Non-stationarity of runoff and sediment load and its drivers under climate change and anthropogenic activities in Dongting Lake Basin

Non-stationarity of runoff and sediment load and its drivers under climate change and anthropogenic activities in Dongting Lake Basin | Scientific Reports

Abstract

Analysing non-stationarity in runoff and sediment load is crucial for effective water resource management in the Dongting Lake basin amid climate change and human impacts. Using the Mann-Kendall test, Generalized Additive Models for Location, Scale, and Shape framework, and Random Forest models, we evaluated non-stationarity and its drivers in the annual runoff and annual sediment load series at eight hydrological stations from 1961 to 2021. These stations include three inflow sites at the Jingjiang Three Outlets (Ouchi, Songzi, and Hudu Rivers), four inflow sites in the Four Rivers basin (Xiang, Zi, Yuan, and Li Rivers), and one outflow site at Chenglingji. Results revealed a significant decrease in annual runoff at the Three Outlets and Chenglingji, while the Four Rivers basin showed no significant trend. The non-stationary models with multiple physically-based covariates better captured non-stationarity compared to single covariate models. Annual rainfall was a key contributor to annual runoff in the Four Rivers basin, while reservoir storage capacity played a more dominant role in the Three Outlets. At Chenglingji station, both factors significantly influenced annual runoff. For annual sediment load, reservoir storage capacity emerged as the most critical factor across all regions. These findings provide a basis for improving runoff and sediment regulation in the Dongting Lake basin.

Introduction

Climate change accelerates both global and regional hydrological cycles, exacerbating the uneven spatial and temporal distribution of water resources 1, 2. Furthermore, in the face of worsened water supply-demand imbalances due to climate change and considering the essential functions of reservoirs for human life, the availability of reservoir water becomes a crucial factor for sustainable development 3, 4.

A total of 24,783 dams with a combined storage capacity of 7,384 km3 were identified in the georeferenced global dams and reservoirs dataset 5. Reservoir construction not only influences the hydrological processes within the watershed but also alters the quantity and quality of suspended matter in the river 6, 7. Consequently, the changes in runoff and sediment load due to the impacts of climate change and human activities have received considerable focus and attention 8, 9, 10.

Research on the variability of runoff and sediment span basins worldwide, including the Yangtze River basin 8, Mississippi River basin 11, Peninsular Indian 12, Yellow River basin 13, Himalayan basin 14, Ningxia Reach 15. Most studies employed various methodologies, such as the non-parametric Mann-Kendall (MK) test, Sen’s slope estimator, wavelet analysis, and the Pettitt test, to investigate the trends, periodicity, and change points in runoff and sediment time series 16, 17, 18. Additionally, different methods identify driving forces and their contributions, including the Budyko hypothesis 11, hydrological model 14, Double Mass Curve 19, hydrodynamic model 20. For example, streamflow in the Mississippi River basin increased by 1.68 km3 year− 1, while the Missouri River’s sediment load decreased by 2.80 million tons per year, with precipitation driving streamflow increases (> 50%) and human activities significantly impacting sediment load (> 40%) 11. In Peninsular India, sediment load in 12 major tropical rivers declined by over 40%, largely due to reservoir and dam construction 12. Affected by climate change and human activities, water discharge and sediment load in the Yangtze River basin generally decreased, with significant declines at Dongting Lake and the Min River basin at rates of − 13.93 × 108 m3 year− 1 and − 1.8 × 108 m3 year− 1, respectively 8. In the Yellow River, anthropogenic activities like terracing and reservoirs drove a sediment load decrease of about 90% 13. Similarly, annual water and sediment discharge at the Ningxia reach of the Yellow River were reduced by at least 16.16% and 51.97%, respectively 15.

These investigations highlight the growing non-stationarity in hydrological processes across various regions, driven by climate change and human activities. Therefore, the stationarity hypothesis has been challenged for reliable water resources risk assessment and planning 21. To address this, researchers have increasingly turned to the flexible Generalised Additive Models for Location, Scale and Shape (GAMLSS) framework proposed by Rigby and Stasinopoulos 22. GAMLSS framework allows for a broader set of distribution functions, the incorporation of more physically meaningful explanatory covariates, and more robust results in determining non-stationarity by comparing stationary and non-stationary models 23, 24, 25. For instance, Xiong, et al. 26 found that rising temperatures and decreasing precipitation contributed the non-stationarity of the Weihe River’s annual runoff series within the GAMLSS framework. Mo, et al. 27 showed that the Log-normal distribution model, using annual precipitation and average temperature as covariates, best explained annual runoff variation in the Xiaoli River Basin. Furthermore, the sediment transport model accurately described nonstationary changes in sediment load in the Jinsha River basin when GAMLSS included precipitation as a covariate 28.

In recent years, the robustness of the machine learning algorithms in capturing hydrological variability has made them essential tools for predicting atmospheric and hydrometeorological variables 29, 30, 31. Applications of machine learning algorithms, such as support vector regression, support vector machine, artificial neural networks, Random Forest (RF), relevance vector machine and long short-term memory algorithms, have become prominent in predicting and analysing water and sediment changes 32, 33, 34, 35, 36. Kumar, et al. 32 showed that the support vector regression, support vector machine, and RF models outperformed a conceptual hydrological model in simulating monthly discharge across 54 peninsular catchments of India, with RF model being the most effective. Sharafati, et al. 35 integrated gradient boost regression, AdaBoost regression, and RF regression to predict daily discharge and suspended sediment concentration in the Mississippi River, finding RF model to be slightly superior. Miao, et al. 34 used six machine learning algorithms, including support vector regression and artificial neural networks, to model annual runoff and sediment discharge in the Yellow River Basin, with all models better at simulating sediment discharge than runoff. In RF regression models, variable importance was assessed by the percentage increase in mean square error (IncMSE) when each explanatory variable was permuted individually while keeping the others constant 37, 38, 39. A larger increase in mean square error indicates a more important variable, making it widely used in attributing runoff and sediment changes 37. For example, Guo, et al. 40 used RF models to evaluate the relative importance of inflow discharge, sediment concentration, and operating water levels in determining wash-load concentrations along the Jingjiang reach. Yao, et al. 41 revealed that in the Wuding River Basin of the Yellow River, the normalized difference vegetation index significantly influenced both runoff and sediment transport, with IncMSE values of 14.63% and 15.93%, respectively. Han, et al. 42 found that crop sown area and vegetation coverage contributed 20.88% and 21.95% to annual runoff and sediment transport in the Ru River Basin from 1989 to 2019.

