Introduction

Dams have been constructed on most of the world’s large and moderate rivers, and the impacts of reservoirs on rivers, deltas, and adjacent seas have been extensively studied in recent decades1,2,3,4,5,6. The retention of sediment in reservoirs and sediment deficits in downstream channels and deltas are among the most important environmental impacts of dam construction. In extreme cases, such as the Nile (Egypt), Colorado (Mexico), and Kizil Irmak (Turkey) rivers, the sediment flux to the sea has decreased to almost zero7,8. Dam construction on the Yellow, Yangtze, Pearl, and Liao rivers of China; the Narmada and Krishna rivers of India; the Mississippi and Rio Grande rivers of the USA; and the Indus (Pakistan), Red (Vietnam), Chao Phraya (Thailand), Orange (South Africa), Limpopo (Mozambique), Volta (Ghana), Rhone (France), Ebro (Spain), Yenisei (Russia), and Ombrone (Italy) rivers has led to 50–95% reductions in sediment discharge2,9,10,11,12,13. Furthermore, sediment retention in reservoirs has resulted in downstream channels and deltas being degraded14,15,16,17,18,19.

In contrast to the numerous studies of reservoir-induced long-term downstream sediment reduction and morphological degradation, less is understood of the impact of episodic reservoir erosion on fluvial sediment loads and channel–delta stability. Scouring events in small reservoirs and the consequent downstream sediment pulses associated with dam removal or breaching (partial removal) and hydraulic flushing have been documented20,21,22,23,24,25, but it remains unclear whether episodic erosion also occurs from large (storage capacity >0.1 km3) silted-up in-use reservoirs during flooding or due to the combined effects of flooding and drawdown, and how such episodic erosion affects downstream sediment transport and morphodynamics. Quantitative assessment of sediment budgets between reservoir erosion and downstream sediment delivery and deposition is particularly lacking. Many reservoirs constructed worldwide over the last 150 years are highly silted-up2,7,9,12,26,27,28,29,30. Many new reservoirs are currently under construction or in the planning stage, particularly in developing countries5,31. As more reservoirs silt up, the increasing frequency and intensity of extreme precipitation events due to global warming32,33 necessitates investigation of whether episodic erosion is likely in silted-up reservoirs under natural or regulated conditions and its potential effects on downstream sediment delivery and morphodynamics.

Here, case studies, particularly on the Yellow River (Fig. 1), were employed to address the above questions with the aims of (1) quantitatively assessing episodic reservoir erosion and its effects on downstream sediment loads and channel deposition; (2) constructing a sediment budget for the reservoir-downstream-channel systems; (3) investigating natural and anthropogenic factors responsible for reservoir erosion events; and (4) providing an overview of the potential of episodic erosion in silted-up reservoirs worldwide. The results aim to enhance our understanding of the role of silted-up reservoirs as fluvial sediment sources under extreme hydrometeorological conditions and associated human impacts.

Fig. 1: Diagrams showing the upstream and downstream reservoirs and hydrological stations in the study area.
figure 1

A Schematic diagram of the studied reservoirs. Dots indicate dam locations; lines represent general directions, rather than actual river channels. Distances in the river courses are shown. B Map of the Yellow River basin showing the locations of major reservoirs, gauging stations, and cities. C Multiyear average water and sediment fluxes at gauging stations along the mainstream of the Yellow River prior to reservoir construction (1952–1959).

Full size image

Results

Erosion of the Xiaolangdi and Sanmenxia reservoirs and its downstream sedimentary and morphological impacts in a flood season

Runoff of the Yellow River during the 2019 flood season was much higher than normal. For example, the monthly water discharge at Sanmenxia Station in July–September 2019 (6090 Mm3) was 2.5 times the multiyear average (2450 Mm3) (Supplementary Fig. S1). Based on daily water inflows to the Sanmenxia and Xiaolangdi reservoirs, there were two major natural flood events. The maximum water inflows to the Sanmenxia Reservoir during the first and second flood events were 4850 and 4490 m3 s−1, respectively, almost six times the water inflow during the dry season (averaging 780 m3 s−1 from January to May). Water outflows from the Sanmenxia Reservoir were generally consistent with water inflows (Fig. 2A). However, there were notable differences between inflows and outflows at the Xiaolangdi Reservoir, where outflows were markedly higher than inflows from March to late July 2019, especially during the 20 days from 22 June to 11 July (Fig. 2B). During this period, the water level decreased by ~60 m and water storage decreased to nearly zero from ~8 km3 (Fig. 2C), with a doubling of water-surface slope (Fig. 2D).

Fig. 2: Temporal variations of water and sediment in the Sanmenxia and Xiaolangdi reservoirs.
figure 2

A Time series of daily water inflow and outflow for the Sanmenxia Reservoir. B Water inflow and outflow, C water level and storage, D water-surface slope, E inflow and outflow SSC, F sediment inflow and outflow, and G erosion and cumulative erosion for the Xiaolangdi Reservoir. H Sedimentation and cumulative sedimentation in downstream reaches between Xiaolangdi and Huayuankou stations. Note: inflow to the Xiaolangdi Reservoir equals the outflow from the Sanmenxia Reservoir; daily inflow SSC data for the Sanmenxia Reservoir and daily inflow and outflow SSC data in January–June and October–December for the Xiaolangdi Reservoir are unavailable.

