Terrain and war: Measuring topographic and land cover heterogeneity in interstate wars,1816–2003

Introduction

From Thucydides forward, the study of interstate war has formed the raison d’etre of international relations. Traditionally, the field has focused on the question of war occurrence, or “why do states go to war?” In recent years, more attention has turned in earnest to the question of war outcomes, or “why do states win or lose in war?” This turn is exemplified in literature surrounding the democratic victory or the role that regime type plays in predicting outcomes. While the debate is compelling and carries broad paradigmatic and policy implications, we suggest that an element of basic discovery research is still necessary as it relates to factors that impact the exercise of power in war. This paper introduces novel terrain data in the War Terrain Indices and Geospatial Representation Dataset (WARTIGER) in response to this need and provides preliminary analysis of the impact of terrain on the exercise of power in interstate war.

The immediate purpose of these data is to furnish the study of war with the necessary evidence to draw a more complete picture of why states win, lose, or draw in interstate contests, but this data also has immediate relevance to the broader study of war writ large. Expanding on the methodological approach presented by Shaver et al. (2016), this dataset presents three novel sets of variables for each of the 94 interstate wars in the Correlates of War (CoW) population between the end of the Napoleonic Wars and the US-led invasion of Iraq (1816–2003), as well as disaggregated presentations of the First and Second World Wars. These variables include measures of approximate total spatial extent of each war in square kilometers, topographic heterogeneity measured by terrain ruggedness index (TRI), land cover heterogeneity measured using a modeled historic land cover dataset, and a trafficability index measured by aggregating the percentage present of land cover types which are relatively easy to traverse. This addition is especially necessary in the study of interstate wars given a dearth of spatially and temporally explicit terrain data. Given the cost of war in blood and treasure, it is dangerous to derive policy and paradigmatic prescriptions from incomplete data. The utilization of spatially and temporally precise topographic and land cover data allows for a deeper exploration of the complex relationship between capabilities, strategy, terrain, and regime type.

What do we know about terrain in war?

The bulk of terrain data in recent interstate war outcomes debate was generated using the New York Times Atlas of the World (New York Times, 1983). Maps from the atlas were analyzed and coded by Stam (1996) and used in nearly every work on the topic of the democratic victory, including Reiter and Stam (1998, 2002), Bennett and Stam (1996, 1998), Desch (2002, 2008), Henderson and Bayer (2013), and Cochran and Long (2017). The New York Times Atlas of the World is a reference atlas intended for a general audience and lacks the requisite information to accurately assess terrain. Land cover is limited to a few categories, and topographic information is absent aside from locations of mountain ranges. Advances in the fields of remote sensing, geographic information science, and data modelling, coupled with technological advancements in digital computation, now allow us to more accurately assess the role of terrain in war outcomes.

Recent work outside this debate reflects the ability to more accurately assess the role of terrain in war, beginning with the enhanced ability to determine spatial location globally. The Militarized Instate Dispute Locations Dataset took a first step in the serious consideration of terrain by providing precise coordinates for the location of militarized interstate dispute onsets and reiterating the importance of space in conflict (Braithwaite, 2010). Similarly, the Uppsala Conflict Data Program (UCDP; previously UCDP/PRIO Armed Conflict Dataset) provided an effective means of spatially locating an expansive set of conflicts globally (Gleditsch et al., 2002). These works, in turn, helped facilitate contributions to understanding the role of terrain in non-interstate conflicts, such as Collier et al. (2009), Fearon and Latin (2003), Tollefsen and Buhag (2015).

Accurate representation of case location allows for the collection of data relevant to the terrain features in a given conflict. However, operational faults in these data led to conflicting results. Whereas Collier and Hoeffler (2004) find that percentage mountainous terrain impacts intrastate conflict, Miguel et al. (2004) finds no such relationship. In response to this confusion, Shaver et al. (2016) provide the most robust measures to date of terrain in interstate and intrastate war through the use of a TRI, as developed by Riley et al. (1999), and land cover measures using GlobCover (Bontemps et al., 2011). The authors present ruggedness and land cover at 1 km² resolution globally and aggregate these data according to political boundaries. The dataset is applicable to a wide range of research designs, an example of which is the authors’ replication of Miguel et al. (2004) with the updated product. In short, the study of conflicts, either interstate or intrastate, has outpaced available terrain data and Shaver et al.’s work effectively responds to the problem.

