The Yemen Project: Open Source Investigations and the Law of War

Short of a public confession by a state to unlawful conduct, what does it take to prove a violation of the international law governing aerial bombardment? The Saudi-led coalition’s bombing campaign in Yemen is now the subject of a major Open Source Intelligence (OSINT) investigation by Bellingcat, an investigative journalism organisation. The fruits of this investigation, released in a dedicated website, provide insight into bombing patterns. These bombings, alongside the blockade on the ports of Aden and Al-Hudaydah and ground operations, have often caused grave harm to civilians, including the destruction of essential civilian infrastructure and specially protected objects. They also raise serious questions about compliance with international humanitarian law (IHL), as recently highlighted in the September 3 Report of the Group of Eminent International and Regional Experts on Yemen (Group of Experts), released by the UN Human Rights Council, which found an array of violations by all parties to the conflict in Yemen.

The Bellingcat investigation can contribute to the legal analysis of the Saudi-led coalition’s air war, allowing a granular analysis of practices that have been alleged to be unlawful. The evolving scale of the analyses, as more are released over time, may be unprecedented. Yet even with access to more post-strike information than has been available to researchers, legal experts, and NGOs in previous (and especially ongoing) armed conflicts, questions remain regarding how those outside government can “prove” violations of the law of armed conflict. Below is an initial reflection on the legal significance of this investigation, in anticipation of a methodical engagement in the coming year by the Global Legal Action Network (GLAN), of which I am a co-founder.

Open Source Investigations and the Yemen Project

In January and February 2019, Bellingcat and GLAN, supported by the New York Times and Channel 4, organised a four day ‘hackathon’ at the King’s College Policy Institute in London. Over thirty investigators, experts in OSINT, followed a traceable methodology of online discovery and preservation to begin investigating dozens of selected air strikes that took place in Yemen, starting from the beginning of the war in 2015. Since then, Bellingcat’s analysts have been working to gradually release reports of investigations into individual incidents. This month, a collection of twenty incidents, covering air strikes that led to presumptively significant (more on that later) civilian casualties, was released. These include air strikes on busy markets (Zabid, al Fayoush, Heifan Junction, Aahim and Al Khamis), a communal vegetable-washing facility, a prison (reportedly for petty criminals), residential buildings in a busy neighborhood and in a village, a bus, and a wedding.

In an era of sophisticated misinformation, there are good reasons to be skeptical, even wary, of intelligence gathered from publicly available sources. Its many sources, which include images and videos released by news agencies as well as those uploaded by individuals, vary in reliability. While satellite imagery, image quality permitting, is a generally accepted as reliable public source of visual information, “user-generated content” (UGC) includes videos and photographs uploaded by unknown, and potentially partisan, sources. Only a rigorous methodology of discovery and verification, including a replicable analysis whose steps can be retraced by any person, can ensure UGC’s reliability and inferential value.

Moreover, OSINT should not be perceived as substituting for primary evidence: witness testimony, bomb fragments, verified photographs and videos of the scene received from witnesses. In addition, OSINT’s revelatory effect should not be seen to travel the physical and moral distance between the technologically equipped investigator and the theatre of war. It can be tempting, but it would be presumptuous, to think one can fully know the conflict reality through the use of technologically advanced tools.

With these limitations in mind, NGOs and UN fact-finding missions have been using OSINT in their research for some years. This includes the widely reported, detailed findings of the Group of Experts (par. 34), which used open source evidence both as an initial source of information and for corroborative purposes. In addition, on 15 August 2017, the International Criminal Court (ICC) issued an arrest warrant for Mahmoud al-Werfalli, partly relying on verified and geolocated open-source video evidence that appears to show him executing captive prisoners in Libya.

Through OSINT methodologies, we have more evidence of the conduct of hostilities during an ongoing conflict than at any time in modern history. OSINT may constitute a distinct and important evolution in the methods and actors involved in the collection of evidence. It can excavate a treasure trove of visual data. Through techniques such as geolocation and triangulation, it can provide the determination and confirmation of key data points. Even if some tools are expensive (such as some types of satellite imagery), OSINT can be used by independent networks of investigators. At the most basic level, OSINT allows the vetting and confirmation of material that is available, and maybe familiar, to the wider public, through (social) media. Unlike an eye-witness report of a strike by investigators on the ground, this material allows, and is sometimes a necessary prerequisite for, much more detailed and thorough applied legal analysis and potentially legal action. Specifically, OSINT may lead to the discovery of information that can be key to answering distinct factual questions that are often necessary to legal analysis of a targeting operation, such as: whether a given strike was launched from the ground or the air, the precise location of a target, the time of day it was struck, or the type of ammunition used in an air strike.