Full size image

During the period of 9 July–10 August (Phase 1, Fig. 2), the water flow through the Xiaolangdi Reservoir was unusually high (Fig. 2B) while the water level was the lowest (Fig. 2C) and the water gradient was the largest (Fig. 2D). As a result, the suspended sediment concentration (SSC) of the Xiaolangdi Reservoir outflow increased markedly despite fluctuations, and was notably higher than the inflow SSC. The highest outflow SSC of 210 kg m−3 was observed in the initial stage of Phase 1 (15 July) when inflow and outflow were 2500 and 2,720 m3 s−1, respectively. The second highest outflow SSC of 170 kg m−3 was observed at the end of Phase 1 (9 August) when the inflow was 3,660 m3 s−1 and the outflow increased rapidly to its second highest level of 4,350 m3 s−1 after a brief reduction. During this Phase 1 period, the mean outflow SSC of 57 kg m−3 was 3.8 times that of the mean inflow SSC (15 kg m−3), and the maximum ratio of daily outflow to inflow SSC reached 17 (Fig. 2E).

The sediment outflow increased with a trend similar to that of the outflow SSC and peaked at the end of Phase 1, whereas the second-highest outflow occurred in the initial stage of Phase 1 (Fig. 2F). Considering the difference between sediment outflow and inflow, the erosion rate peaked at 60 Mt d−1 (Mt: million tons) (693 t s−1) on 9 August, at up to 17 times the sediment inflow (41 t s−1), accounting for 94% of the daily sediment outflow of 735 t s−1. The second highest erosion rate of 31 Mt d−1 (362 t s−1) occurred in the initial stage of Phase 1 (July 10), at 12 times the sediment inflow (30 t s−1), contributing 92% of the sediment outflow of 392 t s−1. The cumulative erosion in the Xiaolangdi Reservoir during the one month of Phase 1 reached 373 Mt, or 3.4 times the cumulative sediment inflow (109 Mt), contributing 74% of the cumulative sediment outflow of 498 Mt (Fig. 2F–G). At Xiaolangdi Station, the annual sediment flux in 2019 was 545 Mt, and the cumulative sediment flux for 2010–2019 was 1566 Mt. Erosion in the Xiaolangdi Reservoir during the flood event of July 2019 thus contributed 68% of the annual sediment flux and 24% of the decadal sediment flux in downstream channels.

After this rapid erosion, the Xiaolangdi Reservoir experienced a period of balanced sedimentation and erosion (10–31 August), when water outflow decreased more rapidly than inflow, and the water slope decreased to half its maximum value due to water impoundment (Fig. 2B–D). Another flood occurred in the Yellow River in September 2019 (Phase 2), during which water outflow from the Xiaolangdi Reservoir was consistent with inflow, and the inflow SSC was markedly higher than outflow SSC. In this second flood event, the sediment outflow from the Sanmenxia Reservoir (i.e., the sediment inflow to the Xiaolangdi Reservoir) was ~80 Mt, while the outflow from the Xiaolangdi Reservoir was ~10 Mt, resulting in sedimentation of ~70 Mt in the Xiaolangdi Reservoir (Fig. 2).

The absence of direct SSC inflow data for the Sanmenxia Reservoir precluded an estimation of daily sedimentation and erosion, but the three major peaks in outflow SSC (i.e., inflow SSC to the Xiaolangdi Reservoir) likely indicate erosion events in the Sanmenxia Reservoir (Fig. 2E). The first outflow SSC peak of 32 kg m−3 (compared to <1 kg m−3 under normal conditions) occurred on 1 July (Fig. 2E) when water inflow and outflow of the Sanmenxia Reservoir were 2660 and 2330 m3 s−1, respectively. These high flows were derived from the first flood event in the river basin. During 28–30 June, the water inflow increased from 735 to 2,620 m3 s−1, and outflow from 589 to 2,620 m3 s−1 (Fig. 2A). The outflow SSC from the Sanmenxia Reservoir was 18.8 kg m−3 on 30 June (earlier SSC data were unavailable) (Fig. 2E). The second outflow SSC peak of 126 kg m−3 occurred on 15 July, when Sanmenxia Reservoir inflow and outflow were 2370 and 2500 m3 s−1, respectively, following a temporary reduction. The third peak of 40 kg m−3 occurred on 18 September, when Sanmenxia Reservoir inflow and outflow were 4650 and 4490 m3 s−1, respectively (Fig. 2A, E). Rapid erosion in the Sanmenxia Reservoir during the 2019 flood season is corroborated by annual sediment load data issued by the Ministry of Water Resources of China34.

River channels downstream of the Xiaolangdi Reservoir exhibited opposing reactions to the rapid erosion occurring in the reservoir. Prior to the flood events, the river water in the downstream channel of the Xiaolangdi Reservoir was clear (Fig. 3A); during the flood events, a large amount of sediment poured down from the Xiaolangdi Reservoir, resulting in a rapid increase in SSC in the downstream channel (Fig. 3B, C); after the flood events, the downstream channel returned to a low SSC state (Fig. 3D). During the flood events, cumulative sedimentation in the 126 km length of downstream channels between Xiaolangdi and Huayuankou stations reached ~280 Mt, which accounted for 75% of the Xiaolangdi Reservoir erosion (373 Mt). The date of maximum daily sedimentation in the downstream channels (9 August, 55 Mt) was consistent with that of maximum daily Xiaolangdi Reservoir erosion (9 August, 60 Mt) (Fig. 2H).

Fig. 3: Changes in turbidity in the downstream channel of Xiaolangdi Reservoir.
figure 3

River channel conditions A Before, B, C during, and D after Flood Period 1 (Phase 1 in Fig. 2).

