 Maximum total affected population by county, disaggregated disaster dataset, for counties with
TAP>100 (Esri, 2015)
The disaggregated dataset of disasters with total affected population greater than 100
persons includes 48,699 unique disaster-county pairs ranging in total affected population
from 100 to 1,213,072. The total dataset, however, included 9,842 disasters, with loss
data for each of the 67 counties in Florida, so a total of 659,414 unique disaster-county
pairs, meaning that 93% of disaster-county pairs had a total affected population less than
100 persons and were not included in the analysis. Even so, this represents a rich dataset
77
of historic and predicted disaster data that will improve the accuracy of the model in
determining where critical commodities should be placed to minimize time and/or cost to
respond.
The disaster data provided estimates losses on a county basis. In order to estimate
the location of disaster, the county seat was used as an estimated location for where the
demand would be located within the county. This location was used because the county
seat is a likely location where supplies would be distributed to individuals, and/or a place
where individuals from the community would come for shelter. Figure 5-3 shows a map
of the counties in Florida with the county seats identified.
Paname Cn
Porl
SI
-Vaf:Jw
te Spnngs
t Avgustne
noll
I4w,
PSaM880:
Lee
Monroe
Figure 5-3 Map of Counties in Florida, noting the county seats (Geology.com, n.d.)
78
5.2.2
Florida inventory and demand data
Inventory data were assumed for this study based on a 2011 case study by Goentzel and
&
Spens for water bottles and shelf stable meals or meals ready to eat (MREs) (Goentzel
Spens, 2011). These data are current as of March 2010.
At present, all of the FDEM's physical inventory of MREs and water bottles is
located in one warehouse in Orlando, but the FDEM has identified nine logistics staging
areas (LSAs) throughout the state that are also evaluated in this study as potential
locations for prepositioning goods. The physical addresses of these LSAs are included in
the model for estimating drive time and distance, but in analysis only the city name is
included for simplicity (Florida Division of Emergency Management, 2013). Figure 4-2
shows the locations of these logistics staging areas, which does include a location in
Orlando which represents the physical warehouse where all stock is currently located.
Duke Field AFS, FL
0
Lakeland, FL
Ocala,
Orlando,
e
Ta
FL
FL
nsaoa FLPunta Gorda, FL
*
Tallahassee, FL
*
West Palm Beach, FL
Esr, HERE, DeLorme NGA, USGS IEsn, HERE, DeLorme
Figure 5-4 Florida Logistics Staging Areas (Including Orlando, the location of permanent warehouse) (Florida
Division of Emergency Management, 2013)
To estimate demand by item, 3L of water per person per day and two MREs per person
per day were assumed based on reviewing Florida's logistics documentation and the
international Sphere Standards for water needs for drinking and food (Florida Division of
Emergency Management, 2013; The Sphere Project, 2011). Finally, it is important to
remember how affected population is translated into units of demand. The estimate for
affected population is translated into units of an item demanded by multiplying the per79
person units of need for the initial push of supplies that a logistics network would need to
support following a disaster by the number of individuals affected. In line with Acimovic
and Goentzel (2015), the initial push of supplies is assumed to be enough supplies for 72
hours (e.g. 2 MREs/person/day, over 3 days equals 6 MREs/person).
5.2.3
Analysis scenarios
Three scenarios were run to evaluate the FDEM logistics network, and are summarized in
Table 5-1. First, the current physical inventory allocation of water bottles and MREs in
Orlando only was run to evaluate the ability of this inventory to meet the needs of
disaster survivors across the state in aggregate and at varying time intervals. Then, the
LSAs were added on and considered as additional locations for placing goods in the event
of a disaster to compare metrics for when all stock is located in Orlando to metrics when
additional locations are available. Second, county level disaster risk data from a training
exercise for Hurricane Wilma was run for two time intervals of warning before the
disaster to evaluate the optimal locations for prepositioning goods, and whether they
changed significantly with more warning. Finally, the current physical inventory
allocation was run against the original disaster risk data with reduced demand in just
Miami-Dade County to evaluate the impact of resilience goals in one county. The
sections that follow describe the results of each scenario analysis.
80
Table 5-1 Florida Case study scenarios
Scenarios Considered
Decision
Current allocation of
inventory
Where to allocate
inventory?
What is the benefit, in
terms of cost to serve
the affected population,
if resilience goals
reduce overall demand?
5.3
*
*
Current allocation of
inventory with option
to stock closer to
disaster events in LSAs
Current allocation of
Where to preposition
inventory in a notice
event (e.g. hurricane)?
Metrics Evaluated
inventory with option
to stock closer to
disaster events in LSAs
for two time intervals
before a notice event
Response Capacity
Index
Ability to meet
demand
of current
vs. optimal inventory
allocation
Data Required
*
e
*
*
*
*
e
Reduction in demand in
Miami-Dade County by
10%
Comparison of current
vs. optimal inventory
allocation
Ability to meet
demand
Time to serve
Ability to meet
0 demand
Comparison of current
vs. optimal inventory
allocation
*
*
Warehouse
locations
Warehouse stock
Disaster data
(Comparison
historic or
forecast)
Warehouse
locations
Warehouse
stock
Disaster data
(forecast) for
notice event
All from base
allocation, plus
alctopu
Resilience goals
for a county or
state
Scenario 1: Current Allocation of Inventory
This scenario analyzed FDEM's current allocation of physical stock of water bottles and
MREs. The key metrics assessed were the ability of the current inventory to meet
demand, the allocation strategies (current and optimal), and the fraction of demand met
when varying delivery deadline cutoff times are imposed.
5.3.1
Ability to meet demand
The first metric assessed for physical inventory is the ability of this inventory to meet
demand. Demand is determined to be the weighted average of demand, as described in
the Methods section, where each disaster has an equal weight or probability of occurring.
With this, large disasters carry no more weight than small disasters. It is important to note
that units of demand are reported in terms of number of items, rather than in terms of
people demanding the items. Table 5-2 includes several metrics to evaluate the ability of
FDEM's physical inventory to meet demand, as predicted using the dataset of disaster
risk data from AIR Worldwide (2015).
81
Table 5-2 Ability of Florida physical inventory to meet demand
Demand
Units
ItemUnis
Item
1,600,000
5,292,000
MRE
Water
Bottle
I
_
_
__
435,909
I
_
_
__
228,223
0.79
(
290,519
_
Demand Met
demad sev ed
demnd srve
~
407,377
_
I
_
_
__
_
Fraction of
disasters served
completely
(6)
0.96
0.93
_
I
_
_
_
_
0.99
_
_
I
_
_
__
_I_
In reviewing these metrics, FDEM looks to be well prepared to meet demand overall
given the level of inventory it currently keeps in stock for both MREs and water bottles.
The fraction of demand served is much higher than was seen in the FEMA analysis, and
the fraction of disasters served completely is nearly 100%. There is a difference between
fraction of demand served and fraction of disasters served completely because y is more
susceptible to outliers than 6. Many small disasters are covered completely by the
inventory in stock, which results in the high fraction of disasters served completely (0.96
and 0.99), but a lower fraction of total demand is served (0.79 for MREs and 0.93 for
water) because there are a few large disasters that skew this result.
5.3.2
Network inventory allocation strategies
In order to evaluate the inventory allocation strategies in FDEM's logistics network, it is
useful to compare the current allocation to the optimal allocation when minimizing time
and the allocation when minimizing cost. While FDEM currently keeps its entire
inventory of MREs and water bottles in one central warehouse in Orlando, additional
warehouse locations that could be added in the locations that Florida currently has
identified as LSAs were considered in order to provide metrics for comparison of the
current centralized strategy to one where stock is allocated to warehouses across the state.
The optimal solutions when minimizing time are very similar to those when minimizing
cost because truck is the sole mode of transport being used in the model, so the model
cannot choose between air and truck when optimally allocating goods to minimize either
time or cost, and the variable time and cost for truck travel are directly related to distance
traveled. This is because all disasters in Florida can be served by truck transport. Figure
5-5 compares the actual allocation of MREs to the optimal allocation of MREs when
82
minimizing time and the optimal allocation of MREs when minimizing cost, and Figure
5-6 does the same for water bottles.
