1 Introduction

Increasing precipitation extremes under climate change amplified (Zhang et al. 2013; Myhre et al. 2019; Moustakis et al. 2021; Gründemann et al. 2022) flooding events globally (Jongman et al. 2012; Hirabayashi et al. 2013; Winsemius et al. 2016; Rodell and Li 2023). In recent years, there have been increasingly frequent floods in the Yangtze River (Jiang et al. 2008; Yu et al. 2009; Fang et al. 2018). To mitigate damage, various engineering measures have been implemented to reduce flood peaks. In the Yangtze River Basin (YRB), the Three Gorges Dam (TGD) and other mountainous reservoirs have been built for flood control. Among reservoirs located in the Yangtze River basin, Three Gorges Reservoir (TGR), with storage capacity of 39.3 billion m3, plays a critical role in flood mitigation (Zhang and Lou 2011; Xia and Chen 2021) (Fig. 1a). Floodwater in the Three Gorges Reservoir Region (TGRR) contribute from three sources: runoff from the TGRR itself, the drainage area upstream of the Cuntan station in the Yangtze River Basin, and runoff from the Wu River Basin. Tributaries within the TGRR contribute up to one-third of the total floodwater inflow from upstream to the TGRR. Moreover, runoff of the tributaries in the TGRR directly drains into the reservoir, significantly influencing reservoir security. Thus, investigating changes in hydrological characteristics of the TGRR is crucial for flood mitigation and security of the Three Gorges Dam.

Climate change and land use and cover change (LUCC) are the prime factors that impact hydrological processes of the TGRR (Zhang et al. 2015; Cheng et al. 2019; Wang et al. 2019a, 2020; Huang et al. 2020). Climate alters the temporal and spatial distribution of precipitation patterns, directly modifying flood regimes (Donat et al. 2016; Gründemann et al. 2022). Evapotranspiration (ET) is more likely to affect surface runoff (Zhou et al. 2013; Wang et al. 2019a). Meanwhile, LUCC alters hydrological processes by modifying transpiration, interception, and conservation (Hurkmans et al. 2009; Wang et al. 2015; Li et al. 2020). This is especially the case during urbanization, where a permeable vegetated land surface has been replaced by an impervious cityscape. This alteration affects the proportions of streamflow and baseflow, leading to increased velocity and flood risk (Zhou et al. 2013; Lin et al. 2022). The impervious area of the YRB increased by 162.46% from 1977 to 2018 (Luo and Zhang 2022). From 1990 to 2010, the forest and/or grass-based vegetation restoration project, namely the “Grain for Green” project since the 2000s, led to a 14% increase in vegetation cover from 1999 to 2009 in the YRB (Liu et al. 2014). The increasing vegetation cover leads to a decline in soil moisture and runoff due to increased ET and rainfall (Li et al. 2018), which causing reducing discharge in the YRB (Yang et al. 2015; Zeng et al. 2024). Therefore, the TGRR responses to compound impacts of urbanization and afforestation call for in-depth investigation.

Methods available for analyzing hydrological impacts can be grouped into three categories: catchment paired experiment, statistical analysis (Zhang et al. 2018a, b), and hydrological models (Zhang et al. 2015). The absence of observations renders the first two methods ineffective in the TGRR (Wang et al. 2011b). To better understand how hydrological processes change and recognize the impact of LUCC and climate change, hydrological models have been generally used (Hurkmans et al. 2009; Zhang et al. 2015; Wang et al. 2020; Gong et al. 2023). With the increased availability of remotely-sensed data, high-resolution meteorological data (Brookfield et al. 2023), hydrological model parameters, and high-quality input data are easier to access, enabling effective hydrological modeling (Sood and Smakhtin 2015).

In the TGRR, none of the rivers are hydrologically monitored over the entire tributary basins (Wang et al. 2011b). Evaluations of model parameters within the data-scare basin is an challenging task (Hrachowitz et al. 2013; Elmer et al. 2021). While there are streamflow observations at the outlet and the ungauged zone encompasses 70% of the area (Wang et al. 2011b). The scarcity of data challenges parameter estimation in hydrological modeling (Zhang et al. 2020). To implement hydrological modeling for simulation in ungauged areas, regionalization calibration method was used, and parameters were transferred from gauged to similar ungauged sub-catchments (Alizadeh and Yazdi 2023). For example, parameters calibrated against to upstream observation at Poyang Lake were applied to non-measured area (Zhang et al. 2017).

Many studies of the YRB have analyzed monthly or annual streamflow observations to analyze the streamflow trend under climate change (Fang et al. 2018; Huang et al. 2024), and/or have evaluated the impact of human activities on streamflow (Zhou et al. 2013; Cheng et al. 2019; Wang et al. 2020). Previous studies on hydrological features in the TGRR used a lumped model without consideration of subbasin-specific hydrological responses (Jing et al. 2004). With the development of distributed hydrological models, these models have bee widely used in flood flow modelling in the TGRR (e.g., GBHM, VIC, etc.) (Wang et al. 2011b, 2019b; Pang and Xu 2012). Very recent studies focused on integration of hydrologic and hydrodynamic models for flood flow modeling (Wang and Yang 2020; Xu et al. 2022). Distributed hydrologic models perform better in simulating rainfall-runoff processes at a large basin scale, whereas hydrodynamic models have higher modelling performance for smaller river basins. Combining these models allows for leveraging their respective strengths (Wang and Yang 2020). Previous studies simulated the flow and water depth in the TGRR by coupled XAJ hydrological model and one-dimensional hydrodynamic model (Wang et al. 2003; Zhang et al. 2020). Compared with the simulation results of a single hydrological model, the model coupled with XAJ and H1DM has a higher accuracy in the TGRR (Zhang et al. 2020). However, existing studies using one-dimensional hydrodynamic models only captured alterations at the outlet of the region, overlooking the spatial distribution of hydrological elements within the region. What’s more, previous researches gave limited consideration to human activities like land use changes which are important factors in the hydrological cycle.

