---
title: Environmental Justice Atlas
url: https://www.emergentmind.com/topics/environmental-justice-atlas-ejatlas
type: topic
---

# Environmental Justice Atlas

The Environmental Justice Atlas (EJAtlas) is a crowd-mapped, curated global database of socio-environmental conflicts in which each record is a conflict or “event” that includes location, basic timing, sector or conflict category, descriptions of impacts, mobilization, outcomes, and lists of actors such as companies and Environmental Justice Organizations (EJOs) [2603.29722]. In the literature, EJAtlas is also described as a global platform documenting environmental justice conflicts at case-level granularity, where entries describe a conflict, actors, impacts, mobilizations, and outcomes rather than continuous environmental monitoring streams [2107.11633]. This combination of case documentation, actor attribution, and thematic classification has made EJAtlas both a knowledge repository on environmental injustice and a research infrastructure that can be analyzed as a relational database of conflicts, corporations, movements, and places [2603.29722].

## 1. Conceptual orientation

EJAtlas is used to document conflicts that are intelligible through core environmental justice dimensions: unequal exposure to pollution and ecological harm, unequal participation in decision-making, and unequal recognition of affected communities and more-than-human entities. In work connecting AI ethics to environmental justice, Schlosberg’s tripartite framework of distributive justice, procedural justice, and justice as recognition is presented as directly usable for EJAtlas-style conflict analysis, because it asks who bears burdens and receives benefits, who has a voice in decisions, and which human and more-than-human actors are treated as worthy of consideration [2501.10390].

This framework aligns with EJAtlas-style documentation of struggles around polluted neighborhoods, industrial facilities, and campaign trajectories. A Kansas City air-quality platform developed with CleanAirNowKC was explicitly interpreted as being “very much in the spirit of EJAtlas”: a community documents local environmental harm, names industrial sources, and builds evidence and narratives to support resistance and policy change [2107.11633]. The same literature also suggests that EJAtlas is increasingly compatible with more-than-human and non-anthropocentric justice, including impacts on animals, ecosystems, and rights-of-nature claims, rather than a narrowly anthropocentric account of harm [2501.10390].

A recurrent implication is that EJAtlas does not merely catalogue pollution incidents. It situates environmental conflicts within broader status orders, institutional exclusions, and territorial asymmetries. This suggests that its analytical value lies not only in recording harms, but also in documenting how conflicts are structured by race, class, coloniality, sectoral development, and ecological degradation.

## 2. Data model and conflict representation

The basic unit in EJAtlas is the conflict event. Each entry includes geographic location and timing, assigns a sector or conflict category, and records impacts, mobilization, outcomes, and involved actors, with particular salience given to companies and EJOs [2603.29722]. The conflict categories explicitly listed in network-based work using EJAtlas include Mining, Fossil Fuels and Climate Justice, Biomass and Land, Water Management, Infrastructure and Built Environment, Industrial and Utilities, Waste, Nuclear, Tourism, and Biodiversity Conservation [2603.29722].

In one cleaned subset restricted to conflicts containing at least one company and at least one EJO, EJAtlas yielded 3,396 conflicts in 164 countries, involving 6,244 companies and 11,231 EJOs [2603.29722]. That structure permits multiple relational projections. Conflicts can be linked to companies and EJOs; companies and EJOs can be linked to one another through co-involvement in conflicts; and countries or regions can be linked through the actors headquartered or primarily associated with them [2603.29722].

The resulting data model is unusually flexible. A single entry can simultaneously function as a qualitative case narrative, an actor registry, a geographic point or polygon, and a node in a larger network of socio-environmental contestation. This is why EJAtlas has been described not just as a map of isolated cases, but as a relational infrastructure that can be converted into multilayer networks of conflicts, firms, and movements [2603.29722].

## 3. EJAtlas as a research infrastructure

When EJAtlas is treated as a relational database, its entries can be formalized as bipartite conflict-actor networks and then projected into actor-actor and conflict-conflict networks. In the global network analysis built directly from EJAtlas, the raw conflict-company and conflict-EJO data are transformed into conflict-company and conflict-EJO bipartite matrices, and then into company-company, EJO-EJO, and conflict-conflict networks, as well as country-country and region-region aggregations [2603.29722].

This transformation changes the analytical object. Instead of asking only where conflicts occur, it becomes possible to ask how firms recur across disputes, how EJOs bridge categories and geographies, and how sectoral patterns differ between corporate and movement networks. The main empirical finding is a stark asymmetry: multinational corporations form a cohesive global network, whereas EJOs are fragmented, often operating in isolated clusters with limited interconnection but forming a robust, decentralized and self-organized global network [2603.29722]. The same study further reports that the firms network shows strong dependence on conflict category, while the EJO network does not depend on conflict category [2603.29722].

