For the categorisation of the research organisation(s), we wanted to understand the context in which the reviewed articles were produced. Therefore, we differentiated between articles originating from universities, research organisations such as the American Association for the Advancement of Science (AAAS), the United Nations (UN), the Red Cross, and various non-governmental organisations (NGO).

To display the geographical locations of the conflict events described in the empirical studies, we used country codes, following the norm 3166 from the International Organisation for Standardisation (ISO) (ISO, 2025). The decision to use country codes is informed by the different spatialities of conflicts, which are difficult to categorise on a map. Country codes were the most uniform classification category available. We acknowledge the complex setting of conflict events and geopolitical tensions in country codes, such as the ISO standard, where changes in country borders and contested territories may not always be accurate and well represented.

The spatial detail with which conflict events and potential human rights violations can be documented via RS data largely depends on the spatial resolution of the data. Following Deroin et al. (2011) we define very high spatial resolution as under 2 m, high spatial resolution as 2–30 m, medium spatial resolution as between 30 and 100 m, low spatial resolution as 100–1,000 m, and very low spatial resolution as above 1,000 m. Many reviewed empirical studies comprise more than one data type and, therefore, different spatial resolution categories. We thus categorised each data type per empirical study. This means that the number of counts in is higher than the number of empirical studies.

The spatial resolution of RS data depends on the data providers: Commercial providers offer data with higher spatial resolution compared to public providers. Therefore, we also categories the provider of the data. We define commercial providers as entities from which the data can be obtained only after payment. This is the case for data providers such as Maxar, Planet, or Digital Globe. By contrast, public providers are providers whose data are openly accessible. For example, they include the United States Geological Survey (USGS) from NASA or the Copernicus data provider from the European Space Agency (ESA). We use the term public, even though publicly available data can also be non-public in countries where websites are censored, with data providers being banned for political reasons. The category public provider may be used differently in other reviews that include research from areas impacted by internet censorship. The classification of providers into commercial and public needs further nuance since commercial providers often offer pro bono access to their data with specific deals, contracts, or Application Programming Interfaces (API)s. Commercial providers argue that researchers and NGOs have “a lot on their plates and a gaping hole in their budgets” (EOS Data Analytics, 2023), and companies such as Planet want NGOs to unlock the power of satellite data (Planet, 2024), which is why they offer selected researchers and NGOs access to their databases. To account for these specific circumstances, we included the category commercial (without costs) in our review analysis. Further, some studies used a mix of data, resulting in a category called commercial (without costs) and public.

Most empirical studies employed multiple approaches, we categorised each approach once; consequently, the number of approaches exceeds the number of empirical studies. To understand what RS observations can point to linkages of human rights violations, the respective observations were categorised, for example, into land cover, land use change, damage or fire, among others. We also wanted to understand which specific human rights violations these categories comprise more deeply. However, in most cases, human rights were not referenced clearly, hindering accurate categorization.

Since an increase in ML approaches has been perceptible in recent years across other research areas using digital data (Pugliese et al., 2021), we were particularly interested in understanding whether this trend is also discernible in conflict documentation and the identification of potential human rights violations with digital RS data. We categorised the empirical studies into those that use and those that do not use ML approaches. Since many studies applied several methods, we categorised each method separately. Consequently, counts outnumbered studies.

Due to the specific context of conflicts, accessibility and ground data are often lacking (FEWS NET, 2024; Raleigh et al., 2023). It is therefore particularly interesting to examine how the studies comprised in our corpus dealt with this difficulty of finding (alternative) ways to validate results. We categorised which data were used for the validation process in the following way: Ground data validating RS analysis in this context refers to data acquired in situ, including interviews, measurements, videos, and photos sent by witnesses. Remote data validating RS analyses are data without contact with actors on the ground, which can consist of data collected via visual image interpretation with satellite data, for example. Other data validating RS analysis denotes data collected for other purposes but fitting the criteria for validating results, including open-source information, such as videos, photos, and posts from social media accounts or other accessible information. Ground data and other data validating RS analysis is a mix of ground data and other data described above. We added the category remote data validating hints from direct sources because, throughout the review process, we understood that RS data were used to validate claims from other sources, such as local contacts. We used the category unclear for cases that lacked a discernible validation process.

