David Russell, University of Florida
In 2016, the Islamic State’s Libya Province (ISLP) operated throughout the country, recording incidents of political violence in 11 of Libya’s 22 districts and boasting territorial control of the city of Sirte and its environs. By 2022, however, the group had fully retreated south into the deserts around Sabha and Murzuq. The case of ISLP is indicative of how Libya’s post-2011 civil wars have seen extreme political fragmentation, producing a conflict landscape populated by armed actors pursuing a wide variety of spatial strategies (Mundy 2018). Actors claiming to represent the Libyan state or attempting to capture it operate in many distinct locations or across great distances. Other actors, such as militias based on ethnic or kinship ties, choose to focus their activities on just one location. But as the line between non-state and state actor is increasingly blurred, exceptions abound: local militias begin to operate far away from their origins, and other militias seldom operate outside single neighborhoods even as they are officially integrated into the state apparatus.
Conflict studies remain limited in explaining these types of wars in part by the complexity inherent in representing actor locations in space. Even with the proliferation of reliable, disaggregated event data on conflict, spatializing armed actors is difficult, as they can move around, operate in multiple places at once, and enter and exit conflict. This paper addresses this complexity by introducing a new conceptual and methodological framework for measuring how the actions of armed actors in conflict produce certain spatial point patterns. These patterns, the actors’ Areas of Operations, can be measured, categorized, and tracked over time, aiming to describe not just where armed conflict actors operate, but how their actions produce a characteristic, measurable point pattern at a given time. This paper applies the Areas of Operations concept to answer the research question: How can event data on political violence typify and describe the changing geography of armed actors’ conflict activities over time?
Concept
The literature on locating conflict actors shows that where actors operate varies over time and matters for conflict outcomes. But using event data to describe where actors operate comes with significant methodological challenges (Kim et al. 2023). The spatial patterns of where actors operate depend on their objectives, capabilities, and strategic choices because space constrains and enables groups with different capabilities and objectives in different ways (Radil and Castan Pinos 2020; Russell and Radil 2022).
The Areas of Operations concept argues that conflict actors have certain relationships with space at any given time, that these relationships with space can be represented as point patterns of conflict event data, and that a typology of these representations of actors’ spatialities can help distinguish actors from each other and identify different characteristic periods in the evolution of armed actors’ strategies. In this paper, the term “Areas of Operations” refers exclusively to the spatial distribution of recorded violent events involving the actor. I do not attempt to model or estimate where actors are present but do not participate in acts of political violence, where actors participated in violence but were not recorded doing so, or where front lines or territorial boundaries exist between armed groups in conflict.
Three characteristics of the spatial patterns of recorded conflict events in which an actor engages can represent the actor’s relationship with space in a given time period. These characteristics focus on how the locations where an actor operates relate to other locations where they operate: in the terminology of spatial statistics, this represents a second-order point pattern analytic approach. While each of the metrics representing the three characteristics provides valuable information about the actor on their own, considering all three together allows for the fullest understanding of the range of possible conflict activity point patterns (Figure 1).
Figure 1: Metrics used to characterize the spatial patterns of actors’ Areas of Operations.
The first metric answers the question: Across how great a spatial extent does the actor conduct operations? It can be measured as the maximum distance between pairs of points representing the events in which the actor participated in a given time period, such that a larger value means that the actor is participating in events over a greater spatial extent. Actors who operate across greater spatial extents are more likely to interact with each other in conflict (Dorff et al. 2020; Kim et al. 2023), and their political goals are likely to be larger in scope.
The second metric answers the question: In how many distinct locations does the actor operate? It is measured as the number of clusters the events in which the actor participates in a given time period; larger numbers mean that the actor is operating in multiple places with gaps of no activity between them. Such point patterns could reflect multiple cells or sub-groups of an actor’s organizational structure alongside a capability to operate in noncontiguous spaces (Prieto-Curiel et al. 2020).
