Conflict Maps: Representing Dispute Dynamics

• Conflict Maps are domain-specific representations that encode conflict, risk, and disagreement through spatial, sensor-based, and cognitive techniques.
• They integrate diverse methodologies such as remote sensing, grid-based event density, and route-level hazard analysis to capture complex conflict dynamics.
• Their practical applications span risk analysis, urban planning, and security, though challenges include scale sensitivity, proxy reliability, and context-specific interpretation.

Conflict Maps (CMs) are not a single standardized artifact but a family of representational structures used to encode conflict, contradiction, or disagreement across domains. In the cited literature, the term can denote a surface texture-like 2D map attached to a building facade and derived from ray-to-model-prior visibility analysis (Hanke et al., 21 Aug 2025), a set of spatially explicit, time-resolved proxies for conflict events and their aftermath derived from satellite sensing (Ren et al., 2020), a gridded typology of conflict intensity and conflict concentration (Walther et al., 2020), or, by interpretation, a stakeholder-specific or policy-governed map of conflict relations, decisions, and evidence (Tosunlu et al., 2023, Kinkelin et al., 2019). This suggests that CMs are best understood as domain-specific representations of conflict structure rather than as one universal cartographic class.

1. Terminological scope and conceptual families

The literature uses “Conflict Map” in both explicit and extended senses. In the most literal explicit usage, CM2LoD3 defines a Conflict Map as a facade-aligned discrepancy representation with the three states confirmed, conflict, and unknown, produced by comparing terrestrial laser scanning rays with an LoD2 building prior (Hanke et al., 21 Aug 2025). In geospatial conflict analytics, closely related objects include fire-anomaly maps, destroyed-settlement footprints, density-based event surfaces, and route-level hazard-exposure maps (Ren et al., 2020, Carranza et al., 27 Jun 2025). Other papers do not name a “Conflict Map” directly but present structurally analogous objects: a map of permeability to violence in a multilayer spectral embedding (Skillicorn et al., 2016), a two-dimensional indicator combining event density with clustering–dispersion structure (Walther et al., 2020), or a conflict workflow graph over reviewers, mediation rounds, and evidence in decentralized authorization (Kinkelin et al., 2019).

A recurring distinction concerns whether a CM represents events, dependencies, discrepancies, or narratives. Event-centered CMs encode where violent acts occur and how they cluster. Dependency-centered CMs encode how conflict propagates through networks, borders, or mobility systems. Discrepancy-centered CMs encode where observation contradicts a prior model, as in facade openings or evidential grid fusion. Narrative-centered CMs encode how stakeholders understand drivers, goals, and influence relations in a conflict situation (Tosunlu et al., 2023).

This plurality also clarifies a common misconception: a conflict map is not necessarily a dot map of incidents. In the surveyed work, conflict can be mapped as a chain of remote-sensing signatures, as a risk surface, as a latent interaction space, as a route-exposure profile, or as a formal graph of values and influences. The representational choice depends on the object of interest: violent events, structural risk, human displacement, sensor disagreement, or stakeholder cognition.

2. Geospatial conflict-event and remote-sensing maps

In human-rights monitoring and political-violence analysis, CMs are often constructed from spatiotemporal event proxies rather than direct observations of violence. A remote-sensing study of the Rohingya conflict uses FIRMS active fire detections from MODIS and VIIRS, Sentinel-1 SAR backscatter, Sentinel-1 interferometric coherence, and externally validated destroyed-settlement locations to detect anomalous burning, razing, and post-destruction construction activity. The workflow combines seasonal anomaly detection, kernel density estimation, mutual information between spatial density surfaces, and unsupervised SAR change detection. Reported values include $I(\text{anomalous}; \text{settlements}) = 0.86$, $I(\text{agricultural}; \text{settlements}) = 0.28$, and $I(\text{agricultural}; \text{anomalous}) = 0.29$; the SAR workflow identified an affected area of approximately $1{,}300{,}000\ \mathrm{m}^2$ with first detected clearance on 2018/01/20 (Ren et al., 2020). The resulting CM layers are not direct observations of violence but conflict-specific proxies: anomalous fire hotspots, destroyed-settlement footprints, and post-conflict construction indicators.

