Assessing the effects of dam regulation on multiscale variations in river hydrological regime and ecological responses

The natural variability of flow regime in rivers plays a vital ecological role in maintaining river morphology, preserving biodiversity, and shaping natural habitats (Shiau & Wu 2004; Wang et al. 2012; Gabiri et al. 2018). However, under changing environmental conditions, the hydrological regimes of many rivers worldwide have experienced significant alterations (Rolls et al. 2018; Wang et al. 2022; Battin et al. 2023). Human activities primarily manifest as flow regulation through dam operations, which directly influence the flow regime of rivers and are considered the most prominent factor. In addition, climate change, characterized by changes in precipitation and extreme weather events, further drives alterations in flow regimes (Li et al. 2015; Stagl & Hattermann 2016; Thompson et al. 2021).

In recent years, under the increasing influence of external drivers, the flow regime of rivers has exhibited complex patterns of change, attracting the attention of numerous scholars (Suen 2011; Taylor et al. 2014; McGregor et al. 2018; van Oorschot et al. 2018; Bestgen et al. 2020). Richter et al. (1996) proposed a method to assess the degree of hydrologic alteration caused by human impacts within ecosystems, known as the ‘Indicators of Hydrologic Alteration’ (IHA). Masikini et al. (2018) introduced a river health assessment approach that encompasses different states of aquatic ecosystems, with flow as a crucial environmental variable. From the perspective of hydrological variability, the flow regime directly influences the vertical and horizontal structures of aquatic ecosystems, and the intra-annual flow processes serve as important stimuli for aquatic life activities (Wu et al. 2023). Therefore, investigations into the flow regime of rivers should not only focus on inter-annual changes but also comprehensively understand the characteristics of intra-annual flow processes and hydrological events, which are of significant importance for adaptive management of rivers in changing environments. Furthermore, climate change accelerates hydrological cycles and increases water-related risks. The IPCC report indicates that global average temperature has exceeded pre-industrial levels by 1.5 °C, further impacting the physical attributes of water security and increasing water resource vulnerability (IPCC 2022). Meanwhile, strong coupling exists between precipitation and temperature, with climate warming leading to more frequent and intense extreme weather events. In addition, dam regulation has been identified as an important stressor on rivers. These factors collectively exacerbate the degradation risks of river health (Zhang et al. 2021; Yang et al. 2024). The coupling of increased dam construction and sustained climate warming pushes river systems to the limits of their resilience. In this context, it is crucial to gain a detailed understanding of the multiscale changes in river hydrological regimes and quantitatively assess the responses of flow regimes and hydrological events to different driving forces.

Like other random time series, streamflow sequences exhibit multi-timescale structures, indicating that the characteristics of hydrological systems change over the study period and contain periodic features and local fluctuations at different scales within the same timeframe (Chen et al. 2023). Entropy is a measure of the disorder or unpredictability of system changes, and entropy theory encompasses diversity. Based on multi-timescale entropy, Zhang et al. (2019) analyzed the complex fluctuation characteristics of water and sediment load at different timescales and qualitatively assessed the impact of reservoir operations. Huang et al. (2019) designed a streamflow diversity index based on Shannon entropy and analyzed the relationship between streamflow diversity and the average aggregation rate of fish, demonstrating that entropy theory is an effective indicator with general ecological significance. Therefore, entropy theory is an effective approach that can be utilized to assess the complexity of hydrological regime changes. In addition, considering the proven significant impact of dams on flow regimes at intra-annual scales, entropy theory can also be applied to reveal the characteristics of dam influences.

The Three Gorges Dam (TGD), located in the upper reaches of the Yangtze River (URYR), is the largest hydropower project in the world. Its impoundment operation has significantly changed the natural flow and hydrological regimes in the downstream of the dam (Wu et al. 2012; Li et al. 2016; Tao et al. 2020; Xiong et al. 2020; Zhao et al. 2021). However, previous studies have mainly focused on the inter-annual variations of river hydrological regimes and assessed the impact of dam construction based on the difference in hydrological conditions before and after disturbance (Chen et al. 2016; Wang et al. 2016; Peng et al. 2022). The flow regime changes and potential mechanisms between natural flow and observed flow affected by dams have not been fully understood. For highly regulated rivers like the Yangtze River, it is essential to separate the effects of dam construction and other driving forces on the hydrological regime to facilitate water resource management. Therefore, utilizing modeling approaches to reconstruct the natural flow regime without dam regulation and conducting a difference analysis between the reconstructed and observed flow regime is crucial for separating and quantifying the impact of the dam at comprehensive timescales.

