Weather,terrain and warfare: Coalition fatalities in Afghanistan

Abstract

The study of conflict increasingly focuses on events and relationships within wars. Among these is the relationship between physical geography and violence. Careful examinations of the relationship between physical geography, especially weather, and events within wars are, however, still few. With increasingly available data on the violence within wars and the physical geography within states, the opportunities for such quantitative analysis have grown. In particular, the ongoing war in Afghanistan provides a useful opportunity to examine this relationship. Using this conflict, we test an argument about how the constraints and opportunities provided by physical geography, in particular daily and seasonal measures of weather, explain combat fatalities. We evaluate our argument with a series of event count models and find consistently significant evidence connecting warm temperatures, decreased visibility and windy conditions to coalition combat fatalities. Alternatively, we find mixed support that the more commonly studied elements of physical geography, distance and rough terrain are connected to these fatalities.

Keywords

Afghanistan armed conflict insurgency physical geography terrain weather

Introduction

Within the study of conflict, measures of rough terrain and distance are frequently examined as possible determinants of violence. These are, however, only a subset of a broader field termed physical geography, which includes terrain, weather, climate, vegetation and soil.¹ The importance of terrain in previous conflict studies suggests that other elements of physical geography may also warrant the attention of researchers. In particular, as weather appears prominently in historical accounts of combat, it too may play an important role in determining wartime violence. Thus, in this paper, we argue for the addition of this element of physical geography, the often-overlooked role of weather, to the quantitative study of wartime violence. By examining daily measures of temperature, visibility and inclement weather, in addition to a seasonal measure of winter, we expand on recent efforts to study the disaggregated elements, in particular the time-varying relationships, within conflicts.

The importance of both weather and terrain has long been recognized by historians and military scholars. Classic works on military strategy, such as those by Sun Tzu ([~500 BC] 1994) and Clausewitz ([1830] 2004), argue that knowledge of terrain and weather is a tactical necessity. Sun Tzu discusses the ways in which a general may use terrain and distance to manipulate combat conditions. Clausewitz describes Napoleon’s concerns with physical geography, the weather and terrains conditions in the Battle of Waterloo (1815). More recently, Winters et al. (2001) examine the impact of temperature, wind, rain and fog across a series of wars and argue for its critical, sometimes decisive, importance in military operations. For example, German armies timed their attacks during the Battle of the Bulge (1944) to occur when visibility was poor. Similarly, the Viet Cong used fog to conceal their movements from US surveillance during the siege of Khe Sanh (1968). Wind has also been a critical factor in battle, such as a during the North African campaign of the Second World War (1941–1943), in which dust storms continually limited the capabilities of both armies.²

Nevertheless, although both physical terrain and weather are frequently argued by military historians and strategists to be important in warfare, only terrain has so far received much attention in the quantitative literature.³ This is probably due to the focus, until recently, upon the country–year level of analysis. Alternatively, the role of climate, a variable similar to weather but more appropriate for country–year analysis, has received some attention in the recent conflict literature.⁴ For instance, Buhaug and Lujala (2005) find that conflict duration increases in regions with monsoon seasons, and both Hendrix and Salehyan (2012) and Raleigh and Kniveton (2012) find a connection between rainfall and increased conflict in African states.⁵ Raleigh (2010) demonstrates that climate change is related to increased communal conflict in some African states. Similarly, conflicts in states affected by El Nino have been shown to be more severe than conflicts in other states (Hsiang et al., 2011).⁶ Yet despite the emerging evidence for the importance of climate in studies of conflict, other elements of physical geography, especially those, such as weather, which vary daily, remain largely unexamined in the quantitative conflict literature. Unlike most time-varying measures used to study conflict events or severity, the meaningful changes in weather occur much more quickly. In particular, as weather can show large variation over a span of just a few days, it is difficult to capture its effects at the country–year level. If some components of weather affect the interactions of combatants, as the historical evidence suggests, then it is important to quantify and test their impact in order to fully understand wartime violence.

In this paper, we study two elements of physical geography, physical terrain and weather, in the ongoing war in Afghanistan. We are particularly interested in how variation in temperature, wind speed, visibility and season affects the occurrence of counterinsurgent fatalities.⁷ As the first three of these elements can vary considerably over short periods of time, it is necessary to use temporally disaggregated data to examine these relationships. Thus we employ daily level data of weather conditions and coalition combat fatalities in Afghanistan. Additionally, we examine other elements of physical geography, terrain and distance, which are commonly held to be important factors in conflict. For these, we use data that are spatially disaggregated to the province level. As these variables can show significant within-country variation, data disaggregated to the sub-national level is appropriate. The benefits of such a disaggregated approach to the study of conflict have been demonstrated in recent works such as Buhaug and Rød (2006), Raleigh and Urdal (2007), Cederman and Gleditsch (2009), Østby et al. (2009) and Nepal et al. (2011). We analyze our data with a set of zero-inflated negative binomial count models.

In developing our argument, our paper proceeds as follows. First, we discuss the mechanisms through which physical geography may affect combat between insurgents and counterinsurgents and, in turn, may affect coalition combat fatalities. Second, we present our hypotheses for each of our weather and terrain variables. Third, we describe the war in Afghanistan. Fourth, we discuss our data. Fifth, we describe our methodology and statistical models. Finally, we present our findings and conclude with a summary and discussion of future research directions.

Physical geography and insurgent warfare

We attempt to examine the combat between insurgents and counterinsurgents and how variation in physical geography affects these interactions and, in particular, counterinsurgent fatalities. As insurgent warfare is often fought through a series of brief, contained attacks, explaining the occurrence of fatalities in such wars requires a focus on small-scale events. Additionally, as insurgents often live amongst and blend in with civilians, it is difficult for counterinsurgent forces to identify and initiate encounters (Galula, [1964] 2005). Thus, most violent events in asymmetrical warfare are initiated by the insurgents (Lawrence, 1929). Moreover, as insurgents are generally the initiators, their decisions to enter combat should depend upon their expectations of success. There is evidence that insurgent attacks occur along spatial and temporal patterns, indicating that there are specific conditions insurgents prefer for carrying out attacks (Siebeneck et al., 2009).⁸ We use this as a starting point in developing our theoretical expectations.⁹

Certain weather and terrain conditions provide potential tactical advantages for insurgents, whereas others provide potential advantages for counterinsurgents. As insurgents generally choose the time and place of encounters, they are better positioned to benefit from the variation in terrain and weather. In combat, insurgent tactics center on quickly inflicting damage, and then successfully retreating (Jalai and Grau, 2001). Prolonged engagements allow advanced counterinsurgent forces to target air strikes against insurgent positions and, given their near monopoly on high-speed transport, quickly bring reinforcements to nearly any battlefield (Biddle, 2002). Thus, insurgents should prefer to strike in places and at times that neutralize these advantages (Gott, 2004). One of the most readily available means to do this is to make use of terrain and weather. For instance, insurgents may use inclement weather to obscure their movements and stage attacks (Dougherty and McNabb, 2007). Consequently, we argue that variance in terrain and weather may advantage insurgents.