A complementary event-data formulation appears in the Spatial Conflict Dynamics indicator. Here the map unit is a uniform $50 \times 50$ km grid, and conflict is represented along two dimensions: conflict intensity as event density and conflict concentration as an Average Nearest Neighbor ratio. The indicator uses a “CI generational mean” threshold of $0.0017\ \text{events per km}^2$ to classify cells as high- or low-intensity, and interprets $CC_{i,t}<1$ as clustered and $CC_{i,t}>1$ as dispersed (Walther et al., 2020). The resulting four-type typology distinguishes high-intensity clustered, high-intensity dispersed, low-intensity clustered, and low-intensity dispersed conflict geographies. This is a CM in which the crucial object is not the single event but the internal spatial organization of violence within a cell.

The same paper shows why scale and representation matter. A $10 \times 10$ km grid would be too sparse, while a $100 \times 100$ km grid would aggregate events too far apart to be assumed related (Walther et al., 2020). This reinforces a general property of CMs: they are sensitive to the spatial support on which conflict is encoded. In remote sensing, this appears as the need to distinguish conflict-generated fires from seasonal agricultural burning; in event grids, it appears as the need to avoid both fragmentation and over-aggregation.

3. Risk surfaces, diffusion spaces, and route-level exposure

Another major CM family models conflict as risk, diffusion, or hazard exposure rather than as discrete observed events. In pastoral-conflict mapping for Cameroon, the Central African Republic, Chad, and the Democratic Republic of the Congo, the task is formulated as a binary cell-level prediction problem over regular grids, with outputs rendered as “Pastoral Conflict Probability” maps. The feature stack combines NASA weather and terrain variables, histogram-based environmental descriptors, and neighboring-conflict features $I(\text{agricultural}; \text{settlements}) = 0.28$0 for $I(\text{agricultural}; \text{settlements}) = 0.28$1. Performance varies by country and resolution: for example, the best $I(\text{agricultural}; \text{settlements}) = 0.28$2 km result for Cameroon reports Precision $I(\text{agricultural}; \text{settlements}) = 0.28$3, Recall $I(\text{agricultural}; \text{settlements}) = 0.28$4, F1 $I(\text{agricultural}; \text{settlements}) = 0.28$5, and AUC $I(\text{agricultural}; \text{settlements}) = 0.28$6, while CAR and Chad peak at $I(\text{agricultural}; \text{settlements}) = 0.28$7 km (Solaa et al., 2024). Here the CM is a country-specific gridded likelihood surface derived from sparse event data plus environmental covariates.

A more relational version appears in the map of permeability to violence, which replaces Euclidean geography with a spectral embedding over three edge types: geodesic distance, border costs, and directed sequential attacks by the same group. Geographic distances are converted into affinities, border permeability can be modeled as $I(\text{agricultural}; \text{settlements}) = 0.28$8, and sequential attack edges encode temporal succession (Skillicorn et al., 2016). In the resulting embedded map, closeness means effective accessibility for violence diffusion rather than literal geographic proximity. This is a CM of strategic space: a location may be remote cartographically yet central in conflict space if borders are permeable and attack sequences recurrent.

A further extension links conflict to displacement routes. Using Somalia’s settlement-level displacement network, the route-reconstruction framework models movements as stochastic walks on a directed weighted graph $I(\text{agricultural}; \text{settlements}) = 0.28$9, where settlements carry hazard-specific displacement counts $I(\text{agricultural}; \text{anomalous}) = 0.29$0. Path-level hazard exposure is then defined as

$I(\text{agricultural}; \text{anomalous}) = 0.29$1

Conflict and drought emerge as the dominant hazards, with conflict becoming less prominent at every next step, while exposure probabilities across trajectories remain widely dispersed (Carranza et al., 27 Jun 2025). This produces a route-aware CM: not “where conflict happens” alone, but “what conflict is encountered along likely displacement pathways.”