What We Know from OSINT

The Yemen Project illustrates the potential of OSINT in the context of a sustained and devastating campaign of aerial bombardment. In some cases, OSINT confirms the very occurrence of an air strike, despite official silence, express denial, or attempts at misinformation or justification. Countering denial and misinformation is particularly crucial for investigators responding to the ostensibly independent Saudi-led coalition Joint Incident Assessments Team (JIAT). The JIAT plays a central role in the coalition’s protestations that its war conduct is lawful.

For example, the analysis of the al Fayoush market strike counters the JIAT’s statement that “the mentioned market was not bombed, and the coalition forces bombed two [militia] sites in the same referred date” 7 and 10km away. Alternatively, it may be that an OSINT analysis will conclusively correct the official account of which target was struck, by identifying the precise location of the strike. This may also counter claims as to the legality of target selection, as illustrated in the difference between a weapons depot and a detention facility or prison for petty criminals, in the al Abs strike. In some cases, OSINT can flatly contradict specific factual claims made by the JIAT – such as the claim that the destroyed al Khamis market building was not affected by what the JIAT determined was a bombing of Houthi targets.

In other cases, OSINT will aid the precise determination of the time of the attack. This may be useful in creating an independent verification of the time cited in NGO reporting. For example, in the case of a strike on a market, the identification of the time can provide information regarding the number of people who were likely to have been present at the time of the attack. In the al Fayoush strike analysis, a calculation of the time based on the shadows visible in photographs, with the aid of freely available software, shows the strike occurred in the morning. Conversely, the analysis of the air strike on the al Aahim Triangle, a commercial location including resstaurants and shops in the Hajja province, refers to existing NGO reports and user-generated videos to confirm the attack occurred in the evening. This, during the month of Ramadan, suggests the location was targeted at one of its busiest times. Finally, a combination of temporal and location analysis confirms that the strike in Sanaban, Dhammar province, was in a residential district, at 10pm, during a wedding celebration.

More specifically, temporal determination combined with visual analysis may confirm both the existence of more than one crater and/or the temporal proximity of two or more air strikes. The latter could suggest the practice of “double tapping,” the practice of attacking a target twice or more within a short period of time, often affecting individuals whose movement was compelled by the first strike, such as first responders. The proximity of the craters and the temporal proximity of the strikes may indicate whether the incident included multiple strikes at different targets or at the same target, or whether the target was hit by a single air strike while there were, in the same timeframe, multiple air strikes in the vicinity. No example is so stark as the air strike on the Office of the Presidency in Sana’a, in which the second strike was captured in a video of a man and a woman trying to rescue a boy from the rubble.

Finally, OSINT can contribute to our understanding of the nature of the victims, to the extent that immediate post-strike visual evidence confirms the presence or absence of armed persons, which could potentially impact the question of civilian status, or the presence of children among the victims.

The Limits of OSINT and the Law’s Thresholds

So we know that air strikes have struck locations at times when large numbers of civilians congregate. We know that they led to massive casualties. We know that the official coalition position, and its purportedly objective investigative unit, sometimes provide erroneous accounts of these same strikes (or in some cases deny that they occurred at all). This may be enough to convince an objective lay observer that civilians and civilian objects were targeted, that the law of armed conflict has been violated, that the violations are serious, and that they may be attributable to decision makers at different levels of the command structure, who may be individually liable. It will be frustrating for those observers, whose good faith and judgement should not be easily dismissed, that, in fact, even the most rigorously-analysed and revelatory analysis derived from OSINT will find it very difficult to meet the law’s thresholds.

In order to find a violation of the fundamental principle of distinction between military and civilian objects, which prohibits launching an attack directed against civilians or civilian objects, at both the level required for a violation of IHL and, even more so, at the level of a war crime, access to information that cannot be adduced through OSINT may be necessary. Especially in the case of aerial bombardment, the distance between the action and its consequences seems to weaken the capacity to draw reasonable inference from the observable results.

For example, even if the analysis of the targeted location and the temporal sequence of the explosions reveals two consecutive strikes, more would be required to conclude this was a deliberate attack on civilians and first responders, as acknowledged in the relevant Bellingcat analyses. The attacking party’s allegations of the presence of fighters at busy markets (as in the al Khamis and Haifan Junction strikes) or the presence of a military target near a civilian location (as in the Bani Qais wedding strike), are very difficult to conclusively disprove. An investigator, if speaking carefully, can therefore only speak in the negative, saying that no military targets or armed men (potentially members of military groups) were discovered in the vicinity through open source evidence. Without access to operational logs, military intelligence, and records of the decisions that could provide an insight into the eye and mind of the military planner or commander, there can only be some space for reasonable inference, vulnerable to counter-claims based on classified information.