1
Actual
" Orlando, FL
" Pensacola, FL Airport
Punta Gorda, L
0 Pensacola, FL
" Ocala, FL
Minimize Cost
Lakel
" Tallahassee, FL
d,
Punta Gorda, FL
Punta Gorda, 11
ELakeland, FL
Minimize Time
West Palm Beach, FL
lL
-
500,000
Duke Field AFS, FL
1,000,000
1,500,000
2,000,000
Units in Inventory
Figure 5-5 Actual and optimal allocation of MREs (minimizing time or cost)
83
Actual
0 Orlando, FL
0 Pensacola, FL Airport
Punta Gor ia, FL
0 Pensacola, FL
0 Ocala, FL
Minimize Cost
MTalahassee,
Lakeh d, FL
FL
Punta Gorda, FL
Punta Gor Ja, FL
Lakeland, FL
West Palm Beach, FL
Minimize Time
_
-
Duke Field AFS, FL
1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000
Units in Inventory
Figure 5-6 Actual and optimal allocation of water bottles (minimizing cost or time)
The actual allocation of both MREs and water bottles is currently all in Orlando, so by
adding eight additional locations where stock may be placed, the allocation shifts
drastically for both commodities. The optimal solutions for minimizing both time and
cost for water bottles and MREs place most of the inventory in West Palm Beach, FL,
with the second highest quantity placed in Punta Gorda, FL, and the third highest quantity
placed in Lakeland, FL. Referring back to Figure 5-1 and Figure 5-2, this is likely
because there is a high concentration of disaster events in the southeastern portion of the
state, and these are some of the largest disasters in terms of total affected population.
While allocating inventory to West Palm Beach would minimize time and cost in the long
run, there may be additional factors that would make locating the majority of stock in that
location permanently a risk, and may be why FDEM currently has 100% of its stock
located in Orlando. West Palm Beach is where the disasters are largest in size and is
where disaster events are the most frequent, so locations in the county are likely at a
greater risk of being flooded or damaged by high winds in a hurricane event, and as such
might not be a safe location for long term storage of critical commodities.
84
In addition to evaluating inventory allocation strategies for existing and optimal
inventory, decision-makers may be increasing or decreasing the system inventory levels
across multiple warehouses. Figure 5-7 and Figure 5-8 show the optimal allocation of
inventory in each warehouse as the total inventory in the system increases for MREs and
water, respectively, minimizing time. The current system inventories of MREs and water
for FDEM are 1.6 million and 5.3 million respectively. Similarly to the evaluation of
actual and optimal allocation of inventory, the evaluation of where to place inventory as
total system inventory increased was evaluated for all potential staging areas in the state
as well as the central warehouse in Orlando.
100%
90%
*Punta Gorda, FL
80%
-
--
West Palm Beach, FL
70%
E Lakeland, FL
Duke Field AFS
50%
" Tallahassee, FL
40%
HOcala, FL
30%
" Pensacola, FL
20%
" Pensacola, FL - Airport
10%
" Orlando, FL
0%
'
sqp dV'
Total Inventory in System
Figure 5-7 Optimal allocation of MREs (minimize time) with increasing total system inventory, with current
inventory indicated by dashed line
When minimizing time, the optimal allocation of inventory as the total inventory in the
system is increased from 10,000 to 10 million units is to place most of the inventory in
West Palm Beach, with Punta Gorda, Lakeland, and Orlando holding smaller fractions of
the overall inventory. This is similar to what is observed with the optimal allocation of
the current inventory, minimizing time.
85
100%
90%
K
*Punta Gorda, FL
West Palm Beach, FL
80%
70%
E Lakeland, FL
60%
0 Duke Field AFS, FL
50%
0 Tallahassee, FL
40%
0 Ocala, FL
30%
0 Pensacola, FL
20%
0 Pensacola, FL Airport
10%
N Orlando, FL
0%
Total Inventory in System
Figure 5-8 Optimal allocation of water bottles (minimize time) with increasing total system inventory, with
current inventory indicated by dashed line
Like the optimal allocation of water bottles when minimizing time and the allocation of
MREs as overall system inventory increases, the allocation of water bottles when
inventory in the system is increasing from 10,000 units to 10 million units consistently
places the majority of inventory in West Palm Beach, FL. The current inventory level of
water bottles is shown here with the dashed line. Orlando makes up about 15% of the
total inventory when inventory levels are around 10,000, but as overall inventory
increases, the amount in Orlando decreases as a percentage of the overall inventory in
stock. The allocation across warehouses as total inventory in the system increases is
nearly identical for MREs and water, as is expected given that the demand for these items
is on a units per person basis, and does not take into account other factors. This analysis
would be useful for decision makers in determining where to allocate new stock into the
system to move towards a more optimal solution.
86
5.3.3
Delivery deadline cutoff times
To better understand how well prepared the current stock is to meet demand, the fraction
of demand served at several time intervals was evaluated. This evaluation assumes, as is
actually the case, that all stock is located in Orlando. It does not consider the LSAs as
potential locations to allocate stock. This evaluation is useful for evaluating how quickly
the demand can be met with the current allocation. Figure 5-9 and Figure 5-10 show the
ability of the current inventory allocation to meet demand over a period of 18 hours after
a disaster for MREs and water bottles, respectively, minimizing time to serve.
0.9
0.8
0.7
0.6
Cu
w
C
C
0.5
0.4
--4-Actual
0.3
0.2
0.1
1 1
0
0
5
15
10
20
Time to serve (hours)
Figure 5-9 Fraction of demand served, actual allocation of Florida MRE inventory, minimize time
87
1
0.9
.0 0.8
0 0.7
0.6
0.5
4
-Actual
0.4
0.3
~0.2
0.1
0
0
5
15
10
20
Time to Serve (hours)
Figure 5-10 Fraction of demand served, actual allocation of Florida water bottle inventory, minimize time
No demand is met in the first six hours because that is the fixed time assumed for a responding organization,
such as FDEM, to obtain and load a truck. After that time, stock is deployed and can serve first the disasters
located closest to Orlando, then farther away. Since the disaster locations are assumed to be the county seat of
each of the 67 counties in Florida as described in the Methods section, an evaluation of the drive distance from
Orlando to any county in Florida explains why the fraction of demand served jumps from nearly 40% at 9 hours
post disaster to over 90% at 10 hours post disaster.
Table 5-3 shows the number of counties within 1, 2, 3, and 4 hours of Orlando.
Some of these counties may be inland and may not have a large number of disasters or a
high demand, but this table shows that within the first ten hours after a disaster, including
the fixed time of six hours, stock leaving Orlando can reach 52 (nearly 80%) of the 67
counties in Florida.
88
Table 5-3 Number of counties within one to four hours driving distance from Orlando, FL
Driving distance from Orlando
(hours)
1
2
3
4
Time after disaster (hours)
(driving distance from Orlando
+ 6 hour fixed time)
7
8
9
10
Number of counties
6
19
20
7
As has been shown here Orlando is a central location and positioning stock in Orlando
still allows FDEM to respond to 40% of the demand within just about 9 hours after a
disaster occurs, assuming a fixed time of 6 hours to mobilize and load trucks. The
optimal solution remains to place the majority of inventory in West Palm Beach, but that
location may be vulnerable to flood or high wind hazards.
5.3.4
Response Capacity Index
The RCI, as explained in the Methods section is an index to evaluate the quality of stock
allocation for an organization or set of organizations providing humanitarian aid
(Acimovic & Goentzel, 2015). The RCI was calculated for FDEM's current allocation of
physical inventory, considering the logistics staging areas as options for allocating stock
to evaluate the balance metric and optimal time to serve. The results are shown in Table
5-4 and calculations for each metric in the index are described in the paragraph that
follows.
Table 5-4 Response Capacity Index for Florida
Item
Optimal
tim
serve
Balance
metric time
Balance
metric cost
T
(OOP )
(A)
(AC)
Fraction of
Fisasters
covered
Fraction of
disasters
covered in
12 hours
RCI
(simple
average)
Water Bottle
7.29
7.67
9.49
9.58
5.48
7.90
MRE
7.11
7.94
9.55
9.90
5.38
7.97
Each of the metrics included in the RCI is scaled from 1-10 according to the best and
worst possible values for that metric. The descriptions below describe what values are
used as best and worst for each of the metrics, and where the actual value fell between
those values.
89
1. Optimal time to serve
(4
0PT)
The best time is 6 hours, the time assumed to acquire
and load a truck, as outlined in the Methods section. For Florida, the worst time is
assumed to be the longest travel time that might exist between a warehouse and a
disaster when traveling by truck. This distance is from Pensacola, FL to Key
West, FL and according to Google's driving distance function represents a driving
time of 12 hours. The values for optimal time to serve per unit shipped for water
bottles and MREs are 7.6 hours and 7.7 hours respectively.
