Urban peacekeeping under siege: attacks on African Union peacekeepers in Mogadishu, 2007–2009

Peacekeepers in cities face particular challenges because cities are densely populated and heterogeneous, encompass multiple terrains and fluid features, and host key assets of political, economic and strategic importance. Attacks targeting peacekeepers in cities constitute a recurrent problem, but how do they affect a peace operation’s activities? We theorise the effects of such violence on three outcomes: patrolling and outreach, use of force, and the establishment of new bases. We explore these dynamics by analysing intracity dynamics of violence and operational activity following attacks on African Union (AU) peacekeepers in Mogadishu, Somalia, from initial deployment in 2007 through 2009. We use the geo-referenced UCDP Peacemakers at Risk (PAR) dataset and extend it by coding specific city sub-locations for incidences of violence, allowing us to analyse the spatiality of violence involving peacekeepers in Mogadishu. The evidence suggests that during its first three years, attacks on AMISOM significantly hampered its ability to spread out in the city and operate effectively, but did not evidently alter wider patterns of violence in the city. Despite these challenges, AMISOM managed to fulfil the core element of its mandate: preventing the overthrow of the Somali Transitional Federal Government. ARTICLE HISTORY Received 12 April 2019 Accepted 4 October 2019


Introduction
Many contemporary peace operations are deployed in high-threat environments, where conflict-related violence persists and without comprehensive or stable political agreements. Peacekeeping has also become increasingly urbanised, which brings its own particular challenges. 1 One key challenge, addressed in this article, is how attacks on peacekeepers deployed in urban environments affect their ability to operate. In a number of cases, peace operations have suffered sustained, deliberate attacks in cities, including in Bosnia-Herzegovina (UNPROFOR), Somalia (UNOSOM II and AMISOM), Haiti (MINUSTAH), Mali (MINUSMA), and Central African Republic (MINUSCA). In such contexts, attacks on peacekeepers may curtail their movement and make them less responsive to crisis situations. In extreme cases, the killing of peacekeepers has led some countries to withdraw their contingents from a mission; for example, the US from UNOSOM II in Somalia and Belgium from UNAMIR in Rwanda. 2 A growing literature has theorised and analysed how urban contexts condition and shape armed conflict and peace-making activities. 3 Meanwhile, capitals and major cities are of particularly high strategic importance for peace operations and are usually peacekeepers' point of entry for deployment. Many important peacekeeping activities necessarily concentrate in cities, and establishing security in the capital is central to enabling national authorities to govern, a critical objective for most peace operations. At the same time, however, urban peacekeeping is subject to major challenges. Armed actors are often difficult to distinguish from the urban civilian population, and the same factors that make the city important to securethe concentration of political power, people, and economic and symbolic resourcesalso tend to make it particularly contested. In turn, contested cities often play key roles in wider peace processes; potentially becoming stumbling blocks and flashpoints for renewed armed conflict. 4 Attacks on peacekeepers in these contexts may have major implications for the trajectory of violence and prospects for peacebuilding, in the city and beyond.
We analyse these dynamics with reference to the African Union Mission in Somalia's (AMISOM) deployment of military personnel in Mogadishu. AMISOM is the peace operation that has come under the most sustained urban attacks in the modern era, principally by the militant group, Harakat al-Shabaab (hereafter 'al-Shabaab'). We theorise how attacks may affect peacekeeping troops' operational activities and reach in the city, focusing on three outcomes of interest: peacekeeping patrols and outreach, use of armed force, and the establishment of new bases. We then conduct a plausibility probe focusing on AMISOM's initial years of deployment, [2007][2008][2009]. Our analysis indicates that AMISOM faced a series of specific urban challenges due to al-Shabaab's use of asymmetric tactics and knowledge of the city's terrain and population, impacting on AMISOM's ability to operate and spread out across Mogadishu. In contrast to the more commonly studied United Nations (UN)-led peacekeeping operations, AMISOM deployed without a ceasefire or peace agreement and was tasked with protecting a set of contested transitional government institutions. Yet the implications of AMISOM's experience arguably extend to a broader set of current and likely future peace operations.
Below, we begin by taking account of existing research on peacekeeping and urban violence. Next, we lay out our theoretical expectations about the consequences of attacks on urban peacekeepers and outline our methodology for probing these expectations. In the subsequent section we present our empirical analysis and assess how the empirical findings relate to our expectations. The concluding section summarises and notes avenues for future research.

