The role of centralization in the efficiency of national humanitarian logistics systems: a warehouse location model-based approach

The increasing frequency of disasters and complex humanitarian crises – such as those resulting in massive displacements and casualties (United Nations Office for Disaster Risk Reduction [UNDRR], 2020; ICB, 2019) – require agile and adaptive relief operations (Leiras et al., 2014). A total of 393 natural hazard-related events were recorded in 2024 alone, having affected approximately 167.2 million people and caused over US$ 241.95bn in direct economic losses (CRED, 2025; UNDRR, 2025). However, international aid systems face a growing gap between escalating needs and constrained donor funding (Humanitaires, 2024; Parliament, 2024), necessitating a reevaluation of existing operational strategies to ensure aid reaches affected populations promptly.

Since logistics functions typically represent over 80% of total humanitarian expenditures (DHL Group, 2024; Gupta et al., 2016; Kovacs and Spens, 2012), achieving efficiency through resource optimization is essential for organizations managing the complex challenges of aid delivery (Zarei et al., 2019). This financial pressure underscores the urgency of exploring logistics models that offer greater cost-effectiveness and improved service levels in highly uncertain and dynamic environments (Zarei et al., 2019; Van Wassenhove, 2006).

Historically, humanitarian logistics have been characterized by decentralized and often noncoordinated efforts, with multiple organizations operating independently in the same region. While this approach offers flexibility and rapid deployment, it can lead to redundancy, inefficiencies and suboptimal resource utilization. Many researchers have advocated for a more centralized model to address these issues (Dohale et al., 2024; Negi, 2022; Kabra et al., 2015), arguing that such an approach could enhance overall effectiveness and reduce costs. However, there is a noticeable gap in the literature regarding the quantification of these potential benefits, limiting the ability of stakeholders to make evidence-based decisions on adopting centralized frameworks.

One of the critical dimensions of moving from a decentralized to a centralized coordination model concerns the strategic location of warehouses, which function as consolidation points for aid distribution. In many humanitarian operations, agencies continue to rely on decentralized distribution networks in which each organization independently manages its own warehouses and distribution routes. Although this arrangement preserves operational autonomy, it frequently results in redundant infrastructure, underused storage capacity and excessive costs – the very inefficiencies highlighted in the broader debate on decentralization. By contrast, a model with centralized coordination, where agencies coordinate efforts and share distribution facilities, offers the potential to reduce duplication, optimize resource allocation and improve overall service levels. Here, centralized coordination refers to an interagency coordination mechanism in which logistics decisions are coordinated centrally, allowing humanitarian organizations to share warehousing resources and jointly allocate storage capacity across the network. While coordination is centralized, the resulting material flows are not predefined; rather, they emerge from the optimization process, which determines the most efficient allocation and movement of supplies throughout the network.

Beyond the structural challenges of centralized coordination, the dynamic nature of humanitarian operations introduces temporal variability as a critical factor. Due to the inherent unpredictability of demand, costs and lead times (Galindo and Batta, 2013), static network plans often become obsolete as a crisis evolves, particularly in conflict zones where territorial shifts and population displacements occur. Consequently, a multiperiod approach is more appropriate for capturing the progression of a disaster (Gelareh et al., 2015), as it allows for adaptive managerial decisions regarding network configuration and demand allocation over a planning horizon characterized by fluctuating requirements and varying involvement of relief agencies.

In this study, we investigate the impact of centralized coordination on in-country (national-level) humanitarian logistics systems, focusing on its economic and operational benefits. Using a multiperiod multiperiod mixed-integer linear programming (MILP) model, we compare two network configurations: (i) a three-echelon network consisting of central distribution centers (CDCs), location distribution centers (LDCs) and shelters, where each agency operates its own decentralized network; and (ii) a three-echelon network based on a centralized coordination strategy, featuring CDCs with higher capacity, LDCs and shelter clusters, shared among multiple agencies. The model also identifies which interagency coordination strategies offer the most operational advantages.

We focus on designing a logistics network for positioning humanitarian warehouses using a centralized strategy between humanitarian agencies across different stages of preparedness and response during a war emergency. The instability inherent in such events often prolongs their duration, leading to extended suffering for affected populations. As a result, identifying safe and economically viable warehouse locations presents a significant challenge for most humanitarian organizations (Yang et al., 2023).

The Mosul Offensive (Iraq, 2016–2017) serves as a case study to analyze how a centralized strategy can optimize warehouse allocation, minimize costs and enhance supply chain performance in a complex emergency setting. The Mosul case is not only relevant to the Iraqi conflict but also representative of broader challenges observed in humanitarian logistics across diverse disaster contexts. Similar issues and inefficiencies have been reported in sudden-onset natural disasters, such as the 2010 Haiti earthquake (Apte, 2009), as well as in protracted crises, including the Syrian conflict and refugee crises in Sub-Saharan Africa (Kian et al., 2022; USAID Office of Inspector General, 2017; Vledder et al., 2019). In each of these cases, the strategic location and coordination of warehouses played a decisive role in shaping both the timeliness and cost-effectiveness of aid delivery. By examining the Mosul Offensive, this study therefore provides insights that are generalizable to other in-country humanitarian logistics systems, where centralized coordination remains critical to overcoming redundancy and resource fragmentation.

Given the challenges of resource fragmentation and the dynamic nature of conflict-driven crises, this study is guided by the following research question: To what extent can a centralized coordination framework improve the economic and operational efficiency of a national humanitarian logistics system compared to traditional decentralized strategy?

Based on this question, we test the following hypotheses:

Consequently, the main objective of this study is to develop and apply a MILP model to quantify the economic and operational impact of centralized coordination within a national humanitarian logistics system, using the 2016–2017 Mosul Offensive as a focal case study.

The remainder of this paper is structured as follows: Section 2 provides a literature review on facility location problems in humanitarian logistics. Section 3 presents the proposed model. Section 4 applies the model to a real-world case study, followed by a sensitivity analysis in Section 5. Section 6 presents key insights and the main contributions and limitations of the study. Finally, Section 7 presents the authors’ conclusions.

