A caravan crossing Asia had to buy food and water every few days from someone, which makes the places that sold them the most important institutions in the subject and the ones most accounts skip over in a sentence.
The previous lesson derived the constraint: a camel eats its own maximum load in twenty-five days, so no caravan can carry the supplies for a journey of months. This lesson follows the constraint to what it implies. If the trade depended absolutely on a chain of supply points, then the history of the trade is largely the history of those points, of where their water came from, who controlled them, and what happened when either failed.
Water that arrives from a mountain
The Tarim Basin is an endorheic basin: rivers run into it and none run out. Rain contributes almost nothing, 16 mm a year at Turfan and about 40 mm at Dunhuang, so every drop that matters is snow and glacier melt from the Kunlun range to the south and the Tian Shan to the north.
That gives the water a distinctive behaviour. It appears in early summer as the snowline retreats, runs down the mountain front in short violent streams, spreads across the alluvial fans where the slope flattens, and sinks. Most Tarim rivers simply end in the sand. A handful, chiefly the Tarim itself and the Khotan River, run far enough to matter, and the Khotan River is famously fed well enough in high summer to be crossed by caravans in the wrong season and impassable at the right one.
An oasis, then, is not a spring in a desert. It is a place on an alluvial fan where a mountain stream is still on the surface, or close enough beneath it to be reached, and where the gradient is gentle enough to spread the water over fields. Draw the ring of alluvial fans around the Taklamakan and you have drawn the position of every significant town in the basin, and the reason the route splits into a northern and southern rim.
Tunnels, because the sun takes the rest
Turfan is the extreme case and the most instructive. The basin floor at Aydingkol Lake is about 154 m below sea level, summer air temperatures pass 45 degrees Celsius, and potential evaporation is around 3 m a year. Water carried in an open ditch across that basin is being poured onto a hotplate.
The response was the karez, called qanat in Persia: a gently sloping tunnel dug from the water-bearing gravels of the alluvial fan down to the fields, with vertical shafts every twenty or thirty metres for spoil removal, ventilation and maintenance. Seen from the air the system is a line of small craters marching down the slope. The water runs underground, in the dark, at a gradient shallower than the land surface, so that it emerges at the surface where it is wanted, entirely by gravity, with no lifting and almost no evaporation. Turfan's network is usually given as something over a thousand karez with a total tunnel length of several thousand kilometres, dug and maintained by hand.
Example. Take a karez delivering a modest 20 litres per second, and compare it with the same water carried 20 km in an open canal 3 m wide across the Turfan basin, where potential evaporation is 3 m a year. How much is saved?
The flow is m³ a year, enough to irrigate on the order of 125 hectares. The open canal has a surface of m², and at 3 m of evaporation a year it loses m³, which is 29 per cent of the whole flow, before counting seepage into coarse gravel, which in that ground can take as much again. The tunnel loses effectively none of either. So the karez is not a quaint alternative to a canal; it is roughly the difference between irrigating 125 hectares and irrigating 60 or fewer, which in a basin where cultivable land is defined by water is the difference between a town and a hamlet.
Now you. Karez tunnels need constant maintenance: collapse, silting, and the shafts filling with blown sand. What does that imply about the relationship between an oasis's political stability and its agricultural capacity?
Answer
It makes them the same thing, on a delay of a few years. A karez is capital that decays, and maintaining it requires organised labour underground in dangerous conditions, which requires either a community that can compel and coordinate its members or a ruler who can. Break that, by war, by depopulation, by a change of overlord that dissolves the labour obligation, and the tunnels silt up while the fields are still nominally there. The reverse also holds: a strong local authority can extend the system and grow the oasis. Two consequences matter for this course. Damage to an irrigated oasis is not proportional to the damage done: kill or scatter enough of the population and the water infrastructure fails, so the loss compounds long after the army has left, which is the argument a later lesson makes about the Mongol destruction of Merv. And an oasis, unlike a farming region fed by rain, cannot be reoccupied cheaply once abandoned, because the newcomers inherit blocked tunnels rather than working ones.
