Open AccessArticle

An Investigation into the Applicability of the SHUD Model for Streamflow Simulation Based on CMFD Meteorological Data in the Yellow River Source Region

Tingwei Bu

^1,2,3,

Chan Wang

^1,2,*,

Hao Chen

^1,2

Xianhong Meng

^1,2

Zhaoguo Li

^1,2,

Yaling Chen

^1,2,3,

Danrui Sheng

^1,2,3 and

Chen Zhao

^1,2,3

Key Laboratory of Cryospheric Science and Frozen Soil Engineering, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou 730000, China

Field Scientific Observation and Research Station of Climate and Environment in the Yellow River Source Region, Lanzhou 730000, China

University of Chinese Academy of Sciences, Beijing 100049, China

Author to whom correspondence should be addressed.

Water 2024, 16(24), 3583; https://doi.org/10.3390/w16243583 (registering DOI)

Submission received: 25 October 2024 / Revised: 30 November 2024 / Accepted: 10 December 2024 / Published: 12 December 2024

(This article belongs to the Section Hydrology)

Download

Browse Figures

Figure 1
Distribution of the Yellow River source region, river system, and the geographic locations of observation stations. "> Figure 2
The unstructured SHUD coarse/fine mesh for the Yellow River source region generated by the rSHUD tool. "> Figure 3
Flow duration curves (a), scatter plot (b), and hydrograph processes (c) of daily observed and simulated streamflow at the Tangnaihai hydrological Station. "> Figure 4
Flow duration curves (a), scatter plot (b) and hydrograph processes (c) of monthly observed and simulated streamflow at the Tangnaihai hydrological Station. "> Figure 5
Hydrographs and scatter plots of daily observed and simulated streamflow at the Tangnaihai hydrological station for 2008 (a,b) and 2014 (c,d). "> Figure 6
Monthly scale (a) and annual scale (b) temperature, precipitation, and observed and simulated streamflow at Tangnaihai hydrological station from 2006 to 2018, with temperature and precipitation as the annual averages from CMFD. "> Figure 7
Hydrographs (a,c,e) and scatter plots (b,d,f) of daily observed and simulated streamflow at hydrological stations in the Yellow River source region: (a,b) Jimai station, (c,d) Maqu station, and (e,f) Jungong station. "> Figure 8
Monthly average values of observed and simulated streamflow (a) and error percentage for simulated streamflow during warm and cold seasons (b) at four hydrologic stations in the Yellow River source region. "> Figure 9
Comparison of precipitation on daily (a), monthly (b), and annual (c) scales between meteorological stations and the CMFD in the Yellow River source region. ">

Versions Notes

Abstract

The simulator for hydrological unstructured domains (SHUD) is a cutting-edge, distributed hydrological model based on the finite volume method, representing the next generation of coupled surface–subsurface hydrological simulations. Its applicability in high-altitude, cold regions covered by snow and permafrost, such as the Yellow River source region, necessitates rigorous validation. This study employed the China Meteorological Forcing Dataset (CMFD) to simulate streamflow in the Yellow River source region from 2006 to 2018, comprehensively assessing the suitability of the SHUD model in this area. The SHUD model excels in simulating monthly streamflow in the Yellow River source region, while its performance at the daily scale is comparable to existing models. It demonstrated significantly better performance in the warm season compared to the cold season, particularly in the middle and lower reaches of the region. Distinct seasonal and regional differences were observed in simulation performance across sub-basins. However, the model encounters limitations when simulating the extensively distributed permafrost areas in the upstream region, primarily due to oversimplification of the permafrost thawing and freezing processes, which points the direction for future model improvements. Additionally, the model’s shortcomings in accurately simulating peak streamflow are closely related to uncertainties in calibration strategies and meteorological data inputs. Despite these limitations, the calibrated SHUD model meets the hydrological simulation needs of the Yellow River Source Region across various temporal scales, providing significant scientific reference for hydrological simulation and streamflow prediction in cold regions with snow and permafrost.

