Outburst flood at 1920 BCE supports historicity of China’s Great Flood and the Xia dynasty

China’s earliest historiographies, including Shujing (Book of Documents) and Shiji (Records of the Grand Historian, by Sima Qian), tell of the Great Flood, a lengthy, devastating flood of the Yellow River. The culture hero Yu eventually tamed this flood by dredging, earning him the divine mandate to establish the Xia dynasty, the first in Chinese history, and marking the beginning of Chinese civilization. Because these accounts laid the ideological foundations for the Confucian rulership system, they had been taken as truth for more than 2500 years until challenged by the “Doubting Antiquity School” in the 1920s. Within a decade, archaeological excavations demonstrated the historicity of the second dynasty, Shang, and the search for similar evidence for Xia began (1, 2). Archaeological fieldwork since the 1950s on the Early Bronze Age Erlitou culture (~1900 to 1500 BCE) has led many scholars to associate it with the Xia (1–6) because it overlaps with the spatial and temporal framework of the Xia dynasty. Traditionally, historians have dated the start of Xia to ~2200 BCE, whereas the government-sponsored Xia-Shang-Zhou Chronology Project adopted the date as 2070 BCE (5), leaving a chronological gap in associating Erlitou with Xia (7–9). Other scholars see Xia purely as a myth fabricated to justify political succession (10, 11).

Scholars also have long sought a scientific explanation of the Great Flood (12–14), with even Lyell mentioning it (15), yet no evidence for it has been discovered. Here, we present geological evidence for a catastrophic flood in the early second millennium BCE and suggest that it may be the basis of the Great Flood, thereby lending support to the historicity of the Xia dynasty. The evidence found in our investigations along the Yellow River in Qinghai Province includes remains of a landslide dam, dammed lake sediments (DLS) upstream, and outburst flood sediments (OFS) downstream (Fig. 1 and figs. S1 to S5) that allow us to reconstruct the size of the lake and flood (16).

• Download high-res image
• Open in new tab
• Download Powerpoint

Field observations (fig. S2B) show that the ancient landslide dam deposits reach an elevation of 240 m above present river level (arl) and stretch for 1300 m (fig. S2A) along Jishi Gorge (Figs. 1A and 3A). We estimate that the saddle of the dam would have been 30 to 55 m lower than the highest preserved remnants, so the lake would have filled to an elevation of 185 to 210 arl [2000 to 2025 m above sea level (asl)] (fig. S2B), impounding 12 to 17 km³ of water (16) (table S1). Based on typical river discharge values, the dam would have completely blocked the Yellow River for 6 to 9 months before overtopping (16). DLS distributed widely upstream of the dam are up to 30 m thick and have a highest elevation of ~1890 m asl (Fig. 1B and figs. S1 and S3A). We interpret this as indicating that the catastrophic breach dropped the water level 110 to 135 m (Fig. 1B), releasing ~11.3 to 16 km³ of water (16) (table S1), tens of times that estimated by a previous study (17). After the breach, DLS infilled a residual lake behind the lowest part of the dam that remained.

• Download high-res image
• Open in new tab
• Download Powerpoint

Outburst flood sediments are found downstream at elevations from 7 to 50 m arl in the lower Jishi Gorge and in Guanting Basin (Fig. 1 and figs. S1 and S4). They are characterized by high-concentration suspension deposition and consist exclusively of angular clasts of greenschist and purple-brown mudrock sourced from Jishi Gorge (table S2). At the mouth of the gorge, where the Yellow River enters Guanting Basin, the sediments reach 20 m thickness and include boulders up to 2 m in diameter (Fig. 1B and figs. S1 and S4, C and D). We also identified the OFS at the earthquake-destroyed prehistoric Lajia site (fig. S5), a settlement of the Qijia culture (18, 19) known for its early noodle remains (20), 25 km downstream from the dam. OFS at Lajia covered the settlement’s last Qijia culture occupation and filled in collapsed cave dwellings (fig. S5, A and B), pottery vessels (fig. S5B), and earthquake fissures (fig. S5C), mixing with pottery sherds (fig. S5D) and other Qijia cultural materials, with heights of up to 38 m arl.

Stratigraphic relationships of the OFS, remnant dam, DLS, loess, and other deposits in Jishi Gorge and neighboring basins, along with destruction features at the Lajia site (fig. S1), allow us to reconstruct and date a sequence of events ending in the outburst flood. First, they show that the damming and outburst flood event occurred during the archaeological Qijia culture period (~2300 to 1500 BCE) after the collapse of the Lajia cave-houses. Ground fissures caused by the earthquake at the Lajia site were entirely filled with OFS (fig. S5C) before silts from surface runoff during the annual rains could enter them, indicating that the outbreak flood must have occurred less than 1 year after the earthquake and collapse of the houses. It is likely that the same earthquake that destroyed Lajia also triggered the landslide that dammed the river, along with widespread contemporaneous rock avalanches whose deposits lay directly beneath the DLS (fig. S3A).