There are various methods for reconstructing the natural flow, primarily divided into distributed hydrological models and data-driven models (Valeh et al. 2021; Ren et al. 2024). Distributed hydrological models based on physical principles have strict data requirements and are limited by boundary conditions of the hydrological system (Ma et al. 2016; Su et al. 2020). Data-driven models have also been proven feasible for flow reconstruction, where the principle of these models is to map independent variables to dependent variables (Gao et al. 2020). However, traditional machine learning methods like artificial neural networks are not specifically designed to handle time series data and cannot directly capture temporal dependencies between time series. In contrast, the long short-term memory (LSTM) model addresses the issue of long-term and short-term dependencies in time series, significantly improving the accuracy of time series prediction (Fan et al. 2021; Cho & Kim 2022). Moreover, it has been shown that streamflow in the Yangtze River basin is closely related to precipitation (You et al. 2023), indicating that the LSTM models are feasible in the URYR.

This study proposes an assessment framework aimed at analyzing the multiscale variations of flow regime and their driving factors, as well as investigating the response of ecological events, with the upstream Yangtze River basin as a case study. The range of variation approach is employed to analyze the variations in environmental flow components (EFCs), inter- and intra-annual flow regimes, and the natural flow regime is reconstructed using the LSTM model. Subsequently, we quantitatively assess the impacts of dam construction, climate and incoming water change on flow regimes, and hydrological events and analyze the responses of fish fry abundance to hydrological process changes under dam regulation. This study provides valuable insights into the mechanisms of river hydrological conditions and their responses to driving factors and offers a scientific basis for dam operation strategies oriented toward ecological restoration.

The Yangtze River is the largest river in China in terms of water resources. It originates from the Qinghai-Tibet Plateau, spanning a length of 6,379 km; the total amount of water resources is 9.6 × 10¹² m³, with a basin area of 1.8 × 10⁶ km², accounting for one-fifth of the total land area of China. The mainstream of the Yangtze River is divided into the upper reaches, stretching from the source to Yichang city, covering a length of 4,511 km and controlling a basin area of 1 × 10⁶ km². This section exhibits distinct characteristics of the East Asian subtropical climate, with distinct flood seasons and dry winters. The flood season typically occurs from May to October each year, with the main flood season occurring from July to August (Liu et al. 2018). Due to the abundant water resources, a number of large dams have been constructed in the URYR, and the river flow regime is seriously affected.

The TGD has a dam height of 181 m and a normal water level of 175 m. The flow regulation began in 2003. The operation of the TGD has had profound impacts on hydrological elements such as runoff, sediment, and water temperature in the downstream reservoir area. The Yichang station is located 37 km downstream of the TGD, which serves as the outlet control station for the URYR. The measured flow regime at the Yichang station is significantly influenced by the TGD.

Meteorological data: Daily meteorological data, including daily average temperature and daily precipitation, from 21 meteorological stations in the URYR were collected for the period 1965–2019. The data were obtained from the National Meteorological Science Data Center (https://data.cma.cn/).

Riverine biological data: Early-stage fishery resource data of the Four Famous Major Carps (FFMC) in the Yichang section of the mainstream of the Yangtze River during the spawning period (May–July) were collected. The dataset includes 37 monitoring records from 1997 to 2003 before the construction of the TGD, 31 monitoring records from 2004 to 2009, and 10 monitoring records from 2014 to 2019 (the TGD has been conducting 14 ecological regulation experiments to promote fish reproduction continuously for 10 years since 2011). The data were obtained from the Yangtze River Fisheries Research Institute (http://www.cjyzbgs.moa.gov.cn/).