Keywords:

SHUD model; cold region hydrology; Yellow River source region; streamflow simulation; CMFD data

1. Introduction

Hydrological models serve as pivotal tools in exploring the hydrological cycle, playing an indispensable role in hydrological simulation, prediction, water resource allocation, and management [1]. These models not only support agricultural production within basins but also play an integral part in water resource management, disaster prevention and mitigation, pollution control, and the high-quality development of the socio-economy [2]. With the advancement of technology, the coupled simulation of surface–subsurface hydrological processes has become a significant trend in the development of hydrological models [2,3,4]. The simulator for hydrologic unstructured domains (SHUD) model, as a multi-process, multi-scale, distributed hydrological model with flexible spatiotemporal resolution, employs the finite volume method to solve coupled surface–subsurface hydrological processes within watersheds. This model, renowned for its multi-process and multi-scale characteristics, along with its adaptable spatiotemporal resolution, uses the finite volume method to accurately depict hydrological physical phenomena within basins. The SHUD model integrates multiple processes such as surface streamflow, snowmelt and snow accumulation, evapotranspiration, groundwater flow, and river evolution and supports bidirectional interaction calculations between surface and groundwater. It has been applied in various basins across different spatial and temporal scales and has demonstrated exceptional performance in regions such as the Vauclin catchment experiment, V-shaped basins, the Cache Creek basin in North America, and the Qinghai Lake basin [5,6,7].

The Yellow River source region, located in the northeastern part of the climate-sensitive Tibetan Plateau, is the main streamflow-producing area and water conservation area of the Yellow River basin. Although the area of the source region accounts for only 15% of the total area of the Yellow River basin, its contribution to the annual streamflow volume of the Yellow River basin exceeds 30%, playing a crucial role in maintaining the water supply of the Yellow River basin [8,9]. Various models have been used to simulate hydrological processes in the Yellow River source region, including lumped models such as the Xinanjiang model [10], HBV [11], semi-distributed hydrological models such as the soil and water assessment tool (SWAT) [12], and distributed hydrological models such as the geomorphology-based ecohydrological model (GBEHM) [13,14] and the weather research and forecasting model hydrological modeling system (WRF-Hydro) [15,16]; land surface process models, such as the community land model (CLM) [17,18,19] and the Noah land surface model (Noah LSM) [20], etc. However, many existing models often omit or simplify the glacial/snow accumulation and ablation, as well as the permafrost freezing and thawing processes when simulating the cryospheric hydrological processes unique to high-altitude basins, leading to limitations in the applicability and accuracy of the models. Therefore, some scholars have been committed to improving these processes within the models to enhance their applicability and accuracy [12,14,20,21,22]. Appendix A Table A1 summarizes an overview of some hydrological models applied to the streamflow models in the Yellow River source region, including both models directly applied to the area and those adjusted with module modifications or optimized parameterization schemes. Although these models have achieved some success in simulating the characteristics of streamflow changes over time, there are still biases in simulating peak flow rates, and the simulation effects at different stations also show differences. The general trend of model simulations is to underestimate the observed peak streamflow and overestimate the observed low flow, reflecting the inevitable errors in the simulation process. When selecting and applying hydrological models, it is necessary to consider the applicability of the model, parameter calibration, data quality, and model complexity comprehensively. By continuously optimizing the model and adjusting parameters, the accuracy and reliability of the model in simulating streamflow in the Yellow River source region can be improved. In terms of monthly scale simulation, the SWAT semi-distributed hydrological model has demonstrated its superiority in application in the Yellow River source region.