Due to the novelty of this field of research, we also classified the reported challenges in the studies into remote, human, machine, data, monetary, or not mentioned. Remote challenges are cases in which the researchers declare that they had difficulty accessing local knowledge about the study area. In line with Lillesand et al. (2002), we consider it important that the given researcher has knowledge of the phenomenon and geographic region under investigation. We therefore foreground the declared difficulties in accessing information from on-the-ground sources in crisis countries, which are complex and sometimes unfamiliar environments for the researchers. Related but different are human challenges. We used the category human challenges when the researchers reflected on their positionality while processing the data and model, indicating potential uncertainties that can arise because of this positionality. Further, we use the category method challenges when the researchers remark on the possibility that an algorithm or a model used in their analysis generated inaccurate results. Connected to this aspect is the category data challenge, which marks cases where data availability was limited. We also used this category when information was missing, for example, the date in a Google Web Map Service (WMS). Economic challenges highlight cases where the researchers of the empirical study state that they could not run a particular analysis due to financial constraints.

The visualisation of the categories was undertaken with Python 3.10.9 in a Jupyter Environment 4.0.11 and QGIS 3.38.0.

For the discussion, the interdisciplinary scope of this review, along with our goal of unpacking both the context and content of the examined studies, necessitates an evaluation and discussion of the results through a critical RS lens. Using critical RS to review results involves not only interpreting the data used and methods applied but also reflecting on the socio-political contexts in which the production, analysis, and application of the empirical study data are embedded. This relational evaluation allows us to shed light on methods and practices of injustice, evaluate the engagement with situated knowledge from the ground, and understand the development of practices in the political and economic production contexts of RS (Bennett et al., 2022a).

The articles in our corpus were published between 2006 and 2023 and vary from reports and journal articles to conference papers. Figure 2 shows a time series of the selected empirical studies and their publishing organisations. What emerges is the collaborative manner of this research field, with four of the categories, namely Academic & Gov & Other, Academic & NGO, Academic & Other, and Other & NGO, containing research that was produced by more than one organisation. Additionally, NGOs played an important role until 2014; from then on, the share of Academic and collaborative work, including Other, increased.

Figure 3 ¹ illustrates the geographical locations of the conflict events described in the empirical studies.

To identify reporting trends, we plotted the number of investigations per country. In Figure 4, Ukraine is the country examined most often, followed by Sudan, Myanmar, Syria, and Nigeria.

The spatial resolution of the data used over time is visualised in Figure 5. We observe that the focus was on very high spatial resolution data during the first years of the period under consideration, whereas in recent years, more high spatial resolution data were used.

Figure 6 shows a stacked bar plot of empirical studies per year and their data origins in different colours. We observe that until 2012, data used in the empirical studies stemmed predominantly from commercial providers, whereas in 2023, the majority of studies sourced data from at least one public provider.

The trend of empirical studies using ML versus non-ML approaches over time is visualised in Figure 7. The figure shows that, over time, there has not been a clear trend towards ML approaches. Of all studies applying ML approaches, ten used supervised algorithms, and one used an unsupervised approach. Most studies used more than 1 ML type: the majority used Random Forests (three times), Support Vector Machines (two times), and Neural Networks (two times). All other types only occurred once, such as Maximum Likelihood or Minimum Distance, among others.

We summarize the non-machine learning (non-ML) approaches in Figure 8. Most empirical studies employed multiple approaches, and we counted each approach only once; consequently, the number of approaches exceeds that of studies using non-ML approaches. Due to the diversity of the empirical studies, the approaches were often described in various degrees of detail.

