How megadroughts ended civilizations — and the ones currently underway

Tell Leilan was not destroyed. That’s what makes it strange.

In 2200 BCE the city occupied roughly ninety hectares in the Khabur Plains of what is now northeastern Syria — a major administrative center of the Akkadian Empire’s northern grain-producing territories. Harvey Weiss of Yale led the excavations that revealed what happened next, or rather what didn’t happen: no destruction layer, no evidence of conquest or fire, no defensive perimeter breached. What the stratigraphic record shows is a city in full administrative operation at one level — grain storage vessels, milling installations, clay tablets prepared for use but uninscribed — and then, above a thin layer of volcanic ash and the windblown desiccation sediments that accumulated over following decades, nothing. The settlement record across the Khabur Plains collapsed by 87 percent. Weiss documented this in Science in 1993. The culprit, now confirmed across multiple independent proxy records, was a sustained and severe drought — the 4.2 kiloyear aridification event — that eliminated the rainfall on which the region’s rain-fed agriculture depended.

No final accounting. No annotated silence. Just objects left mid-task beneath layers of dust.

The objects are the evidence. Not that the collapse was sudden, but that it was ordinary right up until it wasn’t — that a society in full administrative operation had no apparent awareness it was crossing a threshold from which it would not recover. That is what the record shows. The mechanism the record describes is unfolding again — in the American Southwest, the Fertile Crescent, and the Horn of Africa right now.

What a megadrought actually is

The word “drought” covers too much territory to be useful without qualification. A drought year, even a severe one, sits within the normal range of variability that agricultural societies have adapted to manage: surplus storage, diversified cropping, trade. These buffers work, up to a point.

A megadrought is defined by its duration. Conventionally: 20 or more years of sustained precipitation deficit at severity levels outside normal interannual variability. The distinction is not one of degree. A megadrought exceeds the temporal horizon of any stored surplus. No granary system builds reserves against a twenty-year shortfall. No pastoral cycle adapts to two consecutive decades of failed rains. The adaptive mechanisms that complex societies evolved to handle drought variability don’t scale to megadrought timescales, and the evidence shows what happens when they fail.

That evidence comes from physical proxies — archives preserved in the earth that record rainfall without requiring human observers to write anything down. Three families matter here.

Tree-ring chronologies reconstruct the Palmer Drought Severity Index from annual growth records, giving spatial maps of drought severity extending back centuries to millennia. The North American Drought Atlas (NADA) and Old World Drought Atlas (OWDA), developed by Edward Cook at Lamont-Doherty Earth Observatory, are the primary datasets for their respective regions.

Speleothem records — stalagmites in caves — preserve rainfall signals in oxygen isotope ratios. As rainfall decreases, the isotopic composition of water infiltrating the cave shifts in predictable ways; the calcite layers of a growing stalagmite encode this at sub-annual resolution in good specimens.

Marine sediment records work differently. Haug et al. (2003, Science) analyzed cores from the Cariaco Basin, an anoxic marine basin off Venezuela, measuring titanium content as a proxy for riverine runoff. Less rainfall means less sediment washed into the basin and lower titanium concentrations — a riverine-runoff signal preserved at seasonal resolution. Lake sediment cores operate on analogous principles; pollen records add vegetation-change data.

These three evidentiary families draw on different physical processes, involve different geographic archives, and have different temporal resolutions. When they converge on the same drought chronology — as they do, repeatedly — that convergence is not coincidence.

The drought atlases

The North American Drought Atlas reconstructs the Palmer Drought Severity Index across North America from tree-ring chronologies, with some records extending back 2,000 years. The Old World Drought Atlas covers Europe and the Mediterranean on comparable timescales. Both were produced by Edward Cook and colleagues at Lamont-Doherty Earth Observatory; the NADA was published in Science in 2004 and has been updated since. Both are publicly available datasets and are the primary tools for reconstructing pre-instrumental drought variability in their regions. Their resolution limits — geographic gaps where tree-ring chronologies are sparse, temporal limits in regions lacking long-lived tree species — are documented and do not undermine their utility for identifying multi-decadal drought events.