Using hydrological and hydrodynamic models, we attempted to simulate hydrological characteristics in the TGRR and tried to reflect water depth distribution of flood events. The study adopted the distributed hydrological model VIC with a physical mechanism to simulate the rainfall-runoff process under the conditions of complex topography and water system in the TGRR, and at the same time, the hydrodynamic model CaMa-Flood was applied with the VIC runoff results to reflect water depth distribution within the TGRR during the flood events. Based on the results, we analyzed the variations in hydrological elements within the historical period. Considering the impact of land-use changes on hydrological elements, we designed scenarios of land-use changes for different periods. We employed simulations to quantify the influence of land-use changes under various scenarios, exploring the impacts of both climate change and land-use changes on the hydrological feature in the TGRR.

2 Materials and methods

2.1 Study area

TGRR is in the upper Yangtze River (Fig. 1b), and specifically the stretch of the Yangtze River from Cuntan in the mainstream to the location near Wulong in the tributary, extending to the Three Gorges Dam (Shen et al. 2014). The Three Gorges Interval is located at the junction of Chongqing and Hubei provinces with drainage area of about 55,000km2, and over 658 km in length. The altitude of the TGRR is between 3 and 3100 m. The TGRR exhibits an east-high-west-low topography, with Fengjie as the dividing point. Topography upstream to Fengjie is characterized by low hills and undulating landscapes. Regions downstream to Fengjie lies in the Three Gorges along the Yangtze River, including Qutang Gorge, Wuxia Gorge, and Xiling Gorge. TGRR is climatically characterized by monsoon area with annual rainfall ranging from 1000 mm to 1600 mm with 1200 mm on average. Rainfall from June to September accounts for more than 70% of the total annual rainfall, with frequent heavy rainfall events (Wang et al. 2011b).

Runoff in TGRR, upstream inflows from the Yangtze River, and the Wujiang River jointly constitute the three main sources of floods in the TGRR (Wang et al. 2011a). The main tributaries in the TGRR include 47 rivers, such as the Daning River, Xiangxi River, Xiaojiang River, Shennong River, Tangxi River, Meixi River, Zhuyi River, Modao River, Jiwanxi River and Ruxi River (Fig. 1c). The main tributaries with drainage areas exceeding 3,000 km2 include the Yulin River, Longxi River, Xiaojiang River, MoDaoxi River, Daning River, and Xiangxi River.

Fig. 1
figure 1

Map showing the location of the Yangtze River Basin in China (a), the location of TGRR in the Yangtze River Basin (b),Map of the TGRR hydrological stations, and weather stations (c)The LUCC map is from the CNLUCC2015 dataset

Full size image

2.2 Data

In this study, three kinds of input data were used: (1) meteorological forcing data; (2) vegetation and soil data; and (3) topographical data. Data required by the VIC model include the forcing elements of daily rainfall, temperature, and wind from 2015 to 2018 collected at 47 meteorological stations around the TGRR. The meteorological data was interpolated to each grid cell using the linear distance weighted interpolation method. Locations of the meteorological stations can be found in Fig. 1b. Daily discharges at six gauging stations from 2015 to 2018 were obtained from the Hydrological Yearbook of China, including Cuntan station (CT2, which measures the inflow from Yangtze River into the TGRR), Wulong station (measures the inflow from Wujiang River into the TGRR), Yichang station (the outlet of the TGRR), Lianghe station (LH, in the Quxi basin), Wuxi station (WX, in the Daning River basin), and Changtan station (CT1, in the Modao River basin). Daily water level data from 2015 to 2018 at Cuntan, Wanxian, Qingxichang, Shibanping, and Wuxi stations were also obtained from the Hydrological Yearbook of China.

The vegetation classification is derived from the 1 km land cover classification for 1992–1993 from the Advanced Very High-Resolution Radiometer (AVHRR) provided by the University of Maryland (Hansen et al. 2000). Table S1 lists the detailed land cover classes. The land use and cover data is based on China’s Multi-Period Land Use Land Cover Remote Sensing Monitoring Data Set (CNLUCC) (Xu et al. 2018). This dataset with a spatial resolution of 1 km for the years of 1990,1995, 2000, 2005, 2010, 2015, and 2020. Detailed land cover classes are listed in Table S2. Considering the study period of 2015–2018, the land-use data for 2015 and 2020 were utilized. The land cover map derived from AVHRR has been updated by incorporating the construction land class from the CNLUCC dataset, resulting in an updated LUCC dataset. Soil texture is based on the Harmonized World Soil Database (HWSD v1.21)(FAO/IIASA/ISRIC/ISSCAS/JRC 2012) from the FAO, offering detailed soil attributes for two soil layers.

The digital elevation model (DEM) of the catchment originates from GDEM v3 (ASTER Global Digital Elevation Map). The spatial resolution of the DEM is 30 m. This digital elevation model data set is used to extract the channel network. The MERIT Hydro dataset (Yamazaki et al. 2019) is taken for river-floodplain parameterization river length, and river width data in the hydrodynamic models.

2.3 Models and method

The general procedures used for simulation and data processing are summarized in Fig. 2.