Physical geography further affects the interactions between insurgents and counterinsurgents through a series of limitations it places upon the logistics, intelligence gathering and combat capabilities of conventional armies (Keegan, 1993). These limitations typically constrain counterinsurgents much more than the less numerous and less well-equipped insurgents. Large-scale counterinsurgent operations require the regular, scheduled resupply of forces and the operation of many basic security functions. Both difficult terrain and inclement weather create problems for the ground operations of advanced militaries, particularly in maintaining supply lines and insuring the safety of convoys (Rumer, 2006). During inclement weather air transport is much less readily available, forcing counterinsurgent troop and supply movement to rely more heavily on ground transportation. Additionally, harsh weather slows counterinsurgent supply convoys, providing better opportunities for insurgents (Dougherty and McNabb, 2007). Greater convoy traffic at these times increases both the vulnerability of counterinsurgent troops and the predictability of their routes, thus increasing the opportunities for insurgent attacks. Moreover, in addition to the logistical constraints, weather and terrain impose significant limitations on advanced combat operations. For example, severe weather can limit visibility both for troops and air surveillance. Additionally, the availability of counterinsurgent air support is significantly reduced during extreme, windy weather (Rickard, 2004). Counterinsurgent actors are necessarily aware of their increased vulnerabilities during these conditions, but there is only so much that can be done to mitigate or avoid these situations. Unlike the insurgents, who may actively take advantage of weather conditions, the counterinsurgents forces are often constrained.

The argument tested in this paper focuses on the effects of weather when counterinsurgents possess a large technological advantage over insurgents, in particular in airpower and surveillance capabilities. The current conflict in Afghanistan fits this condition, as would many recent Western interventions (e.g. Kosovo and Iraq). That being said, the theoretical mechanisms discussed below may also matter, but perhaps differently, in other types of conflicts. For instance, in conflicts where counterinsurgents do not possess advanced air power or surveillance, insurgents may be more willing to engage in conventional warfare. Also, as a number of our theoretical mechanisms relate to insurgent familiarity with local terrain and weather, these effects may be weaker in conflicts without foreign counterinsurgents.

We now present our individual hypotheses about how weather and physical terrain may affect the combat between insurgents and counterinsurgents and affect counterinsurgent fatalities. We first discuss the reasons behind our selection of Afghanistan as a case study and then present our hypotheses regarding weather followed by our hypotheses regarding physical terrain. For each hypothesis we describe a set of possible mechanisms, relating to the strategic interaction discussed above, for how each of our measures may explain coalition combat fatalities.

Hypotheses

The war in Afghanistan provides a useful case through which to investigate the effects of weather and terrain on counterinsurgent combat. Afghanistan has considerable topographical variation, including many mountainous areas.¹⁰ The southeastern provinces border the Waziristan region of Pakistan, over which the Pakistani government has little control. This area is similar to insurgent-controlled territories located across state borders in other conflicts. For example, in the Ugandan civil war, the Lord’s Resistance Army operated in South Sudan and the Democratic Republic of Congo. The Taliban and other insurgent groups are known to operate freely in the Waziristan region, and frequently enter Afghanistan to stage attacks against coalition forces. Additionally, there is considerable daily variation in temperature and other weather conditions, and at their extremes these conditions may frustrate counterinsurgency operations and aid insurgent tactical operations. For example, in Afghanistan’s dry climate, high wind speeds frequently produce conditions such as sand and dust storms that allow us to test our hypotheses regarding inclement weather.¹¹ Further, as described in more detail below, the disparity in capabilities between insurgents and counterinsurgents is such that the effects of weather, both in terms of opportunities and constraints, are likely to be sizable.

Weather

Extreme weather conditions, such as sandstorms, dust storms and blizzards, should be advantageous to insurgents and detrimental to counterinsurgents. These events include high winds and decreased visibility. Inclement, windy weather is a primary constraint on counterinsurgent air support, as it frequently grounds planes and interferes with detecting ambushes (Young, 2003). This was the case in Iraq in 2005 when counterinsurgent operations were suspended owing to sandstorms that temporarily grounded US aircraft.¹² Additionally, when winds are greater, fewer counterinsurgent soldiers are transported by air, increasing the number of soldiers on the ground and thus increasing the vulnerability of counterinsurgent soldiers. Taken together, we expect that, as weather conditions grow worse, measured by increasing wind speed, counterinsurgent fatalities should increase.

H1: As wind speed increases, counterinsurgent fatalities increase.

Visibility is also likely to affect the interactions between insurgents and counterinsurgents.¹³ Concealment is an important part of insurgent tactics, and insurgents should take advantage of conditions, such as fog and rain, that limit visibility. For instance, insurgents can make use of sandstorms to conceal attacks and cover their withdrawal from battles (Posen, 2003). Iraqi insurgents used the cover provided by a sandstorm to hide their approach before a large-scale rocket attack against coalition forces in 2008.¹⁴ Moreover, good visibility allows counterinsurgent air support and satellite surveillance to locate insurgent ambushes and to provide support in the event of an attack. If counterinsurgent forces recognize an ambush, the insurgent advantage is compromised and counterinsurgent fatalities become less likely (Marques, 2003). When visibility is poor, satellite and drone surveillance is limited.¹⁵ Taken together, we expect that, as visibility improves, counterinsurgent fatalities should decrease.

H2: As visibility increases, counterinsurgent fatalities decrease.