The 4.2 kiloyear event — two empires, one drought

The Akkadian Empire was, by 2200 BCE, one of the most administratively sophisticated states the world had yet produced — a centralized bureaucracy sustained by the grain-producing capacity of its northern provinces. The Khabur Plains were its breadbasket. Rain-fed agriculture in the northern zone supplied the surplus that paid armies, staffed bureaucracies, and funded the monumental construction that signals state power. When Weiss et al. published their Tell Leilan findings in 1993, they documented exactly what happens when that substrate fails: within roughly a generation, the northern provinces depopulate, the administrative network collapses, and the archaeological signature of complex statehood simply stops.

The drought signal at Tell Leilan is not isolated. The 4.2 kiloyear event appears in Porites coral records from Oman, in ice cores from Kilimanjaro and the Guliya ice cap, in lake sediment records across the Middle East and East Africa, and in speleothem records from cave systems in China and India. The geographic spread indicates an atmospheric reorganization that was hemispheric in scale, not regional.

Fifteen hundred kilometers to the southwest, Egypt’s Old Kingdom collapsed into the First Intermediate Period at almost exactly the same time. Stanley et al. (2003, Geoarchaeology) documented strontium isotope evidence from Nile delta cores showing catastrophic Nile flood failure coinciding with the end of the Old Kingdom, approximately 2160–2130 BCE. The First Intermediate Period is not a transitional phase in the standard narrative sense. It is the absence of the administrative state — regional fragmentation, dissolution of centralized redistribution, documented famine.

Neither society knew the other was failing. They were subject to the same atmospheric system, and it broke them the same way: not through conquest or rebellion, but through the failure of the agricultural foundation on which every other function of the state depended.

That is the template. It will repeat.

Liangzhu and the global scale of the 4.2k event

The simultaneous collapse of the Akkadian Empire and the Egyptian Old Kingdom is striking enough. But the 4.2 kiloyear event is associated with a third major civilizational collapse in East Asia: the Liangzhu culture of the lower Yangtze River valley, one of the most sophisticated Neolithic societies in the ancient world, shows signs of collapse around 2300–2200 BCE. The association with the 4.2k event appears in the paleoclimate literature, and recent work has linked hydroclimatic disruption — in this region, anomalous flooding rather than aridification, as the same atmospheric reorganization produced different regional expressions — to the collapse. The precise dating of the Liangzhu collapse and its causal connection to the 4.2k event remain subjects of active investigation, with the clearest evidence pointing to hydroclimatic disruption as the forcing mechanism. The causal pathway in East Asia — flooding from disrupted monsoon circulation rather than the aridification pattern in Mesopotamia and Egypt — differs in direction but not in its consequence: the collapse of an agricultural system driven by the same planetary-scale atmospheric reorganization.

1200 BCE — a networked world that stopped networking

By 1200 BCE the Late Bronze Age eastern Mediterranean was an interconnected trading system of considerable sophistication: Mycenaean Greece, the Hittite Empire, the city-states of Canaan, Ugarit on the Syrian coast, and Egypt bound into a network of grain, copper, tin, and luxury goods cycling through multilingual merchant families whose correspondence survives in cuneiform. The network was not fragile in the sense that individual nodes were weak. It was fragile in the sense that interdependency at that scale means a systemic shock propagates everywhere.

Between roughly 1200 and 1150 BCE, almost every major palace civilization around the eastern Mediterranean collapsed simultaneously. The Hittite Empire disappeared as a state. Mycenaean palace culture vanished. Ugarit was abandoned. Egypt survived but contracted dramatically and never recovered its New Kingdom reach. The standard version of this collapse centers on the “Sea Peoples” — migrating warrior groups whose identity remains contested. The Sea Peoples appear as cause; the collapse is the effect.

The physical record disagrees. Brandon Drake (2012, Journal of Archaeological Science) analyzed eastern Mediterranean sea surface temperatures and pollen records documenting prolonged drought conditions from around 1200 BCE. Kaniewski and colleagues published multiple papers on pollen cores from the Syrian coast and the plain of Antioch — including a 2013 study in PLOS ONE — showing drought onset at approximately this period and its persistence for up to 300 years. The highest-resolution direct evidence comes from Gordion in central Turkey: Manning and colleagues (2023, Nature) combined tree-ring widths with stable isotope records from juniper samples at the Midas Mound site to document a severe, continuous drought coinciding precisely with Hittite collapse around 1198–1196 BCE.