The third metric answers the question: To what extent does the actor concentrate their operations in a single location? It can be measured as the percentage of the actor’s events that year that took place in the cluster with the largest number of events. A higher value indicates that the actor is primarily concerned with its operations in a single place. Beardsley et al. (2015) note that groups lacking strong ethnic ties and the ability to engage in conventional warfare tend to fight in more inconsistent locations. Their activities would therefore tend to be scattered among multiple different locations. On the other hand, places experiencing extreme political fragmentation during civil wars often see the emergence of local protection forces or vigilante militias, as has been the case in Libya (Lacher 2020). These militias tend to focus their operations on the locales in which they emerged and which they seek to protect from unrest and from outside forces.
Combining all three of these metrics in a typology allows for the description of a wide range of possible spatial patterns of recorded violent events involving an armed actor. Likewise, summarizing an actor’s spatial patterns with a single value from a typology lends itself to relating these types to information about the actor: for example, actor types or an actor’s structural position in the social network of conflict rivalries and alliances. A consistent and reproducible typology also enables the study of patterns of sequences over time and the relation of these temporal sequences to other temporally-disaggregated data. This paper’s analysis focuses, however, on showcasing this typology to show how event data on political violence can typify and describe the changing geography of armed actors’ conflict activities over time.
Methods
This paper uses spatially disaggregated event data from the Armed Conflict Location and Event Data (ACLED) project (Raleigh et al. 2010). ACLED details events of political violence, including attribute data on the date and location of the event, the groups involved, fatalities, information sources, and other attributes. While subject to limitations relating to the biases of researchers and how conflict events are reported in media, ACLED is one of the more reliable event datasets on conflict available (Raleigh et al. 2023). While I acknowledge that ACLED data is imperfect, I make no attempt to impute missing data or interpolate unknown actor presence using the available event data, as these tasks lie beyond the scope of this project. The data processing, analysis, and visualization for this paper was completed using the R programming language implemented in RStudio.
ACLED records 8,298 events of political violence in Libya from January 1, 2011 – December 31, 2024. After a process of recoding certain actor names based on domain knowledge, this paper’s analysis identified 402 unique actors represented in the dataset across all years, of which 141 were considered for this paper’s analysis. The 261 actors not considered for the spatial analysis include civilian actors, actors labeled as “unidentified,” actors who participated in fewer than two violent events across all years, and actors who were only active for one year or less. Finally, actors who had only operated in the 100 km buffer zone outside of Libya and not inside of Libya itself were also removed. These removal processes left 141 actors, for a total of 1,974 actor-years, to be included in the main analysis.
For each actor, for each year, this analysis summarizes the actor’s conflict activities’ point patterns by taking three measurements of the spatial distribution formed by recorded events in which that actor participated as described in the conceptual section of this paper. The first measurement is the maximum distance between pairs of points; a larger value means that the actor is participating in events over a greater spatial extent.
The second measurement is how many clusters the actor’s events form during the given year. The third measurement is the percentage of the actor’s events that year that took place in the largest cluster, the cluster with the largest number of events. Both rely on the use of a simple spatial clustering algorithm: points in a given actor-year’s point pattern were grouped together into the same cluster if they were within 80 km of each other. In network terms, a pair of points shares an edge if they are within 80 km of each other, and spatial clusters are defined as the connected components of the network. The buffer value of 80 km was chosen because this is the largest value that prevents Libya’s major cities from being grouped together without any events in between. (Libya is roughly 1,500 km by 1,000 km, for reference.) The case study-specific threshold of 80 km is specific to the unique context of Libya’s human geography and should not be interpreted as universally applicable to other cases.
This analysis used k-means clustering to group the actor-year observations into types of Areas of Operations, which are groups of like observations based on the three primary point pattern measurements detailed above. These measurements were first standardized using min-max normalization. Because actor-years with either a single event or no events could not be analyzed using the spatial clustering techniques described above, these were set aside into their own respective categories. The remaining actor-year observations were classified into one of six categories; a gap statistical analysis using R’s cluster package recommended six as the optimal number of clusters.
