Thanks to Lawfare for allowing me to write about my empirical piece: The U.S. Practice of Collateral Damage Estimation and Mitigation (and thanks to Ben for his kind words about the paper). I was motivated to write about this topic because I sensed a major disconnect between the descriptions of the targeted killing process offered by commentators and what I knew to be the actual practice of the U.S. military in air to ground operations, in particular the rigorous steps followed to avoid civilian casualties through both the collateral damage methodology (CDM) and fairly protective Rules of Engagement (ROE). The paper is based on field interviews, documents released in litigation, training materials, official policy guidance released through FOIA requests, some WikiLeaked documents, press accounts and the standard fare of law review articles and other scholarly sources. The piece is empirical and descriptive; I take up the normative implications of my findings in a separate article entitled Collateral Damage and Accountability, which is not yet available in SSRN.
Much of the commentary about air launched targeted killing--especially the commentary that focuses on a "video game" style of warfare with unaccountable geographically remote pilots dropping bombs at their own discretion--simply does not describe the reality of current combat operations (I directly address the false claims about targeted killing in a forthcoming book chapter). To highlight one example of the reality I describe versus commentary we typically read, just consider the fact that in Afghanistan since at least June 2009, all air-to-ground operations are pre-planned operations unless troops are in an emergency situation requiring close air support (CAS), close combat attack (CCA) or the pilot is acting in self-defense. In both CAS and CCA in Afghanistan, the pilot may not deploy a weapon without ground commander direction, usually through a Joint Terminal Attack Controller (JTAC) (a JTAC is a person who accompanies ground forces and is specifically trained to direct accurate close air support to engage enemy targets while reducing collateral damage and civilian casualties). The pilot’s only discretion in current operations is to decide not to release a weapon, in other words the ground commander owns the battlespace not the drone pilot. Furthermore, Air Force leaders repeatedly emphasize to their pilots that they will not be disciplined for returning to base with all of their bombs on their plane. Air Force leadership will even support the decision of pilots not employ a weapon, even if that decision directly contravenes the orders of the ground commander. This reality is a far cry from the free fire zone of “video game” warriors described by many drone critics.
Because targeted killing operations by UAV are not CAS or CCA, they are pre-planned operations, and as such must be subjected to the military's rigorous collateral damage methodology. That methodology is grounded in scientific evidence derived from research, experiments, history, and battlefield intelligence, and is designed to adapt to time-critical events. The CDM is a planning tool that assists commanders in mitigating unintended or incidental damage or injury to civilians, property and the environment and aids them in assessing proportionality and in weighing risks to collateral concerns. In the context of targeted killing, the CDM takes into account every conventional weapon a UAV could carry.
The first step in the military’s collateral damage estimation process requires military commanders and their subordinates to ensure that they can positively identify, with reasonable certainty, that the object or person they want to affect is a legitimate military target (in the case of persons, one who is directly participating in hostilities). In targeted killing operations, the process of positively identifying a target means that commanders and their subordinates are focusing principally on the identity of the target. This is an intelligence-heavy task that relies on the collective effort of the intelligence community (both military and civilian) to vet and ensure the validity of the target in accordance with IHL and the ROE. Before engaging in an operation, military personnel must inform a commander (or “strike approval authority”) of the assumptions and uncertainties associated with information provided for the operation, including the time sensitive nature of any intelligence relied upon. In practice, what this means is that if positive identification of a target fails, no operation will take place. This is so because in U.S. practice, positive identification satisfies the requirement of ensuring that a target has the status of a legitimate military objective. When doubt arises as to whether an object holds civilian status, there exists a presumption that it does, hence the requirement of positive identification in U.S. operations.
Commentators analyzing military operations generally discuss positive identification (distinction), and then turn their attention to the question of proportionality and balancing of collateral damage against military advantage. The paper differs from the general approach followed by commentators because in the actual practice of modern operations there are a series of scientifically grounded mitigation steps that commanders undertake prior to engaging in any proportionality balancing.
The mitigation steps followed by the military are highly technical, however the best way to understand them is to recognize that the U.S. has an extensive database of weapons effects for nearly every conventional non-direct fire weapon in the U.S. military’s arsenal. That database takes into account the blast, fragmentation and debris patterns of weapons when they explode. This information, combined with knowledge about speed, direction of weapon employment, fusing techniques, and information about terrain and structures can be modeled in a way that allows for very precise estimates of weapons effects. After positively identifying a target, the weapons effect data is converted into a collateral hazard area, and trained specialists will determine whether or not there are any collateral concerns within that area. If so, mitigation techniques must be employed to reduce the likelihood of harm to collateral concerns. For example, the military can predict that a 1,000 pound bomb dropped in a certain place within a structure will kill everyone within Room 1, and everyone in Room 2, and not harm anyone in Room 3. A mitigation technique might be to use a 500 pound bomb, or bury the bomb in the ground prior to detonation, or detonate the bomb in the corner of Room 1, at a point farthest from Room 2. Any of these mitigation techniques might ensure with a high level of certainty that the persons in Room 2 are unharmed.