Urban peacekeeping
Spatial dimensions of armed conflict have long been a topic of research, and the microdynamics of violence have garnered increasing attention. 5 Simultaneously, more peace and conflict research has focused on urban dynamics of violence. Björkdahl describes a shift in the literature whereby 'cities are seen as the new battleground of an increasingly urban world, thus promoting the urbanisation of security and militarism'. 6 This research emphasises how the intersection in cities between density, heterogeneity and large socioeconomic inequalities increases the risk of violence and instability, posing crucial challenges for peacebuilding. 7 However, the dynamics and consequences of attacks on peacekeepers in urban settings remain understudied.
The peacekeeping literature has increasingly addressed the geographic dimensions of operations, 8 including their spatial distribution within host countries and local effects. 9 Recent work suggests that UN peacekeepers tend to deploy to the most dangerous locations where they are needed but with some delay, 10 and that sub-national deployments can deter violence or enforce peace also at the local level. 11 This focus on locallevel dynamics has been facilitated by new disaggregated data on peace operations, 12 including fine-grained data linking peacekeepers to specific sub-national locations. 13 However, the focus has been almost exclusively on UN peacekeeping operations and on variation across different within-country units. Less attention has been devoted to peace operations in urban settings specifically, and their spatial dynamics at the very micro-level. 14 Yet, peace operations typically deploy first to a major city (often a port city) and retain a substantial presence there. In particular, peace operations often have headquarters in the host state's capital city, for political as well as logistical reasons. 15 They require access to infrastructure such as roads in order to fan out in a country, and therefore tend to concentrate around transportation networks. 16 Missions also have to liaise and work with a range of partners to finance and deliver logistical support, which can be especially difficult to procure for peacekeepers operating in contexts of ongoing warfare. 17 The increasing attention to spatial dynamics of peacekeeping has coincided with changing peacekeeping mandates. Traditional UN peacekeeping is premised on not taking sides in an armed conflict. 18 But twenty-first century UN-led and UN-authorised peace operations have exhibited a more assertive approach to impartiality and an increasing concern with supporting the consolidation and extension of state authority, often under the rubric of stabilisation. 19 Assisting a legitimate and self-sustaining government is understood as a necessary part of the peacekeepers' exit strategy. 20 If a peace operation requires a government to support, and a government requires a capital, it follows that capital cities are of critical importance for peacekeepers. However, it also follows that a peace operation's presence in the city will be highly contested when armed actors oppose its mandate, and attacks targeting peacekeepers may serve as a way for armed actors to challenge local power configurations. 21

Consequences of attacks on urban peacekeepers
Peace operations are often mandated to perform multiple and complex tasks in a context of ongoing violence, including stabilisation efforts and even counterinsurgency, presenting particular challenges in urban environments. 22 Notably, fighting in cities can negate many of the advantages modern militaries have over less technologically-advanced adversaries. A city's infrastructure and urban density can become a target of belligerents, but also an important resource that allows armed actors to hide and gain anonymity. 23 Consequently, inhabitants with local knowledge are able to leverage the urban context to outweigh substantial numerical and material forces. 24 The man-made physical terrain of the citybuildings, roads, alleys, markets, subterranean infrastructure etc.affect the ability of peacekeepers to move, communicate, and monitor the surrounding area, as well as provide concealed vantage points for potential attackers. Zones of control and contestation can be closely co-located in the city and borders between them difficult to patrol. 25 Allen also notes 'the pivotal psychological importance of cities as a source of national will and cohesion'. 26 Cities are thus important for their symbolic, not only their material, significance; this is especially true for the capital. The physical manifestations of a contested government, and the peacekeepers that protect these locations, can therefore become key targets of armed groups. Yet few studies have assessed the impacts of attacks on peacekeepers' ability to operate effectively in urban settings.
This article explores these dynamics by theorising that the targeting of peacekeepers deployed in urban environments carries adverse consequences for their operational activities and reach. Further, we expect these consequences will be shaped by the fact that cities are densely populated, heterogeneous, encompass multiple terrains and fluid features, and host key assets of political, economic and strategic importance. 27 The density and heterogeneity of cities implies particular difficulties for peacekeepers to distinguish fighters from civilians, and from the urban infrastructure itself. 28 It also implies that informal institutions and authority structures are often more difficult for outside actors to disentangle than in rural areas. 29 The city's fluidity and permeability allows people to move relatively freely, which creates specific challenges for controlling and predicting armed activity and yields a tactical advantage to armed groups with detailed local knowledge. The city's political, economic and strategic importance increases the risk of attacks intended to make a high psychological impact, and also places a premium on controlling and protecting key locations. These dynamics inform our theorisation of three ways in which attacks may affect urban peacekeepers' operational activities and reach.