2. Balance metric, time (A&) The best possible value here is 1, meaning that the
current and optimal allocations are perfectly in balance. The worst possible value
is considered here to be 2, because it is small and easy to interpret, and no results
&
obtained in the analyses for FDEM's logistics network went above 2 (Acimovic
Goentzel, 2015). To test the bounds of this, a sensitivity analysis similar to that
done for FEMA could similarly be conducted for Florida. The balance metric
values for water bottles and MREs (minimize time) were 1.233 and 1.206,
respectively.
3. Balance metric, cost (Ac) The balance metric for cost is calculated similarly to the
balance metric for time, with the same high and low bounds. The balance metric
values for water bottles and MREs (minimize cost) were 1.051 and 1.045,
respectively.
4. Fraction of disasters covered (6) refers to the fraction of disasters that are served
completely given the current inventory level. This number is robust to outliers.
The worst value is 0, and the best possible value is 1, so the value is simply
multiplied by 10 to scale it for the RCI. The values for fraction of disasters
covered for water bottles and MREs were 0.958 and 0.990 respectively.
5. Fraction of disasters covered in 12 hours
(612)
refers to the fraction of disasters
that are served completely in 12 hours time. This is included in the RCI to serve
as another way of evaluating the overall positioning of warehouses, independent
of total inventory in the system. Similar to the fraction of disasters covered, the
worst value is 0 and the best possible value is 1, so the value is simply multiplied
by 10 to scale it for the RC. The values for fraction of disasters covered in 12
hours for water bottles and MREs were 0.548 and 0.538 respectively.
90
The RCI can be used as a tool to evaluate the tradeoffs between various metrics. For
example, by moving stock closer to the coast and to where many disasters occur, Florida
could increase the fraction of disasters covered in 12 hours, but they may decrease the
fraction of disaster covered overall.
5.4
Scenario 2: Prepositioning of Inventory
An evaluation of allocation strategies for prepositioning stock in advance of a notice
event was conducted to understand how the model could be applied to this type of
decision. The Methods section includes a detailed description of the input for this
analysis. Figure 5-11 and Figure 5-12 below shows a comparison of the allocation of
inventory for two intervals of time before Hurricane Wilma (minimizing time), which
made landfall in Florida on October 24, 2005 at approximately 6:00 a.m. CDT (NOAA,
2005). Report 1 represents disaster risk data estimated on October 21, 2005 at 7:00PM
CDT, and Report 2 represents disaster risk data estimated on October 23, 2005 at 1:00
p.m. CDT.
Actual
d AFS, Fl
H
Orlando, FL
N Pensacola, FL Airport
Tall
West
eeFL
m Beach, FL
E Pensacola, FL
Ocala, FL
Report 1
D Tallahassee, FL
Orlando, FL
all
ee, FL
est Palm
Duke Field AFS, FL
ach, F
ELakeland, FL
Report 2
West Palm Beach, FL
Punta orda, FL
|Punta
Duke Field AF, FL
-
500,000
1,000,000
1,500,000
Gorda, FL
2,000,000
Units in Inventory
Figure 5-11 Optimal allocation of MREs, minimize time, for two prepositioning time intervals
91
I
1
II
Actual
A Orlando, FL
0 Pensacola, FL Airport
Pun
allahassee,
Gorda, FL
* Pensacola, FL
M Ocala, FL
Report 1
M Tallahassee, FL
Lakeland, FL
Duke Field AFS, FL
T
Report 2
s e FL
Punta G rda, FL
.West
ELakeland, FL
Palm Beach, FL
Laelan Fl
E Punta Gorda, FL
-
1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000
Units in Inventory
Figure 5-12 Optimal allocation of water bottles, minimize time, for two prepositioning time intervals
It is interesting to observe how the allocation of inventory changes between Report 1 and
Report 2, which are themselves about 40 hours apart. For MREs, Orlando is not allocated
any stock in Report 1, but as the area of impact of the storm becomes clearer with Report
2, it is added back in as a warehouse location. The changes in inventory levels for water
bottles are less pronounced between Report 1 and Report 2. This is likely because the
overall quantity of water bottles in the system is much larger than that for MREs, so with
a relatively even allocation at the time of Report 1, only small changes are made in
Report 2 to serve the expected demand.
In addition to comparing strategies for allocating inventory, it is also interesting to compare the average time to
serve from the actual allocation of all inventory in Orlando to the optimal allocations for Reports 1 and 2.
Table 5-5 summarizes the benefits, in terms of average time to serve (per unit
delivered) for actual and optimal inventory allocations for both Report 1 and Report 2
(minimizing time).
92
Table 5-5 Average time to serve per unit of commodity for prepositioning scenarios
Water Bottles
MIREs
Percent
Percent
Actual
Optimal
decrease (to
Actual
Optimal
optimal)
decrease
(to
optimal)
Report 1
7.18 hours
7 hours
2.5%
7.62 hours
7 hours
8.9%
Report 2
7.72 hours
7 hours
10.3%
8.4 hours
7.28 hours
15.4%
In the case of water bottles, optimally allocating supplies in advance of a notice event
would reduce the per unit time to serve by 9-15%, depending on when the decision was
made to allocate supplies.
Looking closer at water bottle allocation in Report 1 and Report 2, it is observed
that the biggest advantage of optimally allocating inventory throughout the LSAs in
Florida comes in the first seven hours of the response. In Figure 5-13, at a time of 7
hours, the actual allocation of supplies will serve approximately 11% of demand, while
the optimal allocation will serve approximately 30% of demand.
93
0.35
0.3
41
0.25
eu
0.2
C
0.15
241
-+-Actual
-
C
Cu
5-
WOptimal
0.1
0.05
0
0
2
4
8
6
10
12
14
16
18
20
Time to Respond (hours)
Figure 5-13 Fraction of demand served, minimize time, for water bottles in Report 1 (10/22/2005 at 7PM CDT)
0.8
0.7
41
0.6
0.5
Cu
241
0.4
/l
'4-
+
Actual
C
C
0.3
Cu
I-
0.2
-
aOptimal
0.1
0
0
2
4
6
8
10
12
14
16
18
20
Time to Serve (hours)
Figure 5-14 Fraction of demand served, minimize time, for water bottles in Report 2 (10/23/2005 at 1PM CDT)
In Figure 5-14, which compares the fraction of demand served over time to serve for the
actual and optimal allocations at Report 2, which represents forecasts on October 23,
94
2005 at 1:00 p.m. CDT, approximately 17 hours before landfall of Hurricane Wilma.
Compared to Report 1, optimally allocating water bottles in Report 2 increases the
fraction of demand served at both 7 and 8 hours after the disaster. Additionally, note that
the overall fraction of demand served in Report 2 increases to approximately 70%, as
compared to about 30% when estimates are made in Report 1. A more focused storm
track allows the allocation of inventory to be more precise and meet demand in the areas
most likely to be impacted by the disaster.
5.5
Scenario 3: FDEM Physical Inventory Allocation with Reduced Demand
This section evaluates FDEM's allocation of physical inventory with a 10% reduction in
demand in Miami-Dade County. Miami-Dade County was chosen because it has frequent
disaster activity and has some of the largest disasters by total affected population in the
disaster risk database used. Ten percent was chosen as a conservative estimate of the
impacts of a disaster risk reduction plan in the county. In the future, higher values could
be tested to evaluate the relative impacts of varying levels of demand reduction. This
additional sensitivity analysis would give decision makers an idea of the threshold for
demand reduction that would result in shifting supply, or allow them to meet given
targets for response with their current stock levels. The following sections assess several
metrics related to the allocation of stock given this reduced demand.
5.5.1
Ability to meet demand
Similar to the evaluation of FDEM's physical inventory with the original disaster data,
the ability of the current physical inventory to meet the new reduced demand was
evaluated. Table 5-6 compares a few key metrics for evaluating the ability of FDEM's
current inventory to meet demand when demand is reduced in one location, Miami-Dade
County.