However, there are limitations in the data as it relates to the study of interstate war, both spatially and temporally. Whereas lower-intensity or intrastate conflicts may be spatially limited or ill-defined, interstate war often occurs in relatively larger spatial extents—with the potential to span previously defined political boundaries by their nature. Interstate wars are spatially defined by battles as well as the movement of armies and materiel between fights. Given this, clearly defined spatial extents are crucial in determining which data to assess. Additionally, the issue of temporal variability has not been addressed by the previous work. Human activity has altered the landscape throughout history, and these changes have accelerated rapidly over the last few centuries. Increases in agricultural activity, urbanization, and deforestation have significantly changed the global landscape.

Estimation of these changes is essential to accurately characterize land cover data within war extents. Without explicit spatial and temporal discrimination, users risk substantial measurement errors. The erroneous inclusion of a mountain or the assumption of easily trafficked grassland (that was forested 100 years ago) owing to ill-defined spatial or temporal boundaries may generate data of questionable validity. The WARTIGER dataset effectively addresses these issues with precise spatial and temporal boundaries established using historical and contemporary cartographical representations of wars and the use of modeled historical data.

Defining terrain

In this dataset, terrain is defined as the physical features and characteristics of land within the spatial extent of where a war is fought. This definition is built and operationalized on the foundation of Clausewitz’s (2007, originally published 1832: 142) definition of terrain in war as the “combination of the geographical surroundings and the nature of the ground”. This presents two important elements in the operationalization of terrain in this study: spatial extent and terrain features. Clausewitz writes, “one cannot conceive of a regular army operating except in a definite space” (p. 109). While space has been alternatively conceptualized by a host of prominent thinkers and competing conceptions of space persist, for the purpose of this study, the Cartesian position—that space is defined by length, breadth, and depth—is adopted. This adoption characterizes the handling of spatial extent as length and breadth, and elevation and land cover as depth.

Germaine to this project, this can be taken to mean terrestrial surface. All but three interstate wars—The Naval War, Off-Shore Islands War, and Taiwan Straits War—prominently feature combat on land. Even in these cases, where a combination of naval, missile, and air combat predominantly caused the number of casualties necessary to reach war severity as conceptualized by Small and Singer (1976), the political motivation and consequences of the war relate to control of land. Spatial extent can be taken to mean the boundaries of a conflict as determined by political and military actors during the conflict and estimated by cartographers after the event—or more simply, where a war is fought. The features of land influence every interaction in a war—meaning we must know these features, or the “nature of the ground,” to gauge their impact. In broad terms, we take these features to mean characteristics that may influence the movement of peoples (e.g. armies, or tools of armies—horses, tanks, vehicles), provide cover (e.g. for Fabian and punishment strategies), or change the conditions of war in some other meaningful way.

Measuring terrain

The geographic assessments of topography and land cover, which enable the quantitative measurements of discrete regions on the earth’s surface (in this case, within spatial extents of interstate wars), are grounded in the field of geographic information science and spatial analysis. In the context of geographic information science, spatial analysis refers to the mathematical, statistical, and geometric techniques that can be utilized to assess spatially explicit data. Technological advancements allow for massive amounts of spatial data to be processed and analyzed rapidly using computer systems and increasingly allow for complex analyses of both localized areas and on continental or global scales. For the purposes of this project, this includes the assessment of global elevation and land cover data derived from satellite borne optical and radar sensors and their analysis within a Geographic Information System for spatial extents determined through examination of historical maps and narrative descriptions of the interstate wars in the CoW dataset.

Spatial extent

The first challenge in collecting this data is determining where a war was fought—as well as where it was not fought. States, especially larger ones, often have rather heterogenous terrains and landscapes. For example, the Caucus Region, the site of numerous wars between a host of states, contains vast expanses of mountainous forestland as well as relatively level grass and shrublands. The nature of sovereignty in interstate war suggests that there are maximum boundaries of conflicts at or near the borders of non-participant states. In this sense, interstate wars almost universally take place within participant borders. This may not always be the case in other types of wars but, generally, a violation of a non-participant’s sovereignty through territorial incursion will prompt their participation in the conflict in some form. Thus, determining spatial extent begins by determining war participants as coded by the CoW. War entails combat, movement of armed forces, and positioning and repositioning in light of opposing movement. Combat occurs in fixed locations (battles, skirmishes, etc.) but movement between these points is essential to the outcomes of wars. Combat locations and lines of movement then establish the boundaries of a given war. Casting too wide a net risks including space irrelevant to the war, while too narrow a focus risks excluding space relevant to the course and outcome of a war.