Figure 8 shows the non-ML approaches applied, such as Triangulation, Visual Change Detection, and Visual Image Interpretation, which are among the most frequently used approaches.

Figure 9 shows that most empirical studies document damage characteristics, land use and land cover change, and/or fire patterns with RS data.

As pointed out above, a categorization of the human rights-related terms within the categories was often too diverse. Exemplarily we point that out for Land Use Land Cover Change where genocide, mass violence, war crimes, International Humanitarian Law (IHL), attacks against civilians, ecocide, and general human rights were used. Similarly, human rights-related terms within the empirical studies focusing on Damage Assessment(s) were broad. The mentioned aspects ranged from attacks on civilians, civilians in conflict, intentional burning, property rights, destruction and burning of places, IHL, ethnic cleansing, land clearance, crimes against humanity, war crimes, indiscriminate violence, destruction of civilian infrastructure, forced evictions, forced relocation, cultural artefacts, communal conflict, attack on medical facilities as well as pointing to human rights overall.

Figure 10 depicts that the type of validation data used changed over time: Until 2013, remote data validating hints from direct sources was used predominantly, while from 2014 onwards, remote data validating RS analyses were used more often.

Figure 11 shows that most researchers declared data challenges, while some did not mention any challenges. Human and method challenges are less often declared, with remote and economic challenges being declared even more seldomly.

Figure 2 offers first insights into how our review and the selected empirical studies cover global conflict dynamics. The increase in empirical studies in recent years can be linked to the intensification of conflict violence, particularly conflict-related deaths, since 2019 (Uppsala University, 2025). To better understand the empirical studies’ coverage of specific conflict regions, a comparison with the Uppsala Conflict Data Program (UCDP) from the Department of Peace and Conflict Research (UCDP, 2014) is helpful. UCDP documents all known state-based violence, non-state-based violence, one-sided violence, and the total number of deaths. Such a comparison showcases that several conflicts, along with potentially related violations, were not investigated using satellite data or are not covered in our review. For example, while the UCDP records high death rates between 2006 and 2022 in the conflict areas of Burkina Faso and Mali, our corpus does not include empirical research into any one of these areas. The same is true for Niger, Chad, Yemen, Pakistan, Libya, Colombia, and Brazil.

The analysed empirical studies, therefore, betray that attention is skewed to select conflict areas while other conflict areas and related human rights violations remain neglected. Figure 4 reveals a marked research focus on conflicts in Ukraine, Syria, and Sudan. This focus can partly be explained by the conflict duration. The Russian invasion of Ukraine has been unfolding since 2022, and Russia invaded and annexed Crimea already in 2014 (Kulyk, 2023). Further, the start of the revolution in Syria goes back to 2011 (Al Jazeera, 2018), while Sudan experienced multiple conflict waves: two civil wars before 2011, as well as one conflict in the west of the country from 2003 to 2020, with another violent phase starting in 2023 (African Journal on Conflict Resolution, 2024; Hassan, 2023). Note, however, the difference between publishing date and conflict timeframes, as we use the year of the articles’ publication for our analyses and associated visualisations, not the period of conflict (Section 2). To clarify the relationship between the publishing date and conflict period, we checked all reviewed articles and found that in most empirical studies, the publishing date lies within the time of conflict. Another reason for the noted augmented coverage of Ukraine, Syria, and Sudan using RS data could be information biases in the media landscape of conflict reporting, as most conflict events are ignored (Jakobsen, 2000). This reporting bias likely influences satellite RS researchers’ perception of where research is needed. Arguably, the more a specific conflict is covered in the daily media, the higher the motivation to better understand these conflicts and related, potential human rights violations, translating into an incentive to conduct more analyses. If this is the case, it is especially unfortunate since RS data holds the potential to counteract the attention economy of media coverage with its worldwide geographical reach (Defries and Townsend, 1999).