First, we expect peace operations that come under attack are likelyat least initiallyto scale back their movements and activities. 30 Regular patrols, notably, constitute a cornerstone of modern peacekeeping, allowing peacekeepers to establish a visible presence to deter violence or de-escalate security situations, observe and interact with local communities. 31 Attacks may curtail the frequency and scope of patrols to reduce peacekeepers' exposure to risk or because peacekeepers lack the equipment or training to operate effectively in non-permissive environments. Given that military operations in cities are particularly complex and resource-intensive, 32 such effects may be particularly salient in the urban context. In cities, it can also be difficult for outsiders to know who controls particular institutions and infrastructures, making peacekeepers reliant on local advice. 33 Yet adaptation in response to attacks may also include limiting outreach activities, making peacekeepers less able to interact with local civilians and challenging their ability to gain communities' trust and access information. 34 Adequate intelligencegathering capabilities, including regular and credible information from local communities and counterparts, is also critical to anticipate and prevent attacks. 35 Second, while attacks on peacekeepers may reduce their ability to operate in a general sense, we also expect them to stimulate an armed response, potentially feeding into escalated violence. Most contemporary peace operations have expansive authorities to use military force in defence of the mandate. 36 However, it is an open question whether the peacekeeping force will outperform local armed actors in terms of capabilities and weaponry. Peacekeepers' responses to attacks may thus risk escalating violence, also as peacekeepers responding to attacks may level less controlled armed responses, which raises the risks of collateral damage. 37 The risk of collateral damage is likely to be greater in densely populated urban contexts where belligerents and civilians intermingle, and further compounded if community outreach and intelligence-gathering is hampered (our first expectation, above). With media coverage and other information networks particularly dense in the city, news of collateral damage will quickly reach local and international audiences and may help armed actors mobilise resistance against peacekeepers. More generally, peacekeepers that use armed force may lose their position as impartial arbiters, conceivably opening themselves up for further attacks. Taken together, attacks on urban peacekeepers may increase the risk of an escalatory violence spiral and make stabilisation more difficult.
Our third expectation is that attacks on urban peacekeepers will hamper their ability to establish new bases, thereby lessening their reach. Peace operations typically take considerable time to fully deploy, as well as to establish a presence throughout a mission area. 38 In its initial phases, a peace operation is thus likely to be concentrated in a few key locations. Given the need to secure and protect key locations within the city, a peace operation's ability to spread beyond these locations will likely be heavily impacted by attacks on its personnel, as well as broader dynamics of violence.
These dynamics combined, are likely to carry wider consequences for the conduct of a peace operation. For instance, peacekeepers concentrated to a few bases with their movements and activities beyond these perimeters heavily restricted, should be less able to implement certain authorised tasks, especially those that hinge on peacekeepers' proximate presence, including accessing civilians in need of humanitarian assistance or protection. As accessing local knowledge and buy-in is critical to effective peacekeeping, such restrictions may be particularly impactful in urban environments where those who know a city's infrastructure have a tactical advantage. If attacks on peacekeepers lead to escalated violent interactions where peacekeepers are blamed for causing civilian casualties, civil-military relations are likely to suffer further. Hence, while beyond the scope of assessing systematically within this article, the dynamics theorised here are likely to reduce a mission's ability to implement its mandate.