Fig. 2
figure 2

Flowchart for methodology (four main components included: data used, simulation procedure, scenario design, and result analysis)

Full size image

2.3.1 Models

The Variable Infiltration Capacity (VIC) hydrologic model is used in this study. The VIC model is a macroscale, semi-distributed land surface hydrologic model, which shares several basic features with other land surface models (LSMs) (Liang et al. 1994). The runoff generation scheme in VIC represents both saturation and infiltration excess runoff processes dynamically in a model grid cell through a statistical parameterization of sub-grid heterogeneity in terms of soil and rainfall (e.g., local water holding capacity) (Liang and Xie 2001). This led to better performance in runoff simulation. Runoff is comprised of surface runoff, subsurface flow (baseflow), and interflow. The VIC model has been successfully applied in YRB (Gu et al. 2015; Lu et al. 2018) and TGRR (Wang 2021). VIC 4.1.2 is used in this study. A separate surface routing model is coupled with the VIC model for routing (Lohmann et al. 1996). Utilizing the runoff output from the VIC model, this model generates daily streamflow output at the basin outlet.

CaMa-Flood is a hydrodynamic model and has been used for global- and basin-scale flood simulations (Hirabayashi et al. 2013; Willner et al. 2018; Wang et al. 2021). By parameterizing the subgrid-scale topography of a floodplain, it calculated simulated discharge, velocity, water level, and flood inundation depth by solving the local inertial equation. thus describing floodplain inundation dynamics (Yamazaki et al. 2011, 2013, 2014). The water level and flooded area are diagnosed from the water storage at each unit-catchment using the sub-grid topographic parameters of the river channel and floodplains (Yamazaki et al. 2011). The progression of water storage is determined by solving the water balance equation, taking into account the incoming water from upstream cells, the outflow to downstream cells, and the runoff input for each catchment unit (Yamazaki et al. 2011). Therefore, adopting the model could simulate changes in all stages of flooding, including discharge and water levels. The 1-min water depth data (about 2 km at the Equator) was obtained by downscaling based on DEM (Shin et al. 2020).

2.3.2 Coupled hydrological and hydrodynamic model

Utilizing Digital Elevation Models (DEMs), geographical data, and topographical features of the watershed, this study constructed the distributed hydrological model and the hydrodynamic model in the TGRR. Using these models, hydrological characteristics within the TGRR are simulated, including runoff, streamflow, and water depth. Initially, the VIC model was utilized to generate daily surface and subsurface runoff, including interflow and baseflow. Subsequently, the runoff produced by the VIC model served as the boundary condition for the CaMa-Flood model to calculate water depth.

Meteorological forcing data for the VIC model, including precipitation, temperature, and wind speed data, are sourced from 47 stations around the TGRR, as detailed in Sect. 2.2. These data were interpolated into 0.1 degrees using the inverse distance weighting method. Soil, vegetation, and land cover data that are necessary for the VIC model were detailed and described in Sect. 2.2 (Soil texture is based on HWSD, land cover data is from CNLUCC, and vegetation data is from the land cover classification from AVHRR). For this study, we ran the VIC model version 4.1.2 in water balance mode at a daily time step and at grid resolutions from 0.1 degrees. There are 637 grids in the TGRR. In this study, the routing model developed by Lohmann et al. (1996) is used to calculate the discharge at the basin outlet. The overflow employs the unit hydrograph method, while the channel routing employs the linear Saint Venant equations.

Taking the runoff simulation of VIC as forcing, CaMa-Flood runs daily with a spatial resolution of 0.1 degree. The water depth series is calculated from 2015 to 2018 with the first two years repeated as a spin-up period. The period 2015–2018 was chosen for this study because the meteorological station data and hydrological observation are accessible within this period. The water depth is calculated based on water storage at each grid cell assuming hydro-geometry power-low equations (Yamazaki et al. 2014).

2.3.3 Model calibration and validation

Due to the scarcity of observation data, the real streamflow of the TGRR is not available, thus it is not possible to determine the parameter rate for the whole region based on the interval flow data. According to previous research (Xie et al. 2007), parameters in the TGRR are obtained following this procedure: model parameter is estimated from a limited number of calibrated basins and then regionalized to the whole region. It is noted that naturalized streamflow without reservoir regulation is simulated in this research, thus only hydrological stations without the backwater effects are selected, namely Lianghe hydrological station, Wuxi hydrological station, and Changtan hydrological station. Parameter estimation was performed at these hydrological stations during the periods 2015–2017 (the calibration period) and 2018 (the validation period). A trial and error calibration method is being used and the calibrated values(Park and Markus 2014; Ju et al. 2020) of Binf, Ds_Max, Ws, D2, D3 are shown in Table 1.

Table 1 Calibrated parameters in the VIC model
Full size table

The parameters shown in Table 1 were estimated by calibration for the best fit of model outcomes to observed behavior. Five metrics were used to assess the model performance: Nash–Sutcliffe Efficiency (NSE), Relative error (Re), determination coefficient (R-square), Normalized Root Mean Square Error (RMSE), and Normalized RMSE (NRMSE). The formula for each indicator is given below (Eqs. (1)-(5). The main evaluation metrics as the objective function are Re and NSE. Re reflects the total accuracy of the model simulation, and NSE reflects the fitness between the simulated flow and the observation. According to (Moriasi et al. 2007, 2015), model performance at the basin scale can be evaluated as satisfactory if daily RE < 25% and NSE > 0.5.

$$NSE = 1 - {{\sum \> {{\left( {{Q_{sim}} - {Q_{obs}}} \right)}^2}} \over {\sum \> {{\left( {{Q_{obs}} - \overline {{Q_{obs}}} } \right)}^2}}}$$
(1)
$$\:\text{R}\text{e}=\frac{\overline{{Q}_{sim}}-\overline{{Q}_{obs}}}{\overline{{Q}_{obs}}}\:\:\:\:\:$$
(2)
$$\:RMSE=\sqrt{\frac{\sum\:{\left({Q}_{obs}-\overline{{Q}_{sim}}\right)}^{2}}{\text{N}}}\:\:\:\:$$
(3)
$$\:NRMSE=\frac{\text{R}\text{M}\text{S}\text{E}}{\text{max}\left({Q}_{obs}\right)-\text{min}\left({Q}_{obs}\right)\:}\:\:4$$
(4)
$$\:{R}^{2}=1-\frac{\sum\:{\left({Q}_{sim}-{Q}_{obs}\right)}^{2}}{\sum\:{\left(\overline{{Q}_{obs}}-{Q}_{obs}\right)}^{2}\:}\:\:\:5$$
(5)

To test the efficiency of the parameter regionalization result, a verification was conducted by comparing the observed flow at Yichang station and simulation results from 2015 to 2018. Located in the outlet of the TGRR, Yichang hydrological station has streamflow series data. Combined with the incoming water from Cuntan and Wulong hydrological stations and simulated runoff in the TGRR, simulated streamflow at the outlet of the TGRR is available.