Surface–subsurface process-coupled distributed numerical hydrological models represent an innovative branch in the field of hydrological modeling. Compared to traditional distributed hydrological models, these models employ numerical methods to tightly couple surface and subsurface hydrological processes, capturing hydrological dynamics under complex topography and geomorphology with greater precision and effectively expressing the spatial heterogeneity of watersheds [2]. Although such models represent the cutting edge of hydrological model development, their applicability in complex mountainous regions with cold climates remains relatively understudied. The Qinghai–Tibet Plateau and its surrounding cold mountainous areas, with their unique snow, ice, permafrost, and complex topography, significantly impact regional hydrological cycles. Utilizing surface–subsurface process-coupled distributed numerical hydrological models to conduct streamflow simulation, verification, improvement, and application in these areas is not only a challenging task but also a hot topic in hydrological, atmospheric, and ecological research.

In the field of hydrological simulation, particularly in data-scarce areas, the use of reanalysis data as a substitute for gauged data has become an important means [23]. However, different precipitation datasets can yield varying streamflow simulation effects in hydrological models [16,24,25]. Studies have shown that in the Yellow River source region, climate change is the primary factor affecting streamflow changes, with precipitation being the dominant factor determining streamflow variation, and the sensitivity of streamflow to precipitation changes is higher than its sensitivity to temperature changes [26,27]. Precipitation data from the China Meteorological Administration’s CMFD have shown good performance in the Qinghai–Tibet Plateau and the Yellow River source region, with CMFD’s precipitation volumes being closest to the annual and seasonal observed values in the area compared to data from the Climatic Research Unit (CRU) and European Center for Medium-Range Weather Forecasts Reanalysis 5th Generation (ERA5) [28,29,30,31,32].

This study utilizes CMFD data to drive the SHUD model, assessing the applicability of the SHUD model in the Yellow River source region. By constructing and automating the optimization of SHUD model parameters applicable to the Yellow River source region, this study is dedicated to evaluating the model’s effectiveness in simulating hydrological processes in this area and conducting an in-depth analysis of its strengths and limitations. The results not only provide significant reference for the application of the SHUD model or similar numerical hydrological models in cold mountainous areas such as the Yellow River source but also promote further verification and promotion of the SHUD model in this region, providing empirical support for the model’s continuous improvement. Furthermore, the calibrated SHUD model, suitable for the Yellow River source region, will become a powerful tool to help us understand the patterns and characteristics of streamflow changes in this area under the background of climate change.

This study aims to rigorously evaluate the applicability of the SHUD model in the Yellow River source region. The specific objectives are as follows:

(1): To construct and calibrate the SHUD model for streamflow simulation in the Yellow River source region from 2006 to 2018 using the China Meteorological Forcing Dataset (CMFD) and to assess the model’s performance at daily and monthly scales.
(2): To investigate the seasonal and spatial variability in the SHUD model’s performance across three sub-basins, with a focus on differences between warm and cold seasons and among upstream, middle, and downstream regions.
(3): To examine the model’s capability to simulate high and low streamflow extremes and to identify limitations in its representation of key processes, such as permafrost dynamics and snowmelt.
(4): To analyze uncertainties associated with input data quality, model structure, parameter settings, and calibration strategies and to propose potential directions for model improvement.

The structure of this paper is as follows: Section 1 introduces the research background, including the importance of hydrological models, the development and application of the SHUD model, and the hydrological significance of the Yellow River source region, along with the challenges and objectives of this study. Section 2 provides details on the SHUD model, the study area, the data utilized, and the model construction and calibration process. Section 3 presents the results and discussion, including a detailed analysis of the model’s performance at different temporal and spatial scales, seasonal variations, and associated uncertainties. Finally, Section 4 summarizes the main findings, evaluates the SHUD model’s applicability in the Yellow River source region, and outlines potential directions for future model improvements and applications.

This study is essential for understanding the dynamic changes and hydrological processes of wetlands, particularly the Zoige Wetland on the Qinghai–Tibet Plateau, using the SHUD model, which has not been previously applied in this region. The SHUD model was chosen for its ability to accurately describe the physical hydrological processes in basins, especially those with complex terrain and unobserved characteristics, and its fully coupled nature ensures the continuity and consistency of hydrological simulations. Despite other models showing high R² and NSE values, the SHUD model’s physically based approach and emphasis on process accuracy, including extreme event simulations and spatial heterogeneity, make it particularly suitable for studying the intricate hydrological dynamics of wetlands.