Runoff and sediment load entering Dongting Lake originate from the Jingjiang Three Outlets (Ouchi, Taiping, and Hudu Rivers) and the Four Rivers basin (Xiang, Zi, Yuan, and Li Rivers) ultimately flowing into the Yangtze River through Chenglingji station 43(Table 1; Fig. 44, 45, 46, 47. These studies revealed that human interventions, such as the Jingjiang River diversion and the Three Gorges Project, have significantly accelerated the decline in runoff and sediment load at the Three Outlets, reducing sediment deposition in Dongting Lake 9, 20, 48. In the Four Rivers basin, sediment reduction was evident, but no significant variation in annual runoff was observed 49, 50. Additionally, Gao, et al. 51 found that the flood extremes in the Dongting Lake basin exhibited notable variations and non-stationarity, but the contribution of physical explanatory variables has not been quantified. These studies provide a foundation for investigating the non-stationary characteristics and underlying causes of changes in the Dongting Lake basin amid a changing environment. However, current analyses lack explanatory covariates with physical significance, requiring further exploration of factors indicative of climate change and human activities. Furthermore, existing qualitative or linear trend analyses are inadequate to capture the contributions of climate change and anthropogenic activities to the non-stationarity in the Dongting Lake basin. To address this gap, flexible models that handle nonlinearity and collinearity, such as GAMLSS and RF models, should be incorporated into the attribution analysis of runoff and sediment load non-stationarity in the Dongting Lake basin.

Location of the Yangtze River basin, Dongting Lake basin, hydrological stations, and reservoirs.

In this study, we employed the annual runoff series (ARS) and annual sediment load series (ASLS) from 1961 to 2021 at eight hydrological stations in the Dongting Lake basin. Our objective was to quantitatively differentiate the relative importance of covariates contributing to non-stationarity in these series under the impacts of climatic variations and human-induced activities. To achieve this, we established stationary and non-stationary models based on GAMLSS framework, and RF models, incorporating nine physically-based covariates that capture human and climatic influences in the Dongting Lake basin (Fig. 2). The study was divided into three parts: (1) Application of nine statistical methods and Hurst exponent for identifying temporal trends, change points, periodic patterns, stationarity and persistence. (2) Development and validation of non-stationary models incorporating single and multiple covariates, alongside RF models. (3) Investigation and evaluation of the relative impacts of fundamental factors causing non-stationarity. This research aims to provide a decision-making basis for the effective management of water and sediment resources amid a changing environment in the Dongting Lake basin.

Methodological framework for identifying the non-stationarity for annual runoff and annual sediment load series in the Dongting Lake basin.

Results

Comprehensive change analysis

Interannual variations and change points The linear trend analysis revealed that the ARS and ASLS recorded by the Three Outlets stations predominantly exhibited downward trends over the last six decades (see Supplementary Fig. S1). The ASLS in the Four Rivers basin and Chenglingji station displayed a marked downward trend, while the ARS demonstrated no significant change. The total average annual runoff for the Three Outlets was less than that in the Four Rivers basin, at 753.98 × 108 m3 and 1,683.96 × 108 m3, respectively. Nevertheless, the total average annual sediment load in the Three Outlets amounted to 8200.4 × 104 t, markedly surpassing the 2177.69 × 104 t recorded in the Four Rivers basin. Moreover, the coefficient of variation for the ASLS at the seven lake-inlet hydrological stations were significantly higher than that of the ARS, with the Three Outlets having notably higher ARS variation than the Four Rivers basin (see Supplementary Fig. S2).

Comparing the results of six trend analysis methods, we found consistency in trend direction of the ARS and ASLS at eight stations, confirming the robustness of the trend detection results (see Supplementary Table S1 and Fig. 3). For example, the ARS at Shimen station exhibited no significant downward trend, while the other three stations in the Four Rivers basin showed no significant upward trend. However, the Three Outlets and Chenglingji stations exhibited significant downward trend. At Chenglingji and Guanjiapu stations, discrepancies arose between the MK method and its improved versions (Pre-Whitening MK (PW-MK) and Trend-Free Pre-Whitening MK test (TFPW-MK)). Specifically, the PW-MK method, by removing autocorrelation, may weaken the originally significant trend, rendering it non-significant (P > 0.05). In contrast, the TFPW-MK method, which removes the trend before addressing autocorrelation, still indicates a significant negative trend (P < 0.05). This indicated that the TFPW-MK method provides more reliable trend analysis results, particularly at stations with strong autocorrelation, such as Chenglingji and Guanjiapu, where significant declining trends in ARS and ASLS were observed. The Pettitt test results revealed that the ARS at Three Outlets and Chenglingji stations experienced changes in the 1980s and 1990s, while the ASLS at the seven lake-inlet stations showed mutations in the 1990s (see Supplementary Table S2). However, the ARS in the Four Rivers basin exhibited no obvious changing point.

Spatial distribution of trend (a), change point (b), and stationarity (c) of the annual runoff series, and spatial distribution of trend (d), change point (e), and stationarity (f) of the annual sediment load series at the eight stations in the Dongting Lake basin.

Periodicities and stationarity detection Using wavelet analysis, contour maps were created to display the real parts of the wavelet coefficients for the ARS and ASLS at the eight stations (see Supplementary Fig. S3). By integrating the analysis of wavelet variance plots from eight stations, the highest peak in the image is considered to represent the primary periodicity of the ARS and ASLS variations, while smaller wavelet peaks indicate secondary periodicities. The analysis revealed distinct multiple-timescale periodicities in both ARS and ASLS. Contour maps for the ARS wavelet coefficients at each station identified the most pronounced oscillation period between 27 and 38 years, while ASLS wavelet coefficients indicated predominant oscillations within an 11 to 39-year period, with the majority of stations exhibiting oscillations under 30 years (see Supplementary Table S3). The multi-periodicities of the ARS and ASLS were identified to be less than 18 and 14 years, respectively. The stationarity of the ARS and ASLS was tested using the Augmented Dickey–Fuller (ADF) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests, as shown in Supplementary Table S4 and Fig. 3. The ARS at the Xinjiangkou and Mituosi stations exhibited non-stationary. In contrast, the ARS at the Taojiang, Guanjiapu, and Chenglingji stations were trend stationary, difference stationary, and difference stationary respectively, while the other stations showed stationarity. The ASLS at the Taojiang and Guanjiapu stations exhibited difference stationary, whereas the ASLS at the remaining stations were non-stationary.