Step 1: The daily hydrological and meteorological data are collected. The study period is divided into pre-TGD and post-TGD periods based on the construction of the TGD.

Step 2: Reconstruction of the natural flow regime unaffected by the TGD. Climate change and variations in upstream water alter the inflow regime of the TGD, coupled with further regulation by the TGD, resulting in changes in the flow regime at the Yichang station. In this study, the LSTM model is employed to simulate the natural flow regime.

Step 3: Evaluation of multiscale changes in flow regime and ecohydrological events. This study integrates the IHA and five additional indicators that reflect intra-annual hydrological processes. From a hydrological perspective, five ecologically significant flow events are identified.

Step 4: Quantification of the impact of dam construction on flow regime and ecohydrological events. After the construction of the TGD, changes in the hydrological regimes at the Yichang station reflect the combined effects of dam regulation, climate change, and variations in upstream inflow. The specific impact of the TGD is quantified by analyzing the differences between observed and modeled conditions.

The LSTM network exhibits robustness and the ability to learn and remember information, enabling it to capture long-term and short-term dependencies between input and output data. It effectively solves the vanishing gradient problem. The framework based on the LSTM network can identify the flow process under the condition without dam regulation, providing a prerequisite for quantifying the impact of the TGD. By combining indices that characterize the flow regime at inter- and intra-annual timescales, as well as hydrological event indices, the multiscale effects of the dam on the flow regime can be comprehensively revealed.

The pre-TGD period is divided into calibration period (1965–1990) and verification period (1991–2003) to train the LSTM and then reconstruct the natural flow from 2004 to 2019. The root mean square error (RMSE) and Nash–Sutcliffe efficiency coefficient (NSE) are used to evaluate the performance of the LSTM model. The specific principle is referred to in the study by Graf et al. (2019).

Richter et al. (1996) proposed the IHA index system, which characterizes the inter-annual variations of flow regime from five perspectives: magnitude, timing, duration, frequency, and variability. The framework consists of a total of 33 indicators grouped into five categories. Since zero-flow events were not observed at the Yichang station, the analysis in this study includes only 32 of these indicators (Table 1).

Based on the exceedance probabilities of 25 and 75% during the reference period, the threshold range is reconstructed. Ecosurplus and ecodeficit are defined as the portions above the 75% FDC and below the 25% FDC, respectively. They are then normalized using the method proposed by Wang et al. (2017).

EFCs are defined as the required flow and associated processes necessary to maintain ecological stability in a river system. They encompass five flow events: low flow, extreme low flow, high flow pulse, small flood, and large flood (Table 2). The threshold delineation for these flow events associated with EFC can be found in the study by Richter & Thomas (2007).

The FFMC are economically important fish species in the Yangtze River and China. This study focuses on the FFMC as representative fish species to investigate the impact of hydrological process changes caused by dam regulation on riverine ecological events. The spawning period of FFMC generally occurs from May to July, with a lower limit water temperature of 18 °C for reproduction. Under suitable water temperatures, river flooding becomes an important factor inducing FFMC reproductive activities (Wang et al., 2020a, 2020b). Previous studies have shown that FFMC typically start spawning approximately 0.5–2 days, or even 3 days, after the river begins to rise, and spawning ceases when the rising stops, and the fry abundance of FFMC increases with the magnitude of rise (Li & Xia 2011; Yu et al. 2019). Suitable flood conditions are crucial for FFMC reproduction. From the flow perspective, this study focuses on the maximum rising duration, the rising count (2 days and above), and the changes in the maximum and minimum daily average rising rates from 1965 to 2019 during May to July. Furthermore, the fry abundance, as a biological factor reflecting the scale of FFMC reproduction, is correlated with the rising indicators associated with each spawning event, including the rising duration (days), daily average rising rate (m³s⁻¹day⁻¹), and magnitude of rise (m³s⁻¹).