2. Model and Data

2.1. SHUD Model

The simulator for hydrologic unstructured domains (SHUD) model, distinguished by its multi-scale, multi-process, high temporal resolution, and flexible spatial resolution, has been successfully applied to simulate hydrological processes across various hydroclimatic regions and multiple spatial and temporal scales. These applications range from hourly to centennial timescales and from 1 m² to 10⁵ km² spatial scales. The SHUD model has been implemented in numerous watersheds across North America, Asia, and Africa, encompassing studies on lakes, flooding, and water resources [5,33,34,35].

Compared to traditional hydrological models, the SHUD model employs numerical methods that offer significant advantages in terms of spatiotemporal continuity. Regarding temporal discretization, the SHUD model utilizes the finite volume method to solve the ordinary differential equations of watershed hydrological processes. This includes the simulation of surface water flow, unsaturated and saturated zone flows, and river channel flow, enabling precise calculations of water storage and flux across spatial domains. In terms of spatial discretization, the model employs unstructured triangular grids to represent hillslopes and rivers, with the vertical dimension comprising the soil layer, unsaturated zone, and saturated zone. The SHUD model simulates processes such as vegetation interception, snow accumulation/melt, evapotranspiration, infiltration, groundwater recharge, surface streamflow, baseflow, and two-way flow in river channels. Within the model, vegetation canopy interception is based on the bucket model, snow is addressed using the degree-day model, and potential evapotranspiration is calculated using the Penman–Monteith equation. Surface streamflow and river flow are computed using simplified Saint–Venant equations, while water movement in the unsaturated and saturated zones is simulated using the Richards equation and Darcy’s law, respectively. The model iteratively solves the water states in the surface, unsaturated zone, saturated zone, and river channels at each time step [5,34].

In this study, we employed the SHUD (2.0) model ’s full suite of modules to simulate the hydrological cycle comprehensively, from precipitation to evapotranspiration and discharge, without selectively choosing specific modules. The model’s automatic invocation of all relevant modules ensures an integrated simulation of the watershed’s hydrological processes. The SHUD model’s coupled physical mechanisms cover processes such as canopy interception, surface runoff generation and flow, infiltration and unsaturated zone flow, groundwater flow, river flow, and evapotranspiration. This seamless integration through coupled physical equations captures the consistent interactions within the water cycle. Moreover, the SHUD model’s robust two-way coupling strategy accurately represents vertical and horizontal exchanges between surface water and groundwater, enabling dynamic flow simulations based on real-time hydraulic conditions. This mechanism allows for a multi-scale representation of hydrological dynamics, enhancing the model’s applicability to complex watershed simulations.

Selecting an appropriate spatial discretization method is of paramount importance for ensuring the precision of simulations within the study area and for meeting boundary conditions. Moreover, it significantly reduces the number of hydrological computational units, thereby diminishing the demand for computational resources [36]. The SHUD model employs the rSHUD tool (2.0.1) for spatial discretization of the watershed and preparation of input data, which incorporates the covariance matrix adaptation evolutionary strategy (CMA-ES) for automated parameter optimization. This approach enhances the precision and efficiency of the hydrological model [5,7,34,37]. Within the R programming environment, the open-source toolkit rSHUD is utilized to process spatial data, time series data, and attribute data of the study area. The AutoSHUD program facilitates the extraction, transformation of data, and generation of attribute files, thereby constructing the input files for the SHUD model [5,7,34].