Full size image

Annual deposition and erosion in the seven reservoirs over the past decades

The annual water and sediment supplies from the gauged and ungauged areas all showed rapid decreasing trends over the past seven decades (Supplementary Fig. S2). The water discharges at the Tongguan Station (entrance of the Sanmenxia Reservoir) ranged from 14.9 to 69.9 km3 yr−1, and averaged 33.4 km3 yr−1; the sediment fluxes at the Tongguan Station ranged from 55 to 2995 Mt yr−1 and averaged 911 Mt yr−1. The estimated water discharges from the ungauged areas between Tongguan station and Xiaolangdi Station (immediately downstream of the Xiaolangdi Dam) ranged from 0.17 to 2.70 km3 yr−1, and averaged 0.72 km3 yr−1; the estimated sediment fluxes from the ungauged areas between Tongguan station and Xiaolangdi Station ranged from 0.05 to 44.5 Mt yr−1, and averaged 5.6 Mt yr−1. The total annual water inflows of the Sanmenxia and Xiaolangdi reservoirs (defined as sum of the water discharge at the Tongguan Station and the water supply from the ungauged areas) ranged from 15.2 to 72.6 km3 yr−1, and averaged 34.1 km3 yr−1; the total annual sediment inflows of the Sanmenxia and Xiaolangdi reservoirs (defined as sum of the sediment flux at the Tongguan Station and the sediment supply from the ungauged areas) ranged from 55.1 to 3034 Mt yr−1, and averaged 916 Mt yr−1. The contribution of the ungauged water discharge to the total water inflow into the reservoirs ranged from 0.63% to 4.67%, and averaged 2.04%; the contribution of the ungauged sediment supply to the total sediment inflow into the reservoirs ranged from 0.01% to 1.64% and averaged 0.39%. Thus, the contribution of the ungauged water and sediment supply to the water and sediment flows of the reservoirs were generally low. At the Xiaolangdi Station, the water discharge ranged from 14.4 to 70.2 km3 yr−1 and averaged 33.6 km3 yr−1; the sediment flux ranged from ~1 to 2980 Mt yr−1 and averaged 837 Mt yr−1. The absolute predominance of gauged sediment flux over ungauged sediment supply suggests that annual depositions and erosions of the Sanmenxia and Xiaolangdi reservoirs based on sediment budget are highly reliable.

Based on annual sediment inflow and outflow data, the net erosion of the Xiaolangdi Reservoir in 2019 was 265 Mt (Fig. 4). The net erosion in this reservoir during July–September 2019 was ~300 Mt (Fig. 2G), indicating a cumulative sedimentation of ~35 Mt during the remaining nine months of that year when daily sediment-load data were unavailable. The sediment inflow and outflow of the Sanmenxia Reservoir in 2019 were 168 and 280 Mt, respectively, indicating an annual net erosion of ~112 Mt, with most occurring during the flood events when >85% of the annual sediment outflow from the Sanmenxia Reservoir (i.e., sediment inflow to the Xiaolangdi Reservoir) occurred during the flood season (July–September) (Fig. 2F).

Fig. 4: Annual sedimentation and erosion in the Sanmenxia Reservoir and Xiaolangdi Reservoir.
figure 4

Because the Sanmenxia and Xiaolangdi reservoirs began operation in 1960 and 1999, respectively, sedimentation and erosion are shown for only Sanmenxia Reservoir prior to 1999, whereas sedimentation and erosion are shown for both Sanmenxia and Xiaolangdi reservoirs since 1999.

Full size image

Initial erosion in the Sanmenxia Reservoir was recorded in 1965, only five years after impoundment, with the initial annual erosion of 277 Mt yr−1 contributing almost 40% of the downstream annual sediment load (735 Mt yr−1). The highest annual erosion of 300 Mt was recorded in 1971 (Fig. 4), contributing 18% of the downstream sediment load (1650 Mt). In 2020, Sanmenxia Reservoir erosion (104 Mt) contributed 30% of the downstream sediment load of 344 Mt. In the period 1965–2020, erosion exceeded sedimentation in the Sanmenxia Reservoir, with a cumulative erosion of 1430 Mt (735 Mt yr−1), with yearly alternations between erosion and deposition. More than 40% of the total sedimentation of 1960–1964 (3480 Mt) was eroded in the following 55 years.

For the Xiaolangdi Reservoir, prior to erosion in 2019, sedimentation was continuous for 20 years (1999–2018), with a cumulative sedimentation of 3260 Mt (163 Mt yr−1; Fig. 4). The overall dry bulk density of the rapid sedimentation in the Sanmenxia and Xiaolangdi reservoirs was 0.93 g cm−3 (Supplementary Fig. S3), giving sedimentation volumes in the Sanmenxia Reservoir (1960–1964) and Xiaolangdi Reservoir (1999–2018) of 3.74 and 3.50 km3, respectively. The storage capacities of the Sanmenxia and Xiaolangdi reservoirs were 6.0 and 12.7 km3, and their initial net annual erosion occurred at the time when siltation-induced storage capacity losses were 62% and 28%, respectively.

Annual net erosion events have also been recorded for the Qingtongxia, Yanguoxia, Wanjiazhai, Naodehai, and Gongzui reservoirs (Fig. 5). Initial erosion in these reservoirs occurred 5–26 years after impoundment, when cumulative losses of storage capacity due to previous sedimentation were 43–87% (Table 1). The Qingtongxia Reservoir began operation in 1967, with initial erosion occurring in 1972, when the cumulative storage capacity loss due to rapid sedimentation was 87% (average 105 Mm3 yr−1). The highest annual erosion recorded in the Qingtongxia Reservoir was 28 Mm3 yr−1 in 1981, with a maximum relative increase in annual sediment outflow of 40% (Fig. 5A; Table 1). The Yanguoxia Reservoir began operation in 1961, with initial erosion in 1971 when the cumulative storage capacity loss due to sedimentation was 76% (average 18 Mm3 yr−1). The highest annual erosion recorded in the Yanguoxia Reservoir was 6.1 Mm3 in 1971, with a maximum relative increase in the annual sediment outflow of 52% (Fig. 5B; Table 1). The Wanjiazhai Reservoir began operation in 1998, with initial erosion in 2014 when the cumulative loss of storage capacity due to sedimentation was 50% (average 28 Mm3 yr−1). The highest annual erosion recorded in the Wanjiazhai Reservoir was 7.6 Mm3 in 2014, with a maximum relative increase in annual sediment outflow of 49% (Fig. 5C; Table 1). The Naodehai Reservoir began operation in 1942, with initial erosion in 1951 when the cumulative loss of storage capacity due to sedimentation was 43% (average 8 Mm3 yr−1). The highest annual erosion recorded in the Naodehai Reservoir was 8.8 Mm3 in 1965, with a maximum relative increase in the annual sediment outflow of 59% (Fig. 5D; Table 1). The Gongzui Reservoir began operation in 1971, with initial erosion in 1987 when the cumulative loss of storage capacity due to sedimentation was 71% (average 14 Mm3 yr−1). The highest annual erosion recorded in the Gongzui Reservoir was 3.1 Mm3 in 2000, with a maximum relative increase in annual sediment outflow of 11% (Fig. 5E; Table 1). After the initial erosion events, storage capacities of the reservoirs tended to stabilize even though deposition and erosion occurred alternately (Fig. 5).