Dynamic city-scale CMs can also be process-based. A multiplex conflict model treats each city as a nonlinear bi-stable system with stable states in either war or peace, coupled through geospatial, political, and cultural layers, and reports F1 scores of $I(\text{agricultural}; \text{anomalous}) = 0.29$2 to $I(\text{agricultural}; \text{anomalous}) = 0.29$3 for predicting transitions from war to peace and peace to war at city scale (Aquino et al., 2019). A plausible implication is that some CMs are best interpreted not as snapshots but as dynamical state fields evolving on multilayer interaction networks.

4. Structural-semantic and evidential conflict maps in engineering

In engineering literatures, the term often shifts from political violence to geometric contradiction, sensor disagreement, or resource contention, but the underlying representational logic remains recognizably “conflict-mapped.” In CM2LoD3, a Conflict Map is a facade-aligned image of discrepancy between measured laser rays and an LoD2 prior. Each projected facade location is marked confirmed if the point lies within a tolerance $I(\text{agricultural}; \text{anomalous}) = 0.29$4 of the prior surface, conflict if the ray penetrates behind the modeled facade, and unknown if the point lies in front of the model or is otherwise ambiguous (Hanke et al., 21 Aug 2025). The three-channel CM is then semantically segmented into facade, window, door, and unknown, optionally fused with image-based Mask R-CNN probabilities, and back-projected into CityGML openings. In this setting, a CM is not a map of social conflict but a physics-based discrepancy field that reveals missing facade elements.

Evidential top-view grid mapping provides another explicit conflict quantity. For two basic belief assignments $I(\text{agricultural}; \text{anomalous}) = 0.29$5 and $I(\text{agricultural}; \text{anomalous}) = 0.29$6, conflict mass is

$I(\text{agricultural}; \text{anomalous}) = 0.29$7

Rather than normalize this away with Dempster’s rule, the paper uses a reliability-aware ER rule with

$I(\text{agricultural}; \text{anomalous}) = 0.29$8

so that conflict directly modulates source reliability (Richter et al., 2022). This yields a cell-wise conflict field over the grid, even though the paper does not name it a conflict map. The distinction between conflict, nonspecificity, and discord is especially important here: disagreement between sources is not the same as ignorance or indecision within one source.

In multi-agent path finding, structural-semantic topometric maps serve as conflict-localization devices. Instead of planning on all free cells, Conflict-Based Search is run over sparse regions such as intersections, pathways, and dead ends. Conflicts are then defined over occupancy intervals of regions and openings rather than over single cells and unit timesteps. A region conflict occurs when two agents occupy the same region during overlapping time intervals; an opening conflict occurs when agents attempt opposite-direction traversals through the same opening (Fredriksson et al., 29 Jan 2025). The map abstraction itself therefore encodes likely contention resources.

A related but implicit conflict-map idea appears in learning-guided CBS. There is no explicit CM object, but candidate conflicts are scored by a linear ranking function

$I(\text{agricultural}; \text{anomalous}) = 0.29$9

where $1{,}300{,}000\ \mathrm{m}^2$0 contains 67 features spanning conflict class, pairwise dependency, MDD width, local topology, space-time neighborhood density, and conflict-history counts (Huang et al., 2020). This is effectively a local conflict saliency map over the current search node.

5. Cognitive, workflow, and contested cartographies

Some CM traditions represent conflict not as externally observed events but as perceived structure, governance workflow, or contested representation. In conflict transformation research, a cognitive map is defined as a weakly connected directed graph

$1{,}300{,}000\ \mathrm{m}^2$1

where nodes are concepts and edges are positive or negative influence relations (Tosunlu et al., 2023). These maps function as conflict maps when they encode how stakeholders perceive causes, consequences, and objectives. The paper’s contribution is then a formal transformation pipeline from Cognitive Map to Value Cognitive Map, Ends-Means Map, and Value Tree, moving from descriptive representation to prescriptive decision support.