The last administrative correspondence from Ugarit — tablets found in a kiln where they were being fired when the city was abandoned — describes famine and requests emergency grain shipments from Egypt. Ugarit shows no destruction layer. Like Tell Leilan a millennium earlier, it simply stopped.

Egypt survived because the Nile is a drought-resistant system — flood-based agriculture in an otherwise arid environment is less vulnerable to regional rainfall failure than rain-fed farming. But Egypt’s grain exports, which had been supporting deficit regions through the trading network, stopped. Supply chain failure cascaded through every society whose agriculture depended on external provisioning. The network didn’t buffer the shock. It transmitted it.

Why the Sea Peoples don't explain the collapse

The standard objection to drought as primary cause of the Late Bronze Age collapse is multicausality: the Sea Peoples invaded, internal rebellions destabilized palace economies, supply chains were already disrupted. All of these things happened. None of them is the explanation, because none accounts for the simultaneity and comprehensiveness of the collapse across societies with entirely different internal political situations. More importantly, multicausality doesn't dissolve the drought argument — it specifies it. Drought doesn't need to be the only stressor; it needs only to be the proximate cause that made the other stressors lethal. A society simultaneously managing drought stress, military pressure, and trade disruption is not a society with three problems. It is a society whose fiscal and food-security margins have been eliminated by the drought, leaving it with no capacity to absorb the other two. That is a specific and defensible causal claim, and the evidence for it in the Late Bronze Age record is now substantial.

The Maya — when the drought is the evidence

The Terminal Classic Maya collapse — the abandonment of most major lowland cities between roughly 760 and 910 CE — has been attributed to peasant revolt, epidemic disease, soil exhaustion. The paleoclimate record has progressively made these arguments beside the point, not by ruling them out, but by showing that the drought evidence is now precise enough to match individual political events.

Haug et al. (2003, Science) used the Cariaco Basin titanium record — the marine sediment proxy for rainfall in the Yucatán and northern tropical South America — to reconstruct drought conditions across the Terminal Classic at seasonal resolution. The record shows severe drought events at approximately 810, 860, and 910 CE. These dates correspond with unusual precision to the periods of maximum political disintegration: the collapse of Copán and the southern cities around 800–820 CE, the disintegration of central lowland polities in the mid-ninth century, the terminal abandonments of the late ninth and early tenth centuries. City abandonments cluster, year by year, against the drought chronology.

Medina-Elizalde and Rohling (2012, Science) added a second independent line of evidence using oxygen isotope analysis of stalagmites from Belize cave systems — a speleothem record entirely distinct from the marine sediment proxy. Their findings showed precipitation declines of 25–40% during the peak drought years of the Terminal Classic. Two independent proxy systems, drawing on entirely different physical processes and geographic archives, converging on the same chronology.

The pattern is more specific than “drought caused collapse.” The collapses unfolded across a series of individually severe drought events against a background of general aridification. Between the acute droughts there were partial recoveries — periods where rainfall improved enough to suggest the worst had passed. And then the next drought arrived before the agricultural and fiscal system had restabilized. The cities that survived longest — Chichén Itzá in the northern lowlands, cities in the Yucatán with access to coastal trade — were precisely those with extensive cistern (chultun) infrastructure capable of buffering short-term variability. The geographic pattern of survival is water infrastructure. If the collapses had been caused primarily by political failure, revolt, or disease, that pattern makes no sense.

How stalagmites read rainfall

Speleothem records work because cave drip water inherits the isotopic signature of the rainfall that percolated through the overlying soil and rock. During periods of lower rainfall, the oxygen isotope ratio (δ¹⁸O) of that water shifts in ways reflecting evaporation and soil water dynamics; in good cave systems, this signal is preserved in the calcite layers of the growing stalagmite at sub-annual resolution. The Belize cave records used by Medina-Elizalde and Rohling are particularly well-suited to the Maya question because the regional hydrology is well-characterized and the signal-to-noise ratio in the isotope record is high. The Haug et al. (2003) Cariaco Basin study uses an entirely different method — titanium flux in marine sediment cores as a proxy for riverine runoff — and covers a different geographic footprint. The fact that both records independently converge on the same Terminal Classic drought chronology substantially strengthens the causal inference. When two proxy systems with no physical connection to each other agree on when the droughts occurred, the temporal correlation with political collapse is hard to dismiss as artifact.