For monthly flow, the change degree is 47.4%, indicating a moderate level. The non-flood season months exhibit higher variability than the flood season, with change degrees of 61.2 and 27.3%, respectively, which align with the regulation of TGD. Furthermore, the decrease in the annual average flow is primarily attributed to the influence of climate and incoming water changes, which accounts for 85% of the overall change. Regarding monthly flow, the dam demonstrates a downstream inhibitory effect from July to November, while it exhibits a rising effect in other months. The most significant impacts are observed in May (rising effect) and November (inhibitory effect). However, climate and incoming water changes lead to reduced flow at the Yichang station from May to October, with notable impacts in April (rising effect) and August (inhibitory effect) (Table 3).

Regarding extreme flow conditions, after the initiation of TGD regulation, minimum and maximum flows tend to increase and decrease, respectively, with a rise in base flow index. The change in minimum flow is significant, with a change degree exceeding 67%, while there is no significant change observed in maximum flow. Climate and incoming water change primarily drives the decrease in 7, 30, and 90-day maximum flows, while the reduction in 1- and 3-day maximum flows, as well as the increase in minimum flow, can be primarily attributed to the regulation of TGD. The occurrence of the 1-day minimum flow has shifted from the pre-TGD period of the 57 Julian date to the post-TGD period of the 4 Julian date, with an advance of 53 days and a change degree of 74.8%. This may be related to the early discharge from the dam (−43). There is no significant change in the timing of the 1-day maximum flow, with an average delay of 2 Julian dates.

In order to further investigate the impact of changes in rising processes caused by dam regulation on fish fry abundance, this study utilized an LSTM model to establish a quantitative relationship between key habitat factors and the fry abundance of the FFMC. The feasibility of establishing the LSTM model was supported by the potential connection between fish fry abundance and habitat factors in the natural state without TGD regulation. The calibration dataset consisted of 26 spawning events from 1997 to 2001, including environmental factors and fry abundance, while the validation dataset included 11 spawning events from 2002 to 2003. The LSTM model was trained using 500 hidden neurons, 2,000 epochs, an initial learning rate of 0.35, a descent period of 220, a descent factor of 0.1, and a batch size of 7. The model's fitting accuracy was examined by comparing the expected values obtained from the LSTM model with the observed values. The results showed an NSE of 0.967 and an RMSE of 1.003 × 10⁸, indicating a good performance. Subsequently, based on the key habitat factors, the expected fry abundance of the FFMC in 2004–2009 and 2014–2019 was obtained (Figure 13).

After the construction of the TGD, an obvious decrease in fry abundance of the FFMC was observed. During 2004–2009, the average fry abundance per spawning event was 3.19 × 10⁷, lower than the 4.18 × 10⁸ observed during 1997–2003. However, during the ecological operation period, the average fry abundance was 9.24 × 10⁸, reaching a peak value of 26.39 × 10⁸ in 2019. From a hydrological perspective, the fit between the expected fry abundance obtained from the deep learning model and the observed values was good in the absence of TGD regulation, with an average expected abundance of 4.16 × 10⁸. However, during the post-TGD period, the expected fry abundance of FFMC spawning events in 2004–2009 was significantly higher than the observed abundance, with an average expected abundance of 3.35 × 10⁸, lower than the average abundance in the pre-TGD period. These results indicate that the reduction in flooding events with durations exceeding 6 days and the occurrence of excessive daily average flooding rates in some events are important factors contributing to the decrease in fry abundance. When considering only the influence of flooding processes, the TGD may have caused an average decrease of 19.5% in fry abundance. During the ecological operation period, the average expected larval abundance was 6.03 × 10⁸, lower than the observed values but higher than the expected values for the periods 1997–2003 and 2004–2009. This indicates that the rising water condition changes caused by the ecological release conducted by TGD effectively stimulated the natural reproductive activity of the FFMC. From the hydrological perspective, the ecological operation by TGD resulted in an average increase of 53.2% in fry abundance.