The effectiveness of humanitarian logistics relies heavily on efficient coordination among multiple actors, including nongovernmental organizations (NGOs), government agencies, international organizations and local responders. Coordination is a core mechanism through which information sharing, alignment of activities and avoidance of duplication are achieved in humanitarian operations. Yet, the lack of coordination in humanitarian supply chains has been widely recognized as a major challenge that leads to inefficiencies, delays and wasted resources (Ruesch et al., 2022). Humanitarian logistics operations often involve decentralized decision-making, where each organization operates independently, prioritizing its own objectives rather than collective efficiency. In disaster scenarios, operations tend to be decentralized due to the urgency and complexity of the context; however, the absence of effective centralized coordination among institutions significantly reduces the effectiveness of disaster risk mitigation activities (Quader et al., 2023; Tolessa et al., 2024; Kovacs and Spens, 2012).

The decision between centralized, decentralized or hybrid coordination mechanisms – such as the United Nation’s (UN) cluster approach – is a critical strategic factor that dictates the efficiency and effectiveness of disaster response. Centralized systems, often exemplified by national agencies coordinating multiple responders, aim to reduce redundancies and consolidate resource allocation, though they may face challenges related to communication media bottlenecks (Aros and Gibbons, 2018). In contrast, decentralized schemes can offer greater flexibility and higher operational efficiency in highly uncertain environments (Besiou et al., 2014; López-Vargas et al., 2025).

Recent studies using agent-based modeling have shown that the cluster approach acts as a vital lateral coordination structure. When cluster leads function as information hubs, they significantly accelerate the diffusion of critical data, which is essential for timely humanitarian action (Altay and Pal, 2014). Furthermore, the effectiveness of these mechanisms is closely tied to the design of the first-response teams and the integration of coordination technologies. For instance, the use of peer-to-peer systems and multiagent coordination simulations can help evaluate team configurations to minimize total operation time and overcome the limitations of overloaded communication channels (Hashemipour et al., 2018). By analyzing these mechanisms, it becomes clear that while decentralization favors efficiency, centralized and cluster-type scenarios often demonstrate superior effectiveness in achieving comprehensive requirement coverage (López-Vargas et al., 2025).

One of the primary reasons for the lack of centralized coordination is the diverse and sometimes conflicting agendas of humanitarian actors. NGOs, international agencies and local governments often have different mandates, funding sources and operational priorities, making coordination complex (Tomasini and Van Wassenhove, 2009). For example, competition for donor funding can lead organizations to prioritize visibility over efficiency, discouraging joint efforts that could optimize logistics operations. Moreover, political and cultural differences between actors further hinder information sharing and cooperative decision-making, resulting in suboptimal supply chain performance (Holguín-Veras et al., 2012). Evidence from national-level public-sector supply chains supports the merits of centralizing decision-making and reducing distribution tiers. In a large-scale randomized trial covering 439 health facilities across 24 districts in Zambia, Vledder et al. (2019) found that a two-tier “cross-dock” model – in which health facilities ordered directly from the central agency via a cross-dock rather than through a traditional three-level system – significantly reduced both the frequency and duration of stock-outs of essential medicines. The authors attribute this outcome to improved information flow, clearer accountability and centralized control of key functions (allocation of scarce stock and order-adjustment) at the level of competency in the system.

This finding validates the underlying logic of our framework’s centralized operations structure, where a single coordinating body manages critical logistics assets and information flows to achieve efficiency and coherence in supply networks.

Another major factor contributing to the lack of centralized coordination is the absence of standardized logistical systems and data-sharing mechanisms. Unlike commercial supply chains, which often rely on integrated enterprise resource planning systems, humanitarian organizations frequently use different platforms for tracking inventory, transportation and distribution (Scholten et al., 2014). This inconsistency makes it difficult to achieve real-time visibility of supply chain operations, leading to delays in identifying critical shortages or redundancies in aid distribution. In addition, the lack of interoperability between different logistical systems exacerbates inefficiencies, as agencies struggle to synchronize their operations (Bealt et al., 2016).

The unpredictability of humanitarian crises also makes centralized coordination more challenging. Disasters and conflicts create highly dynamic environments where demand for aid fluctuates rapidly, requiring agile and adaptive supply chain responses (Galindo and Batta, 2013). However, without a centralized mechanism, individual organizations often react independently, leading to competition for transportation assets, storage facilities and local resources. This issue was evident during the Haiti earthquake response in 2010, where multiple relief organizations deployed overlapping supply chains, creating congestion at ports and distribution centers (Apte, 2009).

Despite these challenges, various studies highlight that improved centralized coordination can significantly enhance the efficiency of humanitarian logistics. Cooperative frameworks such as the cluster approach, introduced by the united nations office for the coordination of humanitarian affairs, have demonstrated potential in improving information sharing and joint planning among agencies (Torre et al., 2012). However, these frameworks are often limited by bureaucratic inefficiencies and inconsistent participation by key stakeholders. Without strong enforcement mechanisms, many organizations continue to prioritize their own operations over collective coordination (Balcik et al., 2010).

While structural decentralization remains a major organizational challenge in humanitarian logistics, these coordination issues are compounded by logistical complexities related to the strategic placement of facilities and resources. In particular, the warehouse location problem plays a central role in determining the efficiency and responsiveness of humanitarian operations, especially in contexts affected by conflict or recurring crises. Efficient facility placement can mitigate some of the inefficiencies caused by fragmented operations, yet most models still assume idealized or centralized coordination environments. To understand the current state of research in this area, it is necessary to examine how facility location problems – especially those involving multiperiod planning – have been addressed in the humanitarian logistics literature.

Facility location problems in humanitarian logistics include the identification of sites such as fire stations, emergency shelters, distribution centers, warehouses, debris removal sites and medical centers. Most optimization models for humanitarian logistics have been combined with other logistical challenges, such as prepositioning stock, relief distribution, victim transportation, evacuation planning, resource allocation, goods flow and other operations.