2.2. Study Area

The Yellow River source region, which encompasses the area above the Tangnaihai hydrological station on the main stem of the Yellow River, is nestled in the northeastern part of the Qinghai–Tibet Plateau, with a total area of approximately 123,000 km². The geographical coordinates extend from 32°09′ N to 36°07′ N latitude and from 95°53′ E to 103°24′ E longitude. The topography of this region is characterized by a general elevation that decreases from the west to the east, with significant gradients, and is predominantly composed of alpine terrain, plains, and grasslands, interspersed with lakes, wetlands, glaciers, seasonal frost, and permafrost, among other natural features. The climate of the Yellow River source region is classified as a continental alpine humid type, marked by alternating cold and warm seasons and a distinct separation between dry and wet seasons [38,39]. Between 1979 and 2018, the multi-year average annual temperature recorded for this region was 0.79 °C, with a gradual temperature increase from the northwest to the southeast. The long-term average annual precipitation is approximately 520 mm, predominantly occurring from July to October. Influenced by the terrain and the monsoon climate, precipitation is less than 300 mm in the northwestern part of the Yellow River source basin, while it is more concentrated in the southeastern low-altitude areas. The dry season spans from December to April of the following year, and the main streamflow period extends from May to November, with the Tangnaihai hydrological station recording that the streamflow from June to September accounts for approximately 85% of the annual total [40,41]. As depicted in Figure 1, this study designates three hydrological stations within the Yellow River source region—the Jimai station, Maqu station, and Tangnaihai station—as pivotal nodes, dividing the entire basin from upstream to downstream into three sub-basins: The source area–Jimai, Jimai–Maqu, and Maqu–Tangnaihai. This delineation facilitates a more nuanced investigation of the hydrological characteristics of each sub-basin and their respective contributions to the overall hydrological processes of the basin.

2.3. Data Description

The foundational data for constructing the SHUD model encompasses two principal categories: spatial data and meteorological driving data. The spatial data include elevation, land use, soil texture, river networks, and basin boundaries, while the meteorological driving data comprise wind speed, temperature, air humidity, atmospheric pressure, precipitation, and radiation, among others (Table 1).

The elevation data utilized are derived from the advanced spaceborne thermal emission and reflection radiometer global digital elevation model (ASTER GDEM), with a spatial resolution of 30 m. These high-precision elevation data lay a solid foundation for topographical analysis and hydrological simulation [42].
The land use data are sourced from the USGS Land Cover Institute (LCI) global land use data product, based on MODIS data with a resolution of 0.5 km. This dataset covers 17 primary land cover types [43], with the Yellow River source region’s land use types predominantly consisting of grasslands, followed by water bodies. The land use factors remain constant over the simulation period.
The soil data are obtained from the Harmonized World Soil Database (HWSD), with data within the Chinese territory provided by the Nanjing Institute of Soil Science at a scale of 1:100 [44]. The soil and geological factors remain constant over the simulation period.
The meteorological driving data are based on the China Meteorological Forcing Dataset (CMFD), a high spatiotemporal resolution surface meteorological driving dataset for the China region, covering the period from 1979 to 2018, with a temporal resolution of 3 h and a spatial resolution of 0.1°. This dataset includes seven variables: precipitation, temperature, atmospheric pressure, specific humidity, wind speed, shortwave radiation, and longwave radiation, providing comprehensive meteorological information for simulating hydrological processes [45,46].
The station precipitation data are sourced from the daily value data of the China Ground Climate Data (V3.0) for nine meteorological stations in the Yellow River source region, including Maduo, Dari, Maqin, Jiuzhi, Hongyuan, Zoige, Maqu, Henan, and Xinghai. These data provide a benchmark of actual observations, aiding in the assessment of the accuracy of gridded data.
The streamflow data are derived from the daily streamflow records from 2006 to 2018 at four hydrological stations in the Yellow River source region—Jimai, Maqu, Jungong, and Tangnaihai—as documented in the hydrological yearbook. These streamflow data serve as an essential basis for evaluating the effectiveness of the model’s simulation.