Table 5-6 Ability of Florida inventory to meet demand with 10% reduction in demand in Miami-Dade County
demand
served
Reduced
(Y)
Fraction of
disasters
served
Origial
and
Reduced
0.80
0.94
(8)
0.96
0.99
Fraction
Item
Units
Demand
Original
()
MRE
Water
1,600,000
5,292,000
290,519
435,909
Demand
Reduced
()
278,907
418,486
Demand
Met
Original
Demand
Met
Reduced
Fraction
of demand
served
Original
(y)
223,032
394,191
0.79
0.93
V(
228,223
407,377
95
of
Bottle
Similar to the results observed for FEMA, it is observed here that while the fraction of
disasters covered completely remains the same for the original and reduced demand
conditions, the fraction of demand served (y) increases as demand is reduced in MiamiDade County. This means that the current allocation of inventory is able to serve a greater
fraction of the overall demand. What this shows is that while the overall demand was
reduced, the portion of overall demand that can be met given where stock is currently
positioned (all in Orlando) is greater when demand is reduced in Miami-Dade county.
5.5.2
Network inventory allocation strategies
The actual allocation of physical inventory was compared to the optimal allocation with
original demand and the optimal allocation with reduced demand in Miami-Dade County.
It was hypothesized that a reduction in demand in one geographic location would change
the optimal allocation of inventory across warehouses. While not explicitly evident in
Figure 5-15 and Figure 5-16, reducing demand in Miami-Dade County did change the
inventory levels in the LSAs across the state slightly. It is important to note that FDEM
currently keeps its entire inventory in Orlando, FL, and would only move stock to these
locations in an impending disaster, but these locations are available for short term storage
of stock and as shown in Section 5.3.4, balance metric, do give FDEM significant
opportunity to improve on a number of their metrics.
96
1
I
Actual
" Orlando, FL
" Pensacola, FL Airport
P nta Gorda, FL
" Pensacola, FL
" Ocala, FL
OptTime-RedDmd
Lakelan
" Tallahassee, FL
, FL
Punta Gorda, FL
P nta Gorda, FLI
E Lakeland, FL
OptTime-Orig
West Palm Beach, FL
*Duke Field AFS, FL
Lakelan, FL
500,000
I
I
1,000,000
1,500,000
2,000,000
Units in Inventory
Figure 5-15 Actual vs. optimal allocation of inventory with original and reduced demand for MREs (minimize
time)
1
I
I
Actual
" Orlando, FL
U Pensacola,
FL Airport
" Pensacola, FL
Punta Gorda, FL
" Ocala, FL
OptTime-RedDmd
" Tallahassee, FL
. Punta Gorda, FL
P
ta Gorda, FL
N Lakeland, FL
* West Palm Beach, FL
OptTime-Orig
N Duke Field AFS, FL
I ~~
-
4,000,000
2,000,000
Units in Inventory
97
6,000,000
Figure 5-16 Actual vs. optimal allocation of inventory with original and reduced demand for water bottles
(minimize time)
In the case of both MREs and water bottles, when demand is reduced in Miami-Dade
County, the optimal solution minimizing time moves units of stock out of West Palm
Beach and distributes that stock throughout the other warehouses in the system. In the
case of water bottles, this represents a 26% increase in stock in Tallahassee and a 13%
increase in stock in Orlando when demand is reduced in Miami-Dade County. For MREs,
stock in West Palm Beach is also reduced, and Duke Field AFS (now Eglin Air Force
Base) sees a 12% increase in stock of MREs when demand is reduced. This is interesting
because if demand were reduced further in Miami-Dade, a more even allocation of
inventory across the state might be beneficial. Looking at overall cost savings as a result
of the reduced demand, there is a total savings of $843 for MREs and $3426 for water
bottles. These numbers are insignificant in terms of the cost of an overall response, but do
represent savings that could be factored into a larger accumulation of savings as a result
of mitigation or other efforts to reduce demand.
6 Discussion
The case study analyses for FEMA and FDEM represent an expansion in the application
and function of the Stockpile Capacity Model (Acimovic & Goentzel, 2015) to domestic
response agencies logistics networks, and to disaster scenarios where more than one
location (e.g. county) is impacted at any one time. Additionally, the managers of the
emergency logistics networks studied can use the results of these analyses to make
decisions in the near term. This section will summarize the findings of the FEMA and
FDEM cases studies and build upon them with discussion of how these findings relate to
the central research questions and previous literature, and how they inform the
development of a framework for evaluating logistics and other preparedness decisions.
6.1
Federal Emergency Management Agency
As part of its responsibilities as one of the coordinators of ESF #7, FEMA must
coordinate "comprehensive national incident logistics planning, management, and
sustainment capability" (United States Department of Homeland Security, 2013). A
98
review of the policies and other documentation around disaster response revealed that
performance based metrics for making logistics relevant preparedness decisions were
lacking. The analysis conducted in the case study highlights several metrics that can be
evaluated to assess how well equipped the FEMA logistics network is to meet the needs
of disaster survivors after an event, including the allocation of inventory throughout the
network, the fraction of demand served over time given the current and optimal allocation
of inventory, and the RC. The takeaways from the FEMA analysis include (1) overall
allocation of inventory was well balanced, but including a large portion of stock in the
mid-Atlantic region is sub-optimal based on the historic distribution of disaster risk
across the continental United States, (2) restructuring contracts with MRE vendors to
only two locations as opposed to three, and negotiating shorter wait times for vendor
stock would allow FEMA to serve a greater fraction of overall demand, and (3) that while
reduced demand in one state may not necessarily change the optimal strategy for
allocating critical commodities within the logistics network, it will improve the response
by other metrics.
In all three findings, the optimization model output does not tell the entire story,
but does provide important metrics on which to compare and contrast benefits derived
from logistics decisions. The metrics provide a benchmark for assessing whether the
reasoning for current stock allocation decisions make sense in light of benefits that can be
gained by optimizing allocation of resources. For example, in the case of the allocation of
physical inventory, FEMA's decision to place a large portion of stock in the mid-Atlantic
in two warehouses in Maryland may not be optimal, but it may serve the purpose of
satisfying political or organizational requirements for geographic diversity of stock
allocation, because these organizational factors take into account more than cost and time
to serve, while those are important metrics. Furthermore, while the restructuring of
contract inventory to just two suppliers may represent the optimal solution in terms of
time and cost to serve the disaster affected population, the optimally located suppliers
may not have capacity to increase supply to the desired levels. Additionally, from a
policy standpoint having diversity in suppliers will reduce the risk that a significant
portion of supply is interrupted due to an issue with a given supplier. In the case of
reduced demand, despite not being observed to change the allocation of commodities
99
throughout the logistics network for FEMA, it did increase the fraction of demand served
for both commodities that were evaluated, which is an important metric for emergency
management organizations. It means that they are able to do more, to meet more demand,
to serve more people with the inventory they currently have. The analysis conducted for
FEMA develops metrics that the organization can use to evaluate current and future
logistics planning and preparedness decisions.
6.2
Florida Division of Emergency Management
The FDEM's logistics planning documents have been developed with the intent of
ensuring that the state is well equipped to respond to the needs of its citizens in the event
of a disaster response. The analysis conducted for FDEM suggests that overall, the
FDEM logistics network is well equipped to deliver on its plans. However, similarly to
FEMA, the planning guidance and documentation in Florida lacks metrics on which to
evaluate the response capacity. Using the robust DataCo disaster risk dataset, the case
study assessment of FDEM's logistics network found that (1) the current inventory level
is able to meet a large fraction of the overall demand in the state, but the allocation of
inventory in one central warehouse means that metrics evaluating the time and cost to
deliver goods to the disaster affected population suffer, (2) deploying inventory to
prepositioning sites several days in advance of a hurricane event has the opportunity to
improve the overall time to serve as well as the fraction of demand served, and (3)
reducing demand in one location may impact both the optimal allocation of inventory
throughout the logistics network as well as the fraction of demand that can be served.