Results
The attribute clustering methodology chosen for this analysis categorized actor-years into six possible types based on three metrics describing the point pattern of the conflict events in which the actor participated that year: how many clusters the events formed, the maximum distance between the events (their spatial extent), and the percentage of the events in the largest cluster. Because the spatial clustering analysis would not be meaningful for actor-years with one or zero events, these actor-years were treated separately, resulting in a total of eight possible Areas of Operations types. An overview of these eight types is shown in Figure 2. The order of the type names A through G reflects an increasing mean cluster count, which nearly aligns with increasing mean maximum distance as well.
Figure 2: Areas of Operations types and key summary statistics. Source: ACLED, calculations by author.
The example of the Islamic State’s Libya Province (ISLP) illustrates the potential dynamism of Areas of Operations types well. Over the course of the group’s operations in Libya since 2014, ISLP’s activities fell variously into seven of the eight types over the course of its operations since 2014 (Figure 3). The narrative of the Islamic State’s project in Libya is largely a story of failure: to use its own terms, the Islamic State failed to “remain” or to “expand” in Libya (Ibrahim 2020, 48). After enjoying initial battlefield successes and the capture of territory along the central part of Libya’s coast, ISLP’s extremist ideology led it to alienate its potential allies, attract international intervention, and overreach in trying to operate as both a territory-holding insurgency and a terrorist organization (Lacher 2020, 43–49). The substantial variation in where ISLP operated and how it used space can therefore be seen primarily as a direct consequence of its conflict relationships with other armed actors and its subsequently diminishing capabilities.
In late 2014, jihadists returning to Libya from Syria joined forces with their local counterparts and pledged allegiance to the Islamic State. After attracting fighters and loyalty pledges from existing jihadist organizations amid the chaos of the outbreak of Libya’s second civil war, ISLP began its operations in a scattering of coastal cities. This paper’s methodology categorizes the spatial patterning of the conflict events in which ISLP took part in 2014 as Type E Areas of Operations, characterized by around three spatial clusters a maximum of around 800 km from each other, with fewer than half of events occurring in the largest cluster, indicating little focus on a single location.
In early 2015, ISLP began solidifying its hold on and governance of territory centered around the city of Sirte, a former Gaddafi loyalist stronghold discontent with the overthrow of his regime, while continuing to conduct both territory-seeking operations and terrorist attacks elsewhere in Libya. These events can be grouped into many spatial clusters over a large spatial extent with relatively few events in the largest cluster, meaning that they are best described as Type F Areas of Operations. By 2016, a broad coalition had coalesced around opposition to ISLP, and the group’s strategy began to focus more on defending the core territory around Sirte, which they lost later that year. The continued presence of many spatial clusters of events across a great distance, but with more events occurring in the largest cluster, meant that ISLP’s Areas of Operations in 2016 are best categorized as Type G.
The next two years saw a reversion to Type F Areas of Operations primarily due to lack of focus on the lost core territory of Sirte combined with a persistent presence throughout the rest of the country. By 2019, however, ISLP had shifted its focus into the Fezzan region of southern Libya, again concentrating more on a single cluster of events with continued activity elsewhere, indicating a return to Type G. Since then, ISLP’s engagement in conflict has dwindled, and its Areas of Operations types have diminished in spatial extent and cluster counts through Types D, E, C, and finally to A, with only one ISLP event recorded by ACLED in 2024.
Figure 3: Areas of Operations types for the Islamic State’s Libya Province (ISLP), 2014 – 2024. Note that ACLED does not record any conflict events in which the ISLP participated during 2023. The colors of the points represent different clusters grouped by an 80 km buffer.