The mechanism

Three cases. Separated by centuries and hemispheres, involving societies with entirely different political structures, agricultural systems, and cultural contexts. The same sequence plays out in all of them.

It begins in the periphery. The rain-fed agricultural zones at the margins of the state — not the capital, not the heartland, but the outlying provinces whose surplus production feeds the fiscal system — fail first. In the Khabur Plains: the Akkadian Empire’s northern grain-producing territory. In the Late Bronze Age: the dry-farming zones of Hittite Anatolia and the Levantine coast. In the Maya lowlands: the rain-dependent milpa agriculture of the southern cities.

Fiscal contraction follows. As grain production falls, the state’s capacity to pay armies, maintain infrastructure, staff bureaucracies, and redistribute food through administrative channels collapses. Joseph Tainter, in The Collapse of Complex Societies (1988), argued that complex societies are ultimately sustained by their capacity to deliver returns on complexity — to provide, through their administrative apparatus, more than the extraction costs. Sustained drought eliminates the surplus on which that calculation depends. What looks like administrative failure is the arithmetic of diminishing returns, accelerated.

Elite legitimacy fails next. In every case examined, the collapse of the state’s redistributive function destroys the social contract that justifies elite extraction. The administrative records stop not because scribes are killed but because the institutions that demanded administrative records no longer exist. Tell Leilan preserves this phase: the prepared tablets, the organized storage facilities, the mid-task objects of a society that had not yet grasped that the institution was dissolving.

Depopulation and territorial contraction complete the sequence. The 87 percent reduction in Khabur Plains settlement area is not a statistic describing mass death — it describes mass departure. Migration is the rational individual response to agricultural failure. It is catastrophic for the society that remains.

The mechanism’s essential requirement is not that drought be the only stressor. It is that drought be sustained long enough — roughly two to three decades — that the adaptive reserves of surplus are exhausted before recovery occurs. At that point, additional stressors that would previously have been survivable become lethal. The sequence is not drought → collapse. It is sustained drought → depleted surplus → no margin → collapse on contact with any other pressure.

And the mechanism is flexible. It operates through rain-fed agriculture in peripheral provinces. It operates through supply-chain failure in trading networks. It operates through cumulative depletion of lowland farming systems. The vector differs by society. The logic is the same.

The ones currently underway

The same tools that read the 4.2 kiloyear event can read 2022. Tree rings growing in Arizona and New Mexico are producing annual growth records that extend the North American Drought Atlas into the present. Satellite gravity measurements from the GRACE and GRACE-FO missions detect changes in aquifer mass across regions too large for ground-based monitoring. The evidence is accumulating in the same physical proxies. It is not ambiguous.

Williams, Cook, and Smerdon (2022, Nature Climate Change) published the most comprehensive recent analysis of the American Southwest megadrought. The period 2000–2021 is the driest 22-year stretch in the region since at least 800 CE — 1,200 years of tree-ring drought reconstructions. Approximately 19 percent of the soil moisture deficit is attributable to anthropogenic warming; the rest is natural multidecadal variability, now compounded by it. The Southwest has crossed the minimum duration threshold established by the historical cases. It has been in megadrought for more than two decades.

The Fertile Crescent presents a different configuration. Not one unbroken multi-decade drought, but a structural drying in which catastrophic drought years have become the default and recovery intervals have compressed to the point of disappearing. The 2007–2010 Syrian drought was the worst in the instrumental record (Kelley et al. 2015, PNAS), contributing to the mass rural displacement that preceded the civil war. But that event was an acute episode within a longer arc: rainfall deficits traceable to the late 1990s, then a five-year drought running from mid-2020 through mid-2025, each acute event arriving before the previous one’s damage could be absorbed. World Weather Attribution’s November 2025 analysis found the five-year drought around 31 times more likely over the Tigris-Euphrates basin than in the pre-industrial climate. The Maya case documented exactly this: drought events arriving faster than recovery can follow, exhausting adaptive capacity across multiple cycles rather than through a single sustained blow.

