Armageddon on the Aufidus: Locating the Battle of Cannae

For the historian of ancient warfare, there can be few tasks as frustrating – or as tempting – as trying to locate the sites of ancient battles. Our sources rarely say much about topography, what they say is usually vague or contradictory, and, to make matters worse, two thousand years of earthquakes,  floods, and common-or-garden erosion is generally enough to transform any landscape, especially when assisted by the twin forces of farming and construction. 

Despite these difficulties, military historians over the years have relentlessly sought the locations of the major battles of Antiquity, notable the great clashes of the Second Punic War, Rome’s infamous life-and-death struggle with Carthage at the end of the third century BC. Predictably enough, considering the spell it’s cast on tacticians and military historians over the years, the quest to locate the battle of Cannae has been an unashamed free-for-all. Offhand, I can think of about a dozen different theories of where the battle took place.

Just to put this into context, in the summer of 216 BC the Carthaginian general Hannibal was marauding around Italy with an army of around 50,000 men.  Having been decisively thrashed by him in their three previous encounters, the Romans concentrated on building up their army to a total strength – perhaps, this is debated – of 86,000 men. Hannibal seized the grain stores at Cannae, near Canusium, and the Romans under their consuls Paullus and Varro moved closer, evidently planning to give battle soon. They set up camp by the river Aufidus, across the river from Hannibal’s camp. Hannibal crossed the river and made camp, the next day offering battle to the Romans, who declined. 

The following day the Romans under Varro crossed the river to challenge Hannibal, and Hannibal led his men across the river to face them. Despite being massively outnumbered, Hannibal was able to outflank the Romans and surround them, killing 50,000 men in a manoeuvre – the ‘double envelopment’ – which has been a model to generals ever since.

At the south-western end of the citadel at Cannae there’s a small modern monument, a simple column bearing an inscription from Livy, with a magnificent view of what you might assume is the battlefield. You can see it here, a broad flat plain with the river Ofanto – the ancient Aufidus - shown running across the centre of the shot, rendered clearly visible by the trees and bushes that line its banks. The town in the background is San Ferdinando, which some historians have identified as the site of Hannibal’s camp before the battle. It all looks very straightforward – and if you ask in the museum, or at the small tourist office at the train station, you’ll be told that it is. Unfortunately, as with virtually everything that we touch in the field of ancient history, matters aren’t anywhere near so simple.

Back in the late sixteenth century it seems to have been normal to assume the battle took place on the south-eastern bank of the river. If you ever visit the hall of maps in the Vatican Museum and seek out the map of this area  by the Dominican friar Ignazio Danti you’ll see that not only has Danzi seen fit to mark the site of the battle of Cannae on the southern bank of the river, with the two main camps on the northern bank, but he’s painted both armies deployed for battle with the opposing commanders clearly identified. He does a similar thing with other maps in the series, showing, say, the opposing camps before the Battle of Lake Trasimene, and Caesar about to cross the Rubicon.

If you look at a satellite image of the region round the hill of Cannae, with the river Ofanto – the ancient Aufidus – running in a more or less north-easterly direction from the bottom left corner, you can just about make out the hill of Cannae in the very centre of the shot.

What positive information can we gather from the sources? Well, Polybius 3.113 clearly states that the Roman right wing – the citizen cavalry were positioned by the river, faced by the Carthaginian left wing – the Celtic and Spanish horse. Further, Polybius 3.114 and Livy 22.46 tell us that the Romans faced south, the Carthaginians north. We needn’t be too dogmatic about what Polybius meant when he wrote ‘north’ and ‘south’ here. It can be taken as read, I think, that he meant ‘roughly north’ and ‘roughly south’; it seems unlikely that either Hannibal or the Roman commanders were using compasses to deploy their armies. 

What’s more, both armies had to cross the Aufidus to do battle, since most of their men were camped on the far side of the river; Polybius 3.113 states this unambiguously, while Polybius 3.110 indicates that they must have crossed from west to east, since only a third of the Roman forces were encamped on the eastern side of the river.

It can be helpful to see how people have read the data and attempted to locate the battle over the years. The turquoise box on this black-and-white satellite shot marks the area generally identified as the site of the battle of Cannae up to the early nineteenth century, and still generally pointed to as the battlefield by locals in Barletta and thereabouts. 