Seasonal changes are also likely to affect the military interactions between insurgents and counterinsurgents. Unlike daily variation in weather, which may often constrain counterinsurgents, some seasonal changes may, alternatively, constrain insurgents. In particular, troop movements may be limited in winter owing to heavy ice and snowfall. These conditions frequently render roads, footpaths and mountain passes unusable. Insurgents, who rely much more on light transportation, may be less able to move and fight effectively during the winter months (Shaw and Spencer, 2003). Additionally, in some conflicts, insurgent movements may draw recruits from distant countries with different climates. These fighters may not be conditioned to fight in harsh winters and may be less effective during them (Erikson and Heier, 2009). Taken together, we expect that during winter counterinsurgent fatalities should decrease.

H3: During the winter season, counterinsurgent fatalities decrease.

In addition to a seasonal effect, there are reasons to suspect that extreme temperatures may lead to increased fatalities. For example, many health problems can follow prolonged exposure to high temperatures. Dehydration, sunburn, and heat stroke are all common risks for military forces during the hottest days (Korzeniewski, 2008). These conditions impair soldiers both physically and mentally, leading to slower reaction times and generally lowered levels of concentration (Stulz, 2008). Advanced military forces, generally the counterinsurgents, frequently deploy in full combat gear, and convoy travel can expose these soldiers to high temperatures for long periods with little relief. Insurgents, by contrast, often travel lightly and are more acclimated to severe local temperatures (Grau, 1996). Insurgents may use this to their advantage and purposefully strike on particularly hot days in anticipation of the toll the weather will have taken on counterinsurgent soldiers. Taken together, we expect that, separate from any seasonal effect, as daily high temperature increases, counterinsurgent fatalities should increase.

H4: As temperature increases, counterinsurgent fatalities increase.

Physical terrain

Physical terrain has long been understood to be an important element in warfare. Unlike weather, terrain, in particular measures of distance and mountainous terrain, is commonly examined in quantitative studies of conflict. While the particular importance of mountains varies depending upon which aspect (e.g. onset or duration) of conflict is examined, a number of studies have found evidence that mountainous terrain is positively associated with conflict severity (as examples, Lacina, 2006; Lujala, 2009). Thus it is reasonable to expect that military fatalities, being one aspect of conflict severity, should increase in mountainous terrain. Such difficult terrain may provide an advantage for local insurgents with greater familiarity with the details of these areas, such as hidden passes and cave systems (Bahmanyar, 2004). This local knowledge may mitigate the advantages provided by counterinsurgent surveillance technology by allowing insurgent forces to better avoid detection. Additionally, difficult terrain may allow insurgents to maneuver more easily near counterinsurgent convoys and other targets and to more easily retreat after attacks. Taken together, we expect that, as the percentage of mountainous terrain increases, counterinsurgent fatalities will increase.

H5: As percentage of mountainous terrain increases, counterinsurgent fatalities increase.

Conflict violence has also often been found to increase with proximity to state borders and insurgent strongholds (Hegre et al., 2009; Rustad et al., 2011). In many conflicts, insurgents base much of their operations in neighboring states. As insurgents are often poorly mechanized, their ability to strike far from these bases is likely to be limited (Erikson and Heier, 2009). Thus, most of their attacks should occur closer to their foreign bases. Hegre and Raleigh (2009) find that proximity to state borders and insurgent-controlled areas increases conflict events. Similarly, Buhaug and Rød (2006) find that conflict is more likely in areas near an international border and Salehyan (2007) shows that insurgent control of territory across state borders extends conflict duration. Taken together, we expect that counterinsurgent fatalities should increase in provinces closer to other states providing substantial support to insurgents.

H6: As distance to Waziristan decreases, counterinsurgent fatalities increase.

In sum, we expect that weather will play a sizable role in determining counterinsurgent combat fatalities. Higher temperatures and increased wind speeds should advantage insurgents and increase counterinsurgent fatalities. Conversely, increased visibility and winter months should advantage counterinsurgent forces and reduce counterinsurgent fatalities. Additionally, and in line with the literature’s general findings, we expect that mountainous terrain and proximity to insurgent-controlled territory in other states will advantage insurgents and increase counterinsurgent fatalities. We now provide a background description of the war aims and military capabilities of the primary actors in the conflict.

The war in Afghanistan

The primary counterinsurgent actors in the conflict are the US-led Operation Enduring Freedom (OEF) and the NATO-led International Security Assistance Force (ISAF).¹⁶ In response to the 11 September attacks against the USA, US forces under OEF command initially entered Afghanistan with the stated intention of removing Al-Qaeda operations and bases from the country. The Taliban government, which had supported the presence of Al-Qaeda in Afghanistan, was also targeted for removal. Most coalition forces belonged to the USA and UK militaries and until 2003 most combat operations in Afghanistan were conducted by OEF. The ISAF, created by NATO in December 2001, was tasked with stabilizing the transitional government after the fall of the Taliban. Over time, ISAF mission objectives and troop levels expanded until the majority of combat operations in Afghanistan involved ISAF forces.

Insurgent forces consist of the Taliban, elements of Al-Qaeda and other affiliated groups such as the Haqqani Network and Hezb-e-Islami Gulbaddin. The aims of these insurgent forces may be more disparate than the coalition aims. A number of high-ranking Taliban and Al-Qaeda members accept a radical interpretation of Islam that actively promotes violence against Western forces, while many lower-level soldiers join seeking retaliation for the deaths of family members caused by Western forces (Asfar et al., 2008). The long-term goal of the insurgency is to force the withdrawal of the coalition, after which the Taliban element of the insurgency plans to retake control of the government (Samples and Asfar, 2008).