In northern China, the mechanism operates through a different vector. The North China Plain aquifer system, underlying roughly half of China’s wheat production, is being depleted by extraction at rates documented by GRACE and GRACE-FO satellite gravity measurements (Feng et al. 2013, Water Resources Research). This is not surface drought in the conventional sense — it is the structural exhaustion of the groundwater buffer that allows agriculture to function through dry years. When the aquifer fails, the agricultural system’s exposure to surface precipitation variability becomes catastrophically direct. The 2022 Yangtze drought — the worst in roughly 60 years, with rainfall 45 percent below normal across much of the basin — showed what surface drought severity looks like without that buffer. The aquifer depletion is the drought, operating on a different timescale and through a different vector.

The vector differs. The logic does not.

The Horn of Africa experienced five consecutive failed rainy seasons from 2020 through 2023 — the worst sequence in 40 years. In Kenya and Somalia alone, approximately 6.5 million people faced high levels of food insecurity as of early 2026. This is not a megadrought by the paleoclimate definition. But the mechanism here — cumulative frequency failure, the progressive depletion of agricultural reserves and fiscal capacity through repeated drought cycles with insufficient recovery intervals — is the same process the historical record describes.

The Colorado River Compact and the law of the river

The 1922 Colorado River Compact allocated 7.5 million acre-feet annually to the Upper Basin states and 7.5 million to the Lower Basin — based on an estimated average annual flow of 17.5 million acre-feet that the Bureau of Reclamation projected at the time. Tree-ring reconstructions, beginning with Stockton and Jacoby in 1976 and confirmed repeatedly since by multiple independent studies, established that the compact was negotiated during one of the wettest decades in the Colorado River's multi-century record. The long-term average flow is closer to 12–14 million acre-feet depending on methodology — the allocations exceed actual mean flow by 25–40 percent. The gap has been visible since 2000, when drought conditions began to close the distance between paper allocations and actual flow. The current allocations — enshrined in interstate compacts, international treaties, and decades of infrastructure investment — reflect a river that never existed in the long-term record and does not exist now.

The framing problem

The public vocabulary for current megadroughts is: water security, resource stress, adaptation challenge, climate resilience. They encode a specific assumption — that the appropriate response to sustained drought is management: allocation adjustments, efficiency improvements, desalination investment, demand reduction.

The mechanism the historical record describes is not a management problem.

The mechanism says: at some duration threshold, complex societies don’t manage through sustained drought — they dissolve, specifically because the institutions designed to manage the crisis are themselves dependent on the surplus the drought is eliminating. The question of where that threshold sits for modern states — with global food trade, groundwater extraction capacity, and the fiscal capacity to import what they can no longer produce — is genuinely open. Nobody is claiming Arizona will end like the Akkadian Empire.

But the assumption that these buffers are indefinitely scalable against multi-decade drought in multiple major agricultural regions simultaneously has not been tested. It is an assumption of exceptionalism — the belief that the mechanism which has dismantled every pre-industrial society that encountered it will, this time, encounter something different enough to produce a different result. That assumption might be correct. The basis for it is not the evidence. The evidence describes the mechanism, names its requirements, and shows it operating on current timescales in current geographies. The policy response is being designed as though the evidence says something else.

That is not a scientific uncertainty. It is a choice about which evidence counts.

The grain measures at Tell Leilan are in storage now — in museum collections and in the photographic record of the excavations. The uninscribed tablets are still there. The desiccation layers above the last occupation level are still there. They are not mysterious. They are the physical output of a mechanism we understand in considerable detail: agricultural failure in a rain-dependent region, fiscal collapse, administrative dissolution, depopulation, silence.

The drought that produced them lasted, by the proxy record, somewhere between one and three centuries. The population of the Khabur Plains did not return to pre-collapse levels for roughly three hundred years.

The American Southwest has been in megadrought for twenty-two years. The Fertile Crescent has experienced nearly three decades of structural drying, punctuated by droughts of increasing severity. Northern China’s agricultural aquifer is being depleted at rates that make recharge, at current extraction levels, a matter of decades. The Horn of Africa has not had two adequate consecutive rain seasons in over a decade.