We conducted a literature review on facility location problems in humanitarian logistics, identifying 123 relevant articles. These were analyzed based on disaster type, onset, disaster lifecycle stage and whether they used multiperiod or multi-objective approaches. Of the articles, 7 addressed man-made disasters, 70 focused on natural disasters and 46 did not specify a particular disaster.

Among studies applied to natural disasters, the following categories were covered: earthquakes (34), hurricanes (16), famine (3), floods (4), landslides (3) and epidemics (2). Articles addressing human-made causes primarily dealt with terrorism, with only one study covering armed conflicts.

The analysis of disaster onset revealed 66 articles on sudden-onset disasters, 49 unspecified and 8 on slow-onset disasters. In terms of the disaster stage, 41 studies focused on the preparation and response stages, 38 on response, 31 on preparation and 13 did not define the stage.

There is a clear predominance of models that address single-period horizons, with 93 studies identified, compared to 25 articles that addressed location problems using multiple periods. Regarding the number of objectives, 50 articles considered multiple objectives, while 72 addressed a single objective. Among the multi-objective studies, 49 presented economic objectives and only 1 combined economic and environmental goals. Of the single-objective studies, 35 focused on economic analysis, 29 on service level and 8 on other interests. From the 123 articles, 19 were filtered for analysis based on aspects more specific to this paper (Tables 1 and 2).

The literature review was an important component of this study, as it allowed for an examination of the contexts in which location models are used within humanitarian logistics. This examination made it possible to identify a gap in the existing research: a lack of studies that quantitatively compare centralized and decentralized network location models.

Furthermore, the review facilitated a characterization of 123 relevant articles based on disaster type, lifecycle stage and whether they used multiperiod or multi-objective approaches. The analysis showed a predominance of single-period models over multiperiod ones. This demonstrates that the current study’s approach, which focuses on a quantitative and multiperiod comparison, addresses and helps to fill this identified void in the literature.

The outbreak of a disaster, whether natural or anthropogenic (such as in the case of war emergencies), can lead to many evacuees. Providing shelter and resources is crucial in the post-disaster phase, so the supply chain network supporting humanitarian operations must ensure that agencies can deliver aid to those in need. This paper considers a three-echelon network to structure the postdisaster response. In this study, the following terminology is adopted:

• Main hubs (CDCs): Primary national-level warehouses located in strategic cities (Dohuk, Erbil and Baghdad) that hold long-term prepositioned inventory.

• LDCs: Intermediate nodes whose locations are optimized by the model to serve refugee camps.

• Mobile storage units (MSUs): Temporary and flexible canvas structures (e.g. 10 × 20 m or 10 × 32 m) that can be quickly deployed as LDCs.

• Hard-roof facilities: Permanent masonry warehouses that offer higher storage capacity and lower variable costs but less flexibility for expansion or relocation.

In humanitarian operations, there is often a lack of coordination between the agencies involved, leading to missed opportunities for reducing logistical costs. Figure 1 illustrates a typical network structure, where CDCs are positioned to access major disaster-prone regions, mitigating risks and holding prepositioned supplies. These central DCs are frequently managed by relief organizations that have the resources to maintain long-term inventories (Li et al., 2018).

Some other distribution centers are set up temporarily to support logistics during humanitarian operations, such as LDCs. Essential in humanitarian contexts, these distribution centers serve as consolidation points and are often used to prepare supply kits for beneficiaries, acting as the base for last-mile assistance. However, finding secure and affordable distribution facilities is a common challenge for most relief organizations due to resource constraints (Kara and Rancourt, 2020). Figure 1 illustrates the first strategy used in designing the humanitarian network, known as the decentralized strategy. In this approach, there is no sharing of resources at the local distribution level. Each agency is responsible for a specific shelter and operates its own local warehouse, typically using mobile storage units (MSUs).

The second strategy for designing the humanitarian network is based on centralized coordination between agencies. Known as the centralized coordination strategy (Figure 2), it involves the creation of shelter clusters (demand locations) and the use of both Mobile Storage Units (MSUs) and hard-roof facilities. Refugee camps are assigned to clusters based on proximity or through clustering methods, for example. Each cluster can receive supplies from its MSU or from one of the designated hard-roof facilities, however, the resulting material flows and warehouse utilization patterns are determined by the optimization model, which identifies the most efficient configuration for the logistics network.

In the proposed centralized operations framework, overall coordination and control are exercised by the organization fulfilling the role of the logistics cluster lead, represented by the World Food Programme (WFP) (Altay and Labonte, 2011; WFP, 2025). As the lead agency of the logistics cluster, the WFP is responsible for coordinating logistics preparedness and response activities during large-scale emergencies. This includes managing and allocating transportation assets, warehousing and distribution operations among humanitarian actors. The WFP works in close cooperation with national disaster management authorities, united nations agencies, nongovernmental organizations and other relevant clusters to promote coherence, prevent duplication of efforts and optimize the use of available resources. This governance structure aligns the proposed framework with the coordination mechanisms established under the United Nations cluster system.

The optimization model evaluates two distinct logistics strategies based on the degree of coordination and integration among humanitarian actors. These strategies are operationalized through different material flow constraints and decision-making schemes.

Decentralized coordination and independent flows: In the decentralized scenario, the network reflects a fragmented operational environment where each humanitarian organization or agency manages its own supply chain independently. The decision-making process is localized, and material flows are strictly compartmentalized. Specifically, the model assumes that each warehouse is tied to a specific set of demand points (refugee camps) previously assigned to a particular agency. In this “siloed” architecture (Figure 1), there is no sharing of resources or infrastructure; thus, an agency cannot fulfil the demand of a camp that is outside its institutional mandate, even if it has surplus capacity.

Centralized coordination and open-flow dynamics: In contrast, the centralized scenario aims for the global optimization of the logistics network. The mathematical formulation removes the fixed-assignment constraints between warehouses and camps, enabling an open-flow logic. In this scheme, any warehouse can supply any demand point within the system (Figure 2), provided it represents the global minimum cost. This flexibility allows for “resource sharing” and “demand sharing,” where the central coordinating body can reallocate inventory across the entire network to mitigate local shortages and optimize total operational costs.