Methodological approach
To assess our expectations, we conduct a plausibility probe focusing on AMISOM's first three years in Mogadishu, Somalia. Given our aim to explore the effects of attacks on peacekeepers in urban environments, AMISOM is a suitable case. Having initially been deployed in just the capital city and come under intense and sustained attack, it represents an 'extreme case', useful for exploring the plausibility of theoretical expectations which, if supported, can be tested in other cases. 39 We use the UCDP Peacemakers at Risk (PAR) dataset to analyse the spatial and temporal distribution of attacks on AMISOM personnel from its original deployment in March 2007 through 2009. 40 PAR tracks reported incidences of violence directly involving peacekeepers and resulting in casualties. 41 The data is presented on an event-basis, including information on actors, timing, location, and estimated casualty-tolls. To explore the intra-city dynamics of violence, we extended the dataset by hand-coding the specific locations (geocoordinates) of events within Mogadishuwhich, under the period of study, was the only area in Somalia where AMISOM peacekeepers were deployed. In most cases events could be pinpointed with high precision, such as to a specific building or street, while in some cases events were ascribed to larger neighbourhoods or city divisions. The dataset includes 122 violent events directly involving AMISOM peacekeepers for the time-period, resulting in an estimated 889 fatalities, including 50 AMISOM fatalities. These events included RPG, mortar and IED attacks, ambushes on patrols, suicide bombings, sniper fire, as well as full-scale assaults on AMISOM positions involving hundreds of al-Shabaab fighters, such as in mid-July 2009 when al-Shabaab launched major attacks on AMISOM and the Transitional Federal Government (TFG). 42 To assess our theoretical expectations, we analyse the spatial patterns and intensity levels of violence involving AMISOM over time. We are mindful that tracking violent events based on public reports is susceptible to bias and potential gaps. Notably, constraints on reporting in active war zones are likely to translate into an underestimation of the total number of events and casualties. 43 While peace operations are likely to be closely monitored by journalists and covered in the news media, this may apply unevenly to different types of operations. Furthermore, whereas UN-led peacekeeping operations provide public information on all their fallen peacekeepers, this is not always the case for other peace operations, including AMISOM, which has not released comprehensive fatality and casualty figures. 44 Notwithstanding these constraints, we have no indication of systematic biases related to the within-city locations of violence specifically introduced by the data collection procedures.
To complement the PAR data, we rely on news reports and existing case studies, as well as primary sources such as internal AU documents. To assess our first theorised outcome, we look for evidence of whether attacks were followed by decreased operational activity in terms of patrols, community outreach, etc. For our second outcome, we analyse whether attacks on AMISOM were followed by escalatory spirals of violence, including peacekeepers' use of armed force. For our third outcome, we analyse the spatial reach of AMISOM, notably regarding the establishment of bases, over time. Since AMISOM was confined to Mogadishu during our period of study, we have no point of comparison to contrast urban with rural patterns. We therefore focus on exploring developments within the city, which allows us to probe in more depth how events unfold and take into account untheorised dynamics and developments. The scarcity of public documents on AMISOM operations constitutes a challenge, making it difficult to access detailed information on the processes linking attacks to their expected effects. Despite this limitation, it is important to try and piece together the details of AMISOM's experience both because it has been understudied and as it is likely to foreshadow future peace operations in urban areas.

AMISOM in Mogadishu: peacekeepers under siege
Compared to a broader population of peace operations, 45 AMISOM represents a particular example of 'partnership peacekeeping': it is a regional force led by the AU but authorised by the UN Security Council. While African countries provide AMISOM's personnel, allowances and other costs are paid for by the European Union, and from mid-2009 the UN provided logistical support. 46 The mission's mandate is also closer to counterinsurgency than traditional peacekeeping. Having forcibly installed the TFG into Mogadishu in December 2006, Ethiopia heavily influenced the process of getting AMISOM authorised by the AU to facilitate the withdrawal of its own soldiers from Mogadishu. 47 AMISOM deployed without the consent of all main conflict actors and absent a ceasefire or peace agreement. Instead, when AMISOM deployed into Mogadishu in March 2007 there were approximately 5,000-7,000 Ethiopian soldiers (ENDF) still protecting the TFG from the remnants of the ousted Islamic Courts Union (ICU) and al-Shabaab. 48 From the outset, al-Shabaab clearly stated its opposition to AMISOM, which they saw as supporting the occupying Ethiopian forces (Associated Press, 8 March 2007;Reuters, 8 March 2007).
In this context, AMISOM was established by the AU Peace and Security Council on 19 January 2007 and mandated to '(i) to provide support to the [Transitional Federal Institutions] in their efforts towards the stabilisation of the situation in the country and the furtherance of dialogue and reconciliation, (ii) to facilitate the provision of humanitarian assistance, and (iii) to create conducive conditions for long-term stabilisation, reconstruction and development in Somalia'. 49 The decision document emphasises that the installation of the TFG in Mogadishu comprised a window of opportunity to stabilise the country and restore functioning government structures, underlining the symbolic and practical importance afforded to the capital city. 50 The AU envisaged AMISOM as a short-term mechanism to be replaced by a UN peacekeeping operation within six months. In sum, the ambition was that AMISOM would restore order, enable a withdrawal of Ethiopian forces and then itself be replaced by a UN operation. 51 On 20 February 2007, UN Security Council resolution 1744 subsequently authorised AMISOM to conduct similarly broad tasks. 52 However, the Security Council refused to deploy a UN operation to take over from the AU.