The OEF and ISAF forces possess advanced military technology, including state-of-the-art air power and surveillance capabilities. Examples include the Predator MQ-1 and Reaper MQ-9 unmanned aerial vehicles. These and other unmanned aerial vehicles can detect insurgent movement and instantly relay the information to coalition military forces (Turner et al., 2009). Air transport, such as the CH-47 Chinook helicopter, allows the coalition to move a large number of soldiers in relative safety, while fighter jets such as the F-15 allow coalition forces to quickly strike at insurgent forces. Additionally, OEF and ISAF ground forces are significantly better equipped in terms of both offensive and defensive combat technology. While this provides many advantages, it also means that coalition forces often wear and carry large amounts of gear in extreme heat. For US soldiers in Afghanistan, the total weight of combat gear, weapons and supplies can exceed 59 kilograms.¹⁷ This weight often means soldiers will limit the amount of water they carry, increasing the risk of heat-related fatigue. Coalition forces also have an overwhelming numerical advantage in troop presence. Troop levels have continually increased throughout the conflict, and at the height of the conflict totaled more than 100,000 soldiers (Livingston et al., 2010).

Insurgent military capabilities are vastly inferior and they lack armor and air capabilities. A large number of insurgent weapons and equipment, such as the Kalashnikov assault rifle and RPG-7 launcher, were left after the Soviet occupation and additional arms have been smuggled in from Pakistan.¹⁸ The history of conflict in the region, combined with the recent presence of Al-Qaeda training camps, has produced a number of veteran fighters well versed in guerilla warfare. Additionally, insurgents frequently recruit from the civilian population, where unemployment or grievance against coalition forces is high (Jones, 2007). These recruits receive little formal training, and often assist in only a single attack before rejoining the civilian population, making it difficult for coalition forces to determine who participates in insurgent attacks. The majority of attacks are carried out using improvised explosive devices, rocket-propelled grenades and various small arms.

Data

Our dependent variable is coalition fatalities, and our data include all daily-province counts of US and allied combat fatalities from 7 October 2001 to 31 December 2009 for each of Afghanistan’s 34 provinces.¹⁹ The fatality data are taken from icasualties.org,²⁰ an organization that collects and catalogs all data on coalition fatalities in Afghanistan. From these data, we have removed fatalities caused by friendly fire, accidental death, suicide and illness, and kept only those directly caused by insurgents. All fatalities from both the OEF and ISAF are included in our analysis.

Table 1 displays the descriptive statistics for our dependent and independent variables. We test for a relationship between fatalities and each of our measures of daily high temperature, daily high wind speed, daily average visibility, winter, rough terrain and distance to Waziristan. Our measures of high temperature, high wind speed, and average visibility come from the National Climatic Data Center (www.ncdc.noaa.gov/oa/ncdc.html).²¹ The Center collects daily weather observations from a number of points around the world. We use four weather stations within Afghanistan and nine additional stations near the Afghanistan borders to estimate daily weather conditions for each province.²² Additionally, we include a seasonal measure for winter. In the results presented below, the three coldest months of the year, December through February, are coded with a 1, while all other months are coded as 0. We also test province-level measures of mountainous terrain and distance. The mountainous terrain data are taken from Afghanistan’s Ministry of Rural Rehabilitation and Development (www.mrrd.gov.af).²³ Our data are reported as the percentage of mountainous area in each province. Finally, we include a measure of the distance between Waziristan and Afghanistan’s province capitals.²⁴

Table 1.

Descriptive statistics

	Mean	Standard deviation	Minimum	Maximum
Coalition combat fatalities	0.02	0.21	0	19
Temperature (Celsius)	23.28	10.51	−10.00	60.00
Wind speed (km/h)	24.09	15.14	0	238.18
Visibility (km)	7.67	2.40	0	12.06
Winter	0.25	0.43	0	1
Waziristan distance (km)	394.35	172.58	188.47	744.04
Percentage mountainous	42.44	23.85	0	96.60
Opium (hectares, 1000s)	3.46	11.07	0	103.59
Coalition (1000s)	53.68	47.84	0	145.58
Population (1000s)	672.64	576.00	111.00	3691.40
Area (1000 square km)	19.22	16.41	1.84	58.58

We also include a set of additional variables to help control for possibly conflating factors. In particular, as the insurgent attacks driving these fatalities are constrained by both resources and the availability of targets, we include measures to control for these potential effects. As the number of coalition forces in Afghanistan has changed considerably over the war, we include a measure for the number of these forces. Our measure for coalition soldiers was created using a set of monthly totals of US soldiers and quarterly, sometimes monthly, totals of non-US NATO soldiers. These figures were combined for our coalition measure.²⁵ Second, we include a yearly measure of opium cultivation by province. Opium production may affect both the resources available for insurgents to engage coalition forces and the incentives to engage coalition forces in particular areas. Additionally, the deployment and use of coalition forces may depend upon the insurgent presence and violence in opium-producing areas. Opium cultivation data are taken from the United Nations Office of Drugs and Crime (www.unodc.org).²⁶ Similarly, more populated and physically larger provinces may have more coalition forces present, and as such we include a measure of each. The province population data come from the Central Statistics Organization of Afghanistan (www.cso.gov.af), and province area is taken from the Naval Post Graduate School (www.nps.edu/programs/ccs).

Methodology

We test our hypotheses with a series of zero-inflated negative binomial event count models. The dependent variable we examine is the count of daily, province-level coalition combat fatalities. These fatalities are unlikely to be independently distributed within province days, and instead the counts are likely to exhibit overdispersion. According to our expectations, these fatalities should fail to be independent for two reasons. First, as even loosely connected insurgent groups may operate under some type of command structure, some of the attacks are likely to be at least partially coordinated. Second, fatalities from the individual attacks are necessarily connected with each other. As such, a negative binomial count model should be preferred over a standard Poisson count model.

Additionally, although there have been thousands of coalition combat fatalities throughout the war, most province-days experienced no fatalities. As such, in our data there are significantly more zero observations than would be expected when using a standard negative binominal model. Consequently, instead of using negative binomial event count models, we estimate a set of zero-inflated negative binominal count models. With our data, these zero-inflated models assume that there are two possible causes for the zero observations. They could arise because there was no encounter between coalition and insurgent forces or because the encounter produced no coalition fatalities. To sum, these zero-inflated negative binomial count models allow for both the possible overdispersion within the counts and also for the larger number of zero-counts than would be expected when using standard negative binomial count models.²⁷

One additional methodological issue is the possible non-independence of the observations in our data. To address this, the possible heteroskedasticity and autocorrelation within our data, we estimate each of the models with robust standard errors with the observations clustered on Afghanistan’s 34 provinces.