The cluster approach as a strategic mechanism: In the proposed modeling framework, the clustering process specifically refers to the aggregation of demand. Instead of treating each demand as isolated data points, the model clusters these individual demands into a unified sectoral requirement. This demand-side clustering enables the optimization of a shared logistical backbone, allowing the central coordinating body to serve the collective needs of multiple agencies through a consolidated network of hubs, thereby eliminating the redundancies inherent in independent, agency-specific demand fulfilment.

In practice, the centralization is reflected in the structure of humanitarian supply networks, which typically revolve around a CDC that stores prepositioned supplies and distributes them to regions affected by disasters. Frequently, these CDCs are managed by relief organizations that can afford to maintain long-term inventories (Li et al., 2018). LDCs may be installed temporarily to support logistics during the humanitarian operation. These warehouses play an important role in serving as consolidation points. They can also be used to prepare the supplies kits distributed to the beneficiaries, which are used as the bases for last-mile assistance.

Two models were developed in this study. The first model establishes a baseline for the humanitarian operation under the decentralized strategy. The second model explores the possibility of adopting LDCs as clusters of MSUs or hard-roof facilities shared between humanitarian agencies, following the centralized strategy. Both models assume that the CDCs are already in place. However, decisions regarding the assignment of CDCs to LDCs, the location of LDCs, the allocation of demand locations to LDCs and the type of facility to be built are determined by the model.

In the centralized network model, the objective function (1) minimizes the total cost of distributing the required items to meet demand throughout the entire operational horizon. This includes fixed and variable costs associated with operating the facilities, transportation costs, fixed costs for establishing the facilities and decommissioning costs.

Sets

I: Set of CDCs, indexed by i.

J: Set of candidate locations for LDCs, indexed by j.

K: Set of Demand Locations, indexed by k

T: Set of operation periods, indexed by t.

R: Set of final capacity levels at period $t=1$$R=0,1,…,11$⁠, where R is the set of capacities.

S: Set of initial capacity levels at period t$T=0,1,…,11$⁠, where S is the set of capacities.

Model Parameters

$asj$⁠: Binary parameter equal to 1 if candidate location j can have capacity level s.

q: Average volume of kits.

$dkt$⁠: Families at demand point k in period t.

$fij$⁠: Freight cost from CDCi to candidate LDCj.

$fjk$⁠: Freight cost from LDCj to demand location k.

$djk$⁠: Distance between LDCj and demand location k.

$dmax$⁠: Maximum allowed distance.

$csf$⁠: Fixed operation cost of capacity level s.

$csv$⁠: Variable operation cost of capacity level s.

$crsa$⁠: Opening or expansion cost from level r to s.

$crsd$⁠: Decommissioning cost from level r to s.

$caprs$⁠: Capacity available from level r to s.

M: Large number.

Decision variables

$Xijt$⁠: Flow from CDCi to LDCj in period t.

$Xjkt$⁠: Flow from LDCj to demand location k in period t.

$Orsjt$⁠: 1 if LDCj increases capacity from r to s at time t.

$Wrsjt$⁠: 1 if LDCj keeps capacity level s at time t.

$Zrsjt$⁠: 1 if LDCj decreases capacity from r to s at time t.

$Yjkt$⁠: 1 if demand location k is assigned to LDCj in period t.

Model.

Constraints:

The Constraints 2 represent the flow balance between the income and outcome of each period. The group of Constraints 3 guarantee that all the demand of each period needs to be met, and the Constraints 4 state that the capacity installed at one period needs to be suitable to the quantity of products distributed to the demand points in that period. Constraints 5–9 establish the temporal continuity, enabling the model to change the capacities installed over the periods. equations (5) and (6) capture the status of the network, whether an increase or decrease of capacity, as equations (7) and (8) create the temporal connection between periods. The coverage constraints 10 are also included and linking constraints 11 are created. Constraints 12 impose that each demand point is supplied by only one LDC. Finally, the Constraints 13 and 14 define the boundary conditions of the model.

In the centralized model, the demand has the possibility to be supplied by one MSU cluster (set before the optimization) or by one of the hard roofs, in contrast to the decentralized model, in which each demand point had its own LDC and could only be provided by it at each period. The parameter $ajk$ is in charge to delimit these possibilities of assignment, and the type of facility associated to each candidate location is defined by the $asj$⁠, where each capacity level belongs to one of the two types allowed.

The decentralized model, which makes the baseline, is quite similar to the previous model. It differs by removing the costs associated with the transportation between LDCs and demand points from the objective function. Furthermore, the coverage constraints (10, 11 and 12) are inherently satisfied in the decentralized baseline, as the distance between demand points and their respective LDCs is assumed to be zero.

As previously mentioned, this work aims to encourage important reflections on humanitarian operations by assessing which configuration strategy leads to greater efficiency and better resource utilization in a humanitarian context. The study is based on real data from the humanitarian response operation during the Iraq Civil War. The region, characterized by a history of armed conflicts, wars and ethnic disputes, provides an ideal setting to examine how centralized storage and resource-sharing strategies can achieve economies of scale without compromising demand coverage. To validate the hypotheses, the focus is primarily on the Battle of Mosul, which began in 2016.

To illustrate the national-level humanitarian supply chain analyzed in this study, Figure 3 provides a representation of the system, where the first level corresponds to warehouses (hubs) responsible for storage and centralized activities at the national scale, while the second level consists of the shelter clusters shared among the agencies.

The humanitarian supply chain analyzed in this study is structured as a three-level, national-level in-country system. The first level corresponds to warehouses (hubs) responsible for storage and coordination activities at the national scale. The second level consists of candidate facility locations, including potential camps and cities where intermediate facilities may be established. The third level represents the demand points, corresponding to the camps and affected populations identified in the empirical context. By explicitly defining these three interconnected levels, Figure 2 provides a clear representation of the system under study and serves as a reference for the subsequent discussion on centralized coordination, decision-making and facility location within a national humanitarian logistics network.