Fig. 5: Annual deposition rate and cumulative loss of storage capacity in reservoirs.
figure 5

A–C Qingtongxia, Yanguoxia, and Wanjiazhai reservoirs in the upstream and upper middle reaches of the Yellow River (see Fig. 1A for locations). D Naodehai Reservoir on the Liuhe River, a tributary of the upper Liaohe River. E Gongzui Reservoir on the Daduhe River, a tributary of the upper Yangtze River. Note: sources of deposition/erosion data were as follows: Qingtongxia Reservoir, changes in storage capacity reported by Li67; Yanguoxia Reservoir, Zhang68; Wanjiazhai Reservoir, Ren et al.69; Naodehai Reservoir, changes in storage capacity reported by Han27 and Jiao et al.63; Gongzui Reservoir, Lin65 for 1971–1986 and Xiong and Ma70 for 1994–2007.

Full size image
Table 1 Examples of reservoir erosion after years to decades of sedimentation, and impact on downstream sediment delivery
Full size table

Deposition thickness in 1999–2018 and erosion depth in 2019 in the Xiaolangdi Reservoir

For the Xiaolangdi Reservoir, deposition during 1999–2018 and erosion during 2019 occurred mainly in the mainstream channel. Along the 120-km length of thalweg, the mean thickness of deposited sediment during 1999–2018 was 49 m (2.45 m yr−1), with a maximum thickness of 77 m at sites located 5–10 km upstream of the dam. In 2019, the mean erosion depth along the thalweg was 7.3 m, with a maximum erosion depth of 11 m in the middle reaches of the main river channel (Fig. 6A). In the channel cross-section, deposition during 1999–2018 and erosion in 2019 occurred mainly in the deep channel, with little bathymetric change near the riverbanks (Fig. 6B–C). This rapid erosion occurred mainly during the flood event of July 2019. Bathymetric surveys in the Xiaolangdi Reservoir were undertaken each October, and erosion in 2019 was based on bathymetric survey data of October 2018 and October 2019. Based on the data in Fig. 2G, the erosion depth in Fig. 6 would have resulted from episodic erosion during floods from July to early August 2019. More importantly, maximum erosion depths during the episodic events would have been greater than those shown in Fig. 6 due to the recovery of deposition in September 2019 (Fig. 2G).

Fig. 6: Deposition and erosion along a longitudinal section and two cross-sections in the Xiaolangdi Reservoir.
figure 6

A Temporal changes in the surface sediment profile along the thalweg. B, C Temporal changes in the profile along transects 1 and 2, respectively (see Fig. 6A for locations). Bathymetric surveys in each year were undertaken in October.

Full size image

Impacts of annual deposition/erosion of the Sanmenxia and Xiaolangdi reservoirs on downstream sediment load and channel morphology

Slow net erosion (6 Mt yr−1) occurred in the previously highly silted-up Sanmenxia Reservoir during 1999–2018, while rapid deposition (169 Mt yr−1) occurred in the newly formed Xiaolangdi Reservoir. Most (62%) of the sediment from upstream reaches was retained in the Xiaolangdi Reservoir, which led to a dramatic decrease in sediment outflow and a sharp decline in downstream SSC levels, resulting in channel erosion in reaches downstream of the Xiaolangdi Reservoir. The total erosion rate in the 756 km of downstream reaches was 44 Mt yr−1, offsetting ~25% of the sediment loss due to reservoir deposition. The erosion intensity in the first 126 km reach (0.09 Mt km−1 yr−1) was higher than that in the lower 630 km reach (0.05 Mt km−1 yr−1). Downstream sediment erosion occurred at a much lower rate than reservoir deposition, so sediment discharge to the sea was markedly reduced relative to the fluvial sediment load above the reservoirs (Fig. 7A).

Fig. 7: Sediment transport processes in the study area.
figure 7

Sediment flow, and erosion and deposition in the Sanmenxia (Sanmenxia R.) and Xiaolangdi (Xiaolangdi R.) reservoirs and downstream reaches of the Yellow River in (A) the period 1999-2018 and (B) 2019. Blue, erosion; brown, deposition; positive value, gain of sediment flow; negative value, loss of sediment flow; distance in parentheses = length of river reach.

Full size image

In 2019, the total erosion in the Sanmenxia and Xiaolangdi reservoirs amounted to 377 Mt, which increased the downstream sediment load by 224%. As a result, rapid deposition occurred in channels downstream of the Xiaolangdi Dam: deposition rates were 228 Mt yr−1 in the 126 km upper reaches (1.8 Mt km−1 yr−1) and 9 Mt yr−1 in the 630 km middle–lower reaches (0.01 Mt km−1 yr−1) (Fig. 7B), showing a rapid downstream decreasing trend. Assuming all deposition was contributed by sediment derived from reservoir erosion, 63% of eroded sediment from the two reservoirs settled in downstream channels. This assumption is reasonable considering that deposition is unlikely to occur in these channels without an increased sediment supply from reservoir erosion (see Fig. 7A). Although most sediment derived from reservoir erosion was deposited in the downstream channels, the sediment discharge from the Yellow River to the Bohai Sea increased by 61% relative to the sediment load at the Tongguan Station above the reservoirs (Fig. 7B).