Our corpus review revealed that the use of RS data in human rights studies remains underdeveloped, specifically as it often remains unclear what human rights violations were investigated and in what manner. Similar results on the use of RS and peace and security studies were reported by Avtar et al. (2021). The broadly defined selection criteria used to construct our corpus (Figure 1) may have led to the inclusion of papers that aimed to understand human rights violations only as by-products of armed conflict, not as phenomena apt to be the designated foci of research in their own right. Nevertheless, our review allows us to distil what kind of human rights-specific knowledge is and is not produced in relation to conflict areas, and where satellite data offer an opportunity to understand developments on the ground. For example, Figure 9 reveals that most empirical studies focused on Damage Assessment and Land Use and Land Cover Change analyses. Although the diversity of human rights aspects in these categories hindered our ability to classify them succinctly, Figure 9 nevertheless highlights the production of particular, conflict-related knowledge; everything that is detectable or measurable with digital sensors can be expressed in numbers, such as changes in infrastructures, building destruction, environmental changes and fire patterns. Our findings partially converge with Sticher et al. (2023), who points out two main foci in monitoring conflicts: first, damaged or destroyed buildings, and second, forced displacement, documented via population flows or refugee dwelling structures.

Furthermore, the extent to which legal implications were considered when RS observations were made hinged on the organisation conducting the research. Researchers operating outside of academia, for example, were more prone to take IHL or HRL into account. For instance, multiple empirical studies, collaboratively conducted by the AAAS and other NGOs, were merged into one extensive report that linked several violations (AAAS). Following this first report, the AAAS produced a follow-up report in 2018 (AAAS), explaining that one among several challenges for accountability institutions is to prove widespread, systematic patterns of violations over time. Remotely sensed data, the report continues, can be seen as the solution to this challenge, as RS can document patterns of attacks, changes of destruction, and military movements in inaccessible areas (AAAS). Our results show that academic researchers use more advanced RS analysis methods, as becomes clear when Figure 2 is considered in light of Figure 7. After 2014, when academic researchers started to publish RS research, the use of ML applications increased slightly. Using large datasets, statistical analytics, and ML approaches can help to identify patterns linked with war crimes or other atrocities. On the other hand, non-ML-driven approaches, such as Visual Change Detection, Visual Interpretation, or Triangulation (Figure 8), have mainly been applied by other organisations and NGOs (see Figure 2). Being more qualitative in nature, these non-ML-driven approaches often do not allow for comparable spatial coverage as they are time-intensive and dependent on external circumstances, such as reporting densities of social media channels. At the same time, the studies conducted in this qualitative manner, originating from organisations that are not exclusively academic (Figure 2), are more likely to consider legal implications.

Overall, our review reveals a range of research practices, with some considering legal implications and others not, resulting in varying and often vague human rights knowledge. Concretely, we find that academic researchers who rely on RS data and more quantitative-driven approaches tend not to consider legal implications, whereas other and NGOs, predominantly relying on more qualitative approaches when including satellite data in their work, consider IHL and, at times, HRL.

Critical perspectives on RS research point out that political and economic relations, revealed by national agreements (Zerbini and Fradley, 2018), data access denials (Sue, 2023) or economic costs (Bennett et al., 2022b). Those dependencies impact data analysis practices, specifically in the geopolitically sensitive context of armed conflict (Raymond 2014). Even so, most research in our review did not declare political and economic dependencies but instead elaborated on other types of data limitations. Figure 11 shows which studies disclose data availability and economics as challenges. Some researchers acknowledge (Figure 6) special deals they have with commercial satellite providers. Though such companies advertise the use of open data for NGOs and researchers at universities (Planet, 2024), they act as gatekeepers, causing data accessibility issues. For example, as a reaction to the Gaza-Israel conflict starting in 2023, Planet often did not offer access to its data (Tani, 2023). While the problems that come with the described dependencies are well-known, Figure 6 reveals a continued and firm reliance on satellite data from commercial data providers, at least until 2015.