To determine spatial extent, we first turned to narrative descriptions found in Sarkees and Wayman (2010), as well as Clodfelter (2017) and Dupuy and Dupuy (1986), to determine the general course of the war—including major battles and campaigns. Secondly, we compiled a range of maps detailing these battles and campaigns, as well as the general course of troop movements. Importantly, this allows for discrimination and exclusion of space irrelevant to the course of a war. This task was complicated by the diversity of quality in these sources—largely a result of the historical nature of these conflicts. Whenever possible, we use academic or professional sources. For instance, cartographic representations of major wars such as the First and Second World Wars, as well as all American wars, are available through the US Military Academy (Krasnoborski et al., 2018). When such sources were unavailable, we turned to open-source maps hosted on Wikimedia or elsewhere. While this determination is inherently imperfect and variable in quality, we ensure the most accurate spatial representation possible with available data by verifying that cartographical representation fits the CoW, Clodfelter, and Dupuy and Dupuy narrative descriptions to ensure a minimum level of quality for all determinations of spatial extent. ²

The maps, while usually in digital form, were not spatially enabled to allow for analysis within GIS software. Therefore, the maps were georeferenced (associating the maps with geographic coordinates) using Quantum Geographic Information System and open-source satellite global images provided by Google. The Google satellite image collection, like nearly all web-mapping services, uses a pseudo-Mercator projection, so the gathered maps were all transformed to that coordinate system (WGS 84: EPSG 3857). Maps were georeferenced by associating cities or prominent landscape features (such as peninsulas, volcanoes, bays, etc.) with their locations on the Google basemap, then adjusted using linear or polynomial transformations to match the geographic coordinate system used in the software. Using these georeferenced versions of the source data, we created vectorized polygons in shapefile format representing war extent. This was done by manually digitizing the boundaries that encapsulated the furthest estimated extent of military activities for each war as represented by troop locations, arrows representing troop movement, and battles. An example of this process is shown in Figure 1.

OPEN IN VIEWER

Figure 1.

Top left, Russo-Finnish war source map; top right, geoferenced source map and spatial extent; bottom left, spatial extent and TRI mean; bottom right, spatial extent and satellite image.

In generating these files, we made the decision to act conservatively in the determination of spatial extent—erring on the side of smaller extents, rather than larger, more inclusive extents. For instance, a few kilometers expansion in a given polygon may capture the odd unit’s movement away from the primary fight or even capture an additional small skirmish, but it risks including a landscape feature that would dramatically change the data. These polygons or groups of polygons, representative of a single war’s spatial extent, were used to generate topographic and land cover heterogeneity metrics. We then used these shapefiles to select land cover data from Integrated Science Assessment Model—History Database of the Global Environment (ISAM-HYDE) data (Meiyappan and Jain, 2012). ³

Topographic heterogeneity and TRI

A TRI (Riley et al., 1999) provides a relative measure of an area’s ruggedness and is used to measure topographic heterogeneity for this research. Digital elevation models are the only requisite input data for calculating TRI. Digital elevation models are available from numerous sources at a wide range of resolutions. For our purposes, high-resolution datasets were not required. Therefore, we use the Global Multi-resolution Terrain Elevation Data 2010—jointly produced by the National Geospatial Intelligence Agency and the United States Geological Survey (Danielson and Gesch, 2011). In this dataset, elevation data is presented globally in 30 arc second (approximately 1 square kilometer) pixels with height relative to sea level in meters. TRI for each pixel is calculated by measuring the difference in elevation between it and its eight adjacent pixels. These differences are then squared and averaged, with the square root of this value producing a TRI (Riley et al., 1999: 25). Once the global TRI dataset was processed (see Figure 2), we compiled statistics for each war extent polygon or polygons, including total area, mean TRI, minimum TRI, maximum TRI, and TRI standard deviation. The primary variable of interest is the mean TRI of a given war, but additional variables may be of interest in demonstrating the heterogeneity of ruggedness in a contest. Mean TRI data is also presented as a categorical variable following Riley et al. (1999) as level, nearly level, slightly rugged, intermediately rugged, moderately rugged, highly rugged, or extremely rugged. ⁴ No wars occur in the extremely rugged TRI category and only two wars occur in the highly rugged category.