Phase I: operating in Ethiopia's shadow (March 2007 -January 2009)
Although AMISOM's initial authorised strength was 8,000 troops, only approximately 1,600 Ugandan soldiers actually deployed. They entered a city at war, mainly involving clashes between al-Shabaab, the Ethiopian and TFG forces, as well as other Somali armed groups. 53 Opposition to AMISOM was strong from the outset, and the peacekeepers came under early and repeated attack. The Ugandan vanguard force was attacked during their welcome ceremony at the Mogadishu airport (Associated Press, 8 March 2007). On day two, an AMISOM convoy was attacked by RPGs and small arms fire 54 and mortar fire was directed towards an AMISOM position at the Global Hotel (Agence France Presse, 8 March 2007). The first week also saw an AMISOM cargo plane hit by a missile while landing at Mogadishu International Airport, and a few weeks later, another plane was hit on arrival at the airport, killing all eleven crew but no AMISOM peacekeepers. 55 AMISOM suffered its first fatality on 31 March when a peacekeeper was killed by mortar fire directed near his unit's location in Villa Somalia, the presidential complex where the TFG was located (Agence France Presse, 1 April 2007).
Recurrent attacks on AMISOM were thus a key feature of the mission. Nevertheless, AMISOM had to occupy at least four strategic locations in order to have a chance of implementing its mandate: the airport and seaport, necessary to secure entry points for supplies and equipment; Villa Somalia, the seat of TFG the mission was mandated to protect; and the K4 road junction, the crucial artery connecting the airport to Villa Somalia. Within two weeks AMISOM secured the seaportdespite coming under mortar fireand on 20 March received a shipment of several dozen battle tanks and armoured personnel carriers (APCs). 56 This improved the mission's ability to move around the city and protect its bases. Less than two weeks later, AMISOM conducted its first patrol to Bakara Market, the economic heart of the capital and a key location of militant activity. By May, AMISOM had a company at Villa Somalia, seaport and K4 with the rest of its troops near the Mogadishu International Airport. 57 Aside from securing key locations, however, insecurity early on hampered AMISOM's patrolling; in May 2007, for example, Uganda temporarily suspended its patrols after losing four peacekeepers in an attack (New Vision, 15 July 2007).
The first Burundian contingent began arriving on 23 December 2007 and fully deployed during the first months of 2008, mainly to the Mogadishu University and military academy. 58 Also during 2008, AMISOM received 68 Casspir APCs, which facilitated the mission's movement around the city. 59 After these initial expansion operations, however, AMISOM consolidated its activities around those strategic locations and was not able to deploy to its planned sectors beyond Mogadishu as envisaged in its initial Concept of Operations. 60 Patrolling in armoured vehiclesa consequence of the insecure conditions and to enhance force protectionmade it difficult for AMISOM to conduct extensive community outreach and intelligence-gathering. 61 Figure 1 indicates the locations of violent events involving AMISOM peacekeepers from March 2007 to the departure of Ethiopian troops on mid-January 2009. By then, AMISOM comprised approximately 3,450 troops from Uganda and Burundi. 62 The map also shows the location of AMISOM's main bases and that a large share of the violence took place at or near two of them (the airport and K4 junction). The other location with a high number of eventsmainly in the form of retaliatory mortar attacksis Bakara Market, located northwest of Villa Somalia.
Attacks on AMISOM peacekeepers also resulted in numerous civilian fatalities. Figure 2 shows the distribution of different categories of fatalities in reported events directly involving AU peacekeepers over time. In aggregate, more than half those killed in those The size of the circles represents the intensity of the event (the estimated total death toll, ranging from 0 to 23). Transparent circles denote low location precision (e.g. events assigned only to a larger area within the city). events were reported to be civilians. 63 From accounts reported in connection with violent incidents, al-Shabaab was commonly believed to have caused most of the collateral deaths recorded, but some were the result of AMISOM's use of force.