Non-stationarity analysis

Non-stationary models with single covariates Based on six distribution functions and ten explanatory variables, the stationarity and non-stationarity of the ARS and ASLS across eight stations were examined using eleven models (Table 2). Without covariates, five distributions with the lowest Generalized Akaike Information Criterion (GAIC) values were chosen for each station. The Gamma, Normal, Inverse Gaussian, Log-normal, and Weibull distributions best fitted the ARS at eight stations, while Weibull and Gamma were most suitable for the ASLS (see Supplementary Fig. S4 and Table S5). Furthermore, we concurrently modelled the chosen distribution parameters as functions of the covariates, instead of initially modelling each parameter independently with the others held constant. In this procedure, the stationary model (Model 0) for each station serves as the benchmark for choosing the covariates, and the worm plots of the best non-stationary model with single covariate for the ARS and ASLS at the eight stations are depicted in Supplementary Fig. S4. Although the Model R with Reservoir Storage Capacity (RSC) of the ARS at the Xiangtan and Taojiang stations had the lowest GAIC values, the worm plot of Model P with Annual Rainfall (AR) showed greater performance than Model R (see Supplementary Fig. S5 and Table S6). Thus, the Model P were deemed as the best non-stationary models of the ARS at the Four Rivers basin and Chenglingji stations. The Model t with time of the ARS performed better than other covariates at the Mituosi and Guanjiapu stations. However, the Model R of the ARS was the best non-stationary models with single covariate at the Xinjiangkou station. The eight Model R of the ASLS consistently demonstrated optimal fitting performance, whereas Model t had suboptimal performance compared to other non-stationary models (see Supplementary Fig. S5 and Table S7).

Non-stationary models with multiple covariates The following two strategies and stepwise model selection were used to explore the best combinations of the covariates for the non-stationarity: Strategy A: each distribution parameter varies independently; Strategy B: all distribution parameters vary concurrently with identical covariates. The results of the stepwise model selection under strategies A and B for the ARS and ASLS at the eight stations are presented in Tables 3 and 4, and the worm plots of the best-fitting non-stationary models with multiple covariates are shown in Supplementary Fig. S6. Specifically, the ARS of Shimen and Guanjiapu stations, as well as the ASLS of Xiangtan, Taoyuan, and Chenglingji stations, exhibited improved performance under strategy B. On the other hand, the ARS and ASLS of the remaining stations demonstrated better performance under strategy A. Meanwhile, we observed that the best-fitting non-stationary model with multiple covariates arose from a combination of frequently occurring covariates. For example, the covariates AR, RSC, and time appeared in six out of the seven models for the ARS. The covariates RSC appeared in six out of the seven models for the ASLS, and the covariates time, AR, and Annual Mean Temperature (TEMP) also appeared in three out of the seven models.

To obtain the improvement of the model fitting quality, we subtracted the GAIC values of the best non-stationary model with single covariate from the best non-stationary model with multiple covariates (see Supplementary Fig. S7). Except for the ARS at the Chenglingji station, the best non-stationary model with multiple covariates at the other seven stations showed better performance. The reduction in GAIC values in the Three Outlets were more pronounced than in the Four Rivers basin. However, there was no change in GAIC values for the ASLS at the Xiangtan, Taoyuan, and Chenglingji stations. This indicated that the ARS exhibited greater sensitivity to the combined impacts of climate change and human activities than the ASLS. These results demonstrated that the non-stationary models with multiple physically-based covariates can better reflect the non-stationarity of the ARS and ASLS at most hydrological stations.

Contribution analysis

To ensure the robustness of the modelling results, we first calculated the correlation coefficients between each pair of covariates before constructing the RF regression model and identified those with correlation coefficients greater than 0.8. The analysis revealed that covariates paired with El Niño 3.4 (Niño 3.4) exhibited correlation coefficients ranging from − 0.51 to -0.84. Consequently, Niño 3.4 was excluded from the subsequent modelling process. Therefore, the RF models were established by considering the eight physically-based covariates that influence the non-stationarity of the ARS and ASLS as independent covariates. The ARS and ASLS were used as the respective dependent variables. To evaluate the stochastic importance of the eight covariates, we executed 100 RF models using data from the ARS and ASLS at all stations (see Supplementary Fig. S8). The evaluation of RF models was conducted to obtain the R2 and root-mean-square error values for the ARS and ASLS at eight stations. The average R2 values for the ARS and ASLS at all stations were 0.82 and 0.84 (Fig. 4), respectively. The corresponding average root-mean-square error values were 82.66 and 540.48, respectively. By plotting the average IncMSE of the eight physically-based covariates from the 100 RF models and calculating the Spearman correlation coefficients, we can explore the covariates’ importance and their correlation with the ARS and ASLS (see Fig.

Spearman correlation coefficients between annual runoff series/annual sediment load series and the eight covariates; covariate importance, and R2 values from Random Forest models for annual runoff series (a) and annual sediment load series (b) at eight stations in the Dongting Lake basin. (Covariate importance and R2 values are averaged from 100 Random Forest model runs.)

In terms of the ARS, Chenglingji station received contributions of 26.77% from AR and 22.58% from RSC. The four stations at the Four Rivers basin accrued contributions from AR ranging from 20.06% (Taojiang) to 32.07% (Shimen). In contrast to the significant contribution of AR to the ARS in the Four Rivers basin and Chenglingji stations, AR made a smaller contribution to the ARS in the Three Outlets, accounting for only 1.27–4.32%. The Xinjiangkou, Guanjiapu, and Mituosi stations received 30.75%, 33.77%, and 39.41% from RSC, respectively. The Mituosi and Xinjiangkou stations also received 10.46% and 8.3% from TEMP, while Guanjiapu received 8.55% from Annual Mean Wind Speed (WS). Except for the Taojiang station, which received 12.02% from North Atlantic Oscillation (NAO), the other three station of the Four Rivers basin received ranging from 14.58 to 24.64% from RSC. Additionally, Guanjiapu and Xinjiangkou received 8.55% and 8.3% from WS and TEMP, respectively. RSC and AR exhibited negative and positive correlations with the ARS at all stations, respectively. However, WS, TEMP and other covariates had no significant contribution to the Four Rivers basin. The ASLS at all stations received the most inputs from RSC among the eight covariates, ranging from 16.87% (Taojiang) to 43.97% (Mituosi). Notably, the average contribution values (40.82%) in the Three Outlets were higher than in the Four Rivers basin (24.95%). Additionally, the TEMP contributed between 9.3% and 10.89% to the Mituosi, Xinjiangkou, Chenglingji, and Taoyuan stations, while WS accounted for 9.9%, 8.64%, and 7.77% to Chenglingji, Taoyuan and Guanjiapu stations, respectively. The RSC and TEMP exhibited negative correlations with the ASLS at all stations, while the AR and WS exhibited positive correlations.