Conflict-conflict projections reinforce this distinction. When conflicts are linked through shared companies, category clustering is strong; when they are linked through shared EJOs, inter-category connections are comparatively more prominent [2603.29722]. This indicates that corporate organization is more sector-bound, while EJOs more often work across fossil, mining, land, water, and infrastructure conflicts. A plausible implication is that EJAtlas is particularly well suited to studying not only local conflict intensity, but also translocal coordination, actor recurrence, and the emergence of global structures of resistance and extraction.

The same network analysis also shows why EJAtlas has attracted interest beyond descriptive cartography. It supports standard social network analysis using degree, weighted degree, betweenness centrality, closeness centrality, Gini coefficients on degree distributions, largest connected components, edge density, diameter, average path length, and Jaccard similarity across category-specific subnetworks [2603.29722]. In that sense, EJAtlas has become a platform for computational social science without ceasing to be a repository of activist and case-based knowledge.

## 4. Participatory and localized “micro-atlases”

Although EJAtlas typically operates at case-level granularity, several studies describe local infrastructures that mirror its logic while adding temporal, sensor-based, or neighborhood-scale detail. CleanAirNowKC’s interactive air-quality map is explicitly interpreted as a highly localized, data-intensive “micro-EJAtlas” focused on air pollution in Kansas City, Kansas [2107.11633]. It documents harms, actors, and resistance, but does so through nine PurpleAir sensors, interactive visualization, hazardous-waste overlays, and a reporting system for transient air-quality and other pollution events rather than through static case profiles [2107.11633].

This local system reproduces core EJAtlas functions: identifying affected communities, naming industrial or hazardous infrastructures, recording incidents, and supporting organizing and policy advocacy. It also extends them through pseudo-live PM\(_{2.5}\) and AQI monitoring, line charts, user-adjustable time windows, and map-linked reporting functions [2107.11633]. The literature therefore presents it as a model for local extensions of EJAtlas in which each conflict case could be linked to a living, community-controlled monitoring platform [2107.11633].

Comparable dynamics appear in urban environmental justice mapping outside the air-pollution context. In Chicago, a community-based participatory design process produced a proximity-burden dashboard that aggregated school-level burdens from nearby hazardous sources into a Collective Proximity Burden at the community-area scale, with explicit attention to predominantly Latinx neighborhoods [2112.06119]. In Greensboro, North Carolina, an integrated, intersectional climate vulnerability assessment combined demographic, socioeconomic, health, and environmental indicators at census-tract scale and then used clustering to identify a critically high-risk neighborhood typology characterized by high flood exposure, extreme poverty, poor respiratory health, and aging housing [2601.15675]. These projects are not themselves EJAtlas, but they are methodologically aligned with it in that they convert structurally unequal burdens into spatially explicit, community-usable evidence.

The localized literature also points to a broader design principle: EJAtlas-style mapping becomes more powerful when communities are not only the subjects of mapping but also the co-producers of metrics, reporting functions, and visual representations. This suggests a continuum from global conflict atlas to local monitoring dashboard rather than a rigid distinction between the two.

## 5. Quantitative extensions and methodological interfaces

A major theme in recent work is that EJAtlas can be enriched by quantitative indicators that remain compatible with its conflict-centered logic. The England-wide 3-30-300 study shows how urban nature equity can be operationalized at national scale using building-level tree proximity, LSOA canopy cover, network-based park accessibility, Gini coefficients, and Spatial Error Models linking inequality to deprivation [2510.13861]. The paper explicitly argues that these indicators are directly usable in EJAtlas case descriptions, comparative dashboards, or filters, and that they can contextualize urban conflicts involving tree loss, canopy deficit, park access, and blue-space exclusion [2510.13861].

Global air-pollution inequality research adds a different scale of quantification. Using population-weighted PM\(_{2.5}\) distributions over 80.1 million grid cells in 228 countries or territories, it reports a global PM\(_{2.5}\) Gini Index rising from 0.32 in 2000 to 0.36 in 2020, with most inequality driven by differences between countries, and introduces the “Choking Billion” as the 1 billion people with the highest PM\(_{2.5}\) exposure in 2020 [2307.15669]. The paper argues that EJAtlas cases can be situated within this global distributional backdrop, so that local air-pollution conflicts are also read as manifestations of a broader transboundary environmental inequality [2307.15669].

Other quantitative frameworks are even more tightly aligned with conflict documentation. The Texas coastal flooding study maps hazardous sites, flood hazards, and socio-demographic vulnerability, then estimates threatened population within 1-, 3-, and 5-mile pollutant-dispersion buffers using area-weighted tract overlaps [2301.00312]. That work is explicitly described as a proto-EJAtlas case study because it identifies multi-hazard EJ hotspots where industrial and toxic facilities, flood hazards, and vulnerable communities intersect [2301.00312]. Likewise, heterogeneous causal-effect estimation under bipartite network interference provides a formal framework for evaluating who benefits from emissions-control interventions when treatments occur at facilities and outcomes are measured in downwind communities; in the coal-scrubber application, no statistically significant effect appears in the full population, but significant ischemic heart disease hospitalization decreases are found in communities with high poverty and smoking rates [2304.12500].