Findings

In Table 2, we present the basic results from our statistical models. The first model includes the three daily weather variables, the seasonal winter variable, the variables for mountainous terrain and distance from Waziristan, and the set of controls. The second model includes all of the above except for the seasonal winter variable. The third includes all the variables from the first model except for daily high temperature.

Table 2.

Zero-inflated negative binomial regression for coalition fatalities

	Model 1		Model 2		Model 3
	Coefficient	Z	Coefficient	Z	Coefficient	Z
Fatalities equation
Temperature	0.0548***	6.25	0.0495***	6.93	—	—
Wind speed	−0.0023	−0.61	−0.0021	−0.56	0.0015	0.24
Visibility	−0.0260	−1.08	−0.0218	−0.87	−0.0093	−0.32
Winter	0.2102	1.14	—	—	−0.6823***	−5.78
Waziristan distance	−0.1406	−0.49	−0.1340	−0.47	−0.0690	−0.24
Percentage mountainous	−0.0009	−0.09	−0.0006	−0.06	0.0016	0.15
Coalition	0.0256***	5.11	0.0258***	5.15	0.0283***	5.73
Opium cultivation	−0.0068*	−1.85	−0.0065*	−1.78	−0.0043	−1.28
Population	0.0126	0.06	0.0129	0.06	0.0064	0.03
Area	−0.0571	−0.25	−0.0507	−0.22	−0.0142	−0.06
Constant	−2.3648	−0.7	−2.3649	−0.7	−2.1530	−0.61
Inflate equation
Temperature	0.0110	0.96	−0.0004	−0.04	—	—
Wind speed	−0.0106**	−2.24	−0.0104**	−2.25	−0.0095*	−1.66
Visibility	0.0557**	1.93	0.0537**	1.81	0.0512	1.61
Winter	0.4756**	2.04	—	—	0.2981**	2.13
Waziristan distance	1.1771***	5.34	1.2087***	5.45	1.1727***	5.77
Percentage mountainous	0.0066	0.63	0.0067	0.64	0.0079	0.79
Coalition size	−0.0132***	−3.48	−0.0131***	−3.45	−0.0121***	−3.16
Opium cultivation	−0.0935**	−2.61	−0.0929**	−2.59	−0.0971***	−2.98
Population	−0.2574	−1.14	−0.2608	−1.15	−0.2486	−1.11
Area	−0.2744	−1.21	−0.2676	−1.17	−0.2404	−1.07
Constant	1.1381	0.44	1.2672	0.49	0.9340	0.37
α	3.005***		3.028***		3.543***
N	96,812		96,812		96,812
Nonzero observations	703		703		703
χ ²	607.84		449.19		446.23
Vuong statistic	9.34***		9.44***		9.18***

Standard errors clustered on province.

***

p < 0.01, ** p < 0.05, * p < 0.10 (two-tailed tests).

The upper part of Table 2 shows the results from the negative binomial part of the zero-inflated count model. The estimates are presented as coefficients and can be interpreted as the increase in the log number of fatalities associated with a one-unit increase in each variable. The lower part shows the results from the inflation part of the model. This is used to capture the larger number of zero events, or province-days with no fatalities, in our data than would be implied solely by the negative binomial distribution. These coefficients can be interpreted as coming from an inverse logistic model. In this case, a positive coefficient implies a decrease in the probability of seeing at least one fatality on a given province-day. Thus, for each of our variables, we expect the coefficient from the inflation part to be in the opposite direction of the coefficient from the count part of the model.

The first elements to examine in Table 2 are the estimates for α, the overdispersion parameter. As we argue above, coalition fatalities on the individual province-days should not be independent of each other. Instead, these fatalities should be positively related and, in all of our models, the estimates for α confirm this. They are both positive and statistically significant, indicating that the negative binomial model component is appropriate. Additionally, the Vuong statistic for each model is statistically significant. This indicates that the zero-inflated negative binomial count model is preferred to the standard negative binomial count model.

Next we turn to a discussion of how the different elements of physical geography affect coalition fatalities. As described above, Table 2 shows each variable’s effect on each part of the model, both the negative binomial and the inflation components. Our arguments about the combination of insurgent and counterinsurgent strategies driving these fatalities, however, are not such that we have competing expectations for the separate parts of each model. That is, we do not separately theorize about which factors will affect the likelihood of a fatal encounter and those that will affect the deadliness of the encounters. Instead, we are concerned about how the variables in the aggregate affect the number of coalition fatalities. As such, in our specifications, each variable was entered into both parts of each model. In our discussion we will focus on the combined marginal effect of each variable, but we will note where there is statistically significant evidence that the individual count and inflation components operate differently, for example, making a fatal event less likely but increasing the expected number of fatalities given that at least one occurs. Again, ultimately we are interested in each variable’s total effect on the expected fatality counts, the combination of its effect on the inflation part and its effect on the negative binomial part of each model.²⁸ To see these results, we produce, in Table 3, a set of marginal effects for the three models, and our discussion of the effect of each variable will focus on these marginal effects.²⁹

Table 3.

Marginal effects for predicted counts of coalition fatalities

	Model 1		Model 2		Model 3
	dy/dx	Z	dy/dx	Z	dy/dx	Z
Temperature	0.00020***	2.79	0.00021***	3.1	—	—
Wind speed	0.00004**	2.25	0.00003**	2.27	0.00005**	2.37
Visibility	−0.00036**	−2.35	−0.0003**	−2.26	−0.0003**	−2.08
Winter	−0.00100	−1.35	—	—	−0.0042***	−3.3
Waziristan distance	−0.00572***	−4.06	−0.0055***	−4.34	−0.0052***	−4.26
Percentage mountainous	−0.00003	−0.66	−0.00003	−0.64	−0.00003	−0.54
Coalition size	0.00017***	4.72	0.00016***	4.8	0.00017***	4.54
Opium cultivation	0.00037*	1.75	0.00035*	1.79	0.00039**	2.01
Population	0.00117	0.91	0.0011	0.92	0.00108	0.83
Area	0.00093	0.83	0.00088	0.83	0.00095	0.88

***

p < 0.01, ** p < 0.05, * p < 0.10 (two-tailed tests).