In the reviewed empirical studies, there is a tendency towards the use of very high and high spatial resolution (Figure 5), which enhances the risk of being dependent on commercial providers (Figure 6). Such dependencies can also impact the independence of human rights work. Since most of the commercial satellites are tasked satellites, only data for specific areas of interest is collected; in other words, “tasking is ordering” (Skywatch, 2023). To task a satellite, the satellite provider is sent the coordinates of an area of interest, a time frame, and an interval. However, in the specific context of conflicts, the necessity to task facilitates illustrative over comprehensive documentation: since tasking typically takes place for verification purposes after specific (and often particularly egregious) conflict events become known, the resultant coverage constitutes a probe into conflict events rather than their systematic documentation. Bennett et al. (2022b) attempt to counteract economic and political dependencies during conflict times and improve conflict monitoring, proposing to leverage VHR imagery by archiving it for research and humanitarian purposes, creating standardised conflict-wide mosaics for monitoring, and establishing a sustainably funded public satellite program. The role of commercial satellite companies is further subject of debate, particularly with regard to how these companies assist the military and, therefore, actors potentially violating IHL (Decker, 2023). This concern about commercial satellite providers in times of conflict arises since these companies provide services to both combatants as well as civilians and NGOs (Dunlap, 2023). van Benthem (2023) points out that technology companies increased their support for Ukraine after Russia’s full-scale invasion starting in February 2022, during which these actors assisted the military effort of a party to the conflict. Such assistance, however, raises questions with regard to the status and protection of private sector employees and company infrastructure under the law of armed conflict. Dunlap (2021) reflects on the assistance of satellite companies in Ukraine that provide photo intelligence to armed actors, contemplating in what ways these practices are consistent with international law. Commercial satellite providers fulfil multiple functions in the war: On the one hand, they support armed actors who commit violations, on the other hand, they provide actors with data that documents the violations committed by the armed actors. This example underlines the need for more socio-technical data perspectives, also for satellite data to be admissible as evidence in legal processes concerning alleged human rights violations in conflict zones. Additional uncertainties about the provenance of data from commercial and state sources, and a lack of transparency regarding data acquisition, pre- and post-processing pipelines, create potential differences in the reliability, trustworthiness, and admissibility of data between commercial providers and state agencies. Clarke and Kendall (2019) point out that for digital data to fall within the evidentiary jurisdiction, there is also a need to assess the socio-political context of these digital platforms to understand better how digital data, such as satellite data, can gain evidentiary power. These contexts are yet to be fully explored and are the subject of a current investigation by the authors.

Overall, we find that political and economic dependencies are rarely discussed or analysed in the empirical studies. However, by extracting what challenges in data availability were described in the studies, we were able to trace some of these dependencies. We, therefore, assert that although not declared, dependencies still play an essential role. Conflicts are not unfolding in isolation from the rest of the world but are occurring in a political and economic context. As elaborated on above, the same is true for the production of RS data. We see a need to make the dependencies of RS data on political and economic circumstances more visible and transparent to increase the admissibility of RS research for accountability purposes.