Discussion and conclusion

The non-stationary models and contribution analysis indicated that RSC was the primary drivers in the Three Outlets, with WS and TEMP contributing more significantly than the circulation indices and AR. To examine the correlation between meteorological factors and the ARS in the Three Outlets, we further utilized the six trend test methods to analyse the temporal trends of AR, TEMP, and WS in the Three Outlets over the past 60 years. The trend test revealed that AR showed no significant increase, while TEMP and WS showed significant increasing and decreasing trends, respectively (see Supplementary Table S8). Additionally, the impact of rainfall on erosion is influenced by the intensity and duration of rainfall in the catchment area. However, most sub-basins in the upper reaches of the Yangtze River basin are located in regions with infrequent heavy rainfall 52. This lack of significant increase in AR and the rarity of heavy rainfall may explain the weaker correlation between AR and ASLS. The key anthropogenic activities in the Three Outlets and its upstream areas include: (1) Artificial and natural cutoffs in the lower Jingjiang, Zhongzhouzi, Shangchewan, and Shatanzai during 1967-197253. (2) Increased reservoir construction, such as Gezhouba, Three Gorges, and Xiangjiaba. (3) Long-term soil and water conservation measures. (4) Historical active sand mining. The Cutoff Project in the lower Jingjiang reach has led a rapid decrease of runoff and sediment at the Three Outlets 9, 20. Since 2003, there has been a significant increase in upstream reservoir regulation, enhancing runoff and sediment retention (see Supplementary Fig. S9). Over the past 70 years, soil erosion in the Yangtze River basin has decreased from 622,000 to 346,700 km², a reduction of 44% 54. Finally, unreported sand mining activities in the upper reaches of the Yangtze River may have led to an underestimation of the impact of sand mining on sediment reduction 55. Hence, human activities have predominantly driven the declining trend of annual water and sediment load in the Three Outlets, with the RSC playing a significant role.

Apart from a non-significant decrease of the ARS in the Li River basin, the other three basins exhibited a non-significant increase. AR and RSC were identified as the two primary driving forces. Trend tests’ results at the Four Rivers basin revealed a non-significant increase in AR, a significant downward trend in WS, and a significant increase in TEMP (see Supplementary Table S9). Given the low correlation and contribution to the ARS in the Four Rivers basin, the impact of WS and TEMP on the ARS at the Four Rivers basin were not significant. Furthermore, the increase in RSC and the improvement in water use efficiency likely contributed to the non-significant increase of the ARS in the Four Rivers basin. From 1951 to 2014, the Dongting Lake basin exhibited a tendency of less flooding during the flood season and more drying during the dry season 48. This indicates that reservoir regulation capacity needs further improvement to augment dry-season flow and balance the interannual runoff differences. Water usage analysis from Hunan province over the past two decades showed average total water usage from 2001 to 2021 was 32.5 billion m3 (see Supplementary Fig. S10). Agricultural water usage decreased from 70.3 to 62%, and water usage per unit of GDP dropped from 802 m3 to 70 m3. The average annual water usage accounted for 15.4% of annual runoff. Overall, total water demand in the Dongting Lake basin remained stable, with a decrease in agricultural water usage and improved efficiency. The key factors influencing ASLS in the Four Rivers basin include changes in surface conditions such as reservoir construction, land use/cover, and reduced soil erosion due to conservation projects. Since 1998, ecological initiatives in the Dongting Lake basin have focused on returning farmland to lakes, reforesting cultivated land, and alleviating soil erosion from deforestation and steep-slope cultivation in the upper Four Rivers basin, maintaining a forest land ratio of around 61% 56. Comprehensive soil and water conservation efforts, such as managing sloping farmland and constructing high-standard farmland in western Hunan, have reduced soil erosion over the past 40 years 57. The RSC and major reservoir completion years in the Four Rivers basin from 1961 to 2021 are shown in Supplementary Fig. S9 and Table S1. Although the highest RSC in the Four Rivers basin reached 0.35, they are still much lower than in the Three Outlets. Therefore, the retention of sediment load by reservoirs, the restoration and maintenance of vegetation cover, and the decrease in soil erosion collectively contribute to a further reduction in sediment entering the rivers of the Dongting Lake basin.

The correlations of the ARS between the seven lake-inlet stations and the Chenglingji station were strong, ranging from 0.52 (Xiangtan) to 0.74 (Taoyuan) (see Supplementary Fig. S11). In the Three Outlets and Four Rivers basin, the correlations between adjacent stations were strong. Meanwhile, Shimen station demonstrated a positive correlation with the Three Outlets, but the correlations between Shimen and Xiangtan were weaker. Similarly, Xiangtan and Taojiang exhibited a nonsignificant correlation with the Three Outlets, respectively. On the other hand, the correlations of the ASLS among the seven lake-inlet stations and the Chenglingji station were also strong, ranging from 0.48 (Shimen) to 0.81 (Guanjiapu). The Three Outlets generally exhibited higher correlation coefficients compared to the Four Rivers basin. The correlations of the ASLS among the seven lake-inlet stations were relative stronger than that of the ARS, and were less affected by distance. Therefore, we calculated the correlation coefficients between the ARS/ASLS of the seven lake-inlet stations and the ASLS/ARS of Chenglingji station (see Supplementary Table S10). The results showed highly significant correlations between the ARS in the Three Outlets and the ASLS at Chenglingji, while correlations between the Four Rivers basin and Chenglingji were much lower. The ASLS of the seven lake-inlet stations also showed strong correlations with the ARS at Chenglingji. To further analyse the contributions of water and sediment inflow from the Three Outlets and Four Rivers Basin to Chenglingji station over the past 60 years, the inflow-to-outflow ratio of ARS and ASLS in the Three Outlets and Four Rivers basin were depicted in Fig. 5. On average, the inflow-to-outflow ratio of the ARS and ASLS at seven lake-inlet stations accounted for 88.81% and 338.31%, respectively. The contribution from the Three Outlets of the ARS exceeded 8.98% and 5.88% that of the Four Rivers basin only in 1963 and 1965, respectively. Furthermore, the annual sediment load in the Four Rivers basin exceeded that of the Three Outlets by 3.04% in 1994. In 2006, the annual sediment load in the Four Rivers basin surpassed that of the Three Outlets by 56.87%. There were 9 non-consecutive years during which the inflow-to-outflow ratio of the Three Outlets consistently remained lower than that of the Four Rivers basin from 2006 to 2021. We obtained annual net outflow of the ARS and ASLS by subtracting the corresponding the total outflow at Chenglingji station from the total inflows at seven lake-inlet stations (Fig. 58, contributing to a rise in annual net runoff outflow at Chenglingji station.

Inflow-to-outflow ratio of the Three Outlets and Four Rivers basin, and the net outflow discharge of the annual runoff series (a) and annual sediment load series (b) at the eight stations in the Dongting Lake basin.