The numerical example is based on data from the humanitarian supply chain established in the country. Of the 65 planned camps for internally displaced people (IDP), only the 27 camps that were actually occupied were selected as demand locations. We considered 33 candidate facility nodes for LDCs in the centralized case (in the decentralized case, the demand locations themselves are the candidate locations) and 3 CDCs nodes. Candidate sites for hard roof facilities were selected using the following criteria: (i) the distance between the candidate city and Mosul should be up to 230 km and (ii) the population of the site should be over 40,000 inhabitants.

To determine the candidate sites for the location of MSUs (Mobile Storage Units) in the clusters, a p-median model was used, considering all the facilities planned by the operation. In addition, distances between each pair of locations were calculated using Google Maps. The demand data were extracted from operation reports obtained from the ReliefWeb platform and reports published by Camp Coordination and Camp Management (CCC-MIRAQ). From ReliefWeb, 49 reports of displaced people’s camps related to the Battle of Mosul were found. These reports cover the period from October 2016 to July 2017 and contain data on the planning and occupation of these camps, which were manually consolidated and tabulated. CCC-MIRAQ’s reports, on the other hand, had a different storage structure, with more detailed data on the profile of each camp occupied in 2017. A total of 2,124 files were analyzed using Python libraries to extract relevant information.

The CCC-MIRAQ reports contained a larger number of support facilities for the IDPs located in various regions, suggesting that some were not specifically linked to those displaced by the Battle of Mosul. The two databases were compared, and the installations in the CCC-MIRAQ data that aligned with the response operation for the Battle of Mosul were filtered.

The humanitarian relief chain discussed in this study includes food, water, hygiene kits and other supplies, in line with standards established by the sphere project (2011). An estimated average weight of 109.43 kg per family of six members was used. Furthermore, five sizes of hard roof facilities and six sizes of MSUs were considered, with the assumption that only the MSU warehouses would be expandable within the planning horizon. The costs associated with installations, including opening, expansion, closure and operational fixed and variable costs, were estimated using data provided by humanitarian agents involved in the response operation. Freight costs were estimated using data from developing countries.

The scope of this work is explicitly limited to the analysis of an in-country (national-level) humanitarian logistics system, focusing on the consolidation and distribution of relief products within the national territory of Iraq. These hubs, used as transshipment points for items either donated or acquired for distribution, were situated in three cities: Dohuk, Erbil and Baghdad (Logistic Cluster, 2017). From these hubs, the products were transported by trucks to mobile storage units (MSUs) located near the camps. The most common MSU sizes were 10 × 20 meters, storing between 350 and 500 metric tons and 10 × 32 meters, with a capacity of 500–750 metric tons (Catholic Relief Services [CRS], 2019). Typically, each organization had its own LDC facilities for its operations. The analysis specifically focuses on how the centralized strategy optimizes the mix between temporary MSUs and permanent hard-roof facilities at the LDC level, relative to the supply provided by the three main hubs.

The logistics network established for the operation consisted of multiple independent in-country networks, each with different warehouses and distribution routes. These were often redundant, distributing similar items through separate channels. This decentralized network configuration led to large inventories at the camps, as supplies were stored on-site, resulting in an inefficient use of the operation’s budget. For instance, fixed costs, such as those for management and facility maintenance, could have been reduced by sharing facilities, since many shelter camps were located close to one another.

The model proposed in this study aims to centralize logistics networks at the national level, encouraging the sharing of resources and distribution structures for improved efficiency. In this approach, the final stage of the distribution will be carried out directly from the vehicles, eliminating the need for on-site stockpiling and seeking to reduce costs.

The problem presented in this paper involves determining both the location of LDCs and their monthly assignment throughout the planning horizon. The study instances were solved using Gurobi v9.1 on a computer with 64 GB of RAM, an Intel(R) Core(TM) i9-9900KF processor (eight cores, 16 threads, clocked at 3.6–5.0 GHz) and the Windows 10 operating system. For the mesh sizes of the numerical examples, optimal solutions were obtained within 100 s of processing.

The logistics cost analysis, presented in Table 3, reveals savings of 2.4 million dollars by adopting the centralized coordination strategy. Storage-related savings exceeded transport costs, particularly due to the optimized selection of hard-roof facilities over multiple MSU clusters, which reduced fixed installation and management costs, with the most significant gain found in the fixed installation cost, amounting to around 3 million dollars. Reductions in costs related to opening, expansion, closing and decommissioning also demonstrate decreased requirements for material and human resources to adapt the logistics network over time, simplifying management. However, it also requires increased interaction and coordination between agencies to make the proposed network configuration viable.

Table 4 highlights potential savings when adopting the new joint storage strategy. To verify this, cost apportionment was examined under three strategies: simple, proportional and mixed. The simple apportionment strategy involved uniformly distributing installation and operating costs of the warehouses among the assigned camps. In this approach, the costs of moving goods between facilities were also equally divided among the camps supplied by each warehouse. This method yielded savings for four out of the six management entities observed.

The proportional apportionment strategy followed an absorption costing approach, where installation, operation, transportation, fixed and variable costs were apportioned based on the demand from each camp. This meant that fixed storage costs, as well as opening and deactivation costs, were allocated proportionally to the volume of goods required by each camp. Finally, in the mixed strategy, fixed operating costs, transportation costs and variable operating costs were apportioned according to volume, while fixed warehouse opening and decommissioning costs were equally divided among the camps served by a given warehouse. Across both approaches, the new configuration proved economically beneficial for the three agencies involved – International Organization for Migration (IOM), United Nations Development Programme (UNDP) and United Nations High Commissioner for Refugees (UNHCR) – as well as the Iraq government (Ministry of Migration and Displaced - MoMD and N.Gov).

The results for the centralized and decentralized strategies are compared based on real demand, i.e. the actual occupation levels measured in the camps. In the decentralized strategy (baseline), each camp maintains its own warehouse, and warehouse opening decisions, as shown in Table 5, follow the fluctuations in demand at the camps. In the baseline scenario, the number of open warehouses is significantly higher, with up to 21 facilities operating at the peak during the analyzed period.