OPEN IN VIEWER

Figure 2.

Global topographic heterogeneity.

Land cover heterogeneity and trafficability

The second set of terrain variables, which measure land cover heterogeneity, details the percentage of unique land cover types present within a war’s spatial extent and introduces a trafficability index calculated by aggregating the percentage of land cover types present in a binary classification based on ease of movement as either trafficable or non-trafficable. These variables are calculated using the ISAM-HYDE land cover dataset (Meiyappan and Jain, 2012). The ISAM-HYDE land cover data is derived from the HYDE 3.1 database (Goldewijk et al., 2011). It presents 28 unique land cover classes for the entire planet at yearly intervals, using recent moderate resolution remote sensing data to calculate a baseline and modeling land use and land cover changes based on a variety of historical sources including tree harvest data, agricultural data, and human population data. ISAM-HYDE data is presented at a 0.5-degree resolution (approximately 55 km²) and details the percentage present of each of the 28 land classes for each 0.5-degree grid cell. Although the spatial resolution is substantially lower than that of the elevation data, ISAM-HYDE data was selected because of its high temporal resolution, providing data on a year-by-year basis for every year represented in the CoW dataset. Using data with higher spatial resolution (such as GlobCover) provides only a single snapshot of current or recent land cover and does not account for the vast anthropogenic changes that have altered the landscape over the past two centuries.

We use the war extent polygons to select each 0.5-degree grid cell that falls within the extent of polygons, and then calculate the average percentage for each of the 28 land cover classes present in each war. In addition, we collapse land cover percentages into two broad categories as either trafficable or non-trafficable. Trafficable classes are cover types which can be easily traversed, such as hard-packed terrains, plains, tundra, and cropland. Non-trafficable classes are cover types which are difficult to traverse, such as forests, dense shrublands, and water. Land cover trafficability coding follows Historical Evaluation and Research Organization coding (Dupuy, 1983). For multiyear wars, we select ISAM-HYDE data from the first year of the war given difficulty in acquiring temporally disaggregated maps of uniform quality for each year of each war. In Figure 3, we see the dramatic transformation of trafficability between 1823 and 2003, which necessitates the use of modeled historical data, especially surrounding the boreal and tropics.

OPEN IN VIEWER

Figure 3.

Global trafficability 1816 and 2003.

Conditionality and data limitations

The efficacy of the application of capabilities is partially determined by terrain. Terrain influences nearly every facet of ground action. Certain land cover classes or rugged terrain may impede the movement of forces, while providing cover and protection for others. While the impact of terrain may always be unequal—aiding one state and impairing another, even in close proximity—this inequality is most pronounced in mismatched strategies and uneven capabilities. For instance, if State A engages in a maneuver strategy against State B’s attrition strategy, we might assume that State A enjoys strategic advantage and State B suffers strategic disadvantage.

This is only conditionally true. The classic example of the above strategic advantage is Germany’s blitz through Western Europe. This was partially facilitated by new tools of war in the form of the tank—but the German military was still largely and literally horse-powered. German forces engaged in “sweeping advance[s] which bypassed strong points for later reduction by slower-moving elements” (Dupuy and Dupuy, 1993: 1113). It was this maneuver-vs-attrition strategy which defined early German successes against Polish, Belgian, French, Dutch, Danish, Norwegian, Greek, and Yugoslav forces. In all of these cases and places, with the exception of Yugoslavia and Greece, German maneuver-vs-attrition strategic advantage was facilitated by terrain. In Poland, Belgium, France, Holland, and Denmark, German forces enjoyed considerable allowance from terrain that was level and predominantly trafficable. In Norway, German forces repeated the speed of previous and concurrent successes until they reached the mountainous area surrounding Narvik—where terrain and British and French support only delayed German victory. Germany maintained this success in moderately more difficult terrain in Greece and Yugoslavia. Terrain in the Italian–Greek War largely benefitted Greece’s defensive-mobility strategy. The initial Italian offense, made through Albania, encountered fierce resistance in the rugged Epirus region. After Italian advances stalled, a Greco counteroffensive and maneuver strategy quickly pushed Italian forces back into Albanian territory. The tide would only change when Germany blitzed into Macedonia in relatively easier terrain (with Greek forces tied down in the east). In Yugoslavia, there was little to be done to stop the German invasion despite rugged conditions. Following a coup on 27 March 1941, German forces invaded a mere 10 days later on 6 April. The million strong Yugoslav army failed to mobilize amidst the tumult and suffered an astounding loss exchange ratio of 179:1. Table 1 highlights the ruggedness and trafficability of each of the disaggregated Second World War cases.