To understand whether future variations in ARS and ASLS will persist or stabilize, we calculated the Hurst exponent at eight stations (see Supplementary Table S11). Except for the persistent trend in the ARS at Guanjiapu, other stations exhibited anti-persistence. Integrating the Hurst exponent with trend analysis, the increasing ARS trends in the Xiang, Zi, and Yuan River basins are identified as anti-persistent, suggesting potential future decreases or stabilization. The ARS in the Li River basin may either increase or stabilize, while the decreasing ARS trends in the Three Outlets are anti-persistent. Additionally, the significant decreasing trends of ASLS at all stations are anti-persistent. Three potential drivers for the anti-persistence of ARS and ASLS non-stationarity in the Dongting Lake basin include: (1) The expected increase in annual precipitation and runoff depth in the Yangtze River basin 59, combined with the rising trend of rainstorms in the Dongting Lake basin 52, may increase uncertainty in future runoff predictions. (2) Long-term sediment accumulation in reservoirs reduces their flood control capacity and economic benefits, affecting navigation 60. Maintenance dredging in the Four Rivers basin can help mitigate the ongoing decline in sediment transport in the future. (3) The modern water network construction plan of Hunan province (2021–2035) proposes inter-basin water diversion projects, such as the Yuan to Zi River water replenishment diversion plan 61, which may impact water circulation between basins. In addition, expanding flood control capacity for reservoir like Wuqiangxi will also increase RSC values and sediment retention in the Yuan River basin in the future.

In this paper, we conducted a comprehensive analysis of the variation trends in ARS and ASLS at eight stations across the Three Outlets, Four Rivers basin, and Chenglingji station. Using the GAMLSS framework with single and multiple covariates, as well as RF regression models, we identified non-stationarity and drivers of water and sediment load amid climate change and human activities in Dongting Lake basin, leading to the following conclusions: Firstly, ASLS showed a significant decrease at all stations, while ARS decreased notably at the Three Outlets and Chenglingji, with no significant variation in the Four Rivers basin. Mutation points in the 1980s and 1990s were identified for both ARS and ASLS, showing multiple periodicities. Secondly, the GAMLSS framework and RF models demonstrated good applicability and consistency in detecting the drivers. Specifically, the non-stationary models incorporating RSC, AR, or time as covariates for ARS exhibited optimal performance, while RSC-based non-stationary models were most effective for ASLS. Similarly, RF models also identified RSC and AR as the dominant covariate for ARS and ASLS at all stations. Additionally, multivariate non-stationary models outperformed univariate ones at most stations, showing enhanced sensitivity and accuracy for ARS over ASLS. Variations in TEMP and WS have begun contributing to ARS and ASLS non-stationarity in the Dongting Lake basin. A significant shift from sediment deposition to erosion in the lake area occurred between 2006 and 2021. Finally, Hurst exponent analysis revealed unsustainable trends for ARS and ASLS at seven stations, except for ARS at Guanjiapu. Although GAMLSS and RF models effectively detected non-stationarity, their limitations underscore the need for models that simulate complex basin changes. Future research should integrate distributed hydrological models and more advanced machine learning algorithms to predict runoff and sediment load at multiple scales in a changing environment.

Methods

Study area

Dongting Lake is situated on the southern bank of the middle and lower reaches of the Yangtze River in China (28°30’-30°20’N, 110°40’-113°10’E). Covering a total basin area of 26.28 × 104 km2, the lake is one of the throughput-type lakes that maintains a natural connection with the Yangtze River. The Dongting Lake basin is surrounded by mountains on the east, south, and west sides, while the northern part consists of the Dongting Lake Plain 62. The terrain slopes gradually from the south towards the central and northeastern parts, forming a horseshoe-shaped basin that opens to the north 56. Hunan province spans 21.18 × 104 km2, comprising 80.61% of the entire Dongting Lake basin area. The Dongting Lake basin exhibits distinct seasonal hydrological characteristics and falls within the typical subtropical monsoon climate zone, with an annual precipitation of 1,200-1,450 mm. Based on the average ARS data from 1961 to 2021, the lake’s inflow consists of three main components: 27.47% was derived through water discharge from the northern Three Outlets, sourced from the Yangtze River; 61.34% from the inflow of the south Four Rivers basin, and 11.19% from various small to medium-sized rivers in the lake area, including Xinqiang River and Miluo River. Simultaneously, inflows from the Three Outlets and Four Rivers basin are regulated through West, South, and East Dongting Lake before flowing into the Yangtze River via Chenglingji station.

Data sources and explanatory covariates

The study focused on ARS and ASLS at seven lake-inlet stations in the Three Outlets and Four Rivers basin, and one lake-outlet station spanning from 1961 to 2021 within the Dongting Lake basin. Xinjiangkou station incorporated data from the Songzi River’s Shadaoguan station, and Guanjiaopu station included data from the Ouchi River’s Kangjiagang station. The data resources of the selected nine independent physical covariates, encompassing NAO, Pacific Decadal Oscillation (PDO), Southern Oscillation Index (SOI), Niño 3.4, Arctic Oscillation (AO), AR, TEMP, WS, and RSC were listed in Table 2. Furthermore, we selected 43 reservoirs constructed in the Four Rivers basin and upper reaches of the Yangtze River basin mainstream to characterize the impact of anthropogenic activities (Supplement Table S1). To select covariates more closely related to ARS and ASLS from climate change and human activities, we conducted the preprocessing on the selected covariates as shown in Supplementary Materials. For instance, the annual RSC value was calculated by dividing the total storage of all constructed reservoirs in the upstream of each hydrological station by the corresponding annual runoff for each year.

Trend, change point, periodicity and stationarity analysis

The time trends in ARS and ASLS were estimated using six methods: Kendall rank correlation coefficient test 63, Spearman rank correlation coefficient test 64, 65, MK test 66, 67, PW-MK test 68, TFPW-MK test 69, and Sen’s slope estimator 70. Additionally, the Pettitt method 71 and wavelet analysis 72 were employed to identify change points and reveal multiperiod changes. Trend analysis provides information on the overall direction of a time series, while stationarity tests examine how the mean and variance vary over time 73. Methods commonly used for stationarity tests include the ADF test 74 and KPSS test 75, which were used in our study. The differing results from the ADF and KPSS tests indicate four potential outcomes: trend stationary, difference stationary, non-stationary, or stationary. Detailed criteria for these determinations and more detailed description of the aforementioned are available in the Supplementary Materials.