Obviously, I've enhanced the course of the river; the hill of Cannae should still be clear in the centre. The problem with the obvious identification is that it’s based simply on the assumption that this is a good, handy, flat space that would have been an obvious spot for a fight. Quite right, but unfortunately the sources rule out this location, at least if we take them at face value.

It seems pretty clear, from the aforementioned passages, that the battle must have been fought on the right bank of the river, with the two armies deployed more-or-less at a right angle to it, the Romans facing roughly south, the Carthaginians roughly north. Some nineteenth-century writers attempted to place the battle to the southwest of the hill of Cannae, notably Hesselbarth in his 1874 Göttingen dissertation and Thomas Arnold in 1886. Unfortunately, a cursory inspection of the area – represented with an orange box – ought to have ruled out such ideas – the ground they identified as the site of the battle is rugged and hilly, unsuitable for an infantry battle, let alone a battle involving the use of about 15,000 horses and some fairly sophisticated cavalry manoeuvres.

Attention returned once more to the left bank of the Aufidus, with the American Theodore Dodge placing the battle almost parallel to the river, close to the coast, as represented by the pink box; Dodge wasn’t working so much on the basis of the evidence as he was using what’s termed "inherent military probability". This is the idea that soldiers of whatever era will do what makes military sense, and that bearing in mind certain principles of warfare we can establish what historical generals are likely to have done. Although it ignores the fact that war’s a cultural activity – it varies in aims and methods between cultures – it is a useful tool. However, it’s a tool far too easily abused, as in this case; there’s not a jot of evidence to support Dodge, and plenty of it to refute his theory.

Konrad Lehmann and Hans Delbrück hypothesised that the battle was fought on the river’s left bank, a couple of miles west of the hill of Cannae, as represented by the green box, but Lehmann has the Romans facing roughly north and the Carthaginians roughly south, in direct contradiction of the sources, while Delbrück had the Romans facing roughly east and the Carthaginans roughly west. Again, their theories, however ingenious, lacked any real basis in our evidence. 

De Sanctis, at least, in placing the battlefield on the left bank as shown by the yellow box, with the Romans facing north-east and the Carthaginians facing south-west, had an explanation. Polybius, he believed, thought of Italy as a triangle with its base in the Alps and bisected by the Apennines; the Aufidus thus must have, in Polybius’ mind, flowed in a south-easterly direction; since the Aufidus in fact flows north east, all Polybius’ directions must be corrected – a Roman force facing roughly southeast, for Polybius, would in fact have been facing northeast, with its right flank on the river. 

It’s a clever idea but it doesn’t really follow that Polybius if envisaged Italy as a triangle he must have assumed that all rivers flowing into the Adriatic flowed south-east; besides, considering that Polybius explicitly speaks of the Aufidus having an east bank, it would seem that he believed the river to have flowed from south to north. De Sanctis’ theory, though ingenious, doesn’t hold up.

In 1912, Johannes Kromayer proposed that the battlefield was in fact on the river’s right bank, downstream from the hill of Cannae, on a front over four kilometres wide, with the Romans facing south-west-south and the Carthaginians north-east-north, the Roman right flank and the Carthaginian left flank both resting on the river. The red box here represents Kromayer's theory. Over time this theory gained more and more ground, eventually becoming generally accepted as the one that best fits the evidence, the topographical reality, and military practicality. See, for instance, the definitive book on the battle*. (Cough, cough)

Yes, I'm using the drawing in my book. So sue me.

 In 1981, however, Peter Connolly dropped a very astute bombshell onto this increasingly cosy consensus. It was all very well working on the basis of the texts, he pointed out, but we only have two items of geography to correlate with our literary evidence, and while the hill of Cannae is conveniently stationary, rivers have a habit of moving. What, he asked, if the Aufidus had flowed rather further north in 216 BC than it had done in his day? 

Assuming the river was at the northernmost limit of its floodplain in 216 BC, he placed the main Carthaginian camp at the modern town San Ferdinando with the main Roman camp just over a mile downhill. His proposed site for the battle, shown in pink below, falls down largely because it doesn’t allow for Hannibal having challenged the Romans to battle on the left bank of the Aufidus the previous day. Polybius 3.112 describes Hannibal’s army deploying for battle along the Aufidus, without crossing the river, and if we follow Connolly’s theory there simply isn’t space for them to have done that in the narrow stretch between San Ferdinando and Connolly's hypothetical battlefield.