The mechanism doesn’t require these crises to produce collapse. It requires them to be understood correctly.

At the moment, they’re not.

Gen AI 免責事項

このページの一部のコンテンツは、ジェネレーティブAIの助けを借りて生成・編集されたものです。.

メディア

Şahin Doğdu – Pexels

主な情報源と参考文献

Weiss, H., Courty, M.A., Wetterstrom, W., Guichard, F., Senior, L., Meadow, R., and Curnow, A. “The Genesis and Collapse of Third Millennium North Mesopotamian Civilization.” Science, vol. 261, no. 5124, 1993, pp. 995–1004.

Stanley, J.D., Krom, M.D., Cliff, R.A., and Woodward, J.C. “Nile flow failure at the end of the Old Kingdom, Egypt: Strontium isotopic and petrologic evidence.” Geoarchaeology, vol. 18, no. 3, 2003, pp. 395–402.

Haug, G.H., Günther, D., Peterson, L.C., Sigman, D.M., Hughen, K.A., and Aeschlimann, B. “Climate and the Collapse of Maya Civilization.” Science, vol. 299, no. 5613, 2003, pp. 1731–1735.

Medina-Elizalde, M., and Rohling, E.J. “Collapse of Classic Maya Civilization Related to Modest Reduction in Precipitation.” Science, vol. 335, no. 6071, 2012, pp. 956–959.

Kaniewski, D., Van Campo, E., Guiot, J., Le Burel, S., Otto, T., and Baeteman, C. “Environmental Roots of the Late Bronze Age Crisis.” PLOS ONE, vol. 8, no. 8, 2013, e71004. https://doi.org/10.1371/journal.pone.0071004

Manning, S.W., Kocik, C., Lorentzen, B., and Sparks, J.P. “Severe multi-year drought coincident with Hittite collapse around 1198–1196 bc.” Nature, vol. 614, no. 7949, 2023, pp. 719–724. https://doi.org/10.1038/s41586-022-05693-y

Williams, A.P., Cook, B.I., and Smerdon, J.E. “Rapid intensification of the emerging southwestern North American megadrought in 2020–2021.” Nature Climate Change, vol. 12, 2022, pp. 232–234. https://doi.org/10.1038/s41558-022-01290-z

Kelley, C.P., Mohtadi, S., Cane, M.A., Seager, R., and Kushnir, Y. “Climate change in the Fertile Crescent and implications of the recent Syrian drought.” Proceedings of the National Academy of Sciences, vol. 112, no. 11, 2015, pp. 3241–3246.

Tainter, J.A. The Collapse of Complex Societies. Cambridge University Press, 1988.

Stockton, C.W., and Jacoby, G.C. “Long-Term Surface-Water Supply and Streamflow Trends in the Upper Colorado River Basin.” Lake Powell Research Project Bulletin No. 18. National Science Foundation, 1976.

Feng, W., Zhong, M., Lemoine, J.M., Biancale, R., Hsu, H.T., and Xia, J. “Evaluation of groundwater depletion in North China using the Gravity Recovery and Climate Experiment (GRACE) data and ground-based measurements.” Water Resources Research, vol. 49, no. 4, 2013, pp. 2110–2118. https://doi.org/10.1002/wrcr.20192

World Weather Attribution. “Human-induced climate change compounded by socio-economic water stressors increased severity of 5-year drought in Iran and Euphrates and Tigris basin.” 21 November 2025. https://www.worldweatherattribution.org/human-induced-climate-change-compounded-by-socio-economic-water-stressors-increased-severity-of-5-year-drought-in-iran-and-euphrates-and-tigris-basin/

Cook, E.R., Woodhouse, C.A., Eakin, C.M., Meko, D.M., and Stahle, D.W. “Long-Term Aridity Changes in the Western United States.” Science, vol. 306, no. 5698, 2004, pp. 1015–1018.

Drake, B.L. “The Influence of Climatic Change on the Late Bronze Age Collapse and the Greek Dark Ages.” Journal of Archaeological Science, vol. 39, no. 6, 2012, pp. 1862–1870.