In contrast, the solution provided by the centralized strategy, shown in Table 6, reveals a more streamlined supply chain, reducing the management effort required due to the smaller number of facilities. In this scenario, the peak month for operational installations is November, with 11 facilities running, when two newly planned camps (Surdesh-MODM and Basateen Al Sheuokh-UNDP) became occupied.

The reduction in the number of required facilities is evident when comparing the total number of operational installations in both strategies. The centralized network solution revealed a reduction of 119 installations over the entire period, meaning that, on average, 7.9 fewer installations were used each month. This reduction results in a decrease of 10,000 tons in the system’s installed capacity.

To evaluate the performance of the multiperiod model, an experiment was conducted using the camp occupations observed over a 15-month horizon. A comparative analysis was made between the solution of the multiperiod model and a static model, both using the centralized strategy. In the static model, the average demand over the horizon was used and the decisions remained unchanged throughout the analyzed period.

The selected locations in both models were the same, except for Chamakor-IOM and Laylan. In the static model, Chamakor-IOM was not selected as a candidate, so the demand for that cluster’s camps was assigned to the Al-Hamadaniya warehouse, which also supplied the Hasansham U2-UNHCR, As Salamyiah-2: N.GovUNHCR camps. In the multiperiod model, the Chamakor-IOM warehouse opened in January 2017, maintained its capacity until April 2017, and then its demand was shifted to Al-Hamadaniya. The Laylan-UNHCR location, in contrast, only operated in the first period of the dynamic model, as it was assumed that warehouses active in this period were already operational. After October 2016, the demand from the Laylan-UNHCR cluster was assigned to the Kirkuk hardroof, as in the static model.

An efficiency analysis was also conducted to compare the two models. The installed capacities were evaluated over the 15-month horizon, and the system’s performance in meeting demand was analyzed. Figure 4 shows the relationship between the system’s aggregate installed capacity and demand. Although the total installed capacity in the static model is theoretically sufficient to meet demand, capacity utilization in the multiperiod model was more balanced, ranging from 9% to 68%, whereas in the static model it fluctuated from 1% to 77%. Both models averaged a 53% utilization rate, but the standard deviations were 0.13 and 0.22, respectively, indicating a greater variability in the static model.

Despite having sufficient capacity, the static model exhibited inefficiencies due to the assignments made, as it assumes that demand must be met within a 40 km radius. Comparing Tables 7 and 8 reveals significant idle capacity in the static model, with facilities remaining unused in every period analyzed. On average, three facilities were idle per period. Conversely, the static model also experienced a lack of installed capacity in four warehouses, over 13 of the 15 months of operation, with an average of 1.8 facilities facing capacity shortages each month. The static model maintained 12 warehouses monthly, with a total of 180 operational points over the 15 months, while the dynamic model required only 137 points, with a maximum of 11 warehouses operating in any given month.

In the static case, despite the system having aggregate capacity sufficient to meet demand, performance was hindered by coverage constraints and demand assignments. The inability to redistribute excess capacity from one installation to another with shortages led to unfeasible outcomes, negatively impacting service levels and system security. Consequently, a direct comparison of costs between the two models was not possible, as the cost of unmet demand or noncompliance was not modeled.

This study aimed to analyze the impact of a multiperiod approach in the planning of humanitarian facilities and coordination strategies within humanitarian actions by proposing a new strategy for configuring the logistics network in the case studied. The results indicate significant gains from improved coordination between humanitarian agencies, notably through a reduction in the number of facilities required. This led to greater centralization of the operation, allowing for more efficient resource utilization and economies of scale.

By comparing the static approach to the multiperiod model, the latter demonstrated superior performance in both facility location and demand allocation when temporal and spatial variations in supply points occurred. The static model lacked robustness, rendering its solution impractical as it required frequent plan adjustments and sometimes violated imposed constraints. In contrast, the multiperiod model was more effective in addressing the system’s needs, accommodating variations in demand over time and space while optimizing the use of available resources.

Nevertheless, further investigation into the model using additional demand scenarios would be beneficial. Moreover, extending the model through stochastic optimization or incorporating tactical or operational-level decisions, such as inventory management and the evaluation of potential supply shortages, could enhance its applicability. The event analyzed also revealed critical issues that could disrupt the distribution of supplies, including delays caused by checkpoints or facility breakdowns, which could sever the links between nodes in the network.

Seeking to evaluate the impact of the restriction on the coverage radius of warehouses, experiments were conducted by varying the maximum distance parameter to values between 8 and 200 km. This analysis of costs reveals that transportation costs represent a significant portion of the total cost composition, however, installation costs have a greater impact on decision-making. Through the analysis, it was observed that the total transportation cost has the lowest coefficient of variation, at 8%, highlighting its more stable behavior, while the total operating cost has a coefficient of variation of 47%. When analyzing the ranges, the impact of installation costs becomes evident, varying by millions, whereas transportation costs vary by thousands.

The total cost spent on transportation remains relatively stable, ranging from 2.35 million to 3 million dollars, with an inversely proportional relationship between its two components (costs in the first and second segments). The trade-off between transportation costs in the first and second segments balances the total transportation cost as the radius increases. For example, when comparing the solutions for an 8 km radius and a 200 km radius, there is an 11,000-kilometer reduction in the first segment and an increase of about 21,000 kilometers in the second segment, resulting in a net increase of 10,000 kilometers. Despite the 83% increase in the total distance traveled, the increase in total costs was 27%, as for every kilometer reduced in the first segment there is a reduction of $51.67, while for the second segment, each additional kilometer results in an increase of $58.20.

The graph in Figure 5 shows the total costs obtained for all coverage radius scenarios, with their components highlighted. The results reveal an increase in transportation costs for the second segment (between warehouse and field). Due to the reduction in the total number of installations observed in Figure 6, the second segment becomes more expensive as these installations tend to concentrate demand from a larger number of fields, causing an increase in the total distance traveled in this segment. Furthermore, the first segment (between hubs and warehouses) becomes less significant as the coverage radius increases, indicating that these installations tend to be located closer to the hubs.