Visual interpretation and visual change detection, both based on visual interpretation, were the leading methods used in the empirical studies (Figure 8). Weizman and Weizman (2014, p. 6) explain that, when the event that constituted the action is not captured, visual change detection can establish “a direct line of causality between a singular action and a unique effect.” Therefore, they urge researchers to “fill the gap between the two images with a narrative” (Weizman and Weizman, 2014, p. 6). Several scholars point out that interpreting the visual characteristics of satellite data also comes with risks (Clarke and Kendall, 2019; Herscher, 2014; Lyons, 2012). Among these, the subjective character of visual analysis is the most significant (van Wyk, 2019). Social, political and personal factors can shape the subjectivity (Clarke and Kendall, 2019), and caution is specifically needed when the complex environments depicted in the images are unfamiliar to the researcher (Lyons, 2012). One of the most prominent examples of subjective visual interpretation of satellite imagery in a geopolitical context is the case of Powell’s presentation at the Security Council to press a case against Iraq in 2003. Offering aerial surveillance imagery as evidence, Powell accused Iraq of threatening global order by possessing weapons of mass destruction in violation of UN resolutions. Months later, after the US invasion of Iraq, these accusations were declared false (UN News, 2024). Herscher (2014) explains that what was read into these satellite images was what the US administration was most afraid of. Clarke and Kendall (2019) contend that the aerial evidence used in this example appears to speak for itself, also called the “evidence effect” of visual data or, in other words, “seeing is believing” (Van Wyk, 2019, p. 31). An additional problem with the subjectivity of visual interpretation is the limitation that satellite imagery cannot capture all incidents; for example, crimes such as gender-based violations remain “invisible” (Koenig and Egan, 2021). Even though visual methods abounded in the empirical studies (Figure 7), only a handful of researchers declared human challenges (Figure 10). The failure to reflect on the subjectivity of visual interpretation appears particularly problematic because in the given context of difficult-to-access war zones, RS data, is conceived of as a safe and objective alternative to traditional witness testimony (Marx and Loboda, 2013; Quinn et al., 2018), yet may be subjective, unreliable, flawed, or manipulated (Sticher et al., 2023).

Technological development and enhanced spatial resolution of satellite imagery over time may explain why visual detection has become so prominent; the improvement in spatial resolution allows non-experts to ease the use of RS data as the visual characteristics are fine-grained, containing increasing amounts of detail, and thereby enabling visual interpretation. Figure 2 shows that until 2013, most of the empirical studies were conducted by researchers from the Other and NGO category using very high spatial resolution data (Figure 5), while starting from 2013 onwards, more medium and low-spatial resolution data were applied. This was precisely the time when academic RS researchers, hence experts, entered the field (Figure 2). A AAAS report from 2016 (AAAS) demonstrates that human rights advocates use geospatial technologies in their work. According to AAAS and the United Nations Operational Satellite Applications Program (UNOSAT), progress in this area was made by applying their expertise to select human rights documentation projects in partnership with satellite imagery providers, advocacy organisations, and funders (AAAS). The researchers’ decision to use satellite imagery can be explained by data collection on the ground being arguably expensive, time-consuming, and potentially dangerous (Lavers et al., 2009). Yet, geospatial data are no panacea: If used by researchers with limited expertise, such as human rights advocates, errors are likely to be introduced (DeNicola, 2008). DeNicola (2008) offers a nuanced perspective on the recent inclusion of less experienced users, which he argues can diminish public confidence in RS imagery as a source of knowledge. He further explains that on the one hand, new users with limited expertise in satellite RS may produce inaccurate analyses, on the other hand, these new users are often skilled at gaining public attention.

While human biases are present in machine processes, data collection, and the interpretation of ML results, these biases materialise at levels different from the levels at which biases in human visual image interpretation manifest. Of the studies comprised in our corpus, only two used an unsupervised approach, whereas the rest used supervised algorithms (Figure 7). Supervised machine approaches, in addition to the human interaction, necessitate further human interaction, namely in labelling the training dataset. The interaction of humans with machines has been extensively discussed in other research areas where ML approaches are increasingly being used (Pugliese et al., 2021). Figure 7 shows, however, that a comparable rise cannot be observed in human rights-related research. The algorithm types used are diverse (Figure 8), which can be explained by the diversity of researched human rights characteristics (Figure 9).

Overall, minimising the risk of human misinterpretation from afar hinges on new methodological developments, and ML approaches can offer one such opportunity. However, researchers using data-driven approaches of this kind should never underestimate the continued need for human interaction. Unsupervised and supervised approaches are not immune to economic and political dependencies and, hence, always need to be considered in context.