Conclusion
The example of the Islamic State’s Libya Province underscores the importance of incorporating the fluidity and dynamism of conflict actor identities and objectives in complex, multiparty civil wars. In Libya, actors often do not fall neatly into a single category; there is no single state apparatus, local militias have become incorporated into the dueling state militaries, and some jihadist organizations are hyperlocal while others are transnational. Instead of limiting actor attributes to a static typology, we can represent their conflict behaviors over time as a way of informing our understanding of actor behavior.
This approach helps to show how the Islamic State’s approach to territory in Libya changed in tandem with its capabilities and relationships with other armed actors. Simply mapping ISLP’s activities provides an overview of the broad spatial patterns of the group’s relationship with space: an initial expansion, a retreat south, and a subsequent contraction. Where the Areas of Operations approach shines, however, is in exploring the nuances of how ISLP operated at each stage of its trajectory. For example, ISLP engaged in violence in many places in both 2015 and in 2016, but the core of their operations focused more around Sirte in 2016 due to the pressure that rival armed groups were exerting on them that year. Future work could apply this approach to multiple organizations to test whether armed actors in conflict tend to move through predictable patterns of spatiality as they develop over time.
This paper contributes to our understanding of Libya’s understudied, complex, and potentially still volatile conflicts as the factors that led to Libya’s Second Civil War have not been resolved. In particular, the institutionalization of Libyan militias into the state apparatus mixes private with public interests and maintains divisions within the government based on ideology and tribal and geographic origin; these divisions mean that while levels of violence in Libya have been much calmer since 2021, the potential for another outbreak of civil war remains (Lacher 2023). Understanding where Libya’s armed actors tend to operate and what spatial patterns their activities take can help policymakers identify the fault lines along which country might fracture again.
References
Beardsley, Kyle, Kristian Skrede Gleditsch, and Nigel Lo. 2015. “Roving Bandits? The Geographical Evolution of African Armed Conflicts.” International Studies Quarterly 59 (3): 503–16.
Dorff, Cassy, Max Gallop, and Shahryar Minhas. 2020. “Networks of Violence: Predicting Conflict in Nigeria.” The Journal of Politics 82 (2): 476–93.
Ibrahim, Azeem. 2020. Rise and Fall? The Rise and Fall of ISIS in Libya. US Army War College.
Kim, Sangyeon, Howard Liu, and Bruce Desmarais. 2023. “Spatial Modeling of Dyadic Geopolitical Interactions between Moving Actors.” Political Science Research and Methods 11: 633–44.
Lacher, Wolfram. 2020. Libya’s Fragmentation: Structure and Process in Violent Conflict. I.B. Tauris.
Lacher, Wolfram. 2023. “Libya’s Militias Have Become the State: Dimensions and Consequences of a Consolidation Process.” SWP Comment 44: 1–6.
Mundy, J. 2018. Libya. John Wiley & Sons.
Prieto Curiel, Rafael, Olivier J. Walther, and Neave O’Clery. 2020. “Uncovering the Internal Structure of Boko Haram through Its Mobility Patterns.” Applied Network Science 5 (1): 1–23.
Radil, Steven, and Jaume Castan Pinos. 2020. “The Territorial Contours of Terrorism: A Conceptual Model of Territory for Non-State Violence.” Terrorism and Political Violence 32 (5): 1027–46.
Raleigh, Clionadh, Andrew Linke, Håvard Hegre, and Joakim Karlsen. 2010. “Introducing ACLED: An Armed Conflict Location and Event Dataset.” Journal of Peace Research 47 (5): 651–60.
Raleigh, Clionadh, Roudabeh Kishi, and Andrew Linke. 2023. “Political Instability Patterns Are Obscured by Conflict Dataset Scope Conditions, Sources, and Coding Choices.” Humanities and Social Sciences Communications 10 (1): 1–17.
Russell, David G., and Steven M. Radil. 2022. “Landscapes of Control: The Geography of Counterinsurgency in the Sinai, Egypt.” The Arab World Geographer 25: 98–118.