GAMLSS framework

The GAMLSS framework was developed by Rigby and Stasinopoulos 22 that can evaluate the stationarity and non-stationarity of hydrological series. The GAMLSS assumes that independent observations \(\:{y}_{i}\) for \(\:i\) = 1, 2, 3, …, n, have the probability distribution function \(\:f\left({y}_{i}\right|{\theta\:}^{i})\), where \(\:{\theta\:}^{i}=({\theta\:}_{1}^{i},{\theta\:}_{2}^{i},.,{\theta\:}_{p}^{i})\) represents the p distribution parameters (location, scale, and shape), and p is usually less than 4. In GAMLSS, a maximum of four parameters is often sufficient to meet the requirements of hydrological modelling, and the distribution parameters correspond to the design matrix of the chosen covariate via a monotonic linking function \(\:{g}_{k}(\cdot\:)\). The following equation offers the semi-parametric additive model framework:

where \(\:{\theta\:}_{k}\) are vectors of length \(\:n\), \(\:{\beta\:}_{k}\) is a parameter vector of length \(\:m\), \(\:{X}_{k}\) is an \(\:n\times\:m\) matrix of explanatory covariates, and \(\:{h}_{jk}(\cdot\:)\) denotes the functional dependence of the distribution parameters on explanatory covariates \(\:{x}_{jk}\). This relationship may be linear or non-linear through smoothing elements, and the cubic spline functions and linear functions were employed to construct the non-stationary models. The Log-normal, Normal, Gamma, Inverse Gaussian, Reverse Gumbel, and Weibull distributions were selected for modelling the ARS and ASLS (Supplement Table S2). Non-stationary models with a single covariate were selected based on the lowest GAIC value at k = 3, supported by the Filliben correlation coefficient and worm plots for model evaluation. Furthermore, the best non-stationary models with multiple covariates were chosen using strategies A and B of the stepGAIC function in the R package gamlss. The detailed methodology and code are provided in the Supplementary Materials.

RF model

The RF model, a non-parametric, multivariate regression algorithm based on decision trees, is highly effective in modelling complex nonlinear relationships, managing multicollinearity, and reducing overfitting 37, 76, 77, 78, 79. Its advantages over multiple regression have led to its widespread use in attributing and predicting water and sediment load changes 30, 42. We employed the RF models to assess the relative importance of covariates affecting ARS and ASLS variability in the Dongting Lake basin under changing environmental conditions. Using data from eight stations, we assessed covariate significance via the percentage of IncMSE, with higher values indicating greater importance. Additionally, 100 RF models were developed using data from all stations to assess the sensitivity and uncertainties of covariates’ importance. A more detailed description of model construction and performance assessment can be found in the Supplementary Materials.

Hurst exponent

The rescaled range (R/S) analysis, proposed by the British hydrologist Harold Edwin Hurst 80 which is commonly used for analysing long-term records of hydrological series. Considering the mature implications of the R/S analysis method on the long-term memory properties of the runoff series 81, 82, we employed the Hurst exponent to determine the persistence and non-persistence characteristics of the annual runoff and sediment load series. The detailed introduction of R/S analysis is given in Supplementary Materials.

Data availability

The annual runoff series and annual sediment load series in the Dongting Lake basin from 1961 to 2015 are available from the Hydrological and Water Resources Survey Bureau of Hunan Province. Restrictions apply to the availability of these data, as they were used under license for the current study and are not publicly available. Data are available from the authors upon reasonable request and with permission of the Hydrological and Water Resources Survey Bureau of Hunan Province. Furthermore, the annual runoff and annual sediment load in the Dongting Lake basin from 2016 to 2021, sourced from the Yangtze River Sediment Bulletin (http://www.cjw.gov.cn/zwzc/zjgb/cjnsgb/).