Given that a direct attack on a civilian target cannot be clearly inferred from OSINT in most cases, a legal analyst working to determine whether there has been an IHL violation would need to incorporate potential counter-claims. Such a counter-claim by the attacking side may be, for example, that an armed individual was present in the crowd, raising the possibility that the attack was not directly targeting civilians. And yet, even if there was a military target present, the way that the air strike was conducted could be, in effect, indiscriminate. A charge of indiscriminate targeting alleges that either the weapon chosen was inherently incapable of distinguishing between civilians and combatants, or, more commonly, the way in which it was used showed a clear disregard for the presence of civilians and civilian objects. Evidencing such disregard is difficult and is associated with the absence of necessary precautions. These could include the prior scouting of the area; the visibility of individual civilians to the attacker, which could be estimated through knowledge of the specifications of the technology transferred to the coalition, such as the DAMOCLES (specifications) and sniper targetingpods (feed); questions of weaponeering such as the identification of the weapon used -through discovery of fragments (subject to verification that they have not been moved); the calculation of the foreseeable blast area given the particular choice of weapon, such as the lethal radius of up to 360m of the 2,000lb bombs used on the crowded al Khamis market strike; the measurement of the crater; or the analysis of the pattern of the collapse of a building.

Even with all of this information, OSINT would not, however, reveal the extent of the options available to an attacker at the time of the attack or the military necessity calculation behind the choice of weapon. Here, again, as rigorous as the analysis may be, as careful and reasonable as the inference may be, the notoriously broad and flexible standards allowing states to easily claim they have fulfilled their obligations may provide space for even a generic and weak counter-explanation to undermine extensive evidence of what appears to be unlawful conduct.

The difficulty in disproving an attacker’s claims of the presence of a military target inevitably leads analysts to the famously tricky assessment required by the principle of proportionality. Since analysis of OSINT is rarely able to discount the possibility of the presence of a military target in the proximity of civilians or within or around civilian objects, legal analysts will be required to estimate whether the loss of civilian lives was excessive in relation to the military advantage anticipated. Apart from the inherent difficulties in such calculations, these would require information not available to those without privileged access to the decision maker’s records, including what was available to them at the time of the specific attack. Such information could concern the military necessity part of the scale, for example concerning the military value of the fighters allegedly present, in both official terms and in the context of the attackers’ military strategy. Similarly, the questions of precautions discussed above in the context of indiscriminate attacks, and all the contextual knowledge and military necessity parameters, including potentially information regarding the broader concept of operations would likely need to be available to guarantee a conclusive legal analysis.

The matter is even more problematic when it comes to individual liability for war crimes, as reflected in the absence of successful prosecutions. For example, OSINT will likely be deficient in proving knowledge on the part of the attacker, and classified information may be needed to prove key facts, such that even a wealth of rigorous OSINT will remain vulnerable to unverifiable counterclaims. While this does not mean that courts and international institutions are prevented from drawing reasonable inferences, either in relation to overall conduct or in relation to a particular commander’s actions, this may present an insurmountable obstacle to the prosecution of individuals that commanded and planned the attack. The requirements of knowledge and intentionality, especially in the stringent and limiting standards of the Rome Statute of the ICC, mean that having a more holistic understanding of a particular strike through OSINT does not make up for the lack of access to the decision-making process, and the individual decision maker’s mindset at the time of the attack.

Conclusion: Reasons to Further Engage OSINT

Extensive and rigorous OSINT analysed by non-governmental actors (be they journalists, committees of experts associated with intergovernmental bodies or NGOs) aims to identify, describe, visualize, and present the facts – in this instance related to air strikes in the conflict in Yemen. This rich factual data provides an unprecedented opportunity for applied and more precise analysis of the legality of the conduct of hostilities of a particular party to the conflict.

One may still feel, however, a sense of frustration that an exhaustive presentation of an air strike leading to extensive civilian casualties still does not yield a clear legal finding of an IHL violation (or war crime). This is an invitation to do two things: first, to identify ways in which OSINT can be further improved to illuminate and identify legally salient factors, painting a fuller, more granular, and more authoritative picture of the circumstances in which the attack came about and the attacker’s decision making; and second, to consider the ways in which the infamous indeterminacy of the law of war continues to serve, from the construction of fundamental principles to the establishment of individual liability, the privileged position of the attacking party, as long as its decision-making process is supported by even a veneer of military professionalism.


IMAGE: A Saudi female journalist films damage at a market for vehicles on August 27, 2016 in the Saudi border city of Najran, a week after it was struck by a rocket fired from Yemen. (Photo by FAYEZ NURELDINE/AFP/Getty Images)


About the Author(s)

Ioannis Kalpouzos

Lecturer in Law at City Law School, University of London, Visiting Fellow at King’s College London, Visiting Researcher and Lecturer on Law at Harvard Law School, Co-founder of the Global Legal Action Network - Follow him on Twitter (@YannisKalpouzos).