OPEN IN VIEWER

Table 1.

Mean TRI and trafficability by theater in the Second World War.

Theater	Mean TRI	Trafficability
Poland	39.57 m	70.85%
Belgium	61.86 m	67.81%
France	58.66 m	77.68%
Holland	20.25 m	75.53%
Denmark	9.03 m	70.56%
Norway	160.81 m	25.07%
Greece (Italian)	328 m	56.13%
Greece (German)	232.68 m	61.42%
Yugoslavia	175.52 m	53.67%
Soviet Union	39.27 m	69.94% (23.13%) a

a

When cropland is coded as non-trafficable.

The tide of German victory turned famously in the East, but not immediately. The German–Soviet War is among the most brutal in human history and the landscape contributed to this brutality. At face value, the terrain seems favorable to the German maneuver strategy. However, several factors combined to remove strategic advantage, demonstrating a conditionality of land cover trafficability. The first was the massive spatial extent of the war. We approximate the spatial extent of the Second World War in its aggregated form at a massive 2,190,850 km²—with a conservative estimation of the German–Soviet War comprising 1,148,200 km² or more than half of the spatial extent of all of the Second World War. In this sense, the sheer scope of the war increased the cost of movement. Secondly, the contest was a “fair fight,” with the Germans holding a small but comfortable capabilities advantage (1.63 to 1). ⁵ In practice though, the parity was greater in that German forces were divided between several fronts and tasked with maintaining previous gains, both in the earliest stages of the war in Yugoslavia and Greece, and later following the Allied campaigns in North Africa, Italy, and Normandy. The initial stages of Operation Barbarossa matched approximately 3 million Axis forces against 3 million Soviet forces, with roughly another 1 million forces scattered throughout the Soviet Union. Third, while terrain was nominally level across the spatial extent of the war, the landscape was only conditionally trafficable by land cover class. The dominant land cover classes of various croplands and grasslands transformed in wet weather and under the movement of massive armies into a muddy quagmire—slowing the movement of infantry, horses, tanks, and materiel. Figure 4 shows the difference between trafficability when cropland is handled as either trafficable or non-trafficable.

OPEN IN VIEWER

Figure 4.

Conditional trafficability of the German–Soviet War.

These elements, taken in tandem, demonstrate a unique relationship between capabilities, terrain, and strategy. While the Soviets expected invasion, Operation Barbarossa benefitted from tactical surprise. It commenced with a two-fold plan of attack: German and Romanian forces would blitz toward Kiev and on to the Dnieper Valley in the south, while the other prong would drive to Warsaw then onto Smolensk and Moscow. Finnish forces were to threaten Leningrad from the north. The initial stages of the war between July and November 1941 were among the most impressive campaigns ever: German forces managed to inflict some 3 million casualties (about 1.5 million of these were prisoners), but they did so at a cost of 800,000 casualties (Dupuy and Dupuy, 1993: 1183).

Yet terrain and time caught up to Nazi forces. Unable to deal a killing blow given unexpectedly challenging terrain and facing a seemingly endless supply of Soviet reinforcements, winter set in upon German forces in summer dress. The sheer spatial extent of the war, coupled with a surprisingly harsh landscape, proved a stumbling block to German mobility. Further attempts to regain mobility faltered into sieges in non-trafficable urban settings such as Stalingrad and Leningrad—all while the Soviet war machine slowly rumbled into gear. The impact of terrain on war outcomes is often unique to the spatial and temporal setting of a given war, but this does not discount broad lessons on the impact of certain terrains and landscapes on outcomes. This suggests that the historical narrative of each war should be considered in conjunction with terrain data when assessing outcomes, as an exogenous event, like weather or human modification, can make a normally trafficable environment non-trafficable.