Of course, in light of this precision, the biggest challenge is an intelligence challenge. If the intelligence indicates that the lawful targets are in Room 1, but the intelligence is wrong or has changed and the civilians are moved to Room 1 and the lawful targets moved to Room 3, all of the precision and care in the world will not prevent a civilian casualty incident. (As an aside this was a lesson learned by the military in Afghanistan. One of the JSOTF’s observed that civilian casualty incidents were more likely during CAS strikes on compounds. In response, the commander issued a directive limiting the circumstances under which air power could be used against compounds and the number of civilian casualty incidents dramatically declined. This lesson learned was later expanded to all of ISAF as part of a subsequent tactical directive). Despite the intelligence challenges, the overall record for CENTCOM operations where the CDM was followed has been remarkably good (perhaps by reference lending credence to the CIA’s claims of no civilian casualties, especially if the CIA follows the CDM). Specifically, for CENTCOM:
- Civilian casualties occurred in less than 1% of the pre-planned operations that followed the CDM.
- Within that 1%,
- 70% of the time it was due to failed “positive identification” of a target.
- 22% of the time it was attributable to weapons malfunction.
- In a mere 8% of collateral damage incidents was it attributable to proportionality balancing --e.g. a conscious decision that anticipated military advantage outweighed collateral damage.
In short, in the rare instances when civilian casualties do occur, it’s usually not a matter of consciously choosing to accept civilian casualties as “proportional” and necessary collateral damage; rather it is truly unintentional conduct. Assuming my statistics and description are accurate, the CDM is a powerful tool for minimizing and perhaps eliminating civilian casualties, however it is only a tool and must be understood in the context of a broader system of accountability.
The CDM alone does not tell the full picture of the military’s process as it relates to collateral damage, a point I make in the final third of the paper. To fully understand how the military prevents or mitigates risk to civilians in targeted killings, one must also understand the Rules of Engagement. A simple way of describing the ROE is to think of them as legally binding administrative regulations (violations of the ROE can carry criminal penalties) that supplement the Law of Armed Conflict. Commentators seem to know a lot about the Law of Armed Conflict, but many of them write as if the ROE do not exist, this is a serious omission because the ROE define the circumstances in which certain categories of targets can be attacked, and most importantly the threshold points at which decisions regarding collateral damage must be elevated to a higher level in the chain of command.
In the paper I provide two examples of how the ROE in Iraq circa 2005 and Afghanistan circa 2010 calibrate these decision points. I describe in detail the briefing process for what the ROE term “high collateral damage” targets and other sensitive targets requiring high levels of command approval. I highlight how in Iraq pre-planned strikes that were predicted to result in 30 or more civilian casualties had to be authorized at the Secretary of Defense level or higher, levels below 30 were reserved for senior commanders or on-scene commanders (as specified in the ROE). Contrast this with operations in Afghanistan circa 2010, where if even one civilian casualty was expected in a pre-planned strike that strike would have to be authorized by the Secretary of Defense or higher (although some reports indicate that this authority has since been delegated to the CENTCOM commander). Taken together, the CDM process provides predictions about likely effects, and the ROE specifies the decision authority necessary to authorize certain strikes. The process, as I explain it in the paper, is far more detailed and accountable than that which has been described by most commentators.
I should caution that this blog post differs a bit from the article. I’m making the point here that most critics have largely ignored the levels of accountability and procedural care I describe in the paper, I don’t make that claim in the article mostly because I’m limiting it to an empirical description of the process. I do think it’s important to highlight that many commentators have not fairly described the military’s process despite the fact that most of the documents I rely upon were available on the internet, were released to the ACLU in the al Aulaqi litigation, or were published by WikiLeaks (although synthesizing them and supplementing them with interviews was a big challenge). In some respects the military can be faulted for not adequately explaining their very defensible procedures to the public. In any case, irrespective of your opinion about the merits of targeted killing, I’m hopeful my paper provides the foundation necessary for scholars and commentators to build upon, and I hope it serves as a helpful corrective to the descriptions of state practice currently circulating in public commentary. One final note, the paper is still a draft and I welcome comments and corrections. I can be contacted here.