From Space to the Courtroom: AI Enhanced Satellite Imagery and the Future of Accountability

[Inshira Faliq is a legal fellow at the Fénix Foundation, a Netherlands-based non-profit organisation with a mission to leverage technology to support international justice, peace, and accountability]

Introduction 

Satellite imagery has long been important for protecting and promoting human rights. It provides visual insights that shed light on abuses and violations often hidden due to ongoing conflict or restrictions imposed by governments. By giving human rights and media organisations critical insights into indiscriminate aerial bombardments, attacks on civilian areas, and accurate details about the locations, scale, and conditions of displaced persons in camps, along with assistance in the identification of mass graves and the protection of the environment, satellite data has proven to be an indispensable source of information in the digital age. 

Satellite images are also part of a growing body of digitally derived evidence (DDE) used in international courts and tribunals to prosecute international crimes. The first reported use of satellite data in international criminal proceedings was in 2004 during the Srebrenica trials before the International Criminal Tribunal for the Former Yugoslavia (ICTY) to prosecute Bosnian Serb generals Krstić and Mladic. Since then, satellite imagery has been used as a crucial piece of evidence in many cases before the International Criminal Court (ICC) and other international accountability mechanisms. While still an evolving area of law in many jurisdictions, satellite imagery has also increasingly been used in domestic prosecutions in recent years (also here). 

Recent advances in artificial intelligence (AI) are expected to transform the capabilities of earth observation satellites significantly, pushing the boundaries of how satellite imagery can support climate action and disaster response. Similar or comparable approaches could also be relied upon for human rights monitoring and accountability efforts. This rapidly evolving field holds the potential to reshape the future of justice, peace, and accountability. While AI opens numerous possibilities for the future use of earth observation data, this post explores one key area: enhancements in the clarity and visibility of satellite imagery thanks to advancements in AI-driven satellite technology. After describing this development, the post analyses the implications of such advancements for human rights and accountability purposes.

The Enhancement of Clarity and Visibility of Satellite Imagery Using AI 

Satellite imagery has revolutionised how human rights violations are monitored, particularly in conflict zones where on-the-ground access is restricted. High-resolution satellite imagery is especially valuable in this context due to its high potential probative value—i.e. its ability to prove an issue in criminal proceedings. However, high-resolution images are expensive to obtain. The cost of satellite imagery is largely determined by its spatial resolution, with higher-resolution images commanding higher prices. Further, processing smaller sections of raw satellite data into usable products is often not profitable for private companies, which means that most providers have a minimum order requirement. Typically, users must purchase imagery covering an area of 25 to 50 square kilometers, even if their specific needs are much smaller, making access to satellite imagery very expensive. Low-resolution images are more affordable (and often free). However, they offer limited detail and their potential probative value is therefore significantly reduced. 

In some instances, the challenge is not the cost of high-resolution satellite imagery but rather its availability. There are several reasons why only low-resolution images might be available for a specific region though we’ll focus on the two most common reasons. First, satellite coverage is uneven worldwide. Certain areas, such as urban centres or regions of strategic economic importance, benefit from higher satellite coverage and frequent image captures. In contrast, rural, economically marginalised, or geopolitically less significant areas often have limited satellite coverage, resulting in lower-resolution imagery. Second, cloud cover is a persistent challenge in satellite image capture. Clouds—and even their shadows—can obstruct the visibility of the land surface, rendering the imagery incomplete or unusable. To address this issue, composite images are often created by combining multiple satellite images of the same location over time to produce a clearer representation of the area. However, the presence of clouds can still create significant gaps in the data, complicating efforts to study or monitor changes on the ground.

AI-driven super-resolution (SR) models offer a promising solution for the challenges of availability and affordability. These models are expected to drastically improve the visual quality of satellite imagery, creating high-definition (HD) images of earth surfaces for much lower costs. SR can be explained as ‘the process of enhancing the resolution of images through AI’. SR for satellite imagery is primarily a post-processing technique applied to images after they have been captured and transmitted to earth. Once they have been transmitted to earth, advanced image processing algorithms, often based on machine learning or deep learning techniques, are applied to these images. These algorithms work to enhance the visual quality and details of the original image, creating a higher resolution output. 

This process can increase the spatial resolution of satellite images, effectively enhancing the level of detail visible in the image and making features that may have been indistinct or blurry in the original low-resolution image more precise and defined. This process can make previously blurry or indistinct features—such as military installations, mass displacements, or destroyed villages—more visible. Although SR models were not originally designed for cloud removal in satellite imagery, they have, in certain instances, been applied successfully to remove cloud cover, demonstrating potential in this area. For human rights organisations, this advancement means improved monitoring capabilities at a lower cost. For criminal accountability actors, it could mean an improvement in the accessibility and potential probative value of satellite imagery evidence.

Challenges to Evidentiary Integrity

The use of AI-enhanced satellite imagery in international criminal proceedings raises complex evidentiary questions. Legal systems around the world, both domestic and international, are struggling to respond to the use of AI-generated or AI-enhanced evidence in court. 