What the oasis actually sold
The obvious product is agricultural: wheat, millet, cotton, melons, and above all the grapes of Turfan, whose raisins are attested in the documents and whose vineyards were taxed. But a caravan town's income is not chiefly farming.
It sold fodder, water and stabling, which the previous lesson showed no traveller could do without. It sold fresh animals, and bought exhausted ones cheap, which is a business with a very good margin. It sold lodging, food and repairs. It sold credit: the Turfan and Dunhuang documents are full of loans of grain, cloth and coin, often at interest rates that look startling until you remember the default risk of lending to a man about to walk into a desert. It sold information, on which passes were open and which lord was demanding what. And it sold, or extorted, protection and passage, which is the largest item of all and the one that appears least often in the accounts.
The market dues are what turned a village into a state. A town of a few thousand people sitting where all traffic must pass, selling necessities to people with no alternative, collects a tax on the entire commerce of a continent, and this is why so many Tarim oases were independent kingdoms with royal titles, palaces and diplomatic correspondence quite out of proportion to their size.
Kingdoms of fourteen thousand people
The Hanshu gives figures for each of these kingdoms, and they repay a close look, both for what they say and for how far they can be trusted. Shanshan, the kingdom whose towns included Niya and Loulan, is given 1,570 households, 14,100 persons and 2,912 men able to bear arms. Khotan is given 3,300 households and 19,300 persons with 2,400 able to bear arms. Kucha, the largest in the basin, is given 6,970 households and 81,317 persons with 21,076 able to bear arms.
Example. Check those three sets of figures against each other for internal consistency, and say what you conclude.
Compute the household size and the fighting share for each. Shanshan gives persons per household and per cent able to bear arms. Khotan gives persons per household and per cent. Kucha gives per household and per cent. Household size varies by a factor of two and the fighting share by more than two, and no plausible demography does that between neighbouring towns with similar economies. So the categories are not being counted the same way in each entry, which is what you would expect of figures assembled from different informants at different dates for different purposes: some may count men of military age, others men actually enrolled, others all adult males. The right conclusion is not that the numbers are worthless but that they are order-of-magnitude information. Shanshan had thousands of people and Kucha tens of thousands, and the precision of 81,317 is spurious.
Now you. A modern account writes that "the population of Kucha was 81,317". What is wrong with the sentence, and how would you write it?
Answer
It launders a first century administrative estimate into a modern census figure, and the five significant digits do the laundering. The number comes from one passage of the Hanshu, compiled around 111 CE from earlier reports of unknown date and method, describing a kingdom the compilers had never visited, in a chapter whose purpose was to inform frontier policy rather than to count people. It is also undated: the Hanshu's figures for the Western Regions probably reflect conditions somewhere in the first century BCE, but the text does not say which year, so the sentence is missing a date as well as an error bar. A defensible version reads: "the Hanshu, compiled around 111 CE, credits Kucha with 81,317 people, by far the largest figure it gives for any Tarim kingdom, though its categories are inconsistent between entries and the true figure is best treated as tens of thousands."
The size matters for the politics. A kingdom that can field 2,912 men cannot resist a Han expeditionary force or a Xiongnu raid, and knows it. The consequence, visible throughout the Chinese sources, is a foreign policy of prompt and reversible submission: send a prince as a hostage to whoever is currently strongest, accept a seal and a title, pay, and switch when the balance moves. Han officials wrote about the faithlessness of the Western Regions as a moral defect. It was the only survival strategy available to a town of fourteen thousand people sitting on the only water for three hundred kilometres.
The building that follows from the pace
Between the towns, the same logic produced a piece of architecture. If a caravan makes about 30 km a day, then shelter is useful at roughly 30 km intervals, and that is very close to the observed spacing of the caravanserais built across the Islamic world and Anatolia: a walled courtyard with a single defensible gate, animals and goods stalled below, men above, water in the yard.