2.3.4 Flood events extraction and validation

The study evaluates model efficiency in simulating flood events in TGRR by conducting verification on flood peak discharge and peak time.

Utilizing the Peaks Over Threshold method, this study applied a threshold of 25,000 m3/s to identify significant flood events based on observed streamflow data recorded at Yichang station from 2015 to 2018. However, the average flow from July to the end of August 2018 exceeded 25,000 m3/s, so a threshold of 40,000 m3/s was used. Therefore, we have not taken a uniform flood standard. A total of 10 flood events were extracted from 2015 to 2018 (Table 2 shows the list of flood events). Maximum discharge during a flood event is defined as the flood peak, and the occurrence time of the maximum flow during a flood event is defined as peak time. The study calculates the relative error between simulated discharge and observed values for flood peak and between simulated and observed peak times.

Table 2 Extracted flood events in 2015–2018
Full size table

2.3.5 Analysis of changing hydrological features

To examine annual flow variations, this study computed the monthly averages, maximums, and minimums of discharge in the TGRR from 2015 to 2018, the statistical characteristics of discharge and water depth series are listed in Table S3 (Burgan et al. 2013). Subsequently, linear trend analysis was performed on these monthly discharge series. To explore seasonal variations, the data were partitioned into flood season (May to October) and non-flood season (November to April). The same analysis has been performed on the water depth series.

2.4 Impact of LUCC on hydrological features

The land cover dataset from AVHRR (Hansen et al. 1999) has detailed vegetation information, while the CNLUCC has the latest built-up distribution data. To obtain input with detailed vegetation information and the latest land cover data in 2015 and 2020, we updated the built-up class in the AVHRR by land use type from CNLUCC. Then, we constructed the LUCC2015 and LUCC2020 scenarios and simulated the discharge and water depth under different LUCC scenarios. The analysis compared variations in discharge and water depth under different LUCC scenarios to assess the impact of diverse land uses on hydrological features in the TGRR.

At Yichang Station, flow patterns are impacted by both land use within the TGRR and the inflow from the upstream. To explicitly show the influence of land use changes on hydrological features in the TGRR, the discharge used for comparison involves routing exclusively based on local runoff without considering upstream inflow. Meanwhile, the water level simulated by the CaMa-Flood is still influenced by upstream inflow.

3 Results

3.1 Model calibration and validation

3.1.1 Parameter calibration

A list of parameters in the VIC model for each basin (Table S4) was identified. The metrics of simulated daily streamflow for those basins are shown in Table 3. The NSE between the observed and simulated streamflow at Wuxi station is above 0.6 during the calibration and validation periods, and the Re is below 20%. Moreover, this model performs better in the validation than in the calibration period. The NSE increased from 0.61 to 0.62, |𝑅𝑒| decreased from 0.18 to 0.07, indicating that this model could provide satisfactory performances under out-of-sample situations. Based on the calibrated model from the Changtan hydrological station in the Modaoxi basin, the NSE value during the calibration period is below 0.5 and drops to 0.14 under the validation period with a |𝑅𝑒| value of 1.21. According to the calibrated model against the Lianghe hydrological station in the Quxi basin, the NSE value during the calibration period is 0.31 and drops to 0.16 under the validation period with a |𝑅𝑒| value over 6.61. This indicates that the calibrated model in the Modaoxi and Quxi basins can hardly reflect the actual conditions of the region. The simulated and measured streamflow series from 2015 to 2018 at the Wuxi, Changtan, and Lianghe Stations are displayed in Fig. S1.

Table 3 Performances of calibrated hydrologic models for both calibration and validation periods in three gauged basins
Full size table

Previous studies (Moriasi et al. 2007, 2015) show acceptable model performance at the watershed scale if daily Re < 25% and NSE > 0.50. NSE values at the Changtan and Lianghe hydrological stations fall from 0.30 to 0.40. Additionally, the value of |Re| at the Changtan station is approximately 1.00, whereas at the Lianghe station, it surpasses 6.00. Calibrated models underestimated the streamflow in the Quxi and Modaoxi basins (Fig. S1). The region is prone to recurrent short-duration intense rainfall events caused by strong convective weather. This poses challenges in accurately measuring precipitation of rain gauges, resulting in the underestimation of peak flow. Additionally, the lack of updated vegetation cover data adds to the biases in the model (Zhou et al. 2004; Wang 2021). Considering that only the calibrated model in the Daning River could reflect the actual conditions, parameters calibrated against the Wuxi hydrological station are regionalized to the TGRR.

3.1.2 Validation

Fig. 3
figure 3

Simulated versus measured streamflow in 2015–2018 at Yichang Station

Full size image

Daily discharge at the outlet of the TGRR is well simulated with an NSE of 0.81, an R-square of 0.93, and Re of -2% (Fig. 3). 98 parameter sets with NSE values > 0.60 were adopted to evaluate the uncertainty of the parameters(Wang 2021) (Table S5). Results show that NSE ranges from 0.78 to 0.79, Re values range from − 1% to -2%, and R-square is 0.92. Streamflow series under different parameter sets only results in marginal differences.