To aid our interpretation of the substantive effects, we also produce two figures showing the percentage change in the expected counts of coalition fatalities with a one standard deviation increase in each variable, or a 0–1 change for the dichotomous winter variable.³⁰ These values were calculated for our first model, the fully specified model. Using these tables and figures together, we now move to a discussion of the different relationships.

First, we find considerable evidence for hypothesis 4 that higher temperatures are positively connected to coalition fatalities. In both models 1 and 2, although not significant in the inflation component, daily high temperature is positive and statistically significant in the fatalities equation. Moreover, as shown in Table 3, its marginal effect on coalition fatalities is consistently positive and statistically significant. As seen in Figure 1, a one standard deviation increase in temperature increases the expected number of coalition fatalities by more than 60%. The marginal effects of the two other daily measures of weather, wind speed and visibility, are also always in line with our expectations. In accordance with hypothesis 1, greater wind speed is connected to increased coalition fatalities. In the inflation component of all three models, we see that greater wind speeds increase the probability of seeing at least one coalition fatality, and the marginal effect of wind speed is statistically significant positive in all three models. This is in line with our theoretical expectations and fits with accounts of Taliban and Al-Qaeda ambushes occurring during inclement weather when coalition air support is less available.³¹ In line with hypothesis 2, greater visibility is connected to decreased coalition fatalities with its marginal effect negative and statistically significant in all three models. In particular, visibility is statistically significant in the inflation component of models 1 and 2. In model 3, however, it is not quite statistically significant in the inflation component, but together with its effect in the fatalities equation, again it has a statistically significant negative marginal effect.

Figure 1.

Percentage change in expected coalition fatalities given as mean ± one standard deviation change in independent variables.

To examine the seasonal effect of winter, models 1 and 3 also include a separate measure for the three coldest months of the Afghan year. Its marginal effect is always in the predicted direction. Coalition fatalities decline during the winter.³² The seasonal winter measure, however, is not as consistently significant as the daily weather variables.³³ Its marginal effect on coalition fatalities is only statistically significant in the third model, the model without the daily-level temperature variable. Winter is also statistically significant in the inflation part of model 1, decreasing the likelihood of seeing at least one coalition fatality. However, coupled with its slight positive effect in the fatalities equation, its overall marginal effect on coalition fatalities in model 1 is negative but statistically insignificant. As such, although there is some evidence that a coarse seasonal measure explains coalition fatalities, we fail to find as much evidence for hypothesis 3 as winter does not perform as well or as consistently as the daily wind speed or temperature measures and, in regard to its overall marginal effect, less well than daily visibility.³⁴ It is difficult to determine if reduced insurgent military activity or, alternatively, a possible reduction in coalition troop activity during the winter, is driving this negative, although less robust, result. To sum up, we see that, not only the daily measure of temperature and, to a lesser extent, the seasonal measure of winter matter, but also the much less frequently discussed measures of visibility and wind speed are important determinants of coalition fatalities.³⁵

Together these results illustrate the general and separate effects of weather. There is evidence for the often-reported winter lull in insurgent activity and consequent reduction in coalition fatalities. However, daily measures are not only significant determinants of coalition fatalities, but in a combined model the marginal effect of daily high temperature measure is more significant than the seasonal measure. Consequently, the understanding that winter brings a decrease in combat fatalities is accurate, but absent more elaboration that understanding is incomplete. The winter lull is important, but the daily measures of weather are arguably, especially together, considerably more important in determining coalition combat fatalities.

In contrast to the generally robust results for the weather component of physical geography, the results for our hypotheses about the role of the terrain variables are mixed. In line with hypothesis 6, the marginal effect of distance from Waziristan on coalition fatalities, as expected and consistent with the literature and media accounts (Boot, 2009), is always in the predicted negative direction, consistently statistically significant, and is substantively large. It is never significant in the negative binomial equation, but distance to Waziristan is statistically significant in the inflation component of all three models, increasing the probability of seeing no fatal events. Overall, as shown in Figure 1, a one standard deviation increase in distance decreases fatalities by nearly 50%.³⁶ This supports the argument that insurgents use Pakistan as a staging area or a source of supplies for their operations in Afghanistan.

Alternatively, the effect of mountainous terrain is never statistically significant in either part of any of the models, and its marginal effect is never significant and has a consistently negative, instead of the expected positive, effect on coalition fatalities. In contrast to our expectation described in hypothesis 5 of increased fatalities in mountainous areas, the results are more in line with some recent disaggregated studies which find, albeit in studies of violence against civilians, lower levels of violence in mountainous areas. For example, Balcells (2011) finds that collateral civilian fatalities from bombing campaigns are lower in areas with rough terrain. Additionally, Buhaug and Rød (2006) find mixed evidence for a relationship between terrain and conflict onset and theorize that insurgents may build bases in mountains but fight in other areas. Together, these results suggest that, although mountainous terrain may not directly affect fatalities, there may be some property that increases other conflict metrics, such as onset. Our results may also be unique to this case, as many of the mountainous areas in Afghanistan are remote, inhospitable, and sparsely populated. Neither insurgents nor counterinsurgents may be active in these areas. Additionally, the ease with which insurgents can base their operations in neighboring Pakistan may reduce the need to make use of mountainous areas, and therefore reduce the likelihood of violent encounters in these areas. Regardless, across the different specifications, all three daily measures of weather perform more strongly and more robustly than our measure of difficult terrain.³⁷

Finally, our set of control variables behaved largely according to our expectations. As seen in Table 3, fatalities are strongly positively connected to the number of coalition forces in Afghanistan.³⁸ Moreover, as seen in Table 2, coalition force size is significant in both the negative binomial and inflation components of each model. Also important is opium cultivation. As seen in Table 3, the marginal effect of opium cultivation is positive and statistically significant in all three models.³⁹ This relationship, however, is more complicated than most of our findings. Looking at Table 2, we see, in the inflation component, that across all three models opium cultivation is linked to a statistically significant increased chance of a least one coalition fatality, but as shown in the count component, opium is linked with a decreased number of fatalities given that at least one occurs. Again, this suggests that opium cultivation, unlike most of our variables, may have a more complicated relationship with coalition fatalities. Finally, as shown in Figure 2, both more populated and larger provinces see slightly increased coalition combat fatalities, but the marginal effect of both these variables is insignificant in all three models.⁴⁰

Figure 2.