Nevertheless, Connolly had raised an important point, one that Adrian Goldsworthy was to take up a few years ago in his book Roman Warfare (as shown in orange) and subsequently revisit in his books on The Punic Wars (with the battle located in the general area of the dotted turquoise box) and Cannae (as shown in olive green), where he explained his theory. Goldsworthy clearly changed his mind a couple of times as to the precise location of the battle, but that’s no argument against his basic thesis, which needs to be addressed. 

• First, he says, Kromayer requires Hannibal’s camp to be in the open plain rather than on higher ground at the modern San Ferdinando – of this he says "there could have been no intrinsic value to such a position apart from the pressure it applied on the enemy by its proximity to their camp". 
• Second, he argues that the location where Varro offered battle, according to Kromayer, is no less suited to cavalry combat than that location declined by Paullus the previous day. 
• Third, he points out that we have no reason to believe the course of the Aufidus in 216BC was anywhere near where it is now, and it could well have been much further north as per Connolly’s theory. 
• Finally, he says that Connolly has overestimated – and Kromayer has grossly overestimated – the frontage that a Roman army of 76,000 men would need. It would be possible, he argues, to squeeze an enormous Roman army into a battlefield only two kilometres wide – and with one Roman flank anchored on the river and the other flank anchored on the hill of Cannae or thereabouts, the Romans would be able to thwart any attempts by Hannibal’s cavalry to outflank and surround them.

Where do we start? Well, firstly there’s the issue of Hannibal gaining nothing from being encamped in the plain, barring the pressure that he’d apply on the Romans through being so close to their camp. Frankly, that could be reason enough to camp there – Hannibal wanted to draw the Romans out, after all. He wanted them to fight him, since he was sure that if they fought, he’d win.

What then of the charge that Kromayer’s proposed location for the battle is just as well-suited to cavalry combat as the the plain between the camps, in Kromayer’s thesis? That’s simply not true. The map I've cunningly nabbed from my own book shows that the terrain between the two camps is as flat as a pancake, perfect for cavalry combat, while there are at least a few contours in the area where Kromayer places the battle. But even if that area had been better suited to cavalry combat than the area where Hannibal had offered battle the previous day, Varro would still have had good psychological grounds for offering battle there. The previous day Hannibal had picked his terrain – now it was time for the Romans to pick theirs. This is an example of the dangers of applying "inherent military probability" to cultures that are different from our own.

Third, and now we’re moving into the meat of the Connolly/Goldsworthy hypothesis, what if the Aufidus flowed rather further north in Hannibal’s day than it does now? The first thing that has to be borne in mind is that, yes, the river’s current path almost certainly doesn’t match its route in 216BC. It doesn't follow exactly the same route now, even, as it did when Kromayer drew his map. Rivers move, and over the course of six or seven centuries the river could easily move from one side of its flood plain to the other. 

The Aufidus – or the Ofanto as it’s now called  isn’t a particularly impressive river, after all. It’s neither broad nor deep, and lacks in power – it’s the type of river that would always take the path of least resistance. But there’s no evidence at all for what path it took in Hannibal’s day. Yes, it’s rather arbitrary to assume that the river’s current course matches its course when the battle was fought, but it’s equally arbitrary to assume that its course in 216 BC was along the northern limit of its floodplain. It’s even possible that its course then was slightly further to the southeast than it is now, so that it would have tightly hugged the hill of Cannae.

Certainly, whatever path the river took when the battle was fought, two things are definite. Firstly, its general direction must still have been towards the northeast, and second there's no way it followed the almost straight path we see in Connolly and Goldsworthy’s diagrams – though not in Connolly’s reconstruction, which is a more accurate reflection of reality. Rivers don’t follow straight paths in their old age – they meander wildly, as Gerry Fee taught me in geography class millions of years ago, and whatever course the Aufidus would have followed would have been marked by twists and turns, just as today’s river is, if not more so. 

The only way to find what the course of the Aufidus would have been in 216 would involve a close topographical and geological study of the area, looking for evidence of meanders and oxbow lakes – these would almost certainly have been filled in over time, but they could be found. Once found, they could be cored, and the cored deposits could be dated. Without such a project, any attempt to locate the ancient course of the Aufidus must be regarded with suspicion.