1.
Zhai, R. et al. Larger drought and flood hazards and adverse impacts on population and economic productivity under 2.0 than 1.5°C warming. Earth’s Future 8 (2020).
2.
Kanwar, N., Kuniyal, J. C., Rautela, K. S., Singh, L. & Pandey, D. C. Longitudinal assessment of extreme climate events in Kinnaur district, Himachal Pradesh, north-western Himalaya, India. Environ. Monit. Assess. 196, 557 (2024).
3.
Li, Y., Zhao, G., Allen, G. H. & Gao, H. Diminishing storage returns of reservoir construction. Nat. Commun. 14 (2023).
4.
Di Baldassarre, G. et al. Water shortages worsened by reservoir effects. Nat. Sustain. 1, 617–622 (2018).
5.
Wang, J. et al. GeoDAR: Georeferenced global dams and reservoirs dataset for bridging attributes and geolocations. Earth Syst. Sci. Data. 14, 1869–1899 (2022).
6.
Dong, N. et al. Model estimates of China’s terrestrial water storage variation due to reservoir operation. Water Resour. Res. 58 (2022).
7.
Li, B., Liang, Z., Bao, Z., Wang, J. & Hu, Y. Changes in streamflow and sediment for a planned large reservoir in the middle Yellow River. Land. Degrad. Dev. 30, 878–893 (2019).
8.
Zhao, Y. et al. Assessing natural and anthropogenic influences on water discharge and sediment load in the Yangtze River, China. Sci. Total Environ. 607–608, 920–932 (2017).
9.
Guo, L., Su, N., Zhu, C. & He, Q. How have the river discharges and sediment loads changed in the Changjiang River basin downstream of the Three Gorges Dam? J. Hydrol. 560, 259–274 (2018).
10.
Rautela, K. S., Kuniyal, J. C., Alam, M. A., Bhoj, A. S. & Kanwar, N. Assessment of daily streamflow, sediment fluxes, and erosion rate of a pro-glacial Stream Basin, Central Himalaya, Uttarakhand. Water Air Soil. Pollut 233 (2022).
11.
Yin, S. et al. Long-term trends of streamflow, sediment load and nutrient fluxes from the Mississippi River Basin: impacts of climate change and human activities. J. Hydrol. 616 (2023).
12.
Das, S. Dynamics of Streamflow and sediment load in Peninsular Indian rivers (1965–2015). Sci. Total Environ. 799 (2021).
13.
Wang, S. et al. Reduced sediment transport in the Yellow River due to anthropogenic changes. Nat. Geosci. 9, 38–41 (2015).
14.
Rautela, K. S., Kuniyal, J. C., Goyal, M. K., Kanwar, N. & Bhoj, A. S. Assessment and modelling of hydro-sedimentological flows of the eastern river Dhauliganga, north-western Himalaya, India. Nat. Hazard. 120, 5385–5409 (2024).
15.
Miao, J. et al. Evolution patterns and spatial sources of water and sediment discharge over the last 70 years in the Yellow River, China: A case study in the Ningxia Reach. Sci. Total Environ. 838, 155952 (2022).
16.
Yuan, Y. et al. Quantitative assessment of the contribution of climate variability and human activity to streamflow alteration in Dongting Lake, China. Hydrol. Processes. 30, 1929–1939 (2016).
17.
Pérez Ciria, T. & Chiogna, G. Intra-catchment comparison and classification of long-term streamflow variability in the Alps using wavelet analysis. J. Hydrol. 587 (2020).
18.
Kanwar, N., Singh, J. C. K. K. S. R. L. & Pandey, D. C. Longitudinal assessment of extreme climate events in Kinnaur district, Himachal Pradesh, north-western Himalaya, India. Environ. Monit. Assess. 196, 557 (2024).
19.
Gao, P. et al. Use of double mass curves in hydrologic benefit evaluations. Hydrol. Processes. 31, 4639–4646 (2017).
20.
Zhang, J. et al. Impact of the Three Gorges Reservoir on the hydrologic regime of the river-lake system in the middle Yangtze River. J. Clean. Prod. 258 (2020).
21.
Milly, P. C. D. et al. On critiques of stationarity is dead: whither water management? Water Resour. Res. 51, 7785–7789 (2015).
22.
Rigby, R. A. & Stasinopoulos, D. M. Generalized additive models for location, scale and shape. Appl. Stat. 54, 507–554 (2005).
23.
Chen, M., Papadikis, K. & Jun, C. An investigation on the non-stationarity of flood frequency across the UK. J. Hydrol. 597, 126309 (2021).
24.
Villarini, G., Serinaldi, F., Smith, J. A. & Krajewski, W. F. On the stationarity of annual flood peaks in the continental United States during the 20th century. Water Resour. Res. 45, W08417 (2009).
25.
Zhang, Q., Gu, X., Singh, V. P., Xiao, M. & Xu, C. Stationarity of annual flood peaks during 1951–2010 in the Pearl River basin, China. J. Hydrol. 519, 3263–3274 (2014).
26.
Xiong, L., Jiang, C. & Du, T. Statistical attribution analysis of the nonstationarity of the annual runoff series of the Weihe River. Water Sci. Technol. 70, 939–946 (2014).
27.
Mo, S., Li, C., Xing, H. & Jiang, K. GAMLSS model-based analysis on annual runoff in Xiaoli River Basin. J. Basic. Sci. Eng. 30, 40–49 (2022).
28.
Jin, H., Zhong, R., Liu, M., Ye, C. & Chen, X. Nonstationary analysis of water and sediment in the Jinsha River Basin based on GAMLSS model. Stoch. Environ. Res. Risk Assess. (2023).
29.
Moghaddam Nia, A. et al. Runoff and sediment yield processes in a tropical eastern Indian River Basin: a multiple machine learning approach. Land 12 (2023).
30.
Wang, Y., Liu, X. & Shen, Y. Applicability of the random forest model in quantifying the attribution of runoff changes. Chin. J. Eco-Agric. 30, 864–874 (2022).
31.
Rautela, K. S., Singh, S. & Goyal, M. K. Characterizing the spatio-temporal distribution, detection, and prediction of aerosol atmospheric rivers on a global scale. J. Environ. Manage. 351, 119675 (2024).
32.
Kumar, N., Patel, P., Singh, S. & Goyal, M. K. Understanding non-stationarity of hydroclimatic extremes and resilience in Peninsular catchments, India. Sci. Rep. 13, 12524 (2023).
33.
Rautela, K. S., Kumar, D., Gandhi, B. G. R., Kumar, A. & Dubey, A. K. Application of ANNs for the modeling of streamflow, sediment transport, and erosion rate of a high-altitude river system in Western Himalaya, Uttarakhand. RBRH 27, e22 (2022).
34.
Miao, J. et al. Applications and interpretations of different machine learning models in runoff and sediment discharge simulations. Catena 238 (2024).
35.
Sharafati, A., Haji Seyed Asadollah, S. B., Motta, D. & Yaseen, Z. M. Application of newly developed ensemble machine learning models for daily suspended sediment load prediction and related uncertainty analysis. Hydrol. Sci. J. 65, 2022–2042 (2020).
36.
Liu, X., Zhang, X., Kong, X. & Shen, Y. J. Random forest model has the potential for runoff simulation and attribution. Water 14 (2022).
37.
Singh, N. K., Emanuel, R. E., Nippgen, F., McGlynn, B. L. & Miniat, C. F. The relative influence of storm and landscape characteristics on shallow groundwater responses in forested headwater catchments. Water Resour. Res. 54, 9883–9900 (2018).
38.
Mohr, C. H., Manga, M., Wang, C. Y. & Korup, O. Regional changes in streamflow after a megathrust earthquake. Earth Planet. Sci. Lett. 458, 418–428 (2017).
39.
Grömping, U. Variable importance in regression models. WIRE Comput. Stat. 7, 137–152 (2015).
40.
Guo, A., Chen, D., He, L., Peng, Y. & Duan, G. Reshuffling of the sediment suspension regime in the starving middle Yangtze River downstream from cascade mega-reservoirs. J. Hydrol. 622, 129694 (2023).
41.
Yao, J. et al. Transformative trends in runoff and sediment dynamics and their influential drivers in the Wuding River Basin of the Yellow River: a comprehensive analysis from 1960 to 2020. Water 16 (2023).
42.
Han, J. et al. Variation and influencing factors of runoff and sediment in Ru River Basin based on random forest model. Acta Ecol. Sin. 43, 4190–4201 (2023).
43.
Hu, G., Mao, D., Li, Z., Tian, Z. & Feng, C. Analysis on the runoff characteristics in and out Dongting Lake in recent 60 years. Scientia Geogr. Sinica. 34, 89–96 (2014).
44.
Hu, G., Mao, D., Li, Z., Zeng, Y. & Feng, C. Characteristics and driving factors of runoff and sediment changes fluxes into the Dongting Lake from three outlets in Jingjiang River during past 60 years. J. Nat. Resour. 29, 129–142 (2014).
45.