Discussion

In 2019, the time series of daily water outflow of the Sanmenxia Reservoir in 2019 was in good agreement with the time series of water inflow (Fig. 2A), suggesting that few water-level management practices were conducted in this reservoir in 2019. Thus, nearly all of the 112 Mt erosion of the Sanmenxia Reservoir in 2019 could be attributed to natural floods, whereas little erosion was attributable to drawdown and artificial floods. In contrast, considerable water drawdown was conducted in the Xiaolangdi Reservoir in 2019. In fact, more than 95% of the water storage was released from March to July (Fig. 2C). During this period, the water outflow of the Xiaolangdi Reservoir was evidently higher than the water inflow, suggesting an artificial flow in the reservoir. Most importantly, the water outflow was ~2 times higher than the water inflow from the end of June to mid-July, resulting from the combination of natural and artificial floods. Because water inflow of the Xiaolangdi Reservoir was almost equal to the water inflow of the Sanmenxia Reservoir, we roughly estimated that natural flood-induced erosion in the Xiaolangdi Reservoir would have approximated to the erosion in the Sanmenxia Reservoir. That is, ~40% of the erosion in the Xiaolangdi Reservoir in 2019 (265 Mt) was attributable to natural floods, whereas ~60% was ascribed to drawdown and artificial floods. Specifically, the mean water inflow and outflow of the Xiaolangdi Reservoir during 22 June–12 July were 1760 and 3860 m3 s−1, respectively, compared with 813 m3 s−1 during the half year prior to 22 June (Fig. 2B). This suggested a moderate natural flooding and a severe drawdown-induced artificial flooding during the period from 22 June to 12 July. Based on the sediment budget of inflow and outflow, the cumulative erosion in the Xiaolangdi Reservoir during this 21-day period was 113 Mt (Fig. 2F), which undoubtedly should be attributed mainly to drawdown practices. Between 13 July and 10 August, the water level and water storage capacity were at their lowest, and less artificial flow was formed, which led to a high consistency between water outflow and inflow (Fig. 2B–C). During this period, a severe natural flooding with a highest water inflow of 4570 m3 s−1 occurred, and the cumulative erosion amounted to 224 Mt (Fig. 2F). This rapid erosion should be attributed to the combined effect of drawdown and natural flooding. After that, water impoundment practices were conducted, and a compensatory deposition amounting to ~70 Mt occurred, despite a second severe flooding. Therefore, in fully silted reservoirs, like the Sanmenxia Reservoir, considerable erosion will most likely occur during episodic natural flooding events35, unless erosion has already occurred during previous floods. In half silted reservoirs (such as the Xiaolangdi Reservoir), erosion depends on the combination of natural floods with water management practices. Drawdown prior to natural flooding and maintaining lowest water level and storage during the flooding are effective practices to cause erosion. On the other hand, water impoundment during flooding will reduce erosion or even lead to deposition. Water regulation has been applied in many reservoirs worldwide to reduce sedimentation and maintain storage capacity20,28,36.

Our results of the case study from seven reservoirs indicate that the sedimentation ratio (sedimentation volume/storage capacity) at which initial annual net erosion occurs ranged from 28% to 87% with an average of 60% ± 21% (Table 1). Therefore, we propose a critical threshold of sedimentation ratio as ~30% to assist in assessing the risk of sediment erosion in reservoirs. Generally, a higher sedimentation ratio corresponds to an increased risk of reservoir erosion during floods. Compared to coastal sediments, sediments deposited in reservoirs are generally low in compaction and high in water content37,38,39, because of rapid deposition of fine-grained particles in a short history. For example, the sedimentation histories prior to initial erosions in the seven reservoirs ranged from 5 to 20 years (Table 1). As shown in Supplementary Fig. S3, the overall dry bulk density of the sedimentation in the Sanmenxia and Xiaolangdi Reservoirs is 0.93 g cm−3. In the Songtao Reservoir in South China, dry bulk density is only 0.61 g cm−337. In comparison, the dry bulk density of coastal sediments is generally 1.2–1.5 g cm−338,39. A lower critical shear stress for erosion is often related to a lower bulk density of relatively freshly deposited muddy sediments40,41. Consequently, reservoir sediments can be more susceptible to erosion when exposed to flowing water. Water depth in reservoirs decreases with siltation. In the Xiaolangdi Reservoir, for example, the channel bed level increased by more than 80 m at a location near the dam and by ~50 m in the middle reaches due to siltation from 1999 to 2018 (Fig. 6A2), corresponding to a sedimentation ratio of 28%, prior to the initial erosion occurred in 2019 (Table 1). An increase in bed level means a decrease in cross-section of the river. Under a specific water inflow, it is expected that current velocity may increase in a channel where the cross-section has been reduced. In other words, the time that water inflow stays in a reservoir is proportional to the amount of water stored15, so reduced water storage implies higher velocity of flow through the reservoir. The sediment-carrying capacity of flowing water is proportional to the cube of the flow velocity42,43. The water inflow of reservoirs during flood events can be several times greater than normal inflow (see Fig. 2A for examples). With not only much greater water inflow but also much higher flow velocity, the total sediment-carrying capacity would increase evidently in highly silted reservoirs during flood events. Drawdown-induced artificial flooding and water-depth decrease would further increase erosion in silted reservoirs. Thus, the sedimentation ratio for initial reservoir erosion may be associated with factors such as sediment compaction, grain-size distribution and hydrodynamic threshold, and the difference in this sedimentation ratio among reservoirs could be influenced by local conditions of meteorology, hydrology, sediment property, reservoir design, and even reservoir management.