In high-security configuration management, the closest analogue to a CM is not a spatial map but an auditable conflict workflow graph. The TANCS system organizes proposal, review, authorization, and policy-driven mediation rounds over reviewer decisions, commit messages, second-channel confirmations, and reviewer replacement. Conflicts arise when multi-party authorization yields non-unanimous or otherwise unresolved reviewer input, and the paper notes that this structure can be reinterpreted as a conflict map over proposals, reviewer states, evidence, and escalation steps (Kinkelin et al., 2019). A similar reinterpretation is possible for Open RAN Near-RT-RIC, where xApp conflicts can be modeled over xApps, controlled parameters, KPIs, parameter groups, utility functions, and MNO-assigned weights, with mitigation by Nash Social Welfare Function or weighted Equal Gains optimization (Wadud et al., 2023).

The social-movement literature adds a different corrective: maps are not neutral. In the Stop Cop City / Defend the Atlanta Forest case, official, media, grassroots, and individual mapmakers used online maps to frame what the conflict was “about,” what harms counted, and which stakeholders were visible. The corpus comprised 45 unique map images initially, 32 retained for analysis, 166 unique keywords, and 375 keyword assignments across 26 high-level themes (Harris et al., 29 Apr 2025). The paper’s central point is that maps in contentious politics are active participants in the conflict: they legitimize, contest, omit, and circulate. For CM theory, this means representation itself can be part of the conflict process.

6. Methodological themes, limitations, and open questions

Across domains, three methodological themes recur. First, CMs are often proxy-based. Fire detections, SAR change, event density, route-level hazard composition, or sensor conflict mass are not conflict itself but measurable signatures associated with conflict processes (Ren et al., 2020, Carranza et al., 27 Jun 2025, Richter et al., 2022). Second, CMs are strongly scale-dependent. Grid size, Voronoi spacing, temporal bin width, or region abstraction can change whether conflict appears fragmented, clustered, or diffusely connected (Walther et al., 2020, Kushwaha et al., 2022). Third, CMs are frequently multimodal: stronger representations arise when multiple evidence channels agree, whether thermal anomalies with SAR destruction, CM geometry with textured facade segmentation, or xApp parameter overlap with KPI degradation and operator priorities (Ren et al., 2020, Hanke et al., 21 Aug 2025, Wadud et al., 2023).

Several limitations are equally recurrent. Remote-sensing conflict proxies can be confounded by seasonal agricultural burning, settlement morphology, and false positives in coherence thresholding (Ren et al., 2020). Route-level displacement CMs reconstruct plausible rather than observed paths (Carranza et al., 27 Jun 2025). Evidential fusion CMs depend on well-calibrated source reliability (Richter et al., 2022). Cognitive and workflow CMs rely on elicitation quality and policy assumptions (Tosunlu et al., 2023, Kinkelin et al., 2019). Contested cartographies reflect unequal access to geospatial tools and platform visibility (Harris et al., 29 Apr 2025).

A further caution comes from predictive conflict mapping. An unsupervised typology of African conflict avalanches built from climate, geography, infrastructure, economics, and demographics identifies three archetypes—“major unrest,” “local conflict,” and “sporadic and spillover events”—but finds that specifying conflict type negatively impacts the predictability of fatalities, duration, and other measures of conflict size. Reported mutual information between type and size measures stays below $1{,}300{,}000\ \mathrm{m}^2$2 bits (Kushwaha et al., 1 Mar 2025). This indicates that visually or conceptually coherent CMs need not be strong predictors of escalation severity.

The broader implication is that Conflict Maps are most reliable when treated as structured analytic representations whose semantics, scale, uncertainty, and intended use are explicit. In some literatures they are used to detect and monitor; in others to classify, route, arbitrate, or persuade. Their common denominator is not a fixed geometry but a representational function: to externalize conflict relations in a form that can be compared, validated, and acted upon.