Yang, M., Mao, D., Liu, P. & Liu, W. Quantitative assessment on influence of runoff and sediment evolution of Dongting Lake and human activities to the runoff in the period of 1951–2015. Soil. Water Conserv. China, 38–43 (2019).
46.
Liu, P., Mao, D., Wang, J. & He, Q. Analysis of variations of total flood season’s runoff into Dongting lake from four rivers in Hunan Province and its factors of driving force since 1990s. Eng. J. Wuhan Univ. 48, 771–777 (2015).
47.
Mao, D. et al. Analysis on annual variation law and its causes of runoff and sediment flowing in and out of the Dongting Lakes. J. Water Resour. Water Eng. 28, 32–39 (2017).
48.
Yu, Y. W. et al. Hydromorphological processes of Dongting Lake in China between 1951 and 2014. J. Hydrol. 562, 254–266 (2018).
49.
Hu, G., Mao, D., Zhang, W., Xu, Y. & Zhang, M. Cycle characteristics and its impact factor analysis of water and sediment of Four Water Basin in Hunan Province. Resour. Environ. Yangtze Basin. 23, 986–995 (2014).
50.
Feng, C., Zeng, J., Yang, L., Zhou, W. & Liu, Z. Variation of annual runoff, sediment discharge and water level in Xiangjiang River. J. Hydroecology. 44, 56–64 (2023).
51.
Gao, Y. et al. Analysis of the nonstationarity characteristics and future trends of flood extremes in the Dongting Lake Basin. J. Hydrol. Reg. Stud. 44, 101217 (2022).
52.
Wang, L. & Xing, W. Analyses on characteristics of rainstorm in the Yangtze River Basin in 1960–2017. China Rural Water Hydropower, 14–21 (2019).
53.
Dai, Z., Mei, X., Darby, S. E., Lou, Y. & Li, W. Fluvial sediment transfer in the Changjiang (Yangtze) river-estuary depositional system. J. Hydrol. 566, 719–734 (2018).
54.
Lu, J. et al. Evolution of the main channel of the Yangtze River. River. 2, 283–306 (2023).
55.
Tian, Q. et al. Declining sediment discharge in the Yangtze River from 1956 to 2017: spatial and temporal changes and their causes. Water Resour. Res. 57 (2021).
56.
Yang, L. & Deng, M. J.l, W. & 0ue, H. f. Spatal-emporal evolution of land use and ecological risk in Dongting lake Basin during 1980–2018. Acta Ecol. Sin. 41, 3929–3939 (2021).
57.
Li, Z., Li, J., Wang, L. & Li, Y. Spatial-temporal variation and attribution analysis of soil erosion in DongtingLake Basin in recent 40 years based on CSLE-TLSD coupling model. J. Soil Water Conserv., 1–9 (2023).
58.
Yang, L. et al. Four decades of wetland changes in Dongting Lake using landsat observations during 1978–2018. J. Hydrol. 587 (2020).
59.
Zhou, J. et al. Projection of China’s future runoff based on the CMIP6 mid-high warming scenarios. Sci. China Earth Sci. (2023).
60.
Yang, Y. et al. Changes of divergence and confluence relationship between Dongting Lake and the Yangtze River after the operation of the Three Gorges Project and its impact on the waterway depth. Front. Earth Sci. 10 (2022).
61.
Zhao, C. & Yang, J. Preliminary exploration of a new water and energy storage mode for the Yuanjiang River to Zijiang River water replenishment diversion plan. Hydropower New. Energy. 38, 13–16 (2024).
62.
Liu, M. & Wang Landscape pattern change and its driving forces in middle and upper reaches of Dongting Lake watershed. Chin. J. Appl. Ecol., 1317–1324 (2008).
63.
Kendall, M. G. A new measure of rank correlation. Biometrika. 30, 81–93 (1938).
64.
Spearman, C. The proof and measurement of association between two things. Am. J. Psychol. 100, 441–471 (1987).
65.
Gauthier, T. Detecting trends using spearman’s rank correlation coefficient. Environ. Forensics. 2, 359–362 (2001).
66.
Mann, H. B. Nonparametric tests against trend. Econometrica: J. Econometric Soc., 245–259 (1945).
67.
Kendall, M. G. Rank correlation methods. 4th Edition, Charles Grifin, London (1975).
68.
Von Storch, H. Misuses of statistical analysis in climate research. Analysis of climate variability: applications of statistical techniques. von Storch H, Navarra A (eds). Springer-Verlag: Berlin, Germany, 11–26 (1995).
69.
Yue, S. & Wang, C. Y. Applicability of prewhitening to eliminate the influence of serial correlation on the Mann-Kendall test. Water Resour. Res. 38, 4-1-4-7 (2002).
70.
Sen, P. K. Estimates of the regression coefficient based on Kendall’s tau. J. Am. Stat. Assoc. 63, 1379–1389 (1968).
71.
Pettitt, A. N. A non-parametric approach change-point problem. Appl. Stat. 28, 126–135 (1979).
72.
Morlet, Arens, G., Fourgeau, E. & Glard, D. Wave propagation and sampling theorypart Il: sampling theory and complex waves. Geophysics. 2, 222–236 (1982).
73.
Wang, W., Van Gelder, P. & Vrijling, J. K. Trend and stationarity analysis for streamflow processes of rivers in western Europe in the 20th century. Proceedings: IWA International Conference on Water Economics, Statistics, and Finance Rethymno, Greece. London: IWA, 810 (2005).
74.
Dickey, D. A. & Fuller, W. A. Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74, 427–431 (1979).
75.
Kwiatkowski, D., Phillips, P., Schmidt, P. & Shin, Y. Testing the null hypothesis of stationarity against the alternative of a unit root: how sure are we that economic time series have a unit root? J. Econom. 54, 159–178 (1992).
76.
Breiman, L. Random forests. Mach. Learn. 45, 5–32 (2001).
77.
Koch, J. et al. Modeling depth of the redox interface at high resolution at national scale using random forest and residual gaussian simulation. Water Resour. Res. 55, 1451–1469 (2019).
78.
Nelson, N. G. et al. Revealing biotic and abiotic controls of harmful algal blooms in a shallow subtropical lake through statistical machine learning. Environ. Sci. Technol. 52, 3527–3535 (2018).
79.
Cho, E., Jacobs, J. M., Jia, X. & Kraatz, S. Identifying subsurface drainage using satellite big data and machine learning via Google earth engine. Water Resour. Res. 55, 8028–8045 (2019).
80.
Hurst, H. E. Long-term storage capacity of reservoirs. Trans. Am. Soc. Civil Eng. 116, 770–799 (1951).
81.
Xu, J., Chen, Y., Li, W., Ji, M. & Dong, S. The complex nonlinear systems with fractal as well as chaotic dynamics of annual runoff processes in the three headwaters of the Tarim River. J. Geog Sci. 19, 25–35 (2009).
82.
Wu, L. et al. Quantitative assessment of the impacts of climate change and human activities on runoff change in a typical karst watershed, SW China. Sci. Total Environ. 601–602, 1449–1465 (2017).
83.
Wu, J. & Gao, X. A gridded daily observation dataset over China region and comparison with the other datasets. Chin. J. Geophys. 56, 1102–1111 (2013).
84.
Wu, J., Gao, X., Giorgi, F. & Chen, D. Changes of effective temperature and cold/hot days in late decades over China based on a high resolution gridded observation dataset. Int. J. Climatol. 37, 788–800 (2017).
85.
Song, C. et al. A comprehensive geospatial database of nearly 100 000 reservoirs in China. Earth Syst. Sci. Data. 14, 4017–4034 (2022).

Acknowledgements

This research was supported by the Joint Fund for Regional Innovation and Development of National Natural Science Foundation of China (Grant No. U19A2051), the Research Foundation of the Department of Natural Resources of Hunan Province (Grant No. 20240101ST), the Philosophy and Social Sciences Fund of Hunan Province, China (Grant No. 21YBA035), the Key Scientific Research Project of the Education Department of Hunan Province, China (Grant No. 23A0049), the National Natural Science Foundation of China (Grant No. 42001024), and the Construction Program for First-Class Disciplines (Geography) of Hunan Province, China (Grant No. 5010002).

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

About this article

Cite this article

Wang, T., Mao, D., Li, E. et al. Non-stationarity of runoff and sediment load and its drivers under climate change and anthropogenic activities in Dongting Lake Basin. Sci Rep 14, 24333 (2024). https://doi.org/10.1038/s41598-024-74952-x

-
Received:
-
Accepted:
-
Published:
-
DOI: https://doi.org/10.1038/s41598-024-74952-x