2.4. SHUD Model Construction

This study employs the ensemble SHUD model, the rSHUD tool, and the AutoSHUD automated hydrological simulation system to accomplish data preparation, model construction, and result analysis for the Yellow River source region (refer to the structural flowchart in the literature by Shu [7]). The following outlines the detailed steps and methodologies:

1.: Data Processing

Initially, we constructed data subsets and performed regularization treatments, which included preprocessing of catchment boundaries, digital elevation models (DEMs), and river network data to ensure the accuracy and consistency of the data. Subsequently, we processed soil/geological and land use data to extract key hydrological parameters and formatted meteorological data. This involved extracting information from CMFD data, unit conversion, and serialization, thereby providing the necessary inputs for the model.

2.: Spatial Discretization

This study constructs an unstructured grid for the Yellow River source region using the rSHUD model, as depicted in Figure 2. In the Zoige area and its vicinity, smaller triangular grids are generated to enhance representation precision, while larger triangular grids are employed in areas beyond Zoige to ensure computational efficiency. To delve deeper into wetland issues, localized grid refinement simulations were conducted without adversely affecting the overall simulation. The focus of this study is to verify the computational efficacy of the model, with detailed exploration of wetland issues to be addressed in separate research endeavors.

The simulation of the Yellow River source region encompasses 4038 triangular units, with the encrypted area containing 1902 grid units, averaging approximately 8.82 km² per triangle; the non-encrypted area’s triangles average about 49.89 km². The river network comprises 1266 rivers, extending to a total length of 8037 km. This refined grid design affords the model with the necessary spatial resolution to capture the intricate hydrological dynamics within the study area.

3.: Model Calibration

The optimization of over 50 parameters of the SHUD model was conducted using the CMA-ES algorithm (detailed in Appendix A Table A2), aiming to achieve a globally optimal solution. The specific optimization process is as follows:

(1): Parameter Initialization: Based on the soil, geological, and land use data of the Yellow River source region, initial parameter values are set for each triangular grid of the model to ensure these values reflect the region’s topography, soil, and vegetation characteristics.
(2): Parameter Adjustment Range Setting: Based on experience and related studies, an adjustment range is set for each parameter to simulate the possible variations of the parameters under actual conditions.
(3): Parameter Sampling and Simulation: Random sampling is conducted within the parameter space to generate multiple sets of parameter collections for driving the SHUD model for simulation.
(4): Simulation Result Evaluation: The simulated streamflow data are compared with the actual data, and the Nash efficiency coefficient (NSE) is used as the objective function to evaluate the simulation effectiveness.
(5): Covariance Matrix Update: The covariance matrix of the parameter distribution is updated according to the simulation results to guide the selection of subsequent parameter collections.
(6): Iterative Optimization: The sampling and evaluation process is repeated until the NSE value reaches the preset target threshold, ultimately determining the optimal parameter set.

This study selects the daily scale efficiency coefficient as the key evaluation criterion for parameter calibration. Extensive research has found that the precision of daily scale streamflow simulation often falls short compared to monthly scale simulation, which aligns with the conclusions of the existing literature [47,48,49,50,51,52]. In the calibration process, the streamflow data from the Tangnaihai hydrological station from 2006 to 2010 are used as the calibration target, with 2000 to 2005 serving as the model’s spin-up period, 2006 to 2010 as the calibration period, and 2011 to 2018 as the validation period. The discussion and verification of the simulation results utilize the streamflow data from four hydrological stations: Jimai, Maqu, Jungong, and Tangnaihai, with Tangnaihai being the final outlet of the Yellow River source region, and its streamflow simulation results play a decisive role in assessing the overall simulation effectiveness of the SHUD model in the Yellow River source region.