Qualitative comparison is performed between simulated water depth and observed water level data to access inaccessible reference datum data from hydrological stations. Under the effect of the TGD’s backwater, the seasonal trend between observed and simulation was found at hydrological stations located at the mainstream. For the Shibanping station and Wuxi station, located outside the scope of the reservoir backwater effects, observed water level data aligns well with the trend of simulated water depth data, especially in the Wuxi station. Overall, the above qualitative comparison can reflect the credibility of the water depth simulation results to a certain extent. Detailed information on water depth validation is in Supplementary Text 1 and Fig. S2.

Table 4 Comparison between observed and simulated flood events characteristics
Full size table

According to Table 4, the relative error between simulated and observed peak flows ranges from − 2.5 to 40.0%. Simulations mostly overestimated the peak flows, which may be influenced by the storage of hydraulic projects and water withdrawals (The TGRR has numerous water withdrawal sites, dams, and small hydropower stations). Bias in flood peak time ranges from 2 to -5 days. Notably, the bias is more pronounced for flood events in August, exceeding three days in deviation. This discrepancy may be attributed to the fact that August falls within the primary flood season in the TGRR (Yu et al. 2009), being characterized by intermittent rainfall. The overall high flow during this period could result in misjudgments of flood peaks.

3.2 Changing flood characteristics

3.2.1 Streamflow in 2015 to 2018

The monthly average, maximum, and minimum flow at the Yichang station increased from 2015 to 2018 (Fig. 4). Among them, the monthly maximum flow increased most significantly with a rising rate of 1808.28 m3/s/a, followed by the monthly average flow of 1218.12 m3/s/a. The least significant was the monthly minimum flow (751.44 m3/s/a). This is in line with the changes in the observed flow (Pang et al. 2022). According to observation, the monthly maximum flow rose at 1498.32 m3/s/a, followed by the monthly average flow of 1065.48 m3/s/a. Monthly minimum flow with a least increase rate of 667.44 m3/s/a.

In 2015–2018, long-term monthly maximum streamflow is 36858.15 m3/s, deviating by 3.95% from observation; multi-year monthly minimum streamflow is 5483.88 m3/s, deviating by -12.61% from observation; and multi-year average streamflow is 13694.56 m3/s, deviating by -0.06% from observation. In January, the average flow is the smallest at 6,275.36 m³/s, deviating − 11.61% from the observed value. Conversely, in July, the average flow is the largest at 27,240.81 m³/s, with a deviation of 3.77% from the measured value. The monthly average flow exhibits a seasonal trend, initially rising and declining throughout the year, aligning with observations at Yichang station in the Yangtze River Basin. Rainfall, predominantly occurring in summer and autumn, is the prime driver for streamflow (Yu et al. 2009).

Streamflow during flood seasons increased from 2015 to 2018, and the increase rates of monthly average, maximum, and minimum flow were 1613.82 m3/s/a, 2002.80 m3/s/a, and 1175.46 m3/s/a, respectively. The observed discharge displayed an increasing trend with rates ranging from 775.92 m³/s/a to 1542.78 m³/s/a, and the most notable surge occurred in the monthly maximum flow of 1542.78 m³/s/a, aligning with the simulation results. Streamflow in non-flood seasons also increased, however, the monthly minimum flow decreased. In the 2015–2018 (November-April) monthly average, maximum discharge showed an increasing trend during the non-flood season, with an increasing rate of 207.00 m3/s/a and 480.12 m3/s/a. Minimum flow series showed a decreasing trend, with a rate of -13.80 m3/s/a. According to observation, measured data showed that the monthly discharge during non-flood season showed an increasing trend with an increasing rate varying between 192.78 m3/s/a to 392.22 m3/s/a.

Fig. 4
figure 4

Streamflow series at the outlet of the TGRR from 2015 to 2018 (inflow from upstream is excluded)

Full size image

Discharge without the upstream inflow was also analyzed, and monthly mean maximum and minimum flow increased from 2015 to 2018 (Fig. 4). Increasing rates of monthly average, maximum, and minimum flow were 124.56 m3/s/a, 34.92 m3/s/a, and 8.40 m3/s/a, respectively. Monthly mean maximum and minimum flow during non-flood season showed a decreasing trend, with decreasing magnitude of -146.47 m3/s/a, -127.20 m3/s/a, and − 87.60 m3/s/a, respectively. In contrast, monthly mean maximum and minimum flow during flood season showed an increasing trend, with rates ranging from 40.44 m3/s/a to 191.76 m3/s/a. Therefore, incoming flow to the TGR increased continuously from 2015 to 2018, while minimum flow during the non-flood season decreased. The inflow is seasonally varied, with the average flow increasing from January to July and then decreasing, which is consistent with observation. Runoff within the region increased in 2015–2018 and decreased during flood season.

3.2.2 Water depth in 2015 to 2018

Fig. 5
figure 5

(a) Mean water depth series from 2015 to 2018 in the TGRR, (b) The maximum water depth series from 2015 to 2018 in the TGRR

Full size image

From 2015 to 2018, the average water depth in the TGRR increased, with maximum, minimum, and average values of 0.44 m, 0.05 m, and 0.12 m, respectively. (Fig. 5(a). The changing rates of water depths during this period were 0.16 cm/a, 0.47 cm/a, and 0.11 cm/a for the maximum, minimum, and average water depth, indicating a consistent upward trend (Pang et al. 2022). However, the monthly average water depth in the TGRR interval during the flood season (May-October) decreased. The decreasing rate in the monthly maximum water depth is the highest (-0.61 cm/a), followed by monthly average water depth (-0.48 cm/a) and monthly minimum water depth (-0.21 cm/a). Water depths exhibit two peaks, occurring in July and October, respectively. In 2017, the average water depths within the region in the autumn flood events exceeded those in the summer flood events. The monthly average water depth during the non-flood event showed an increasing trend, with the most significant rise in the monthly maximum water depth at 0.60 cm/a. Following this, the monthly average water depth increased by 0.33 cm/a, and the monthly minimum water depth increased by 0.17 cm/a.