Erosion depths of a few meters to >10 m during reservoir flushing events have been reported in previous studies. For example, erosion depths of 1.0–2.3 m along the mainstream in the Cachi Reservoir, Costa Rica, were surveyed during a 1-month flushing in 199644, and erosion depths of 4–14 m were measured in lakes Aldwell and Mills, USA, within 1 year of dam removal22. Flood–drawdown-induced episodic erosion in the Xiaolangdi Reservoir in 2019 yielded a maximum erosion depth of 11 m along the main channel, likely greater than that in most cases of episodic reservoir erosion, with >10% of previously deposited sediment being eroded with an increase in reservoir net storage capacity of 4–5%. Initial erosion in the Sanmenxia Reservoir during 1965 constituted 8% of the prior total deposition, and the maximum annual erosion in 1971 constituted 10% of prior net deposition (Fig. 4), increasing reservoir net storage capacity by 13% and 11%, respectively. Sediment outflow of the Xiaolangdi Reservoir during episodic erosion in 2019 increased by more than an order of magnitude compared with inflow (Fig. 2F). Outflow SSC levels from the Xiaolangdi Reservoir during 2019 episodic erosion events, with peak values of up to 210 kg m−3 and an event-averaged value of 57 kg m−3, were higher than those recorded during many other episodic reservoir erosion events. For example, in a flushing operation in the Valgrosina Reservoir, Italy, the outflow SSC peaked at 80 kg m−3, with an average of 4 kg m−345. Peak SSC values of 11–17 kg m−3 were recorded during dam-flushing events in the Arc–Isère River system, France24. Peak SSC values of 21–28 kg m−3 were recorded during flushing of reservoirs in the Kurobe River, Japan20. A maximum SSC of 850 kg m−3 was once recorded during breaching of the Condit Dam in the White Salmon River, USA21, much higher than the peak SSC of 210 kg m−3 recorded in the Xiaolangdi Reservoir during 2019 episodic erosion events. In extreme cases, peak SSC levels during the flushing of reservoirs may exceed 1000 kg m−328. Episodic erosion in silted-up reservoirs can thus be very intense, with extremely turbid outflow.

Episodic reservoir erosion events may have severe ecological impacts in downstream reaches. Such erosion usually forms fluid mud with SSC levels exceeding 10 kg m−346 in downstream river channels (Fig. 3B, C). Episodic erosion of the Sanmenxia and Xiaolangdi reservoirs in 2019 led to the formation of fluid mud at Xiaolangdi Station (gauging outflow of the Xiaolangdi Reservoir) for 27 days (10 July-11 August, excluding 1 and 3–7 August, avg. SSC 69 kg m−3) (Fig. 2E). At Huayuankou Station, 126 km downstream, fluid mud persisted for 32 days (11 July-13 August, excluding 5 and 12 August, avg. SSC 26 kg m−3). Even at Lijin Station (tidal limit), 630 km farther downstream, fluid mud lasted 33 days (14 July-15 August, avg. SSC 18 kg m−3), based on daily SSC data collected by the Yellow River Conservancy Commission (YRCC). These findings add to our previous understanding from other reservoirs worldwide that the initial period of erosion, with a duration ranging from a few hours to several days depending on the size of the reservoir, is characterized by extremely high SSC levels24,28,45. Due to the lack of specific ecological surveys, the environmental impact analysis in this study has certain limitations. Nevertheless, previous research has demonstrated related effects. High turbidity reduces photosynthesis in phytoplankton47, thereby reducing primary productivity; and the combination of fluid mud, which smothers benthic organisms and clogs gills, and anoxia may kill virtually all organisms in a river48,49. Similar acute negative effects have been described for invertebrates45. Meanwhile, rapid deposition caused by extremely high SSC levels may destroy benthic communities where organisms tend to reside in the uppermost 10 cm of the sediment layer50; in many cases of episodic reservoir erosion, downstream channel deposition may exceed this threshold thickness21,24,34. Many natural reserves in the lower Yellow River and Yellow River Delta51,52 would inevitably be affected by fluid mud and rapid deposition. Of course, to some extent, episodic reservoir erosion can alleviate the sediment starvation in estuarine deltas if it is well arranged53.

Reservoir sediment erosion and its ecological impacts on downstream environments under episodic flooding and drawdown conditions have been reported in previous studies, either based on field measurements or numerical simulations54,55,56. Based on the sedimentation ratio of 30% proposed earlier in this study—above which reservoirs are at risk of erosion under episodic flooding and drawdown conditions—a map showing the global distribution of 57 large reservoirs (storage capacity > 0.1 km³) with more than 30% loss in capacity has been generated. Many reservoirs worldwide are fully or >30% silted-up, especially in the Northern Hemisphere, such as in China, India, Japan, United States and Iran (Fig. 8; Table 2). It indicates that episodic flooding and drawdown events may pose a widespread threat globally, and their impacts on the ecological environment of downstream river channels should not be overlooked. The lack of hydrological measurements and bathymetric surveys in most of the world’s rivers4 means episodic erosion in other silted-up reservoirs may remain undocumented. Perhaps the use of advanced GIS, geostatistical methods and remote sensing could be a good option to identify changes in topography and terrain in river drainage basins and networks during extreme events57. The above examples indicate that episodic erosion events in silted-up reservoirs during extreme floods and/or water-regulation periods may be common, with considerable impact on river sediment transport and fluvial environments. More reservoirs are being constructed or planned, particularly in developing countries such as China and Brazil31, so it is expected that an increasing number of reservoirs will become silted up58. Moreover, extreme flooding will likely increase under global warming59,60. It follows that episodic erosion in reservoirs worldwide may become increasingly common and more intense in coming decades. Strengthening coordinated water regulation between upstream and downstream reservoirs and increasing water discharge during flood periods can help reduce SSC. This, in turn, can mitigate sediment deposition in downstream channels and the ecological impacts of turbid water, allowing more sediment to reach the estuarine delta and alleviating the ongoing erosion caused by sediment deficiency35,49. Considering the differences in reservoir function types (e.g., hydropower, water supply, and sediment-regulating reservoirs), reservoir morphology, sediment characteristics, hydrology, and geology, improvement of management practices (e.g., drawdown timing, water discharges) varies across reservoirs54,61.