Interpretable machine learning adds yet another interface. ML4EJ trains Random Forest and XGBoost models for PM\(_{2.5}\), urban heat, and flood risk in six U.S. metropolitan counties and shows that social-demographic features are the most prominent urban features shaping hazard extent, while infrastructure distribution and land cover are relatively important for urban heat and air pollution respectively [2310.02476]. The paper further reports limited transferability across regions and hazards, underscoring the context dependence of environmental injustice mechanisms [2310.02476]. For EJAtlas, this kind of result is methodologically significant because it links qualitative conflict narratives to structural urban-feature patterns without reducing conflict to a single socioeconomic variable.

Taken together, these studies indicate that EJAtlas can function as the narrative and relational layer in a broader analytical stack that includes remote sensing, inequality indices, spatial econometrics, cluster analysis, causal inference, and interpretable machine learning. They also show that such integration need not erase conflict narratives; rather, it can add exposure gradients, vulnerability typologies, and counterfactual policy estimates to case documentation.

## 6. Limits, biases, and contested issues

The most explicit limitations identified for EJAtlas as a dataset concern coverage, comparability, and the interpretation of ties. The global network study notes that Latin America, Europe, and North America are relatively well covered, whereas parts of Asia, the Pacific, Russia/Central Asia, and the Middle East are under-represented; conflicts before 2000 are also under-represented [2603.29722]. It further cautions that co-presence of actors in a conflict is not proof of cooperation, alliance, or explicit coordination, and that country assignment for EJOs via majority rule plus an “International” category can misrepresent genuinely transnational organizing [2603.29722].

A second set of limitations concerns the interface between EJAtlas and quantitative methods. Proximity burden metrics are proxies rather than full exposure or health-risk assessments, and their interpretation is affected by the Modifiable Areal Unit Problem and by the choice of classification scheme [2112.06119]. Global PM\(_{2.5}\) inequality metrics capture distributional disparity rather than injustice per se, since they do not encode discrimination, consent, or political power [2307.15669]. Interpretable machine-learning models identify strong associations among urban features and hazard exposures, but they do not resolve causality and show limited transferability across cities and hazards [2310.02476]. Climate-vulnerability typologies at census-tract scale remain cross-sectional and can miss within-tract heterogeneity and temporal dynamics [2601.15675].

A third set of issues arises when EJAtlas is extended through community-controlled digital infrastructures. The Kansas City air-quality platform exposes challenges around mobile usability, interpretability of multi-interval AQI glyphs, incomplete sensor coverage, API rate limits, and unresolved questions about privacy, moderation, and retaliation risk in incident reporting [2107.11633]. These are not specific to EJAtlas, but they matter when atlas-like systems move from static case sheets toward real-time or participatory reporting environments.

The literature therefore treats EJAtlas as powerful but not exhaustive. It excels at documenting contested cases, actors, and mobilizations, but it requires supplementation when the analytical question concerns continuous monitoring, high-frequency temporal dynamics, source-receptor causality, or national-scale exposure baselines. Conversely, quantitative systems require EJAtlas-like narrative and political context if unequal outcomes are to be interpreted as environmental justice rather than as mere spatial variation.

## 7. Prospective development

Several studies propose directions in which EJAtlas could expand. One is tighter integration with local “micro-atlases,” such as community-controlled sensor networks, incident-reporting systems, and interactive dashboards that can provide fine-grained temporal evidence of ongoing harm [2107.11633]. Another is the incorporation of structured environmental inequality indicators such as 3-30-300 attainment rates, canopy-cover deficits, tree-access Ginis, PM\(_{2.5}\) exposure deciles, or climate-vulnerability typologies as case-level contextual variables [2510.13861][2307.15669][2601.15675].

A further development concerns substantive scope. The environmental ethics of AI literature argues that AI-related mining, data-centre siting, energy and water use, and multispecies harms should be legible as environmental justice conflicts and therefore analyzable in EJAtlas-style terms of distributive justice, procedural justice, and recognition [2501.10390]. This suggests that EJAtlas can also serve as a framework for emerging ecological distribution conflicts whose causal chains include digital infrastructures rather than only conventional extractive or industrial projects.

Finally, EJAtlas is increasingly portrayed as an infrastructure for connecting place-based struggles to larger structural patterns. Network analysis reveals recurrent corporate actors and decentralized movement architectures [2603.29722]; global air-pollution inequality situates local conflicts within planetary exposure hierarchies [2307.15669]; urban and regional vulnerability studies identify latent hotspots where future conflicts may emerge [2301.00312][2601.15675]. The resulting picture is of EJAtlas not as a static inventory, but as a modular platform through which conflict narratives, actor networks, exposure metrics, and community-generated evidence can be assembled into a more comprehensive cartography of environmental injustice.

Source: https://www.emergentmind.com/topics/environmental-justice-atlas-ejatlas