The maximum, minimum, and average values of the maximum water depth in the TGRR from 2015 to 2018 were 18.92 m, 6.39 m, and 2.33 m (Fig. 5(b)). The rising rate ranges from 0.34 m/a to 0.08 m/a. As the average water depth in the region, the maximum water depths within the region during the autumn flood events exceeded those in the summer flood events in 2017. The maximum water depth in TGRR was the highest in 2016. Monthly maximum water depths decreased during flood seasons, with changing rates of -0.11 m/year, -0.10 m/year, and − 0.15 m/year for the monthly maximum, minimum, and average water depth. The total water depth in the TGRR during non-flood season shows an opposite trend, with the highest decreasing magnitude of monthly minimum water depth.

3.3 LUCC’s effect on hydrological features

3.3.1 Land use/cover change

The land use types in the TGRR were mainly woodland grassland, grassland, grassland, and cropland types (Fig. 5). In 2015, there were 29 sub-basins where the primary land cover type was cropland (mainly found in the north and northeast regions), 10 sub-basins with woodland grassland, and the rest 9 sub-basins with woodland (mainly found in the south and southwest region). From 2015 to 2020, the main land cover types changed in the Zhuyihe, Changtanghe, and Baolonghe basins. The primary land use type of the Baolonghe basin changed from arable land to woodland grassland, the opposite change occurred in the Zhuyihe basin, and the main land use type of the Changtanghe basin changed from arable land to urban and buildings. Land cover proportions in the TGRR in different periods are shown in Table S6.

Land-use changes from 2015 to 2020 in the TGRR are shown in Fig. 6. The percentage of evergreen needles, deciduous broadleaf, mixed forests, and woodland remains the same from 2015 to 2020, at 4.2%, 2.9%, 2.2%, and 9.6%, respectively. Urban and built-up areas expanded evidently, increasing from 1.0% in 2015 to 1.6% in 2020. This growth includes a notable conversion of 42.9% from cropland and around 6.0% from grassland, mainly in the southwest region, encompassing basins like Wubuhe, Lixiangxi, Changranhe, and Chaoyanghe. The percentage of water bodies increased from 0.3% in 2015 to 0.4% in 2020, with growth occurring around the mainstream of the Yangtze River and Longxihe basin, mainly from cropland (94.7%) and grassland (5.3%). The percentage of grassland has seen a marginal increase from 6.9% in 2015 to 7.0% in 2020. This is mainly attributed to the conversion from cropland, predominantly occurring in the mainstream and northern watersheds (such as Qianjiang, Yuling, Xijiang, and Tangxi Basin).

Remote sensing data suggested that woodland grassland, cropland, and forest had been decreasing. The proportion of woodland-grassland within the TGRR declined from 18.6 to 17.9%, marking the most significant relative decrease. The loss of woodland grassland was mainly converted into Cropland, which occurred in the Qixianjiang, Jiuwangxi, and Daning basins. Cropland decreased from 54.2% in 2015 to 54.1% in 2020. The main types of diversion from cropland are woodland prairie and grassland, concentrated in the Xiaojiang, Tangxi, Meixi, Daning, and Modaoxi basins. At the same time, forest, woodland grassland, and grassland were mainly converted into cropland.

Fig. 6
figure 6

LUCC transfer map of region for 2015–2020

Full size image

3.3.2 LUCC impact on discharge

The multi-year annual average flow at Yichang station is 13596.6 m3/s under the 2020LUCC scenario, which is slightly higher than that under the 2015LUCC scenario (13593.3m3/s). The multi-year annual maximum flow is 51171.5m3/s under the 2020LUCC scenario, which is higher than that under the 2015LUCC scenario (51165.5m3/s). Therefore, the streamflow variation at Yichang station across different land use scenarios is not discernable. The disparity in average annual flow remains within 5.00 m³/s, while the difference in maximum annual flow is limited to 8.00 m³/s.

Comparison of the discharge excluding upstream inflow under different scenarios shows that land use changes have a greater impact on discharge during the flood season. In flood season, variations in mean flow between different LUCC scenarios ranged from 3.70 to 7.80%, while the maximum flow differences ranged from 5.70 to 8.50%. In non-flood season, variations in mean flow between different LUCC scenarios ranged from 0.50 to 2.20%, while the maximum flow differences ranged from 0.70 to 2.88%.

Differences in monthly average and maximum flows under LUCC scenarios during the flood seasons (from May to October) in 2015–2018 are shown in Table S7. Notably, the most significant difference occurred in July, surpassing 8%. Nevertheless, regional runoff disparities and discharge at the outlet of the TGRR between land use scenarios remain relatively modest, all falling below 12.00 m³/s, with relative differences amounting to less than 4‰. The results indicate that land use changes contribute to increased runoff in the TGRR, leading to a rise in the inflow of the TGR. However, the impact is relatively limited compared to factors like rainfall.

3.3.3 LUCC impact on water depth

Under LUCC2015 scenario, daily mean water depth in the TGRR ranged from 0.05 m to a maximum of 0.44 m from 2015 to 2018, with a multi-year average of 0.13 m, and notably higher values in 2016. Under the LUCC2020 scenario, the daily mean water depth in the TGRR remains within the same range as that in the LUCC2015 scenario, with a multi-year average of 0.12 m and notably higher values in 2016. This may be explained by the extreme precipitation in 2016 within the TGRR, leading to severe flooding in the middle and lower reaches of the Yangtze River. Between LUCC2015 and LUCC2020, the maximum mean water depth in the TGRR area remained stable, while the multi-year average decreased by 7.70%.