Given that satellite imagery is now a well-established form of digital evidence, guidelines are available for its use, at least at the international level. The ‘Leiden Guidelines on the Use of Digitally Derived Evidence in International Criminal Courts and Tribunals’ ( also known simply as the Leiden Guidelines) provide ‘elements which should be considered before submitting DDE to an international criminal court or tribunal’, focusing on the ICC Rome Statute’s three admissibility requirements: relevance, probative value, and prejudicial effect. 

However, rooted in pre-AI case law, Leiden Guidelines may no longer adequately address the novel challenges posed by AI technologies. For instance, Leiden Guideline C6 considers satellite imagery authentic and reliable if corroborated by an expert or witness. At the same time, Leiden Guideline C5 notes that manipulating or distorting satellite images does not necessarily preclude their admissibility. Additionally, according to the guidelines, the absence of information about the method used to create the satellite imagery does not automatically diminish its probative value or evidentiary weight (see Leiden Guideline C5). While these provisions align with traditional methods of obtaining satellite imagery, the Leiden Guidelines do not account for the new challenges introduced by AI-enhanced imagery.

AI-enhanced technologies, in this instance the SR model that increases the clarity of the satellite image, can fundamentally alter the way satellite imagery is processed. For example, in increasing the clarity of the image AI models may ‘fill in’ the image by artificially reconstructing details in order to deliver an image with more refined clarity. This essentially means that the AI inputs details into the satellite image that were previously not present. Such alterations could result in images that fail to accurately represent on-the-ground realities, raising questions about their authenticity in legal contexts. Further, current SR models may sometimes struggle to differentiate between different types of objects, such as water bodies, building areas, and other complex structures, potentially generating images that are susceptible to misinterpretation. 

Biases in AI training datasets further compound these challenges. As mentioned previously, certain regions of strategic or economic significance—such as urban centres, resource-rich areas, or those with geopolitical importance—typically benefit from higher satellite coverage. This could result in more frequent and higher-quality satellite imagery from these areas being used to train AI models. In contrast, remote, rural, or economically marginalised regions frequently experience sparse or lower-resolution satellite coverage. For instance, while there is abundant satellite imagery of oil installations and specific urban areas in northern Libya, there is comparatively little coverage of rural areas in South Sudan, largely due to the limited commercial interest in acquiring such images. Such disparity could, therefore, create an imbalance leading to the underrepresentation of these areas in AI training datasets. When datasets disproportionately emphasise certain geographic regions or features, AI-processed satellite images may overrepresent urban structures while underrepresenting terrains or rural landscapes. Such biases can lead to potential distortions, particularly in human rights investigations. For instance, critical details—such as mass graves, displaced populations, or damaged infrastructure—may be misrepresented or missed entirely. In conflict zones, where biases inherent in AI models could reflect the interests of states, private companies, or organizations, these distortions could have serious consequences for accountability efforts.

These limitations are particularly problematic when satellite imagery is introduced as evidence in international courts and tribunals, where the integrity and reliability of evidence are paramount. As a result, the Leiden Guidelines, which were developed without accounting for the AI dimension of satellite imagery, may not apply in this new context. If and when AI-enhanced satellite imagery is used in international criminal cases, international courts and tribunals will likely adapt their approach to such evidence, necessitating a reassessment of evidentiary standards and practices. To remain relevant and effective, the Leiden Guidelines must also evolve, incorporating considerations specific to AI-enhanced satellite imagery. This evolution is critical to strike a balance between leveraging technological advancements and safeguarding the integrity of evidence in international criminal proceedings.

Conclusion 

The integration of AI into satellite technology offers transformative opportunities for human rights monitoring and accountability but also presents significant challenges. On the one hand, AI-driven advancements, such as enhanced resolution, provide unprecedented precision and speed in identifying human rights violations, including military movements, mass displacements, and environmental destruction. These innovations can provide critical evidence to hold perpetrators accountable and strengthen global justice efforts. On the other hand, the introduction of AI into satellite imaging raises significant concerns about the reliability and authenticity of these images. Algorithmic biases, the inadvertent creation of artificial details, and the risk of misinterpreting ground realities complicate the evidentiary value of AI-enhanced satellite imagery. 

To address these complexities, there is a need for clear, comprehensive, and updated guidelines specifically tailored to the use of AI-enhanced satellite imagery in international criminal proceedings. As courts in future may increasingly rely on this technology, judicial directives that explicitly outline admissibility criteria and usage protocols are essential to prevent legal uncertainties and maintain evidentiary integrity. While the Leiden Guidelines provide a foundational framework for the use of satellite imagery in international courts, as new case law emerges, they must evolve to reflect the unique challenges and risks introduced by AI-driven advancements. Incorporating practical solutions for mitigating bias, verifying authenticity, and ensuring equitable representation in AI training datasets is essential to maintain the credibility of this evidence. Proactively updating these frameworks will help bridge the gap between cutting-edge technology and traditional evidentiary standards, ensuring that AI-powered satellites enhance justice rather than compromise it. With robust safeguards and a strong legal framework, AI-powered satellite technology has the potential to serve as a revolutionary tool to uphold justice, peace, and accountability in the digital age. 

Photo attribution: Satellite image created by ChatGPT