This study revealed that the inter-annual flow regime variability at the Yichang station was 59.9% according to the RVA method. The concentration of intra-annual flow regime changes decreased, and entropy value, ecosurplus, and ecodeficit showed an increase, with ecodeficit experiencing a higher growth rate than ecosurplus. Compared to the pre-TGD period (1965–2003), the post-TGD period (2004–2019) experienced enhanced hydrological stability of the river, with a trend toward a more uniform distribution of EFCs. Furthermore, we quantified the potential impacts of climate and incoming water changes, as well as the effects of dam construction. The TGD emerged as the primary driver of monthly flow variations at the Yichang station during the non-flood season months (from November to the following April), with an average impact weight of 67.46%. The effects of climate and incoming water changes on flow regime alterations were not as significant as those caused by TGD, making TGD the dominant factor influencing changes observed in 20 IHA indicators. Qualitative analysis by Wang et al. (2016) indicated that the variability and seasonal distribution of flow are significantly regulated by the TGD. Wang et al. (2023) found that the flow regulation by the TGD during June to August is the main factor driving changes in high- and low-flow events. Qian et al. (2022) studied the impacts of climate and human factors on streamflow in the Yangtze River Basin and found that human activities were the dominant factor influencing streamflow changes from 2001 to 2020, with impact weights ranging from 0.48 to 0.61. The conclusions drawn in this study are consistent with previous research. We further discovered that the TGD has led to an advance in the date of the 1-day minimum flow (43 Julian date), resulting in increased complexity and smoothing of the intra-annual flow regime. The decrease in frequency of extreme low-flow and small flood events is also primarily attributed to the regulation by the TGD. However, the increase in maximum flow conditions, the increase in complexity of the intra-annual flow regime, and the increase in ecodeficit are mainly associated with climate change.

The construction of TGD has altered flow events in terms of magnitude, timing, frequency, and other aspects, leading to changes in hydrological processes, which can pose a threat to riverine biodiversity and ecological structure (Sun et al. 2024). From the hydrological perspective, intra-annual flow processes are vital ecological events in river systems. Increased flow complexity implies higher water fluctuation, which can weaken stratification and facilitate nutrient exchange between different water layers (Wang et al., 2020a, 2020b). In addition, rising water temperatures due to increasing air temperatures lead to elevated nutrient loads in water, promoting changes in algal populations and becoming one of the driving factors behind shifts in water environmental conditions. The decrease in the frequency of extreme low-flow events alleviates survival pressures for macrophytes, algae, and fish (Zeng et al. 2018). However, significant changes in small flood events can reduce the river's self-purification capacity, decrease lateral connectivity between rivers and floodplains, result in fragmented habitats on a large scale, and impact the natural reproduction of drift-spawning fish species (Yi et al. 2010; Zhao et al. 2021). Previous studies have demonstrated that the construction of the TGD has significantly reduced the biodiversity of aquatic organisms in the middle reaches of the Yangtze River, while sediment load has decreased by nearly 90%, leading to a sharp decline in total phosphorus concentration and further affecting the survival of planktonic organisms (Li et al. 2013; Zhou et al. 2013; Wang et al. 2016). In summary, the flow responses triggered by dam construction disrupt the ecological balance of rivers and facilitate new population structure dynamics. Adaptive management policies need to be formulated based on conservation goals to address these impacts.

For fish spawning, the rising process in rivers serves as an important stimulus signal, particularly under suitable water temperature conditions (Shen et al. 2018). Previous studies have indicated that after the regulation of the TGD, the majority of FFMC spawning grounds in the reservoir area disappeared, and FFMC shifted their spawning to the upstream reach above the Cuntan station. This is the main reason for the drastic decline in FFMC abundance in the Yichang section after 2003 (Zhang et al. 2012; Ding et al. 2020). In this study, we found that the altered rising process caused by the TGD regulation, including a decrease in the duration of rising and an increase in the daily average rising rate, disrupted hydrological signals for fish behavior. It is another important factor contributing to the reduction in fish fry abundance. When considering only the impact of the rising process, the TGD may lead to an average decrease of 19.5% in FFMC fry abundance. Moreover, the increased monthly average flow during the spawning period caused by the TGD is unfavorable for the safe drift of fish eggs, thereby affecting the hatching success rate (Hung et al. 2022). However, the ecological operation by TGD has been proven to partially meet the hydrological requirements for fish reproduction, promoting the formation of reproductive peaks, and mitigating the adverse impacts of dam construction on fish resources in the Yangtze River. This study found that, from a hydrological perspective, the rising water condition changes caused by TGD's ecological releases led to an average increase of 53.2% in larval fish abundance.