Beyond conditionality, there are certain permanent features of the land which structure space: lakes, rivers, valleys, hills, etc. These features are often present in the easiest of terrains, yet their importance is profound. An entirely level and trafficable place may be easy to traverse until one of these points is reached—i.e. a river without a ford requiring either a permanent bridge, an ad hoc structure (such as pontoons), or tactical diversion to an easier crossing. The tradeoff in capturing such an expansive measure of landscape is losing the peculiar. These are often prominent and defining features of specific battles, such as fords on the Rappahannock during the Battle of Chancellorsville, but do not define entire wars or landscapes. Thus, this measure is in keeping with the general theme of measuring terrain in war writ large.

Our terrain data offers a snapshot of the full spatial extent, rather than tracking its evolution over time. The tides of war often change. For instance, in the 1934 Saudi–Yemeni War, Saudi forces used mobile armor to route Yemeni forces across the desert landscape. However, once Yemeni forces were backed into Sanaa, the mountains slowed Saudi movement. In this sense, the impact of terrain was unequal over time. This is a general problem that exists in every case—but is largely an unavoidable one at this stage of data collection. Regardless, this simply suggests an intuitive point: just as war cannot be divorced from place, it cannot be divorced from time. In turn, it is also worth considering changes in how wars are fought, especially in the meeting of technology and strategy. On the one hand, at least in contests between matched powers, the changes probably develop in tandem. In asymmetrical war, the technological inequalities are likely more pronounced.

While speculative, there are real-world examples. Israeli air superiority and tactical surprise in 1967 partially enabled their decisive victory, but by 1973 mobile and static surface-to-air missiles and infantry shoulder-mounted missiles neutralized this superiority. Terrain, too, may mitigate these advantages. Non-trafficable terrain (e.g. dense forests) or rugged terrain (e.g. mountains) limits disparities in technology and weaponry. Operation Rolling Thunder in Vietnam dropped more ordinance than all of that in the Second World War combined but did little to break the North Vietnamese Army, just as insurgents in Afghanistan have evaded American air strikes in rugged mountains. The cost of movement, especially in easy terrain, should decrease with mechanization (Bueno de Mesquita, 1981: 104; Dupuy, 1985, 16). ⁶ In difficult terrain, this point is more ambiguous and probably conditional. Helicopters and paratroopers may quickly advance but tanks, trucks, and other vehicles still cannot climb mountains and pass through dense forests, swamps, or other obstacles (at least not without the presence of roads). To an extent, we believe technological parity to be implicit in Composite Index of Nation Capability (CINC) data, especially in per capita energy consumption, but should be considered when using this data.

In a similar vein, the question of power projection should also be considered as power decays over distance. Boulding (1963) argued that the further a state is from the place it seeks to exercise its power, the weaker it will be. This occurs for a number of reasons: distance compounds organization and command problems, lowers morale, increases domestic dissent, and weakens soldiers and equipment (Bueno de Mesquita, 1981: 41). It may also exacerbate unfamiliarity with terrain, leading to less effective strategy selection. There is some evidence to suggest that this impact is less pronounced in recent history. This would assume, as Boulding did, that innovation in transportation and air and missile capability have mitigated the loss of strength gradient (Boulding, 1965). Martin (2016) suggests that today there is not a loss of strength gradient, but rather a “loss of time gradient.” Specifically, Martin argues that with proper afloat-support logistics—and their speedy use—power is not lost with distance. On the contrary, Webb (2007) suggests that only with the use of forward-positioned bases can a state mitigate the loss of power by proximity to a target. Regardless, proximity may change the conditional impact of trafficability, extending or contracting supply lines, familiarity, and change interaction with terrain.

Discussion

In this section, we introduce a brief discussion on the use of WARTIGER data in studying interstate war outcomes through the use of a simple multinomial logistic regression model to predict the odds of winning, losing, or drawing in war, with lose as the reference category. Additionally, we demonstrate the intervening role that terrain plays in impacting the efficacy of the application of capabilities using the predictive margins command in Stata. The following model is deeply informed by various works in the democratic victory debate, especially Stam (1996), Reiter and Stam (1998), and Bennett and Stam (1998). The model uses the single nation-state as the unit of analysis and includes relative state and alliance capabilities measured by the Composite Index of National Capabilities (Singer, 1987), regime type scores measured by PolityIV (Marshall et al., 2013), initiation coding following Sarkees and Wayman (2010), strategic advantage and disadvantage coding from Stam (1996), and three variables from the WARTIGER data, including mean TRI, percentage non-trafficable, and spatial extent. Original strategy coding is applied to wars after the mid-1980s using Clodfelter (2017) and Dupuy and Dupuy (1993).