Similarly to FEMA, the analysis of FDEM's logistics network does provide useful
metrics but there are a few key tradeoffs that are likely to factor into decision-making that
are not reflected in the optimal solutions. First, allocating all stock in one central location
in Orlando may not allow FDEM's logistics network to meet demand as fast as would be
possible with a more disaggregated allocation of supply, but the geography and nature of
hazards in Florida may make locating stock in multiple warehouses not feasible or
prudent. The scenarios do not analyze, for example, the potential risk of a warehouse in
West Palm Beach to damage in a hurricane event. That is analysis that could show that
from a risk standpoint, Orlando actually is the best option for placing stock. Interestingly,
the findings from the scenarios where focused storm tracks and projected affected
100
population data in advance of a hurricane were used to preposition stock represent the
best of both worlds. The results of this analysis support keeping stock in a centrally
located warehouse for most of the year, as FDEM does in Orlando, and highlights the
benefits of prepositioning stock to pre-identified locations closer to the impact of a notice
event. This is not possible for certain types of hazards, such as tornadoes and
earthquakes, but is feasible for hurricanes and some flood events. The findings related to
reduced demand in Miami-Dade county resulting in a shift of inventory out of West Palm
Beach shows that as demand changes, the best option for allocating stock may also
change. In this case, the warehouse that holds the most stock in the optimal allocation,
West Palm Beach, loses stock to other locations throughout the state, spreading the
existing inventory out a bit better. This is intuitive, but does show a real impact of
reduced demand as a result of resilience or preparedness work that reduces demand for
critical commodities following a disaster.
6.3
Cross-Case Findings
The types of decisions explored using the case study analyses fall into the broad category
of preparedness decisions, decisions that are made in advance of identified needs. Given
the definition of humanitarian logistics as set forth by (Thomas & Mizushima, 2005),
which emphasizes that the field is concerned with the planning as well as the
implementation of a flow of goods to meet the beneficiary's requirements, these
decisions are a key part of the function of a response organization's logistics network. In
this definition of humanitarian logistics, efficiency and cost-effectiveness of the flow of
goods is also emphasized. Despite the sometimes unique constraints placed on the supply
chain in a disaster response, efficiency and cost-effectiveness remain important
considerations. The metrics assessed in this study across both cases were directly related
to minimizing time or cost to meet the needs of disaster survivors.
The analyses used simulated disaster forecast data to evaluate the implications of
preparedness decisions, and focuses on three types of decisions: allocation of resources,
structuring of contracts with private partners or outside vendors, and decisions related to
preparedness goal setting and implications. This section discusses the cross case
similarities and differences in the findings for each of these decisions, and relates the
findings of this study to previous work.
101
6.3.1
Allocation of critical commodities
The findings related to allocation of critical commodities represent an addition to the
discussion on logistics preparedness decisions. Even if the solutions suggested by the
model are not feasible or desirable given current policy, physical or other constraints,
they provide a starting point on which to make informed decisions about allocating
inventory. While an organization may not completely agree with the optimal allocation of
goods because it concentrates all stock in one area of the state or country (e.g. TX and
GA for FEMA and West Palm Beach for FDEM), decision makers now have better
intuition and metrics to consider alternate strategies. The results also quantify the impact
of suboptimal solutions that are driven by policy or organizational justifications to better
inform the debate around such decisions.
This ability to make decisions on the basis of a benchmark or starting point is a
complement to the frameworks already in place, such as the NRF (United States
Department of Homeland Security, 2013), in which the importance of planning,
operational coordination, and mass care services, among other objectives of the Response
mission area. The ability to include a benchmark for delivering critical commodities to a
mass care operation, for example, would facilitate the application of the NRF or other
similar state or local framework because there would be a way to quantitatively assess
whether a logistics network as currently organized and stocked was able to deliver on the
core capabilities outlined in the NRF to provide certain services or goods. For example,
the balance metric (Acimovic & Goentzel, 2015) applied in this study to FEMA and
FDEM is a quick and intuitive indicator of how out of balance, in general, the allocation
of stock across a logistics network is, and the analysis results suggest what specific
actions can be taken to improve the overall balance of stock within the network. This
metric as well as the other metrics applied in this study is not intended to be static
measures of the allocation of inventory, which is important to note when suggesting that
they complement a framework like the NRF. Similarly to the fact that the NRF serves as
a guide to response but can be tailored to scale up or scale down for a given disaster, the
metrics used here to evaluate the allocation of inventory throughout a logistics network
represent a framework, or a toolkit from which a response organization can utilize the
most applicable metric or set of metrics to assess a given issue. The metrics can also be
102
updated as the allocation of inventory throughout a logistics network shifts, for example
when stock is deployed to a disaster but unused stock must be returned to warehouses
following the response.
The case for using the model output in making prepositioning decisions is strong,
with the findings from the FDEM case study clearly presenting the benefits in terms of
time to serve and fraction of overall demand met. This is in line with intuition, as well as
previous findings that the time to respond when goods are located closer to the disaster
affected population the time to respond is lower (Duran et al., 2011). Also, the results
from this study reinforce the findings in Lodree Jr. et al. (2012), which were from the
perspective of a commercial retailer but used a similar methodology as was used here.
What this study adds to both studies mentioned here is a set of metrics and a framework
that decision makers can use this to strengthen the case for prepositioning goods. It also
extends the application of the methodology to a state logistics network (in addition to the
international context explored by Duran et al. (2011) and the private context in Lodree Jr.
et al. (2012)).
Lodree Jr. et al. (2012) have a similar question of where to allocate supplies in
advance of a notice event, but for a commercial supply chain, and arrive at a similar
conclusion that prepositioning is beneficial to the overall ability to respond. In addition,
Lodree Jr. et al. (2012) conduct a more in-depth sensitivity analysis to estimate the
impact of cost parameters on the expected benefit of prepositioning and also reformulate
the approach to the problem, using a Percentage of Demand Scenario solution Approach
in addition to the two-stage stochastic linear programming model to increase efficiency in
obtaining results and reduce computation time, an important consideration if this
approach were to be used in practice. In this study, the model formulation to evaluate
prepositioning decisions had a computation time of only a few minutes, as the number of
disaster scenarios was reduced significantly to just two scenarios (Report 1 and Report 2)
from the statewide analysis for FDEM's logistics network that considered the entire
catalog of disaster scenarios, so a reformulation of the model or approach was not
necessary to conduct the analysis. However, when conducting the analyses for statewide
allocation decisions, the computation time for an individual scenario was between 20 and
50 h. As these analyses would be performed to inform decisions that would be made well
103
in advance of an event, this length of time can be accommodated. In the case of
prepositioning decisions, output is needed more urgently.
6.3.2
Impact of partnerships in meeting needs
The inclusion of the private sector in planning for disasters has been identified as an
important consideration for emergency managers (Flynn & Prieto, 2006), and the analysis
conducted with the FEMA contract inventory only further supports this theory. Hurricane
Katrina remains a constant reminder of what can go wrong in disaster response, but as
Horwitz (2008) finds in a review of private sector and Coast Guard engagement in the
Hurricane Katrina response, public emergency management organizations can benefit
from and learn from the policies of private organizations in their response to this historic
hurricane event. In his study, he cites the ability of retailers including Wal-Mart and
Home Depot to deliver critical commodities to disaster survivors much faster than FEMA
was able to do at the time.
While this study did not explicitly focus on the capacity of Wal-Mart or Home
Depot, it does suggest a methodology for assessing the involvement of any private
organization in the response in addition to the stock already in the logistics network for
the responding organization. This type of analysis is relevant as many responding
organizations will have at least some organic or physical supply that is either owned
outright or is stored in an agency's warehouse through a VMI agreement, but will need to
initiate contracts prior to an event to provide additional capacity. The metrics from and
results of the analyses conducted for FEMA can inform long term structuring of contracts
to optimize on time or cost to serve the disaster affected population. Where should
strategic surge contract supply be located to ensure that it is able to serve a large fraction
of the disaster-affected population across the country?
6.3.3
Resilience goals
The decisions modeled in the case study scenarios can be classified as preparedness
decisions. They are made assuming that a disaster will occur and that it will impact
people and property such that demand for commodities such as water and food will be
generated. There is, however, another closely related emergency management function
that impacts preparedness decisions and can sometimes be confused with preparedness.
104
This is the mitigation function. As Haddow et al. (2011) point out, mitigation is different
from preparedness because mitigation attempts to "prevent the event altogether" through
methods that reduce or eliminate the hazard itself (e.g. flood control structures) or the
negative consequences as a result of the hazard (e.g. elevation of homes), whereas
preparedness seeks to "improve the abilities of agencies and individuals to respond to the
consequences of a disaster event once the event has occurred."
The scenarios modeled in this study aim to quantify the benefit to or impact on
the logistics network as a result of activities that reduce demand for critical commodities
in a disaster, which could fall into the categories of mitigation or preparedness, or might
be classified as resilience goals. As the term is being used here, resilience goals refer to
goals that might fall under preparedness, mitigation, or another function of emergency
management but share the common goal of improving a community's ability to return to
a pre-disaster state quickly and effectively following a disaster. Here, the assumption is
being made that these types of goals might reduce the demand for critical commodities,
and scenarios are run to evaluate the impact of reduced demand on the logistics network.