Civilians were often killed in the crossfire, including from misdirected missiles fired towards AU peacekeepers. For instance, in the mortar attack on the Ugandan vanguard force mentioned above, around 12 civilians were killed (Agence France Presse, 8 March 2007). On another occasion, al-Shabaab fighters attacked AMISOM soldiers in the K4 junction area, and stray bullets and mortars hit nearby houses, killing civilians (Shabelle Media Network, 27 May 2008). In response to such attacks, AMISOM forces often retaliated heavily, leading to casualties among its armed opponents but also civilian deaths. By creating resentment and reducing cooperation by some communities, 64 collateral damage undermined the peacekeepers' ability to carry out other mandated tasks, and served as an important mobilisation opportunity for insurgents. 65 AMISOM was frequently accused of applying disproportional levels of force, engaging in indirect fire practices, and failing to take sufficient steps to avoid civilian casualties. 66 For example, in response to mortar attacks on Mogadishu airport by al-Shabaab, AMISOM (alongside TFG and Ethiopian forces) fired shells targeting Bakara Market, resulting in many civilian fatalities (Shabelle Media Network, 22 September 2008). Indicative of the particular difficulties of operating in the urban context, AMISOM faced challenges because al-Shabaab regularly used civilians as human shields, deliberately fired on AMISOM forces from residential areas of Mogadishu, and mingled among the civilian population. 67 At least initially, however, AMISOM's failure to establish control did not appear to lead to worsened insecurity in Mogadishu; rather, a few months after AMISOM's deployment, there was a relative lull in the fighting in Mogadishu. ENDF and ICU/al-Shabaab. This fighting was considerably more intense than the attacks directed against AMISOM, with the ENDF alone suffering an estimated 840 fatalities in Somalia between late 2006 and January 2009. 68 During this time-period, no further African states volunteered troops for AMISOM. It was reported that six AU member states considered the idea but ultimately decided against it, in part because of the dangerous conditions in Mogadishu. 69 As a result, AMISOM failed to generate its authorised strength, 70 and was therefore unable to expand much beyond its initial locations and took a more defensive approach to its duties. 71 Overall, as Figure 3 illustrates, after a violent peak in early 2007, the situation in Mogadishu fell into an uneasy status quo.
Phase II: working alone with the TFG During 2008, Ethiopia drew down its forces in Mogadishu to about 2,000 troops. 72 AMISOM's deployment was always intended to facilitate an Ethiopian withdrawal since the presence of ENDF troops acted as a major source of al-Shabaab recruitment. 73 By mid-January 2009, the Ethiopian government decided it was time to leave Mogadishu, so it declared 'mission accomplished' and withdrew its forces on 13-14 January (Tigray Online, 4 January 2009). Since AMISOM still comprised well below its authorised troop strength and the TFG forces were disorganised, upon the ENDF's withdrawal many of their positions were quickly taken over by al-Shabaab and other militias, in some cases putting them in close proximity to AU peacekeepers. 74 Cognisant of the risk of being attacked as they withdrew if information fell into the wrong hands, the ENDF had not shared final details of their withdrawal with AMISOM. 75 With the Ethiopians' departure, AMISOM became al-Shabaab's principal target along with the TFG (All Africa, 15 January 2009). At that stage AMISOM still only had around 3,450 troops, and could do little more than consolidate and protect their existing bases. 76 Even areas under AMISOM and TFG control were not regularly patrolled, and after dark peacekeepers retreated to base, leaving insurgents to move relatively freely throughout Mogadishu. 77 Other tasks, such as facilitating humanitarian assistance, were severely hampered; during the period under study, the humanitarian conditions in Somalia worsened. 78 In order to make up for the obvious deficit in AMISOM's troop strength, in   December 2008 the AU had called on the UN Security Council to deploy an International Stabilisation Force as well as a logistical support package to AMISOM. 79 Although the proposed force never materialisedbecause it failed to attract the required troopcontributing countriesthe UN did provide AMISOM with a logistical support package in late 2009: the UN Support Office for AMISOM (UNSOA). 80 With greater security assistance from the United States in particular, Uganda and Burundi gradually increased the size of their contingents and by the end of 2009 AMISOM's deployed strength had increased to nearly 5,500. 81 AMISOM's increased operational capacity also correlated with more violent events involving the peacekeepers, as illustrated by Figure 4. During this period, which is much shorter than the time-period in Figure 1, violence involving AMISOM was more frequent, more intense, and more geographically dispersed in the city, in part because the larger numbers of AU peacekeepers enabled the force to expand its patrols and take a more proactive approach to its duties. While escalating violent interactions are in line with our expectations about a spiral of violence, this dynamic was thus linked to greater rather than reduced operational capacity on behalf of the mission, contrasting with our first theorised outcome. While AMISOM retained a defensive posture throughout the timeperiod under study, the increased capacity to respond to attacks during 2009 also likely contributed to the overall intensity of these violent interactions.