The other costs, except for variable operating costs, also show a significant reduction related to the number of warehouses installed. The fixed operating cost has the greatest impact on the total cost and remains practically stable starting from a radius of 119 km, where the average number of installations stays below four per period. From a distance of 148 km, this average number of installations remains constant (Figure 6).

Variable costs also decrease but are not tied to the number of supply points, as they depend on the volume of inputs moved in each type of installation. Thus, the total volume handled (demand) does not vary between scenarios, and the reduction in variable cost is attributed to favorable mechanization conditions in hardroof warehouses, which become more prevalent in decisions as the coverage radius increases. When analyzing the variable operating costs, it is observed that, despite the downward trend, this component shows some peaks associated with the warehouse type decisions presented in Figure 6. This leads to a change in the categories and sizes of installed warehouses, causing fluctuations in variable operating costs. Variable costs are widely dispersed but remain on the scale of thousands, similar to the opening and closing costs, playing a smaller role in the composition of the total cost.

Finally, the opening and closing costs also show a downward trend with fluctuations due to the increased use of hardroof warehouses, which offer less flexibility for expansion and reduction. In this study, it was assumed that the capacity of these candidates would not be expandable or reducible, as this would require new service negotiations and contracts. In addition, the construction of masonry warehouses was not a practice adopted in the analyzed operation due to its temporary nature. Despite their high values, the costs of opening, expansion, closing and capacity reduction are on the order of thousands, having a lesser impact on the total cost composition.

Figure 7 presents the number of installations in each period of the analyzed horizon for all coverage radius cases. A greater variation in the number of installed warehouses can be observed over the 15-month horizon for shorter distances, with a more stable pattern as the radius increases. This behavior is due to the increased use of hardroof warehouses, which limits the network’s ability to adapt to demand, i.e. the opening and closing of installations are restricted. Being less flexible for expansions, masonry warehouses remain operational throughout the planning period without alterations. Thus, the larger the coverage radius, the fewer warehouses are installed, with a predominance of hardroof installations that have higher capacity levels.

The results clearly indicate that centralized coordination can significantly improve the efficiency of national (in-country) humanitarian supply chains. Specifically, the multiperiod model led to a reduction in the number of required facilities, optimizing network design within a national system and promoting economies of scale. This centralized configuration enabled better resource utilization while maintaining operational flexibility.

The analysis confirmed the economic benefits of adopting centralized strategy configurations. The static centralized model yielded a cost reduction of approximately $294,076 (3.1%) when compared to the multiperiod decentralized strategy. More significantly, the multiperiod centralized strategy achieved the most substantial savings, reducing total costs by $2.43million (approximately 25.7%) relative to the decentralized configuration. These cost reductions stemmed primarily from more efficient use of storage infrastructure, particularly in the fixed installation and operational costs, as well as lower deactivation and expansion expenses. The multiperiod model’s ability to dynamically adapt warehouse capacity and locations over time proved essential to maintaining high service levels while minimizing idle infrastructure and underused capacity.

A comparative analysis with the static model further emphasized the robustness of the multiperiod approach, particularly under dynamic conditions. The static model, while simpler, was unable to respond effectively to temporal and spatial variations in demand and supply, often resulting in infeasible or suboptimal plans. In contrast, the multiperiod model demonstrated superior adaptability, enabling more accurate facility location and demand allocation decisions that evolved with the crisis timeline.

The robustness tests were essential for validating our model’s performance under different operational policies. By analyzing various warehouse coverage radii, the tests confirmed the critical trade-off between transportation and installation costs, revealing that facility costs are the dominant driver in design of national humanitarian logistics networks operating within a single country. This analysis not only reinforces the reliability of our findings but also provides deeper insights into how the optimal network structure adapts to service-level constraints, thereby strengthening the practical implications of our study.

These findings provide quantitative support for the adoption of centralized coordination frameworks in national-level humanitarian logistics – an argument long supported by theory (Dohale et al., 2024; Negi, 2022; Kabra et al., 2015) but rarely tested in complex, real-world scenarios. Previous studies have attempted different approaches to quantify the benefits of more centralized humanitarian logistics models (Coskun et al., 2024; Singh et al., 2018) – the current study expands this body of research by offering a dynamic, multiperiod framework and applying it to a high-complexity case, thereby reinforcing and enriching existing empirical evidence. Our results validate the operational value of interagency coordination, offering data-driven insights for future logistics strategies and policies. In practice, centralized control within national-level humanitarian supply networks is typically exercised by the logistics cluster, led by the WFP, which coordinates logistics preparedness and response activities among humanitarian actors. This arrangement ensures governance alignment with the coordination mechanisms established under the United Nations cluster system.

From a theoretical standpoint, centralized coordination strategies in humanitarian logistics are most effective when they are responsive to demand variability and structured proportionally to the needs of affected populations. Studies such as those by Balcik et al. (2010) and Van Wassenhove (2006) argue that centralized systems can achieve economies of scale and coordination efficiency, but must be flexible enough to adapt to changing conditions on the ground. A proportional allocation of centralized resources – where facilities serve areas based on their actual or forecasted demand – avoids the pitfalls of rigid or over-centralized systems that may lead to unmet needs or underused resources. Dynamic models, such as multiperiod formulations, align with recommendations from Galindo and Batta (2013), who highlight the necessity of demand-driven, adaptable logistics networks in crisis scenarios. By integrating proportionality and temporal flexibility, humanitarian operations can maximize both efficiency and responsiveness.

However, several avenues remain for further research. Testing the model under a wider range of demand scenarios, including both high-variability and low-variability conditions, would help assess its robustness more thoroughly. In addition, extending the model to include stochastic optimization could better capture real-world uncertainty and enhance decision-making under risk. Future versions of the model might also integrate inventory control mechanisms, lead time variability or supply disruptions, such as those caused by checkpoints or warehouse malfunctions.