Percentage change in expected coalition fatalities given as mean ± one standard deviation change in the independent variables.

Conclusion

This paper has examined the effects of physical geography on warfare in Afghanistan. Making use of a daily, province-level data set of coalition combat fatalities together with detailed weather and physical terrain data, we examined how physical geography affects coalition fatalities within the conflict. Supporting our argument that insurgent behavior is driven by the opportunities presented by variation in weather, we demonstrated the importance of this under-examined element of physical geography. To our knowledge, this is one of the first systematic inquiries into weather’s varied roles in explaining the fatalities within conflicts.

In line with media and military accounts, we find some, but not overwhelming, evidence that winter itself limits insurgent military action and leads to decreased coalition fatalities. More significantly, we see, separate from winter’s seasonal effect, considerable evidence that warmer temperatures bring increased coalition fatalities. We also find evidence that other much less examined measures of weather, visibility and wind speed, significantly affect combat fatalities. We see evidence that increased visibility advantages coalition forces, decreasing the number of fatalities, and that inclement weather, measured by wind speed, advantages insurgent forces and leads to increased coalition fatalities. Taken together, temperature, wind speed and visibility all play sizable roles in explaining coalition fatalities.

In contrast to our weather findings, we see mixed evidence for the importance of physical terrain. As expected, we find strong evidence for the role of distance from Pakistan, in particular the Taliban-dominated area of Waziristan, in explaining coalition fatalities. Provinces nearer Waziristan see a sizable increase in the number of coalition fatalities. Alternatively, our measure of difficult terrain, the province’s percentage of mountainous terrain, was consistently insignificant and even negatively related to coalition fatalities. This stands in contrast to media coverage of the war that often reports on the mountainous terrain advantage provided to the insurgents in Afghanistan. As mentioned in the previous section, this result, or non-result, may be the product of a more complicated relationship between insurgent behavior and difficult terrain. This insignificant finding does conform to the mixed and sometimes conflicting results from the literature on civil wars and events within them.

Given our findings and the increasing availability of weather data for conflicts across the globe, we argue that researchers should pay closer attention to the relationship between weather and violence within war. In particular, as few measures vary as frequently and as greatly over both time and space, detailed time-varying measures of weather may be especially useful for studying the events within conflicts. Nevertheless, as our findings are from a single conflict in a single country, examination of these relationships in other regions and conflicts is required before any generalizations can be made. It is possible that the components of weather matter differently in other conflicts. Further research may find that weather’s importance depends upon the counterinsurgents’ relative technological advantage, especially its advantage in air power. Additionally, it is possible that some of the relationships examined here, particularly the relationship between terrain and fatalities, operate differently with government counterinsurgents, who are more familiar with local conditions, than with foreign counterinsurgents. Taken together, there are a number of reasons for an increased focus on physical geography, and particularly weather, in the study of combat violence.

Footnotes

Acknowledgements

We would like to thank Michael Langlois, Susumu Suzuki, the editor of Conflict Management and Peace Science, and the anonymous reviewers for their valuable comments.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Notes

References

Afsar

Samples

Wood

(2008) The Taliban: An organizational analysis. Military Review 88(3): 58–73.

Bahmanyar

(2004) Afghanistan Cave Complexes 1979–2004: Mountain Strongholds of the Mujahideen, Taliban, and Al Qaeda. Fortress Study Group Series, no. 26, ed. Cowper

Bogdanovich

. Oxford: Osprey.

Balcells

(2011) Continuation of politics by two means: Direct and indirect violence in civil war. The Journal of Conflict Resolution 55(3): 397–422.

Benini

Moulton

(2004) Civilian victims in an asymmetrical conflict: Operation Enduring Freedom, Afghanistan. Journal of Peace Research 41(4): 403–422.

Berman

Felter

Shapiro

(2009) Do working men rebel? Insurgency and unemployment in Iraq and the Philippines. NBER Working Paper 15547.

Biddle

(2003) Afghanistan and the future of warfare: Implications for army and defense policy. Foreign Affairs 82: 31–45.

Boot

(2009) How we can win in Afghanistan. Commentary, 2 November.

Buhaug

(2010) Climate not to blame for African civil wars. Proceedings of the National Academy of Sciences for the United States of America 107(32): 16477–16482.

Buhaug

Gates

(2002) The geography of civil war. Journal of Peace Research 39(4): 417–433.

10.

Buhaug

Rød

(2006) Local determinants of African civil wars, 1970–2001. Political Geography 25(3): 315–335.

11.

Buhaug

Lujala

(2005) Accounting for scale: Measuring geography in quantitative studies of civil war. Political Geography 24(4): 399–418.

12.

Buhaug

Gates

Lujala

(2009) Geography, rebel capability, and the duration of civil conflict. Journal of Conflict Resolution 53(4): 544–569.

13.

Carter

(2011) Explaining insurgent violence: The timing of deadly events in Afghanistan. Civil Wars 13(2): 99–121.

14.

Cederman

L-E

Gleditsch

(2009) Introduction to special issue on ‘Disaggregating civil war’. Journal of Conflict Resolution 53(4): 487–495.

15.

Clausewitz

([1830] 2004) On War. New York: Knopf.

16.

Condra

Shapiro

(2012) Who takes the blame? The strategic effects of collateral damage. American Journal of Political Science 56(1): 167–187.

17.

Dougherty

McNabb

(2007) Combat Techniques: An Elite Guide for the Modern Infantry. New York: St Martin’s Press.

18.

Erikson

Heier

(2009) Winter as the number one enemy? Lessons learned from north Afghanistan. Rusi Journal 154(5): 64–71.

19.

Fearon

Laitin

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

20.

Findley

Young

(2007) Fighting fire with fire: How (not) to neutralize an insurgency. Civil Wars 9(4): 378–401.

21.

Galula

([1964] 2005) Counterinsurgency Warfare: Theory and Practice. Westport, CT: Praeger.

22.

Giustozzi

(2008) Koran, Kalashnikov and Laptop: The Neo-Taliban Insurgency in Afghanistan. New York: Columbia University Press.

23.

Gleditsch

(2012) Whither the weather? Climate change and conflict. Journal of Peace Research 49(1): 3–9.