There are some limitations to this study. While we considered the influence of climate and incoming water changes on flow regimes, we did not separate the effects of incoming water changes caused by the construction of cascade reservoirs on the Jinsha River. Due to the considerable distance from the Yichang station, the influence of these reservoirs weakens gradually due to factors such as the merging of tributaries (You et al. 2023). Consequently, their impact on the Yichang station is not significant. This perspective is supported by the study conducted by Wu et al. (2023).

On the one hand, while this study focused primarily on the impact of climate and incoming water changes on river flow patterns, it did not account for other potential factors such as riverbed morphology and land use that could also affect the flow response at the Yichang station. This is because the Yichang station is located approximately 43 km downstream of the Three Gorges Reservoir. From the TGD to Yichang, there are no other major tributaries converging, meaning that the Yichang station is primarily influenced by upstream water inflow and the TGD. Consequently, the influence of other factors such as riverbed morphology and land use within this stretch of the river is relatively minor (Lai et al. 2023). Since the input data for the LSTM model consisted of upstream water inflow and meteorological data, we interpreted the influence of external factors as minor model errors through a comparative analysis of flow rates between the variant and baseline periods from the LSTM simulation results (Yang et al. 2023). Therefore, the attribution results obtained in this study for the hydrological conditions at the Yichang station can be primarily attributed to the regulation of the TGD, as well as upstream water inflow and climate change impacts. On the other hand, the attribution analysis of flow regime changes based on the LSTM model and model parameters are a significant source of uncertainty. We trained the hyperparameters to minimize model fluctuations while maximizing accuracy, and we used the mean method to process the output results to reduce the impact of uncertainty. Under these constraints, the reconstructed flow results at the Yichang station during the reference period show good fit with the measured values (Figure 4). These results indicate that the model performance can effectively evaluate the impact of dam regulation on river flow regimes and the response of ecological events.

A framework is proposed to quantitatively assess the multiscale impacts of climate and incoming water changes, as well as dam construction, on flow regimes and ecological events in regulated rivers, with the URYR as a case study. The framework utilizes an LSTM model to reconstruct natural flows unaffected by TGD, synthesizing hydrological indicators that characterize inter-annual and intra-annual flow regime changes. It examines the multiscale evolution characteristics of the river and quantitatively assesses the responses of fish abundance events to hydrological process changes.

The study results indicate that after the regulation of the TGD, the flow regime of the river has undergone moderate changes on an inter-annual scale, with a value of 0.60 according to the RVA method, where 16 indicators have experienced moderate or higher levels of change. Furthermore, the homogenization of environmental flow events has resulted in enhanced hydrological stability of the river, where the frequency of low-flow events has increased while the frequency of extreme low-flow and small flood events has significantly decreased, indicating a high degree of change.

From a hydrological perspective, the reduction in flood events lasting for more than 6 days, as well as the excessive daily flood rate during some flood events caused by TGD regulation, played a crucial role in the decline of larval fish abundance. When considering only the impact of flood events, TGD resulted in an average decrease of 19.5% in larval fish abundance. However, the changes in water level conditions caused by TGD's ecological releases led to an average increase of 53.2% in observed larval fish abundance compared to the expected values. Climate and incoming water change was identified as the primary driver behind the decrease in annual flow and monthly flow during the flood season, contributing significantly to the increased complexity of the flow regime and ecodeficit, with contribution rates of 70.6 and 86.3%, respectively. The findings contribute to the understanding of the driving mechanisms of flow regimes and the ecological response characteristics under changing environments, providing valuable insights for water resource management and ecosystem protection.

This study was supported by the National Natural Science Foundation of China (51779094), Science and Technology Program of Guizhou Province (KT202008), and Henan Province Science and Technology Innovation 676 Talent Program (16HASTIT024).

HW and XB planned the research, designed the experiments, and revised the manuscript. HY performed the experiments. LH and XJ analyzed the data. XB and HY wrote the draft. HW and WG checked and improved the manuscript. All authors read, revised, and approved the final manuscript.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.