What then of the idea that a Roman army of 70,000 infantry and 6,000 cavalry could have been squeezed into a plain two kilometres wide? Well, it’s possible. It does seem terribly convenient, though, doesn’t it? The Roman army can, just about, be squeezed into a frontage two kilometres wide… and the gap between the hill of Cannae and the river Aufidus can, just about, be stretched to two kilometres wide. It does rather look as though the facts are being forced to fit the theory here. Besides, it would have been very tricky to deploy an army squeezed so tightly together into such a narrow corridor.

What about the suggestion that with their flanks protected by the river and the hill the Romans would have been well-protected, at least in theory, from Hannibal’s cavalry? Well, the first thing you might wonder is "why doesn’t Polybius or any of our other sources for the battle even hint at the Romans having adopted such a position?" That doesn’t disprove Goldsworthy’s theory, by any means, since ancient writers are often far from forthcoming on the topography of battlefields, but Goldsworthy’s asking us to accept that Polybius was willing to relate the positions of the armies at Cannae to one of the battle’s crucial landmarks, but not to the other. Such a half-silence would be curious, to say the least.

We also need to remember that the encounter at the Aufidus had been invited in the first place by Hannibal’s seizure of Cannae’s citadel, as is described in Polybius 3.107 Hannibal had since then moved the bulk of his army to the left bank of the Aufidus (Polyb. 3.111.11), but it seems unlikely that he would have altogether abandoned the most important and defensible strategic point in the vicinity. For what it’s worth, Plutarch suggests that Hannibal had control of the heights around Cannae when he describes a rather feeble joke made by Hannibal while on a hilltop viewing the Roman army being deployed (Plut., Vit. Fab. Max. 15.1)

If Hannibal did control the hill of Cannae it would have been suicidal for the Romans to have anchored their left flank on it, exposing their men to a potential bombardment of missiles from above – such missiles might not just include the heavy stones that Hannibal’s Balearian slingers could hurl for great distances, but even such simple things as rooftiles! Again, this doesn’t definitively refute Goldsworthy’s thesis, but again it should cause us to pause before swallowing it.

One thing that’s crucial to remember in evaluating this hypothesis is that the hill of Cannae is not an isolated hill. Rather, it’s part of a ridge about fifty to sixty metres high You can pick out Cannae on this shot, just about, by squinting and looking for the memorial column. 

These hills are quite steep with something in the region of a 45 degree slope. All very well, you might think, but bear in mind how little space Goldsworthy allows for the Roman frontage – somewhere in the region of two kilometres – 360 metres for the 2,400 citizen cavalry at the river, 1050 metres for the infantry in the centre, and 540 metres for the 3,600 allied cavalry at the foot of the hills. This seems terribly constricted. 

It’s reasonable to assume that the cavalry by the river were squeezed tightly together –  Polybius 3.115 and Livy 22.47 relate how with little room to manoeuvre the cavalry battle by the river turned into a barbaric melee with all fighting being done at close quarters. But what of the cavalry on the other flank? Polybius 3.116 gives no hint that the cavalry there lacked room to move; on the contrary, it seems as though that part of the battle was characterised by the typical skirmishing and repeated sallies so typical of the Numidian cavalry; there’s certainly no suggestion of any close combat between the Numidians and the Allied cavalry, something which surely would have happened had they been wedged between the infantry and the steep slopes of Cannae and the adjoining hills. No, this part of the battle took place in the open plain.

So where does this leave us? Goldsworthy’s theories don’t really hold up, but can we simply return to Kromayer? Not quite. Kromayer’s theory that the battle was fought on the plain where the uplands near Cannae slope gently towards the sea at Barletta is still the best theory to date, fitting all the facts we have. Unfortunately, the river moves, and until we know exactly what course it took in 216 BC we can never be fully sure where the battle was fought. 

As it happens, even that may be too optimistic; part of me suspects that unless we find a big pile of spearheads with ‘made in Carthage’ written on them we’ll never know for certain, but that’s just me. 

– Modified from a talk first given in July 2005.
________________________________________________________________________
* Also available in Italian, and a second Italian version at that. I hear it's good.