Under LUCC2015 scenario, daily maximum water depth in the TGRR ranged from 2.35 m to a maximum of 19.09 m from 2015 to 2018, with a multi-year average of 6.49 m and notably higher values in 2016. For LUCC2020 scenarios, daily maximum water depth ranged between 2.33 m and 18.92 m from 2015 to 2018, averaging 6.39 m over the multi-year span. The same as LUCC2015 scenario, the daily maximum water depth was higher in 2016 than in other years under the LUCC2020 scenario. In the LUCC2015 scenario, the maximum water depth in the TGRR was 8.91‰ lower than that in the LUCC2020, while the average value displayed a 15.4‰ increase.

Table 5 shows seasonal variation in average and maximum water depth difference between LUCC scenarios. During flood seasons, water depth difference ranges from − 0.35%∽-3.53% in average value and − 0.33%∽-2.58% in maximum value. During non-flood seasons, water depth difference ranges from − 1.38%∽-2.86% in average value and − 1.04%∽-2.47% in maximum value. Reduction in water depth due to LUCC was more pronounced in March-June when the relative change rate in mean value exceeded 2% in absolute terms.

Table 5 Comparison of monthly average and maximum water depth across multiple years under the 2015LUCC and 2020LUCC scenarios
Full size table

In the LUCC2015 scenario, the average water depth in the region was 0.02–0.58 cm lower than that in the LUCC2020. For the maximum water depth, the difference is 0.40–11.38 cm. The difference in water depth between the LUCC2015 and LUCC2020 scenarios shows evident spatial variations. Compared with LUCC2015 scenarios, the overall water depth in the region increased during flood events 1, 2, 5, and 10 in the LUCC2020 scenarios, while it demonstrates contrasting change during flood events 4, 7, and 8. Notably, for flood events 3, 6, and 9, water depth in the Daning basin exhibited an apparent increase, although the overall water depths still decreased. Land cover changes led to over 0.4 cm increase in mean water depth during flood events 1, 2, and 10. In LUCC2015 scenarios, maximum water depth surpassed by over 1 cm in flood events 3, 4, 6, 7, 8, and 9 compared with LUCC2020, with the most significant difference observed in flood6, exceeding 11 cm. Based on this classification, this section analyzes water depth differences due to different land use scenarios for Flood Events 1, 7, 8, and 9 (Fig. 7), with water depth differences for the remaining flood events plotted in Fig. S3.

Fig. 7
figure 7

Difference in water depth between LUCC2015 and LUCC2020 scenarios in flood events 1,7,8,9 (Note: water depth map is at 1-min, and water depth has not changed is not represented in the figure)

Full size image

In LUCC2015 scenarios, the average water depth was reduced by 0.58 cm (-1.73%) in flood event 1 (peaked on 07/01/2015) compared with LUCC2020, and the maximum water depth was 0.37% higher. In the north and northeast regions, there is an apparent reduction in water depth. The reduction in water depth takes over a 10 cm reduction in the river section from Cuntan Station to Qingxi Station in the mainstream of the Yangtze River and the Xiaojiang basin. In the midsection of the region, the difference in water depth is less pronounced between different scenarios.

In LUCC2015 scenarios, the average water depth was reduced by 0.03% in flood event 7 (peaked on 08/31/2017) compared with LUCC2020, and the maximum water depth was 22.40% higher. For flood event 8 (peaked on 09/28/2017), the average water depth was reduced by 0.07% compared with LUCC2020, and the maximum water depth was 32.30% higher. Increased water depth was found in the river section from Changshou to Shuangjiang Station in the mainstream of the Yangtze River and the Yulinhe, Longxihe, and Quxi basin. In contrast, the Qijiang, Wubuhe, Lixiangxi basin, and the northwest region of the TGRR were found to have decreased water depth.

For Flood Event 9 (peaked on 10/07/2017), the distribution of water depth difference is similar to Flood Events 7 and 8. In LUCC2015 scenarios, the average water depth was reduced by -0.64% compared with LUCC2020, and the maximum water depth was over 1000% higher. The maximum water depth difference is 4.71 cm.

4 Discussions

4.1 Comparison with previous studies

Previous studies revealed the flow trend in UYR (Upper Yangtze River, including the TGR and the upstream region), and explained the reasons for this decline due to climate change, human water consumption, and reservoir impoundment (Wei et al. 2014; Shi et al. 2022). According to our results, both the runoff and water depth in the TGRR and the UYR exhibited an increasing trend in 2015–2018 which could be attributed to precipitation anomaly. According to our result, streamflow during summer and autumn was characterized by decreasing trends and increasing trends in spring and winter, which is consistent with previous studies (Jiang et al. 2008; Huang et al. 2024). Furthermore, this study reveals that water depth shows the same seasonal pattern.

Previous studies have attempted to reflect hydrological conditions by analyzing only streamflow change in the TGRR. while this study provides a comprehensive evaluation of streamflow and water depth to reflect flood security conditions. Compared to past research (Wang et al. 2019b; Cheng et al. 2024), this study enables the reflection of flooding dynamics of the TGRR explicitly by the flood depth map. Some studies also turned their views on flood dynamics by applying hydrological-hydrodynamic models, while, currently the use one-dimensional hydrodynamic model limited the knowledge (Wang et al. 2003; Zhang et al. 2020). In this study, by introducing the 2D hydrodynamic model, this study could take flood depth mapping that explicitly reflects flood heterogeneity. This method has been widely adopted in previous studies in the LMRB and other basins to reflect flood dynamics (Kim et al. 2021; Sun et al. 2024).