Al-Shabaab attacked AMISOM using a range of mainly asymmetric tactics including suicide bombings, IEDs, and sniper-fire. 82 The highest confirmed death toll of AMISOM personnel in a single event saw 19 peacekeepers killed in an attack involving multiple The size of the circles represents the intensity of the event (the total death toll, ranging from 0 to 81). Transparent circles denote low location precision (e.g. events assigned only to a larger area within the city). suicide bombers at their base at Mogadishu airport in September 2009 (Reuters, 17 September 2009;Reuters, 18 September 2009). But al-Shabaab also used more conventional tactics against AMISOM, exploiting their knowledge of the urban terrain and the density of the city's infrastructure. This included some large-scale offensives, but also vicious street-fighting with AU peacekeepers, whose positions were sometimes as close as 50 metres. In part this was because al-Shabaab occupied some of the former Ethiopian positions but also because it could mass fighters quickly via a network of tunnels it had dug throughout the city. 83 Using the urban environment to their advantage, al-Shabaab also employed pit-traps (camouflaged pits several feet deep) to snag AMISOM tanks and armoured vehicles. 84 Further, al-Shabaab continued its tactic of firing at AMISOM forces from within civilian-populated areas of the city.
AMISOM responded to these attacks in different ways. Ambushes on its patrols were often met with indiscriminate return fire, as were mortar barrages. 85 As noted earlier, peacekeepers face particularly difficult challenges when armed groups are willing to use civilians as human shields, deliberately fire from residential areas, and mingle among the city's civilian population. For AMISOM, Bakara Market epitomised these challenges: simultaneously the city's most important commercial hub and therefore difficult for civilians to make do without, and a long-standing militant stronghold and weapons-trading node, it had played a central role in the struggle for control over Mogadishu during the UN and US operations in the early 1990s and underlined the difficulties of striking against armed groups while avoiding civilian casualties. 86 Attacks also placed further restraints on civilmilitary interactions. In response to infiltration of its main base, the Burundian contingent commander declared, ' . . . now, not a single Somali enters the camp, not a single car.' 87 Despite tighter security procedures, however, Somalis continued to enter AMISOM's main base to run kiosks and receive medical treatment.

Consequences of attacks on AMISOM in Mogadishu
We theorised that attacks on urban peacekeepers would affect their operational activity in three main ways, related to patrolling and outreach, use of force, and the establishment of new bases. These expectations are largely borne out in our analysis. At the same time, AMISOM's experiences reflect the complex dynamics of these processes, with some outcomes closely interlinked but also playing out partly in sequence, and partly at different levels of analysis.
Firstly, we note that after managing to secure a few key locations during its first months, AMISOM peacekeepers struggled to move and operate effectively throughout Mogadishu due to the insecure conditions. In line with our expectations, accounts suggest that initially AMISOM took a defensive approach to its duties, focusing almost entirely on protecting Villa Somaliathe seat of the contested governmentand its own supply lines. With regard to operations such as patrols and civilian engagement, AMISOM's experiences illustrate the particular challenges that result when facing armed groups that are knowledgeable of the local urban terrain, adept at hiding within the urban population and infrastructure, and skilled at using asymmetric tactics. Following strengthened international assistance and increased troop contributions in the aftermath of Ethiopia's departure, AMISOM was able to conduct more extensive patrolling. Yet, as late as March 2009, 'AMISOM was primarily occupied with protecting its bases and supply lines to the port and airfield'; 88 and even during 2010, peacekeepers were still largely unable to conduct foot patrols. 89 Also in line with our expectations, the spatial distribution of AMISOM's bases and activities remained constrained to a few locations throughout the period under study. To a large degree this can be attributed to the slow deployment of troops and military enablers; yet this was, in turn, partly related to the challenging and dangerous conditions facing peacekeepers already deployed, which deterred many potential troop-contributing states. 90 Hence, the outcome here hinged primarily on politico-strategic considerations rather than at the level of on-the-ground operations. Importantly, however, increasing levels of violence against AMISOM did not propel Uganda and Burundi to withdraw their forces. Instead, both states responded to these challenges by increasing their contingents, enabled in part by the eventual expansion of security force assistance programmes by external partners.