Reflecting prevailing findings in the interstate war outcomes literature, increases in capabilities and alliance partner capability contributions increase the odds of victory, as do initiation, strategic advantage, and increases in PolityIV scores. ⁷ As expected in these models, terrain variables are not significant in prediction of victory alone, given that they serve as a constant in these models amongst opponents in war. Put differently, opponents share the same terrain values in this simple model and serve as a control. However, in draws, where there is a consistent outcome across opponents, an increase in ruggedness promotes draws. If we assume that the primary influence of terrain is on the strategic application of capabilities, the real test is in the interaction of strategy and terrain. A summary of model results is shown in Table 2.

OPEN IN VIEWER

Table 2.

Outcomes model.

	Variable	B	SE
Win	State Capabilities	3.24***	0.66
	Alliance Capabilities	4.61***	0.73
	Polity IV	0.07**	0.02
	Initiator	1.15***	0.01
	Strategic Advantage	2.03***	0.54
	Strategic Disadvantage	−2.90***	0.50
	Percentage Non-Trafficable	−0.01	0.00
	Mean TRI	−0.01	0.02
	Total Area	0.00	0.00
	Constant	−1.61**	0.52
Draw	State Capabilities	1.45	0.98
	Alliance Capabilities	4.04***	0.98
	Polity IV	0.09**	0.03
	Initiator	0.44	0.51
	Strategic Advantage	0.012	1.14
	Strategic Disadvantage	−3.10***	0.83
	Percentage Non-Trafficable	0.016	0.01
	Mean TRI	0.05**	0.02
	Total Area	−0.02	0.01
	Constant	−3.28***	0.83
	Number of Observations: 315Log likelihood: −206.47Pseudo R2: 0.32

***

Significant at p < 0.001; ** significant at p < 0.01; * significant at p < 0.05.

mlogit WLD2 Concap AlliAsst polityIV NTraff TRIMean10 count1000 ib(2).Initiator ib(0).Winstrat ib(0).Losestrat, baseoutcome (2). B, Coefficient; SE, standard error.

Using the predictive margins command in Stata, we demonstrate an important point. As terrain becomes either increasingly rugged (in 50 m increments between 0 and 700 m; see Figure 6) or non-trafficable (in 5% increments between 0 and 100%; see Figure 5), the benefit of strategic advantage decreases. ⁸ Similarly, the negative impact of strategic disadvantage is less pronounced in such environments. Disaggregating strategy into its base categories following Stam (1996) as attrition, maneuver, and punishment demonstrates that the utility of maneuver strategies ceases above 20% non-trafficability, as seen in Figure 7. ⁹

OPEN IN VIEWER

Figure 5.

Predictive margins of strategic advantage × percentage non-trafficable.

OPEN IN VIEWER

Figure 6.

Predictive margins of strategic advantage × ruggedness.

OPEN IN VIEWER

Figure 7.

Predictive margins of maneuver strategy × percentage non-trafficable.

If we can, cautiously, assume that a war’s outcome is the product of capabilities, strategy, regime type, and terrain, then we must understand their complex relationship. This study is complicated by the diversity of an already limited number of cases. While capabilities are consistently the strongest predictor of outcomes in statistical models across studies, numerous wars demonstrate the deterministic limits of capabilities. By classifying where wars are fought by terrain ruggedness and trafficability into one of four categories, low ruggedness–high trafficability, high ruggedness–high trafficability, low ruggedness–low trafficability, or high ruggedness–low trafficability, we gain some insight into this relationship (see Figure 8). ¹⁰

OPEN IN VIEWER

Figure 8.

Scatter plot of wars by TRI mean and trafficability.