The high publication rate of academic researchers starting in 2013 (Figure 2) can be interpreted as the academic RS community understanding the importance of digital human rights research. However, this involvement of academic researchers increased the risk of new methodological limitations caused by the uncertainties within this research group’s validation methodologies. We observe a change in the validation methods (Figure 10): Until 2013, when researchers from the category Other & NGO were most active (Figure 2), RS information was used to validate pre-existing human rights claims (for example, from NGO networks on the ground). Following a shift in the prevalence of organisations around 2013, alongside human rights researchers, academic researchers entered the field. From that point on, a change is evident: RS information was no longer confined to validation but was instead used as a source for getting first insights (Figure 10). The RS analyses in recent studies were mainly validated with the help of remote data, for example, by visually interpreting a change over time with very high-resolution satellite data. Other analyses use other sources, for instance, UNOSAT’s visual damage data, news reports, reports, or statistical data. In addition to the limitations of visual detection discussed above, the reliance on news reports and other sources from conflict areas is potentially problematic, too, as it comes with a high risk of forgery (Tokarczyk, 2023).

Some researchers stated that validation was impossible, whereas others refrained from disclosing any validation mechanisms (Figure 10). But validation is a critical component in an analysis cycle (Loew et al., 2017). Validation is “the consistency check of the RS data with reference measurements which are assumed to be representative of the truth, at least within their own reported uncertainties” (Loew et al., 2017, p. 780). As Leaning (2010, p. 202) points out, “in crisis areas, information is not prospectively or routinely gathered.” This illustrates the complexity of conflict documentation–and yet, we want to emphasise that data quality should not be compromised even if information is lacking. The above-described validation mechanism changes raise concerns about whether current practices of using visual image interpretation data or user-generated data reliably reveal the truth to the extent that remote results can be validated based on these practices. An alternative to the current practice of validation could be the information triangulation process described by Trager et al. (2022), who used an internal validation process with data from within their empirical RS study and triangulated this process externally with a literature review of known, publicly available surveys and interviews. This shows that alternative validation methodologies, when accessibility to the area of interest is limited, are possible. We emphasise the need for alternative robust validation methodologies beyond those used in our corpus.

The establishment of truth and knowledge varies in different epistemic communities (Figure 2), and so does their outcome. We acknowledge this diversity as a strength in our corpus, while also recognising the difficulty of disciplinary exchange when different documentation and transparency standards are in place. The review process (Figure 1) demonstrated that only a small number of articles met the criteria for linking the RS work with human rights work, but research beyond the selected corpus, shows potential to reveal human rights violations. For example, RS reflectance studies of buried animal corpses could further aid the detection of human mass graves (Molina et al., 2022). Also, patterns in the natural environment in times of conflict, detected with RS, can raise questions of responsibility. For example, when armies interfere in agricultural production or the environment itself is used as a weapon (Negash et al., 2023). Long-term conflict monitoring with RS can also present new insights about the duration of conflict events and their impacts, which is relevant for questions of human rights accountability. Conflict events can affect food production and result in long-term changes in land management (Demissie et al., 2022), and the impacts on ecosystem services often go beyond the duration of conflict events (Negash et al., 2023). Linking environmental observations with human rights can therefore offer new pathways (Zwijnenburg and Ballinger, 2023).

Such a collaboration can offer valuable opportunities but also raises new challenge; new technological data and tools, such as RS, create and shape new forms of knowledge (McPherson et al., 2019). This can result in unequal accessibility and raises questions about the ethical use of RS. Levinger (2009) reviewed two projects focused on genocide prevention, where non-experts could utilise satellite data to document the destruction of villages in an attempt to prevent future attacks in the Darfur region of Sudan. He advises that such data practices are not meant for the general public but rather for policy practitioners due to open ethical questions. Ethical questions are of particular relevance when satellite RS data are applied to understand incidents without physical access to the area of interest (Harris, 2013). Conflicts are embedded in contexts, and these contexts need to be well understood before analysis (Miall, 2007). It is important to empower those affected (Bennett et al., 2024), collaborative work between RS researchers and local stakeholders can enable context sensitive research. Overall, we see a need for more collaborative research beyond the traditional epistemic communities, to unite insights from the fields of human rights and RS.