Contrary to our expectations, we did not find evidence of violent spirals as a result of attacks on peacekeepers during the first phase, which saw a relative lull in fighting. In other words, early attacks on AMISOM did not trigger extensive retaliation and set off an escalation of violence. On the other hand, it was Ethiopian forces rather than AMISOM that was al-Shabaab's principal target during this period. However, as Ethiopian forces drew down, aggregate casualties from anti-peacekeeper violence increased significantly from late 2008, and AMISOM's retaliatory fire targeting al-Shabaab's presumed locations often caused civilian casualties. On multiple occasions during this later phase, attacks on AMISOM personnel or bases led to heavy-handed responses and prolonged fighting, often resulting in high civilian casualties. 91 This increased popular dissatisfaction with AMISOM, made it more difficult for peacekeepers to collaborate with the city's inhabitants, 92 and increased pressure on the mission to reduce the levels of civilian harm. 93 The mixed results for this theorised outcome reflect the complexity of violent dynamics: On the one hand, violence can be indicative of peacekeepers having poor command over the situation (not posing an effective deterrent and not wielding force effectively), but it can also be a sign of peacekeepers taking a more proactive stance and challenging insurgents.
At least partly as a consequence of the challenges outlined above, the implementation of some components of AMISOM's mandate suffered, including wider stabilisation, dialogue-promotion, and facilitating humanitarian assistance. 94 Nevertheless, AMISOM managed to successfully implement both of its central military objectives: to facilitate the withdrawal of Ethiopian forces and therefore reduce this element of al-Shabaab's recruitment; and to protect the TFG. Furthermore, while attacks hampered AMISOM's activities during its early phases, over time the mission appears to have adapted to these conditions. Throughout the period under study, AMISOM remained well below its authorised strength and stuck in an effective stalemate with al-Shabaab. However, a series of AMISOM and TFG offensives eventually forced al-Shabaab's main forces out of central Mogadishu in August 2011 and its environs in early 2012. 95 Conclusion AMISOM's early experiences in Mogadishu support some of our expectations and qualify others. Attacks against AMISOM appear to have reduced the mission's ability to operate, at times set off escalatory spirals of violence, and impacted the spatial concentration of peacekeepers, with AMISOM long restricted to a handful key bases. Al-Shabaab's attacks, which utilised their knowledge of the city's terrain and population and used asymmetric tactics to their advantage, resonate with existing literature on the particular challenges of urban peacekeeping. The analysis underscores the uneven distribution of both violence and peacekeeper presence across the city. While our analysis focuses on violence involving peacekeepers, future work could assess the spatial patterns of broader dynamics of conflict-related violence in cities where peacekeepers are present in order to assess their ability to deter violence within the city. An important conditioning factor, or alternative explanation impacting the mission's ability to operate, was its limited operational strength and capabilities. Our analysis indicates that AMISOM's geographical concentration was in large part down to its lack of troops, which remained well below its authorised strength for the entire period under study. However, the insecure operating environment played into this constraint, creating both tactical-and strategic-level impacts. The small number of peacekeepers that initially deployed and the failure to generate its authorised strength prevented AMISOM from conducting more effective operations, and likely emboldened al-Shabaab. Although they remained on the defensive during 2007 to 2009, AMISOM was later able to push al-Shabaab out of Mogadishu and extend its area of operations following a substantially increased troop strength, as well as increased logistical support from the UN. It is plausible that such adaptation takes longer when a mission comes under early and intense attack.
In terms of the generalisability of our findings, we re-emphasise that AMISOM deployed without either a ceasefire or peace agreement and that there was a lack of widereaching consent for the peacekeepers' presence. If a peace operation is contested from the outset, it should be less surprising to see immediate or near-immediate attacks following its deployment. While AMISOM here represents an 'extreme case' compared to conventional UN peacekeeping, it is notable that peace operations are increasingly deployed to high-threat environments characterised by the absence of a comprehensive or stable political agreement and with 'little or no peace to keep'. 96 Moreover, like AMISOM, regional and UN-led peace operations are increasingly tasked with supporting the consolidation and extension of state authority in the face of armed opposition from non-state actors. 97 In the AMISOM case, the TFG relied on peacekeepers to protect its representatives and positions, as well as vital installations in Mogadishu. While this study has illustrated the plausibility of our theoretical expectations, further probing the conditions under which attacks hamper peacekeepers and intensify violence is warranted. Future research could also extend our analysis by studying effects of attacks on other peace operations in other urban settings and assessing urban-rural dynamics comparatively to assess whether effects are indeed more pronounced in the city.