Reduced demand might occur in a number of ways in addition to resilience goals,
mitigation, or preparedness decisions, so the insight derived from these scenarios apply
broadly to cases where demand is reduced.
Reduced demand as a result of resilience goals may not change the allocation of
inventory in the system, but, as observed in both the case of FEMA and FDEM, it will
allow a logistics network to serve more of the overall demand with its current stock. In
the case of FDEM, the reduced demand in Miami-Dade county did shift the optimal
allocation of inventory slightly among Florida's LSAs, so there is the potential that if a
large enough consistent shift in demand were observed, optimal allocation of inventory
throughout a logistics network would shift.
6.4
Framework
The framework outlined in Figure 6-1 encompasses the assessments conducted in this
thesis through the FEMA and FDEM case studies and positions these assessments as part
of the broader set of preparedness decisions. Further, it suggests how these decisions and
the RCI and related metrics could contribute to an overall disaster resilience index,
similar to the one proposed by Cutter et. al (2010) or the Resilience Capacity Index
105
proposed by Foster (2011), including the other aspects of such an index that would
compliment the research done here.
The Resilience Capacity Index developed by Berkeley's Institute of
Governmental Studies (Foster, 2011) combines twelve equally weighted indicators, four
in each of three dimensions to understand how a region might respond to a future stress.
Similarly, the Resilience Index indicated in the framework would combine metrics from a
set of indices or indicators for different phases of the disaster response lifecycle, such as
preparedness, mitigation, response, and recovery into one metric for comparing the
resilience of a given government, commercial, or other organization or entity to its own
identified benchmarks/goals, its own past performance, or other similar entities or
organizations. One of these incorporated indices or indicators into an overall disaster
resilience index would be the RCI (Acimovic & Goentzel, 2015), as an example. What
the RCI and related metrics used in this study and the disaster resilience index included in
the framework do is provide tools for decision makers and leaders to evaluate their
decisions within disaster lifecycle and set goals, benchmarks, and standards. The lack of
widely used standards currently limits critical evaluation and accountability within
organizations in the disaster mitigation, preparedness, response, and recovery space, and
makes it difficult for comparisons to be made across and/or within entities involved in
disaster response.
Within the logistics decisions for preparedness, the framework breaks down the
types of decisions into decisions relevant for a public emergency management
organization and those relevant to private organizations participating in disaster response
through public private partnerships, vendor managed inventory arrangements or other
contract mechanisms. It is split this way because the two sectors have different
responsibilities when it comes to meeting the needs of disaster survivors after an event.
Public emergency management organizations are mandated by state or federal statutes to
serve disaster affected populations, so have several key decisions to consider, whereas
private organizations may choose to be involved in disaster response, committing goods
or agreeing to provide additional capacity to public emergency management
organizations through contracts.
106
A number of outside factors impact the logistics decisions for both entities but
one could argue that the most significant factor in planning decisions for delivering
critical commodities after a disaster is the expected demand for these items. This input is
highlighted here as it relates to other preparedness decisions that might be expected to
change demand. The RCI, as shown here, distills the results of an analysis of various
logistics decisions into one figure and is suggested as part of a future disaster resilience
index.
PREPAREDNESS DECISIONS
Rsduces Demand for Critical Comrnnowts
Laqisks Dodsios
Pr*paednew DOcMOIsi
0Othff
Ir
a
C
p
dt
d
ds
MllraMliktoyrodng
xdwkMasucture
Promoe hrdual readines
atus
s
Graphic
-emiinn
&
Puttic E!1w9eKy Mnament
e
em
loaino mnd
em
gsies
df used s
S-Wreourk
aNbembn(s with pb&ateyConact
I
d ongattgons fh
Private Orgaizations
s
e ova NOlutf CdCal coemodtai ey
* e
- Pnewsitank.o
jdcxiwPonncnun, eic.
esponse Plans
Fth
' Respcows
Capacitaem
*
indexm
0t0wlrsdlces
rfoMmgauon,
Planning. etc.
Figure 6-1 Framework for assessing logistics and other preparedness decisions. Graphic designed and illustrated
by Nicole Seelbach (Seelbach & Seelbach, 2015).
In addition to the framework presented in Figure 6- 1, the analysis conducted in the case
studies can be summarized into a table listing the logistics decisions informed, the
metrics evaluated, and the data needed to conduct the analysis. Table 6-1 shows this.
107
These metrics could also be a complement to the benchmarks in the categories of
institutional resilience or infrastructure resilience as put forth by Cutter et. al (2010).
Institutional resilience is defined there as containing "characteristics related to mitigation,
planning, and prior disaster experience", whereas infrastructure resilience is defined as
"an appraisal of community response and recovery capacity" (Cutter et al., 2010).
Indicators within these categories currently cover benchmarks like shelter capacity (as
measured by the number of vacant rental units in a community), mitigation (as measured
by the percent of the population participating in CRS for Flood, the percent of the
population covered by a recent hazard mitigation plan, and the percent of population in
Storm Ready communities), and recovery (number of public schools per square mile), but
do not include benchmarks specifically related to a community's ability to provide
individuals with critical commodities like water and food after a disaster, which the
metrics applied here do. Some of the benchmarks might serve as proxies for the ability to
provide these commodities (e.g. previous disaster experience), but none explicitly do so.
Table 6-1 Logistics decisions and associated metrics evaluated and data required for analysis
Decision
Metrics Evaluated
e
Where to allocate inventory?
e
e
Data Required
Response Capacity Index
e
Ability to meet demand
Comparison of current vs.
optimal inventory allocation
e
e
Warehouse locations
Warehouse stock
Disaster data (historic and
forecast)
All from base allocation, plus:
Where to preposition
inventory in a notice event
(e.g. hurricane?)
*
C omparison of current vs.
optimal inventory allocation
Ability to meet demand
Time to serve
e
e
Potential prepositioning
locations
Cone of expected disaster
impact several days out
from storm
Should I enter into a contract
with the private sector?
Private sectorperspective:
How much am I improving
the response?
*
*
Current allocation strategy for
contract stock vs. optimal
allocation
Demand served by hour
Where would an additional
unit of stock be the most
valuable?
108
All from base allocation, plus:
Contractual private sector
agreements
Locations of private sector
What is the benefit, in terms
of cost and time to serve the
affected population, if
resilience goals reduce overall
demand?
*
*
Ability to meet demand
Comparison of current vs.
optimal allocation of inventory
All from base allocation, plus
* Resilience goals for a county
or state
Similar to Cutter et. al (2010), this study uses metrics to evaluate an aspect of a
community's resilience. However, it adds to the findings of that study in that it uses data
that are not yet publicly available, and speaks to one specific area of overall resilience,
rather than Cutter et. al (201 0)'s proposal of a comprehensive set of benchmarks for
assessing disaster resilience. The area of overall resilience covered by the metrics and
index used in this study and applied to the overall framework is not explicitly included in
Cutter et. al (2010)'s set of benchmarks, so could serve to inform an added indicator or
set of indicators in the set proposed there.
6.5
Future Research
This thesis focused only on answering the questions/decisions that are classified in the
framework as logistics decisions, and also employed the RCI in assessing logistics
decisions. It also included assessment of the impacts on logistics decisions to changing
inputs through resilience goals and potentially reduced demand. It did not focus on the
other preparedness decisions listed in the framework, nor did it aim to precisely estimate
the impact of these other preparedness decisions on overall resilience or on demand. This
section discusses the impact of the model and data limitations on the research, suggesting
additions that would enhance the work done here.
6.5.1
Impact of model and data limitations
The case studies were subject to limitations in terms of the data available for input as well
as limitations due to the current stockpile capacity model formulation. Limitations to
input data are addressed in detail in the future research section, but the key challenge is
the quality of disaster risk data and the translation of disaster risk data into persons
affected. The FDEM analysis benefit from a more robust disaster risk dataset, but the lack
of approved and validated methods for determining affected population may have
resulted in over or under estimating the total affected population for that analysis. That
said, the methods used in the FDEM case study analysis are intuitive and simple, aligned
109
with the private sector assessments used by the insurance industry, and produced
reasonable estimates of total affected population. The FEMA analysis uses data with a
different methodology for estimating total affected population, and includes fewer
historic/predicted disaster events, so could benefit from a robust dataset of disaster
events.