It is important to note that the scope of this study is primarily focused on the economic and operational benefits of centralized coordination. We acknowledge that a comprehensive evaluation of humanitarian logistics should also include critical social-based performance dimensions. Therefore, a limitation of the current model is the absence of metrics such as response time, equity of relief delivery and effective coverage. Future research should aim to integrate these vital social indicators to provide a more holistic assessment of supply chain strategies, ensuring that efficiency gains do not compromise the effectiveness and fairness of aid distribution.

Building on the findings of this study, further work could also enhance the model’s practical application and replicability. For instance, the development of graphic objects, such as maps showing the spatial and geographic distribution of the proposed supply network, would offer a clearer visualization for decision-makers. Moreover, complementing the current analysis by strengthening the robustness tests with alternative scenarios – representing different scales of the emergency, varying levels of demand, or challenges in accessibility – would greatly aid in adapting the model to other complex humanitarian contexts.

All findings should be interpreted within the context of national (in-country) humanitarian logistics systems, as cross-border or global supply chains may face additional institutional and operational constraints not captured in this model.

Moreover, the growing scarcity of humanitarian funding – exacerbated by donor fatigue and shifting geopolitical priorities (Humanitaires, 2024; Parliament, 2024) makes the adoption of cost-effective and scalable logistics strategies more urgent than ever. In this context, our findings are especially timely. The demonstrated benefits of centralized coordination underscore that such approaches are not only theoretically attractive but increasingly necessary for operational sustainability.

Recent advances in technology offer promising solutions to address coordination challenges in humanitarian logistics. Blockchain, artificial intelligence, simulations and cloud-based supply chain platforms have the potential to enhance transparency and streamline coordination efforts (Dubey et al., 2020; de Brito et al., 2021). For instance, blockchain technology can facilitate secure and real-time data sharing among humanitarian organizations, reducing information asymmetry and improving decision-making. However, widespread adoption of these technologies remains limited due to financial constraints, lack of technical expertise and resistance to change within humanitarian organizations.

While the lack of coordination remains a persistent challenge in humanitarian logistics, research suggests that centralized coordination, standardization and technological integration can significantly improve efficiency. Future studies should explore the implementation of centralized models that balance agency autonomy with collective optimization. In addition, policymakers and humanitarian organizations must work toward developing robust governance structures that enforce coordination and incentivize information sharing. Addressing these gaps is essential to ensuring that humanitarian aid reaches affected populations in the most effective and efficient manner possible. Our research contributes to the literature on the centralization versus decentralization debate (e.g. Drabek and McEntire, 2003, Thompson, 2015) in humanitarian operations by providing empirical evidence that, in the context of warehouse location decisions, stakeholders may benefit from adopting a centralized strategy.

This study aimed to analyze the impact of a multiperiod modeling approach on the planning of humanitarian facilities and interagency coordination strategies at the national level during complex emergencies. By proposing a new configuration strategy for the logistics network – applied to the Mosul Offensive case study – we demonstrated the potential benefits of cooperative planning among humanitarian organizations.

Although the analysis was conducted in the specific context of the Mosul Offensive, the findings carry broader implications for national humanitarian logistics in other disaster scenarios. The need for centralized and strategically located warehouses is not confined to armed conflicts but extends to natural disasters, pandemics and refugee crises worldwide (Kian et al., 2022; Stauffer et al., 2016; Vledder et al., 2019). In each of these contexts, evidence suggests that a more centralized strategy to facility management can reduce costs, enhance service levels and strengthen operational resilience. Consequently, the Mosul case can be regarded as a representative example that demonstrates how lessons learned in one setting may inform the design of more efficient supply chains in other complex emergencies.

Beyond its theoretical contribution, this study offers several practical implications for humanitarian organizations and policymakers. First, the results demonstrate that centralizing warehouse operations at the regional or national level can significantly reduce costs and infrastructure redundancies. For practitioners, this means that agencies can enhance efficiency by consolidating distribution centers and Mobile Storage Units, rather than operating in isolation. In practice, adopting such an approach would require stronger interagency coordination mechanisms, clear agreements on cost-sharing and standardized procedures for facility management. These steps can help overcome the persistent challenges of duplication and competition for limited resources that have been widely documented in humanitarian supply chains.

Second, the findings suggest that dynamic, multiperiod planning should be integrated into humanitarian logistics decision-making. The ability to adjust warehouse capacity and locations over time ensures that resources are deployed proportionally to shifting needs on the ground. For relief agencies, this implies investing in data collection, forecasting and flexible contracting arrangements that allow for capacity expansion or decommissioning as crises evolve. Such practices could improve both the cost-effectiveness and responsiveness of humanitarian operations, offering a more sustainable model at a time of growing financial constraints and increasing disaster frequency.

Third, the study contributes to bridging the gap between conceptual modeling and field application in humanitarian logistics. By offering a practical, adaptable and empirically tested framework, this research supports the transition from ad hoc, reactive logistics systems to strategic, evidence-based planning. We hope it will inspire humanitarian organizations, donors and policymakers to embrace data-informed coordination models in the design of future relief efforts.

Finally, while this study provides quantitative evidence of the benefits of centralized coordination, future research should continue to explore how centralized models can be adapted and scaled across different national humanitarian contexts. Humanitarian logistics is characterized by high uncertainty, political complexity and dynamic interactions among diverse stakeholders, factors that are often difficult to capture fully in mathematical models. Qualitative approaches – such as case studies, interviews and participatory action research – can help reveal organizational, cultural and political barriers to coordination that influence the feasibility of centralized systems. Combining such insights with quantitative modeling could provide a more holistic understanding of when and how centralized coordination strategies can be effectively implemented in real-world humanitarian operations.

The authors gratefully acknowledge the financial support provided by the Brazilian Coordination for the Improvement of Higher Education Personnel (CAPES), the Brazilian National Council for Scientific and Technological Development (CNPq), and the São Paulo Research Foundation (FAPESP). Their financial support was essential for the successful completion of this research.