The majority of wars occur in the “easiest” category of low ruggedness–high trafficability with some 53 wars. This gives reason for the primacy of capabilities in explaining outcomes as the exercise of capabilities is easiest in such environments. Lacking rugged topography or non-trafficable land cover, force may be more effectively applied against opponents in these settings, all while there is less opportunity for hiding or being shielded from such force. In the remaining three categories, there exists a relatively even distribution of wars. Fourteen wars occur in the high ruggedness–low trafficability class, 15 wars in the low ruggedness–low trafficability class, and 20 wars in the high ruggedness–high trafficability class. ¹¹ Among these classes, we also see the majority of surprise or upset outcomes, where a state facing a gross disadvantage in capabilities wins out. In this sense, we see that terrain plays an essential intervening role in determining outcomes—at times facilitating strength or humbling the capable, alternatively, exposing the weak or sheltering them. This supports the basic Clausewitzian proposition that capabilities as the means, applied strategically, and impacted by terrain, combine into a state’s ability to realize power in war.

For instance, in easy terrain, Nazi Germany and Israel have arguably been the most successful states in war in the twentieth century (Desch, 2002). Both states demonstrate masterful execution of maneuver strategies across easy terrain, with Nazi victories across Western Europe and Israeli victories in the Sinai. Both states failed to achieve victory in difficult terrain, with Germany falling on the muddy Russian steppe and Israel fighting to stalemate in 1969 ¹² and 1982. Similarly, facing a gross disadvantage, Chadian forces overwhelmed Libyan forces in the War of the Aouzou Strip in 1986. The war, alternatively known as “the Toyota War,” was defined by the Chadian use of pick-up trucks (gifted by France) to disrupt and route Libyan forces across the flat and hard-packed theater.

In difficult terrain, the full exercise of capabilities is increasingly challenging. The case of American involvement in Vietnam clearly demonstrates this point. Despite the severity of Operation Rolling Thunder, the largest bombing campaign in human history, the strength of the American military could not pierce the expansive tropical canopy. On the ground, American forces were limited by challenging tropical forests and water-logged croplands. In a similar vein, Soviet forces only gained pyrrhic victory over Finnish foes in the winter of 1939–1940—where the landscape was heavily forested, pockmarked with water, and entirely frigid. While the US enjoyed substantial capabilities advantages in the Pacific Theater of the Second World War, the sheer power projection needed, coupled with brutal rugged and non-trafficable landscapes—not to mention Japanese resolve—made the war one of gain by bloody inches. Similarly, Indian and Pakistani forces could hardly deploy their full strength in the rugged and glacial Himalayas in Kargil, depending instead on sporadic artillery and the boldness of soldiers willing to scale cliff faces across various peaks like Tiger Hill. While wildly different in their composition, these settings make the movement of large groups challenging—inherently limiting the full application of capabilities.

Conclusion

In this article, we introduce the WARTIGER dataset with the goal of bringing terrain data in the study of interstate war up to par with increasingly complex research designs. To this end, we introduce three sets of variables for a total of 112 cases between 1816 and 2003: spatial extent, topographic heterogeneity measured using a TRI, and land cover heterogeneity measured using a trafficability index. WARTIGER data is set apart by its careful spatial and temporal treatment of terrain specific to interstate war. We argue that the importance of using spatially and temporally precise data cannot be understated. This is especially true given its central role in the growing democratic victory literature. As we demonstrate, terrain is an essential element in the outcomes puzzle as it impacts the strategic application of capabilities.

Our knowledge of war outcomes will always be limited by its rarity, but potential consequences demand our attention. The spirited efforts throughout the interstate outcomes literature show great promise. However, the gravity of war also demands our caution. To derive paradigmatic and policy prescriptions from incomplete data may cost lives. This risk may be especially true of the democratic victory given a temptation to advocate the spread of democracy by force under the assumption it will end in victory. Similarly, assuming a state enjoying gross capabilities advantages will enjoy victory may be hubris and quagmire her Nemesis. We hope that our additions to this research will ultimately fill in some of these gaps, but we recognize that our work is only one step in understanding the complex puzzle of outcomes. Future research should use this data to explore the causal relationship between capabilities, strategy, and terrain in predicting outcomes but is broadly applicable to a wide range of research designs, including those relating to battlefield effectiveness, strategy selection, and others. In addition, our work may be a resource in the historical study of individual wars by explicitly detailing their terrain, especially in disaggregated forms of land cover. In relation to the democratic victory, future work should consider if democracies are superior in strategically executing capabilities in light of terrain. Of course, it is always worth remembering that the only sure way to avoid defeat in war is to avoid war itself.

Supplemental Material

Research Data