In terms of the impact of limitations of the stockpile capacity model on results,
the most notable limitation was observed in the FEMA analysis for the contract stock.
Because the optimization will force all available inventory, regardless of when it
becomes available into the warehouse locations with the least restrictive time delay, it is
not possible to impose a delay on warehouse availability within the model. In the case of
FEMA, this limitation was not relevant because conversations with logistics personnel
revealed that the assumption that contract stock will be available immediately or shortly
after a disaster event is realistic. However, a model enhancement would improve the
ability of the stockpile capacity model to assess varied availability contract terms.
Finally, two improvements to the RCI are suggested. First, a method for conducting
thorough sensitivity analysis on each of the metrics is needed. The FEMA case study
presents a method for how this would be done for the balance metric time and cost
metrics, by creating a matrix of potential locations and moving all contract stock to the
plausible extreme locations in the matrix to evaluate the bounds of the balance metrics.
This method worked for FEMA, but a generalizable method is needed that can be used
regardless of the unit of analysis. Second, focused discussions with potential users of the
RCI is needed to evaluate both the range of metrics that are appropriate to include in the
RCI as well as the weighting of each of the metrics relative to the others. Should
additional metrics be included in the RCI? Should the current and potential additional
metrics all be weighted equally, or should individualized weights be developed? Focused
discussions with logistics experts in emergency management, private sector, and other
organizations who would use the RCI is needed to ensure that the index is robust and can
be applied across the range of organizations who would use it.
6.5.2
Complimentary future research
here (e.g. mitigate risk to flooding) and others would impact demand for critical
110
'
Future research could explore how preparedness decisions such as the ones mentioned
commodities in a disaster response, and also how these decisions themselves might be
measured in an index. The creation of other indices to measure aspects of the disaster
response lifecycle including preparedness, response, recovery, and mitigation would
supplement the response capacity index in creating a comprehensive resilience index.
Additionally, future indices could evaluate preparedness decisions for other hazards,
supplementing the research on natural hazards done here.
In addition to research to inform the overall framework, additional research on a
few key inputs to this study would improve not only the conclusions made here but future
analysis in this space as well. The key inputs that would benefit from additional focused
research include (1) disaster risk data (2) affected population data.
In this study, disaster risk data were obtained from EM-DAT, a publicly available
dataset with data on disasters that have occurred all over the world, as well as from
DataCo, a firm specializing in catastrophe risk quantification and decision analytics
serving the risk and insurance industry, among others. The data from EM-DAT had the
benefit that it included information from disasters that had occurred in the past, and thus
might better be able to describe the scenario. On the other hand, the EM-DAT data is also
limited by the fact that it only represents historic events. It therefore cannot predict future
events that have a larger impact than events that have occurred in the past. The DataCo
dataset included both historic events and events generated stochastically from a catalog of
characteristics of hurricane events. The limitations of the DataCo dataset were that it did
not include an estimate of affected population for each event, and it only included loss
estimates from hurricane wind and storm surge. Each dataset had merits and drawbacks,
but a more comprehensive set of data on disaster events, applied to the analysis done in
this study would render the results even better for informing preparedness decisions.
As presented in the FEMA case study, a comparison of the disaster affected
population as reported by EM-DAT and the population as assumed using the DataCo
disaster risk data was done, and the results point to challenges in reaching a commonly
agreed upon methodology for determining the total number of people that were or will be
impacted by a disaster event. EM-DAT's figures, based on the injured, homeless, and
affected populations as reported by governments and media, compared to the simple
assumption used here to relate residential property damage to affected population
III
produced different estimates of affected population for the same historic events. The
hypothesized reasons why these figures are different is covered in the FEMA study, but
the broad question of how to assess affected population remains. FEMA estimates total
affected population using a combination of Hazus and census data for the area that has
been affected in a disaster (FEMA Staff 3, 2015), while EM-DAT relies on reported
estimates and when that is not available on relationships between damaged homes and
average family size (Centre for Research on the Epidemiology of Disasters, 2015), and
DataCo relies on a hazard threshold to make a binary affected or not affected decision on
a county-by-county basis, as described in the methods section. These are all very
different ways to arrive at the same desired output, and as such produce different
numbers. Preparedness decisions rely on these assumptions, so a commonly agreed upon
method for making assumptions about affected population would help to bring
organizations involved in disaster response to a common foundation for making these
decisions.
112
7 Bibliography
Acimovic, J., & Goentzel, J. (2015). Models, Metrics, and an Index to Assess
HumanitarianResponse Capacity. Retrieved from http://ssrn.com/abstract=2584560
Acimovic, J., & Graves, S. (2015). Making Better Fulfillment Decisions on the Fly in an
Online Retail Environment. Manufacturing & Service OperationsManagement,
17(1), 34-51.
Arvis, J. F., Saslavsky, D., Ojala, L., Shepherd, B., Busch, C., & Raj, A. (2014).
Connecting to compete - trade logistics in the global economy - the logistics
performance index and its indicators. The World Bank.
Arvis, J. F., & Shepherd, B. (2011). The air connectivity index: Measuring integration in
the global air transport network.
Balcik, B., & Ak, D. (2014). Supplier selection for framework agreements in
humanitarian relief. Productionand OperationsManagement, 23(0), 1028-1041.
doi: 10.111 1/poms.12098
Bergqvist, R., & Pruth, M. (2006). Public / Private Collaboration in Logistics: An
Exploratory Case Study. Supply Chain Forum, 7(1), 104-114.
Business Response Task Force. (2007). Getting Down to Business: An Action Planfor
Public-PrivateDisasterResponse Coordination.
Caplice, C., & Sheffi, Y. (1994). A Review and Evaluation of Logistics Metrics. The
InternationalJournalof Logistics Management, 5(2), 11-28.
Centre for Research on the Epidemiology of Disasters. (2015). EM-DAT: The
OFDA/CRED International Disaster Database. Retrieved March 26,2015, from
www .emdat.be
Chen, J., Chen, T. H. Y., Vertinsky, I., Yumagulova, L., & Park, C. (2013). PublicPrivate Partnerships for the Development of Disaster Resilient Communities.
Journalof Contingenciesand Crisis Management, 21(3), 130-143.
doi: 10.1111/1468-5973.12021
Council of Supply Chain Management Professionals. (2013). Supply Chain Management
Terms and Glossary. Retrieved from
https://cscmp.org/sites/default/files/user-uploads/resources/downloads/glossary2013.pdf
Cutter, S. L., Burton, C. G., & Emrich, C. T. (2010). Disaster Resilience Indicators for
Benchmarking Baseline Conditions. Journalof HomelandSecurity and Emergency
Management, 7(1), Article 51. doi: 10.2202/1547-7355.1732
113
Darcy, J., & Hofmann, C.-A. (2003). According to need? Needs assessment and decisionmaking in the humanitariansector. London, United Kingdom.
DataCo. (2014). ClimateCastReal-time Analysis of Hurricane Wilma.
DataCo. (2015). Disaster Risk Data.
DataCo Staff. (2015a). Electronic Communication 3.
DataCo Staff. (2015b). Electronic Communication 4.
Davidson, A. L. (2006). Key PerformanceIndicatorsin HumanitarianLogistics. The
Massachusetts Institute of Technology.
Donahue, J. D., & Zeckhauser, R. J. (2006). Public-Private Collaboration. In M. Moran,
M. Rein, & R. E. Goodin (Eds.), The Oxford Handbook of Public Policy (pp. 496525). Oxford: Oxford University Press. Retrieved from
http://www.hks.harvard.edu/fs/rzeckhau/public-private-collaboration.pdf
Duran, S., Gutierrez, M. a., & Keskinocak, P. (2011). Pre-Positioning of Emergency
Items for CARE International. Interfaces, 41(3), 223-237.
doi: 10.1287/inte. 1100.0526
Edmonson, A. C., & McManus, S. E. (2007). Methodological Fit in Management Field
Research. Academy of ManagementReview, 32(4), 1155-1179.
Esri. (2015). ArcGIS Web Platform. Retrieved April 1, 2015, from
http://www.esri.com/software/arcgis
Federal Emergency Management Agency. (2011). National DisasterRecovery
Framework.