Fig. 8: Global distributions of large reservoirs (storage capacity >0.1 km3) with > 30% loss in capacity and examples of other severely silted-up reservoirs.
figure 8

This figure is created based on Minocha and Hossain30 and Table 2.

Full size image
Table 2 Examples of severely silted-up reservoirs worldwide
Full size table

Conclusions

Episodic reservoir erosion events and their downstream sedimentary impacts were identified on the basis of data for seven reservoirs in China. Maximum reservoir erosion depth may exceed 10 m during a flood event, increasing the downstream sediment load by one to two orders of magnitude, with peak SSC levels of up to 210 kg m−3. Sediment derived from reservoir erosion during a flood event may contribute not only most of the annual sediment flux, but also ~25% of the decadal sediment flux in downstream reaches. Threshold siltation ratios at which initial erosion occurred in the studied reservoirs were in the range of 28–87%, with such erosion occurring 5–26 years after impoundment. Natural floods alone can cause severe erosion in reservoirs, like the example of the Sanmenxia Reservoir where an erosion of 112 Mt was observed in the flooding 2019. However, regulated drawdown and artificial floods can effectively increase reservoir erosion during natural flooding events. ~60% of the Xiaolangdi erosion was attributed to drawdown and artificial flooding, whereas ~40% was attributed to natural floods. Our results indicate that episodic erosion can occur in partially silted reservoirs, with natural floods and regulation-induced low water depths providing favorable conditions. We supposed that the mechanisms responsible for reservoir erosion are that rapidly silted loose fine-grained sediments are easily erodible under high sediment-carrying capacities during flooding and drawdown. Reservoir erosion-induced extremely high SSC levels and related rapid deposition in downstream channels may adversely affect river ecosystems and socioeconomics. It is anticipated that reservoir erosion will become more frequent worldwide in the near future, as an increasing number of reservoirs will likely silt up, and the frequency of extreme floods may increase with global warming. Our findings aid understanding of the positive and negative effects of reservoirs on river systems, and may promote improvements in river management.

Methods

Study area

Seven large reservoirs in China were examined as case studies, including the Yanguoxia (0.23 km3 storage capacity; 36.2°N, 103.5°E), Qingtongxia (0.61 km3; 37.9°N, 106.3°E), Wanjiazhai (0.90 km3; 39.9°N, 111.5°E), Sanmenxia (6.0 km3; 35.1°N, 111.6°E), and Xiaolangdi (12.7 km3; 35.3°N, 113.2°E) reservoirs on the Yellow River; the Naodehai Reservoir (0.17 km3; 42.6°N, 121.5°E) on the Liuhe River; and the Gongzui Reservoir (0.31 km3; 28.8°N, 102.9°E) on the Daduhe River (Fig. 1A). Those on the Yellow River are located in downstream order along the mainstream of the Yellow River (Fig. 1B), which originates on the Qinghai–Tibet Plateau at 4200 m above sea level and flows 5460 km eastward to the Bohai Sea (Fig. 1B). The basin has a catchment area of 750,000 km2 and currently supports a population of 110 million people. The average annual precipitation and potential evaporation within the Yellow River basin are 460 mm and 1700 mm, respectively. The water discharge and sediment flux from the Yellow River to the sea were 47 km3 yr−1 and 1400 Mt yr−1, respectively, prior to dam constructions, which begun in the 1960s2,10,39 (Fig. 1C). Tongguan Hydrologic Station is commonly acknowledged as being the entrance to the Sanmenxia Reservoir34,62. Sanmenxia Station lies between the Sanmenxia Dam and the Xiaolangdi Reservoir, and Xiaolangdi Station is located immediately downstream of the Xiaolangdi Dam. Sanmenxia and Xiaolangdi stations are thus ideal for respectively gauging inflow and outflow of the Xiaolangdi Reservoir. Huayuankou Station is ~200 km downstream of Xiaolangdi Station, and Lijin Station (the tidal limit) a farther ~600 km downstream (Fig. 1A). The Liuhe River is a tributary of the Liaohe River27,63, which is a major river in Northeast China with a length of 1345 km and a catchment area of 229,000 km2, supporting a population of 35 million. Its water discharge and sediment flux to the sea were ~5 km3 yr−1 and 40 Mt yr−1, respectively, before declining in the 1960s53,64. The Daduhe River, SW China, is a secondary tributary of the Yangtze River65, which is one of the world’s largest rivers, with a length of 6400 km, a drainage area of 1.8 million km2, 900 km3 yr−1 of water discharge, and 500 Mt yr−1 of sediment discharge prior to dam construction, supporting a population of 450 million66.

Data source

Bathymetric data from the Sanmenxia and Xiaolangdi Reservoirs and water and sediment fluxes at the gauging stations along the Yellow River were compiled by the Yellow River Conservancy Commission (YRCC) of the Ministry of Water Resources of China (MWRC) (http://www.mwr.gov.cn). Sources of deposition/erosion data for other reservoirs were as follows: Qingtongxia Reservoir, changes in storage capacity reported by Li67; Yanguoxia Reservoir, Zhang68; Wanjiazhai Reservoir, Ren et al.69; Naodehai Reservoir, changes in storage capacity reported by Han27 and Jiao et al.63; Gongzui Reservoir, Lin65 for 1971–1986 and Xiong and Ma70 for 1994–2007. All the available Sentinel-2 Level-2A and Landsat-8 Collection 2 Level-2 images with cloud cover <20% were acquired from the Google Earth Engine (GEE) platform for the period of 1 June to 31 August, 2019. Both datasets were spatially clipped to the study area boundary and exported as GeoTIFFs in WGS84 (EPSG:4326) at native resolution (10 m for Sentinel-2; 30 m for Landsat-8)71. Most data of global reservoir capacity and their losses in capacity were obtained from Minocha and Hossain30.