The form is a direct expression of the constraints. One gate, because the cargo inside is worth more than the building. A courtyard, because animals must be unloaded, watered and kept together. And the spacing is set by the camel, not by the terrain or the ruler: the Seljuk sultans of Anatolia in the thirteenth century built a chain of hans at roughly a day's march apart, and the Safavid state repeated the exercise across Iran in the seventeenth. Rulers built them because a caravan that reaches shelter each night is a caravan that arrives, and a caravan that arrives pays dues.
Example. Caravanserais on the great routes sit roughly 30 km apart, and rulers built them at their own expense and often charged nothing for a night's stay. Why would a state pay for something it does not charge for?
Because the return arrives as tax rather than as rent. A caravan that reaches a defensible walled compound each night is a caravan that arrives with its cargo, and a cargo that arrives pays market dues, customs and the multiplier of local trade, all of which the ruler collects. Charging for the shelter itself would collect a trivial sum and would divert traffic to a rival's road, whereas free shelter attracts traffic away from competitors, which is the real prize when routes have alternatives. There is also a reason that has nothing to do with revenue: a chain of garrisoned buildings a day's march apart is a military communication system and a way of holding a road, and it doubles as a display of the ruler's reach in country where his authority is otherwise abstract. This is the same logic behind free ports and state-funded roads generally, and it is worth noticing that the private version, an inn charging for beds, coexisted with it everywhere and served the traffic the state was not courting.
Now you. Why did so many oasis rulers, whose kingdoms could not have resisted anyone, nonetheless mint their own coins?
Answer
Because coinage is cheap sovereignty and useful infrastructure at the same time. Issuing coin declares that there is a ruler here whose name is worth something, which matters when your foreign policy consists of being recognised as a legitimate party by whichever great power arrives next, and a king with a coinage is a king rather than a headman. It is also profitable, since the difference between the metal's value and the coin's is seigniorage, collected from every merchant who converts. And it is genuinely useful to the trade: a market where visitors arrive with a dozen unfamiliar currencies needs a local unit, and the ruler who supplies it captures the exchange business as well. The Sino-Kharosthi coins of Khotan, carrying inscriptions in both Chinese and an Indian script, are the clearest illustration, since they are a small kingdom's currency addressed simultaneously to two enormous cultural spheres, which is what these places were.
When the water moved
Oases fail. Niya was abandoned around 400 CE, with houses left standing, documents left in place and orchards left to die, which is the pattern of a slow withdrawal rather than a sack. Loulan, on the edge of the Lop Nur, was abandoned in the fourth century as well. Sven Hedin, who found Loulan in 1900, proposed that the Lop Nur was a wandering lake that shifted position over centuries as sediment blocked one channel and opened another, and that the town died when its water went elsewhere.
The honest position is that the cause is not settled and probably was not single. The rivers of the southern Tarim demonstrably shift, and the desert is demonstrably advancing at some sites. But the fourth century is also when Chinese garrisons and the subsidies that came with them were withdrawn from the region during the collapse of the Western Jin, which is precisely the moment when a marginal irrigated settlement loses both the labour and the money that keep its channels open. Climate, hydrology and politics all point the same way at the same time, and no evidence yet separates them cleanly. Anyone who tells you an oasis died of climate change is choosing among live hypotheses.
Further west the same logic operated on a grander scale. Merv, on the Murghab river in what is now Turkmenistan, was one of the largest cities on earth in the twelfth century, fed by a canal network drawing off the river and administered by a dedicated official with, according to the geographers, thousands of workers under him. Its walls enclose several hundred hectares, and it is the clearest case in the subject of an enormous city that exists only because a river was engineered. A later lesson returns to what happened to it in 1221.
The model of the last two lessons is now complete: a physical constraint, a chain of supply points, a set of small polities living off passage dues. It predicts which goods could cross and which could not. The next lesson tests that prediction against what was actually found in the ground and written in the market lists, and the test half succeeds and half fails, in a way that redirects the whole subject.