24.

Gott

(2004) In Search of an Elusive Enemy: The Victorio Campaign. Fort Leavenworth, KS: Combat Studies Institute Press.

25.

Grau

(1996) The Bear Went Over the Mountain: Soviet Combat Tactics in Afghanistan. Washington, DC: National Defense University Press.

26.

Hegre

Raleigh

(2009) Population size, concentration, and civil war: A geographically disaggregated analysis. Political Geography 28(4): 224–238.

27.

Hegre

Østby

Raleigh

(2009) Poverty and civil war events: A disaggregated study of Liberia. Journal of Conflict Resolution 53(4): 598–623.

28.

Hendrix

Salehyan

(2012) Climate change, rainfall, and social conflict in Africa. Journal of Peace Research 49(1): 35–50.

29.

Hsiang

Meng

Cain

(2011) Civil conflicts are associated with the global climate. Nature 476(3): 438–441.

30.

Jalai

Grau

(2001) Afghan Guerilla Warfare: In the Words of the Mujahideen Fighters. St Paul, MN: MBI.

31.

Johnson

Mason

(2006) Understanding the Taliban insurgency in Afghanistan. Orbis (51)1: 71–89.

32.

Jones

(2007) Afghanistan’s Local Insurgency. Santa Monica, CA: The Rand Corporation.

33.

Jones

(2010) In the Graveyard of Empires: America’s War in Afghanistan. New York: WW Norton.

34.

Keegan

(1993) A History of Warfare. New York: Knopf.

35.

Korzeniewski

(2008) Environmental risk factors in the territory of military operations in Iraq and Afghanistan. Polski Merkuriusz Lekarski 145(1): 5–8.

36.

Lacina

(2006) Explaining the severity of civil wars. Journal of Conflict Resolution 50(2): 276–289.

37.

Lawrence

(1929) The science of guerilla warfare. In Encyclopedia Britannica, 14th edn. Chicago, IL: Encyclopedia Britannica.

38.

Lind

Moene

Willumsen

(2009) Opium for the masses? Conflict-induced narcotics production in Afghanistan. CESIFO Working Paper 2573.

39.

Livingston

Messera

O’Hanlon

(2010) Afghanistan Index: Tracking Variables of Reconstruction and Security in Post- 9/11 Afghanistan. Washington, DC: Brookings Institution.

40.

Lujala

(2009) Deadly combat over natural resources: Gems, petroleum, drugs, and the severity of armed civil conflict. Journal of Conflict Resolution 53(1): 50–71.

41.

Lujala

(2010) The spoils of nature: Armed civil conflict and rebel access to natural resources. Journal of Peace Research 47(1): 15–28.

42.

Marques

(2003) Guerilla warfare tactics in an urban environment. Thesis, Army Command and General Staff College. Fort Leavenworth, KS.

43.

Middleton

(1986) A geography of dust storms in south-west Asia. International Journal of Climatology 6(2): 183–196.

44.

Naval Postgraduate School (n.d.) Program for Culture and Conflict Studies. Afghanistan Provincial Reports Data. Available at http://www.nps.edu/programs/ccs/ExecSumm.html

45.

Nepal

Bohara

Gawande

(2011) More inequality, more killings: The Maoist insurgency in Nepal. American Journal of Political Science 55(4): 886–906.

46.

Østby

Nordås

Rød

(2009) Regional inequalities and civil conflict in Sub-Saharan Africa. International Studies Quarterly 53(2): 301–324.

47.

Posen

(2003) Command of the commons: The military foundations of US hegemony. International Security 28(1): 5–46.

48.

Raleigh

(2010) Political marginalization, climate change, and conflict in African Sahel states. International Studies Review 12(1): 69–86.

49.

Raleigh

Kniveton

(2012) Come rain or shine: An analysis of conflict and climate variability in East Africa. Journal of Peace Research 49(1): 51–64.

50.

Raleigh

Urdal

(2007) Climate change, environmental degradation, and armed conflict. Political Geography 26(6): 674–694.

51.

Rickard

(2004) The employment of airborne forces in modern asymmetrical warfare. Canadian Army Journal 7(3): 111–123.

52.

Rumer

(2006) The U.S. interests and role in Central Asia after K2. The Washington Quarterly 29(3): 141–154.

53.

Rustad

Aas

Buhaug

Falchs

Gates

(2011) All conflict is local: Modeling sub-national variation in conflict risk. Conflict Management and Peace Science 28(1): 15–40.

54.

Salehyan

(2007) Transnational rebels: Neighboring states as sanctuary for rebel groups. World Politics 59(2): 217–242.

55.

Samples

Afsar

(2008) The evolution of the Taliban. Thesis, The Naval Postgraduate School, Monterery, CA.

56.

Shaw

Spencer

(2003) Fighting in Afghanistan: Lessons from the Soviet Intervention, 1979–1989. Defense and Security Analysis 19(2): 177–188.

57.

Siebeneck

Medina

Yamada

Hepner

(2009) Spatial and temporal analyses of terrorist incidents in Iraq, 2004–2006. Studies in Conflict and Terrorism 32(7): 591–610.

58.

Sivall

(1977) Synaptic–climatological study of the Asian summer monsoon season in Afghanistan. Geografiska Annaler 59(1): 67–87.

59.

Stulz

(2008) Combating heat injury in desert wars. Medicine and Science in Sports and Exercise 40(5): 61–83.

60.

Sun Tzu ([500 B.C.] 1994) The Art of War, translated by Sawyer

. Boulder, CO: Westview Press.

61.

Turner

Adair

Hamel

(2009) Optimizing deadly persistence in Kandahar: Armed UAV integration in the joint tactical fight. Colloquium [US Army and Marine Corps Counterinsurgency Center] 29(1): 1–19.

62.

United Nations Office of Drugs and Crime (n.d.) Data for Afghanistan opium assessments. Available at http://www.unodc.org/afghanistan.

63.

Winters

Galloway

Jr Reynolds

Rhyne

(2001) Battling the Elements: Weather and Terrain in the Conduct of War. Baltimore, MD: Johns Hopkins University Press.

64.

Young

(2003) The Speed of Heat: An Airlift Wing at War in Iraq and Afghanistan. Jefferson, NC: McFarland.