Federal Emergency Management Agency. (2014). Emergency Supply List. Washington,
DC: U.S. Department of Homeland Security. Retrieved from
papers2://publication/uuid/F716783F-CAAA-474A-B634-49E8BF74003B
Federal Emergency Management Agency. (2015a). National Preparedness Cycle.
Retrieved May 6, 2015, from http://www fema.gov/national-preparedness-cycle
Federal Emergency Management Agency. (2015b). Whole Community. Retrieved May 6,
2015, from https://www.fema.gov/whole-community
FEMA Staff 1. (2015). Electronic Communication 1.
FEMA Staff 2. (2015). Personal Communication 1.
114
FEMA Staff 3. (2015). Electronic Communication 2.
Florida Division of Emergency Management. (2013). State of Florida Unified Logistics
Section: Logistics Operations and Management Training Course. Retrieved from
http://www.floridadisaster.org/Response/Logistics/Index.htm
Flynn, S. E., & Prieto, D. B. (2006). Neglected Defense: Mobilizing the Private Sector to
Support Homeland Security.
Foster, K. A. (2011). Resilience Capacity Index. Retrieved November 15, 2014, from
http://brf.berkeley.edu/rci/
Fugate, C. (2009). Florida Emergency Management Handbook. Safety Science.
doi: 10.1016/s0925-7535(00)00018-7
Geology.com. (n.d.). Florida County Map with County Seat Cities. Retrieved April 20,
2015, from http://geology.com/county-map/florida.shtml
Goentzel, J., & Spens, K. (2011). Humanitarian Logistics in the United States: Supply
Chain Systems for Responding to Domestic Disasters. In M. Christopher & P.
Tatham (Eds.), HumanitarianLogistics: Meeting The Challenge of Preparingfor
and Responding to Disasters(1st ed., pp. 141-165). Kogan Page Unlimited.
Google. (2014). The google distance matrix API.
G6rmez, N., K6ksalan, M., & Salman, F. S. (2010). Locating disaster response facilities
in Istanbul. Journalof the OperationalResearch Society, 62(7), 1239-1252.
doi: 10.1057/jors.2010.67
Guo, P., Wang, Y., & Liu, F. (2014). Optimal Deployment of Emergency Supply
Inventory for Recurring Disasters with a Humanitarian Relief Objective. Available
at SSRN, 1-40. Retrieved from http://ssrn.com/abstract=2515149
Gustetic, J. (2007). A Frameworkfor Understandingand Designing Partnershipsin
Emergency Preparednessand Response. Massachusetts Institute of Technology.
Haddow, G. D., Bullock, J. A., & Coppola, D. P. (2011). Introductionto Emergency
Management (4th ed.). Elsevier. Retrieved from
http://app.knovel.com/hotlink/toc/id:kpIEMEOOOC/introductionemergency/introduction-emergency
Hoffmann, J. (2012). Ad Hoc Expert Meeting on Assessing Port Performance. Geneva,
Switzerland: UNCTAD.
Horwitz, S. (2008). Making HurricaneResponse More Effective: Lessons from the
PrivateSector and the Coast Guardduring Katrina.Mercatus Policy Series.
115
Kay Roberts, J. (2015). Innovative Activities Potentially Aligning with R!se. Boston,
MA.
-
Kovics, G., & Spens, K. M. (2011). Trends and developments in humanitarian logistics
a gap analysis. InternationalJournalof PhysicalDistribution& Logistics
Management,41(1), 32-45. doi:10.1108/09600031111101411
Kunz, N., & Reiner, G. (2012). A meta-analysis of humanitarian logistics research.
Journalof HumanitarianLogistics and Supply Chain Management, 2, 116-147.
doi: 10.1108/20426741211260723
Lodree Jr., E. J., Ballard, K. N., & Song, C. H. (2012). Pre-positioning hurricane supplies
in a commercial supply chain. Socio-Economic Planning Sciences, 46, 291-305.
doi:0038-0121
National Council for Public-Private Partnerships. (n.d.). Glossary of Terms - NCPPP.
Retrieved January 12, 2015, from http://www.ncppp.org/ppp-basics/glossary-ofterms/
National Hurricane Center. (2005). Wilma Graphics Archive. Retrieved April 27, 2015,
from http://www.nhc.noaa.gov/archive/2005/WILMAgraphics.shtml
NOAA. (2005). Hurricane Wilma. Retrieved April 27,2015, from
http://www.srh.noaa.gov/mfl/?n=wilma
Osborne, S. P. (Ed. . (2000). Public-PrivatePartnerships:Theory and practice in
internationalperspective. London: Routlege.
Ozbay, K., & Ozguven, E. E. (2008). Stochastic Humanitarian Inventory Control Model
for Disaster Planning. TransportationResearch Record: Journalof the
TransportationResearch Board, 2022,63-75. doi: 10.3141/2022-08
Pedraza Martinez, A. J., Stapleton, 0., & Van Wassenhove, L. N. (2009). Field Vehicle
Fleet Management in HumanitarianOperations:A Case-BasedApproach.
Fontainebleau, France.
Robert Wood Johnson Foundation. (2013). National Health Security Preparedness Index.
Retrieved November 18, 2014, from http://www.nhspi.org/the-index/
Seelbach, L. (content), & Seelbach, N. (graphic design and illustration). (2015).
Framework for assessing logistics and other preparedness decisions.
State of Florida Division of Emergency Management. (2006). County Logistics Planning:
StandardOperating Guideline. Retrieved from
http://floridadisaster.org/Response/Logistics/Index.htm
116
State of Florida Division of Emergency Management Unified Logistics Section. (2013).
State of Florida Unified Logistics Plan.
State of Florida Office of Economic & Demographic Research. (2015). Population and
Demographic Data. Retrieved April 29, 2015, from
http://edr.state.fl.us/Content/population-demographics/data/
Stuart, I., McCutcheon, D., Handfield, R., McLachlin, R., & Samson, D. (2002).
Effective case research in operations management: A process perspective. Journalof
OperationsManagement, 20, 419-433. doi: 10.10 16/SO272-6963(02)00022-0
The Florida Legislature. Emergency management powers; Division of Emergency
Management, Pub. L. No. 252.35 (2014). United States. Retrieved from
http://www.leg.state.fl.us/Statutes/index.cfm?App-mode=DisplayStatute&Search_
String=&URL=0200-0299/0252/Sections/0252.35.html
The Sphere Project. (2011). The Sphere Project:HumanitarianCharterand Minimum
Standardsin HumanitarianResponse (2011 Editi.). Rugby, United Kingdom:
Practical Action Publishing.
Thomas, A., & Mizushima, M. (2005). Logistics training: necessity or luxury? Forced
MigrationReview, 22,60-61.
Tomasini, R. M., & Van Wassenhove, L. N. (2009). From preparedness to partnerships:
case study research on humanitarian logistics. InternationalTransactionsin
OperationalResearch, 16(5), 549-559. doi: 10.111 1/j.1475-3995.2009.00697.x
U.S. Department of Homeland Security. (2014). Response FederalInteragency
OperationalPlan.
Ukkusuri, S. V., & Yushimito, W. F. (2008). Location Routing Approach for the
Humanitarian Prepositioning Problem. TransportationResearch Record: Journalof
the TransportationResearchBoard, 2089, 18-25. doi: 10.3141/2089-03
United States Department of Homeland Security. (2013). National Response Framework.
United States Geological Survey. (2001). Elevations and Distances in the United States.
Retrieved April 16, 2015, from
http://egsc.usgs.gov/isb//pubs/booklets/elvadist/elvadist.html
Van Wassenhove, L. N. (2005). Humanitarian aid logistics: supply chain management in
high geart. Journalof the OperationalResearch Society, 57(5),475-489.
doi: 10.1057/palgrave.jors.2602125
Van Wassenhove, L. N., & Pedraza Martinez, A. J. (2012). Using OR to adapt supply
chain management best practices to humanitarian logistics. International
117
Transactions in OperationalResearch, 19,307-322. doi:10.1111/j.14753995.2011.00792.x
Waller, M., Johnson, M. E., & Davis, T. (1999). Vendor-managed inventory in the retail
supply chain. Journalof Business Logistics, 183-204. doi:Article
Webster, J., & Watson, R. (2002). Analyzing the Past To Prepare for the Future: Writing
a Literature Review. MIS Quarterly,26(2), 13-23.
Yin, R. K. (2009). Case Study Research Design and Methods. (L. Bickman & D. J. Rog,
Eds.) (Fourth.). Los Angeles: Sage, Inc.
118