The impoundment and land cover changes impact hydrologic processes (Wang et al. 2019a). This study assessed the impact of LUCC on hydrological features in TGRR by designing scenarios. During the period spanning from 2015 to 2020, notable land cover changes within the TGRR, attributed to urbanization and the Cropland to Forest Conversion policy, include an increase in urban and built-up areas (+ 0.6%) and water bodies (+ 0.1%), a decrease in woodland-grassland (-0.7%) and cropland (-0.1%). This led to slight increasing runoff and inflow of the TGR, and reduced daily mean and maximum water depth in this whole region (particularly in the flood seasons). In addition, the water depth difference between LUCC2015 and LUCC2020 scenarios shows evident spatial variations and it varies with flood events. Reductions in water depth are observed in the northeast region (s.t. Qijiang, Wubuhe, Lixiangxi basins), particularly in the river section from Cuntan Station to Qingxi Field Station in the mainstream of the Yangtze River. Conversely, increased water depth during LUCC is noted in the northern region (Yulin, Xiaojiang, and Huangjin subbasins).

4.2 Impacts of LUCC on hydrological processes in the TGRR

The “Grain for Green Project” is a country-wide ecological program to convert marginal cropland to forest, which has been implemented in China since 2002. As an important ecological zone, the TGRR has adopted this policy to reserve forest resources and ecological carrying capacity. Previous research discussed the impact of policy implementation which leads to increasing vegetation on hydrological processes (Han et al. 2022; Lan et al. 2024). In the Guansihe basin in Mianyang, Sichuan province, cropland conversion to forests significantly decreased runoff (Yu et al. 2015). Increased significantly vegetation cover area due to the implementation of the GGP leads to reducing blue water and surface runoff while increasing green water (Han et al. 2022). Large-scale vegetation greening leads to national hydrological drought (Lan et al. 2024). Contrastingly, the arid Loess Plateau experienced a decreasing runoff after this measure’s implementation, while the flow in the upper and middle sub-basins increased by about 15% and 20% (Qiu et al. 2011).

In this study, based on 2015 and 2020 land use data, this study designed scenarios with different land uses and analyzed the effect of land use change on streamflow and water depth. Results showed that land use change in the area led to increasing runoff, as opposed to previous studies. Expanding construction land may explain this trend: from 2015 to 2020, the region witnessed the most rapid urban development, and the most noticeable change in land use was a conversion of cropland and grassland to urban. Also, the dominant cover type changed from cropland to urban and built-up in the Changtang River basin. There are three subbasins where the main land cover type changes from cropland to grassland, urban area. grassland to cropland. As a result, despite the growing forest leading to a redistribution of surface and subsurface runoff and a subsequent decrease in runoff volume, the expansion of impervious surfaces within the interval overwhelmed these effects, resulting in increasing runoff (Zhou et al. 2013).

4.3 Limitations and applicability

This study tested responses of hydrologic variables to land use change and changing climate, yet there remains room for improvement. To enhance the precision of simulating hydrological characteristics within the TGRR, future studies could incorporate precise rainfall data with higher resolution, both temporally and spatially. While calibrated models yielded satisfactory results and parameter uncertainty was estimated, it should be emphasized that the calibration methods used in this study were relatively simple. Adapting state-of-the-art calibration techniques could lead to better performance. Furthermore, collecting data from more hydrological stations and long periods could significantly improve regional parameter estimation. In this study, the assessment of simulated water depth is qualitative, and the focus remains on investigating relative water depth both in time and space. Obtaining the topographic data of the measured area to calibrate the hydrodynamic model (including hydraulic parameters) can obtain a finer and more accurate distribution of bathymetry within the region and enable the quantitative analysis of water depth during flood events. Despite the limitations, this study could reflect changes in streamflow and water depth during 2014–2018 and assess flood security conditions in the context of climate change, LUCC, and rapid urbanization.

Coupling hydrological-hydrodynamic models to simulate hydrological characteristics and reflect flood dynamics can be adopted in other regions. It’s noted that considering that runoff is an important input that determines simulation performance. In this study, VIC is selected considering that it is suitable in humid basins, and has been examined its validity in this region. the hydrological model selection and parameter estimation are essential.

5 Conclusions

Runoff in the TGRR has a significant impact on the flood control and safety of the Yangtze River basin. However, due to the lack of measured data in the region, the understanding of the hydrological characteristics of this region is limited. In this study, the hydrological characteristics of the TGRR are simulated using the VIC and CaMa-Flood models. Calibrated parameters performed reasonably well in representing the natural and observed discharge in the region. The NSE for daily discharge is 0.81, 𝑅𝑒 is -2%, and the R-square is 0.93. Simulations mostly overestimated the peak flows (relative error ranges from − 2.5 to 40.0%), and the bias in flood peak occurrence time is 2 to -5 days.

Increasing rates of monthly average, maximum, and minimum flow were 124.56-8.40m3/s/a from 2015 to 2018. The monthly average flow exhibits a seasonal trend, initially increasing and later decreasing throughout the year, aligning with observations at Yichang station. Results show an increase in runoff in the TGRR during 2015–2018, accompanied by a decrease in runoff during the flood season (-146.47 m3/s/a~-87.60 m3/s/a). Between 2015 and 2018, the TGRR observed an increase in water depth, with change rates for the average depth spanning from 0.11 cm/a to 0.47 cm/a and for the maximum depth, fluctuating between 0.08 m/a and 0.34 m/a. While flood season saw a decrease in average water depth, change rates varied from − 0.21 cm/a to -0.61 cm/a.

Based on 2015 and 2020 land use data, this study designed scenarios with different land use maps. From 2015 to 2020, the region witnessed the most noticeable change in land use was a conversion of cropland and grassland to urban. The dominant cover type changed from cropland to urban and built-up in the Changtang River basin. There are three subbasins where the main land cover type changes from cropland to grassland, urban area. grassland to cropland. Results showed that land use change caused a slight increase in runoff and a 7.70% decrease in the multi-year average water depth, while an increase in water depth was observed in the northern TGRR.