Estimating water and sediment supply from ungauged areas

There is no gauging station in the 12,000 km2 of tributaries between Tongguan and Xiaolangdi stations. To better understand erosion and deposition in this mainstem reach, we estimated water and sediment inflows that originated from the ungauged tributaries using an analogy method. There are three gauging stations, namely Hejin (catchment area of 38,700 km2), Heishiguan (catchment area of 18,600 km2), and Wuzhi (catchment area of 12,900 km2), on the tributaries, either higher than Tongguan Station or lower than Xiaolangdi Station (Fig. 1B). The ratio of the ungauged area (12,000 km2) to the total gauged area of these three stations (70,200 km2) is 0.171. Thus, we calculated water and sediment inflows from the ungauged tributaries as follows:

$${Q}_{-{ungauged}({Tongguan}{{{\rm{\hbox{-}}}}}{Xiaolangdi})}=\,0.171({Q}_{({Hejin})}+{Q}_{({Heishiguan})}+{Q}_{({Wuzhi})})$$
(1)

where Q-ungauged (Tongguan–Xiaolangdi) represents the water inflow from the area of ungauged tributaries between Tongguan and Xiaolangdi Stations, and Q(Hejin), Q(Heishiguan), and Q(Wuzhi) are the water flows at Hejin, Heishiguan, and Wuzhi Stations, respectively, and

$${Q}_{s-{ungauged}({Tongguan}{{{\rm{\hbox{-}}}}}{Xiaolangdi})}=\,0.171({Q}_{s({Hejin})}+{Q}_{s({Heishiguan})}+{Q}_{s({Wuzhi})})$$
(2)

where Qs-ungauged (Tongguan–Xiaolangdi) is the sediment inflow from the area of ungauged tributaries, and Qs (Hejin), Qs (Heishiguan), and Qs (Wuzhi) are the sediment flows at Hejin, Heishiguan, and Wuzhi Stations, respectively.

Quantifying deposition and erosion in reservoirs

The Sanmenxia and Xiaolangdi reservoirs began operating in September 1960 and October 1999, respectively36,62. For the period of 1960–1998, the sediment outflow that was subtracted from the sediment inflow for the reach between Tongguan and Xiaolangdi stations may have reflected mainly deposition (negative value indicates erosion) in the Sanmenxia Reservoir, assuming that deposition or erosion in the non-reservoir reach between the Sanmenxia Dam and Xiaolangdi Station was negligible. Thus, we calculated the annual weight of deposited sediment in the Sanmenxia Reservoir as follows:

$${D}_{({Weight\; in\; Sanmenxia\; Reservoir})} = \, {Q}_{s({Tongguan})}+{Q}_{s-{ungauged}({Tongguan}-{Xiaolangdi})} \\ - {Q}_{s({Xiaolangdi})}$$
(3)

where D(Weight in Sanmenxia Reservoir) represents the annual weight of deposited sediment in the Sanmenxia Reservoir.

For the period of 1999–2020, the sediment outflow subtracted from the sediment inflow for the reach between Tongguan and Xiaolangdi stations reflects the net weight of deposited sediment in the Sanmenxia and Xiaolangdi reservoirs. Since 1999, annual volume of deposited sediments has been surveyed in the two reservoirs, using a bathymetric approach34. We calculated the total volume of deposited sediment as follows:

$${D}_{({Total\; volume})}={D}_{({Volume\; in\; Sanmenxia\; Reservoir})}+{D}_{({Volume\; in\; Xiaolangdi\; Reservoir})}$$
(4)

where D(Total volume) represents the total volume of deposited sediment, D(Volume in Sanmenxia Reservoir) represents the volume of deposited sediment in the Sanmenxia Reservoir, and D(Volume in Xiaolangdi Reservoir) is the volume of deposited sediment in the Xiaolangdi Reservoir.

Then, we calculated the total weight of deposited sediment in the two reservoirs as follows:

$${D}_{({Total\; weight})}={Q}_{s({Tongguan})}+{Q}_{s({Tongguan}-{Xiaolangdi})}-{Q}_{s({Xiaolangdi})}$$
(5)

Next, we made a double-mass plot of the cumulative volume of deposited sediment and cumulative weight of deposited sediment (Supplementary Fig. S3). This plot shows a strong correlation (R2 = 0.99), suggesting that the annual deposition data obtained using the two approaches are highly credible. The plot also suggests an overall dry bulk density of 0.93 g cm−3. Using this value, we converted the annual volume of deposited sediment to an annual weight of deposited sediment.

Finally, we adjusted the converted annual weight of deposited sediments based on the ratio of D(Weight in Sanmenxia Reservoir) to D(Weight in Xiaolangdi Reservoir) and the ratio of D(Total weight) to the sum of D(Weight in Sanmenxia Reservoir) and D(Weight in Xiaolangdi Reservoir), where D(Weight in Sanmenxia Reservoir) is the weight of deposited sediment in the Sanmenxia Reservoir, D(Weight in Xiaolangdi Reservoir) is the weight of deposited sediment in the Xiaolangdi Reservoir, and D(Total weight) is the total weight of deposited sediment in the two reservoirs calculated using the sediment-budget approach, following the principle that the sum of the adjusted D(Weight in Sanmenxia Reservoir) and D(Weight in Xiaolangdi Reservoir) is equal to D(Total weight).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.