To date the outburst flood, we collected carbon samples for accelerator mass spectrometry (AMS) ¹⁴C dating (16). Seventeen charcoal samples from the OFS and the only charcoal sample from a layer overlying the OFS (fig. S1) indicate that the age for the flood is between 2129 and 1770 cal. BCE [95% confidence interval (CI)] (Fig. 2A and table S5) (16). Charcoal samples from DLS upstream of the dam (fig. S1) yield calibrated ¹⁴C results (95% CI) spanning 2020 to 1506 BCE (Fig. 2A and table S5), demonstrating that the DLS is coeval with or younger than the outburst flood and confirming that it is fill from the remnant lake. The best dating for the flood comes from the Lajia site (16), because it was destroyed within 1 year before the outburst flood. Radiocarbon determinations of bone samples from three human victims, aged 6 to 13 years old, in collapsed Lajia dwellings (Fig. 2B) agree to within uncertainty (Fig. 2A and table S5), consistent with that of two victims reported previously (21) as well. Because the radiocarbon calibration curve is linear in this region and the bones are the same age, we use the inverse variance weighted mean of the three measurements. This yields a calibrated age with a median of 1922 ± 28 BCE (1 SD) and a 95% CI of 1976 to 1882 BCE (Fig. 2C). To simplify this range, we use 1920 BCE to indicate the approximate date of the flood.

• Download high-res image
• Open in new tab
• Download Powerpoint

We estimate the peak discharge of the flood in two ways. Empirical formulas considering the volume of the lake and the height of the dam lead to estimates ranging from 0.08 to 0.51 × 10⁶ m³s⁻¹, with large uncertainties (16) (table S3). We also reconstruct the flood channel cross section from detailed surveys in Guanting Basin and use Manning’s equation to estimate a peak discharge of 0.36 to 0.48 × 10⁶ m³s⁻¹ (16) (fig. S6 and table S4), consistent with the dam break estimations (16) (table S3). The calculated peak discharge of ~0.4 × 10⁶ m³s⁻¹ is more than 500 times the average discharge of the Yellow River at Jishi Gorge. This ranks globally among the largest freshwater floods of the Holocene (22).

We do not explicitly model the inundation and effect of this outburst flood in the lower reaches of the river, but analogous events demonstrate that outburst floods from landslide dams can propagate long distances. In 1967, an outburst flood with a volume of just ~0.64 km³ propagated at least 1000 km along the Yalong-Yangtze Rivers (23), so the Jishi prehistoric outburst flood, with a volume of ~11 to 16 km³, could have easily travelled more than 2000 km downstream. The Jishi flood would have breached the natural levees of the Yellow River, resulting in rare, extensive flooding. It is possible that this outburst flood was also the cause of a major avulsion of the lower Yellow River (Fig. 3A) inferred from archaeological data, with a previously estimated date of ~2000 BCE (24, 25). Widespread destruction of levees and deposition of tributary mouth bars may have destabilized the main river channel, leading to repeated flooding until a new river channel was established. Extensive flooding on the lower Yellow River plain would have had a great effect on societies there. We argue that this event and its aftermath likely would have survived in the collective memories of these societies for generations, eventually becoming formalized in the received accounts of the Great Flood in the first millennium BCE. In fact, early texts such as the Shujing and Shiji even record that a place called Jishi (the same characters as the gorge where the outburst flood began) was where Yu began his dredging of the Yellow River; whether this is a coincidence will require further historical geographical research.

The ~1920 BCE flood shares the main characteristics of the Great Flood described in ancient texts. Apart from its huge peak discharge, the secondary flooding on the lower plains may have been long-lasting, just as the Great Flood remained uncontrolled for 22 years until it was managed by dredging (rather than by blocking breaches in natural levees). There is also the issue of whether the Great Flood could have been caused by exceptional meteorological flooding, but a speleothem record shows a generally weakened Asian summer monsoon from 8000 to 500 years before the present (26), and proxies from lake and loess records also indicate that a cool, dry climate regime begins 2000 BCE along the lower Yellow River (27), so this would be unlikely. Furthermore, the early textual records make no mention of frequent, extreme storms related to the Great Flood.

The discovery and reconstruction here of the massive outburst flood originating in Jishi Gorge provide scientific support that the ancient Chinese textual accounts of the Great Flood may well be rooted in a historic natural event. They also shed light on the potential historicity of the Xia dynasty itself, as Yu’s founding of the dynasty is directly tied to his achievements in controlling the Great Flood. According to the Shiji, Yu’s father labored unsuccessfully for 9 years to tame the flood before Yu took over for 13 more years. Yu’s success led to his mandate to become founding king of the Xia 22 years after the flood started. If the Jishi Gorge outburst flood of ~1920 BCE is the natural cataclysm that came to be known as the Great Flood, then we can propose a new beginning date for the Xia dynasty, ~1900 BCE. This date, some 2 to 3 centuries later than previous reckonings (1, 2, 5), is compatible with the 1914 BCE date proposed by Nivison based on astro-historiographical evidence (28). This 1900 BCE date for the founding of the Xia coincides with the beginning of the Erlitou culture (6), so this finding also supports the arguments that the Erlitou culture is the archaeological manifestation of the Xia and that the Erlitou site was a Xia dynastic capital (1–3). This outburst flood is also coincident with the major sociopolitical transition from Neolithic to Bronze Age in the Yellow River valley (2, 6, 29) (Fig. 3, A and B), suggesting that the concurrence of these major natural and sociopolitical events known through the geological, historiographical, and archaeological records may not simply be coincidence but rather an illustration of a profound and complicated cultural response to an extreme natural disaster that connected many groups living along the Yellow River.