4.: Model Execution

(1): Parameter Settings: Parameters required for the AutoSHUD modeling include the maximum computational unit area, basin boundary simplification threshold, river simplification threshold, simulation days, and the number of computational units, etc.
(2): Basin Unit Division and Grid Generation: The AutoSHUD program automatically prepares buffers, divides basin units, and generates unstructured triangular grids based on these parameters, while calculating the area and grid number of each unit.
(3): River Network and Meteorological Station Processing: River network processing includes river classification and flow direction verification, while meteorological station coverage adopts the Thiessen polygon method.
(4): Spatial Attribute Generation: After matching spatial data with the computational grid, the program generates spatial attributes for each computational unit, such as soil classification, geological classification, land use classification, and meteorological station codes.
(5): Hydraulic Parameter Calculation: Information such as riverbed slope is calculated from the river file and DEM-derived data, and the intersection topology relationship between the river and the triangular unit is achieved by the program’s spatial computation functions. Soil- and land-cover-related hydraulic parameters are calculated using the pedotransfer function (PTF) empirical function, such as horizontal and vertical hydraulic conductivities, saturated water content, etc.
(6): Model Initial Conditions and Operation Control Parameter Generation: Ultimately, the AutoSHUD program automatically generates model initial conditions and operation control parameters based on the number of model units and modeling parameter settings, completing the preparatory work for the SHUD model construction.

2.5. Model Evaluation Metrics

To comprehensively assess the performance of the model in streamflow simulation, this study employs a variety of evaluation metrics [53,54], each of which evaluates the model’s performance from different perspectives. These metrics are complementary and provide a multi-dimensional analysis of the model’s various characteristics, as shown in Appendix A Table A3.

Coefficient of Determination (R²): This is a statistical metric that measures the goodness of fit of the linear relationship between the model’s predictive values and the observed data, with values ranging from 0 to 1. The closer R² is to 1, the better the model’s predictive performance [55]. A high R² value indicates that the model’s predicted trends are highly consistent with the actual observed data, suggesting that the model excels at fitting the overall trend of the data. Although an R² value above 0.5 is considered acceptable, its major drawback is oversensitivity to extreme values. Consequently, R² is more commonly used in conjunction with regression plots for display purposes rather than for the evaluation or optimization of hydrological models.
Kling–Gupta Efficiency (KGE): KGE is a comprehensive evaluation metric that takes into account the bias, variance, and correlation between the model’s predicted values and the observed values. KGE ranges from $- \infty$ to 1, with KGE = 1 indicating perfect agreement between model predictions and observed data [56]. By integrating these critical dimensions, KGE offers a nuanced benchmark for evaluating model efficacy, adeptly balancing the assessment of errors under low-flow and high-flow scenarios. This metric is particularly adept at providing a comprehensive evaluation of model performance, ensuring a fair appraisal that is not skewed by extreme values.
Nash–Sutcliffe Efficiency (NSE): NSE is a dimensionless metric used to evaluate the degree of fit between the model’s predicted values and the observed values, interpreting the model’s residuals. NSE ranges from $- \infty$ to 1. When NSE reaches 1, it indicates perfect consistency between the model’s predictions and the observed values. If the NSE value is between 0.5 and 1, the model’s simulation effect is generally considered satisfactory. However, when the NSE < 0, it typically indicates that the model’s simulation effect is not satisfactory, and further adjustments or improvements to the model may be necessary. However, NSE is overly sensitive to peak flows and tends to underestimate variability [57,58].
Percent Bias (PBIAS): This metric assesses the percentage bias between the model’s predicted values and the average of the observed values, indicating whether the model systematically overestimates or underestimates the observed data. The closer the PBIAS value is to 0, the smaller the model bias and the better the model’s simulation effect.
Root Mean Square Error (RMSE): RMSE is a standard measure of the differences between model predictions and observations. It is sensitive to larger deviations, with higher values indicating greater discrepancies. Lower values suggest higher prediction accuracy. However, RMSE primarily reflects prediction precision and does not reveal performance in trend capture or variability. When used in conjunction with other metrics like R² or NSE, RMSE can provide a more comprehensive view of model performance.

In the assessment of model performance, the monthly scale standards proposed by Moriasi et al. [59] are referenced as follows: when

NSE \leq 0.5

, the simulation results are considered unacceptable; when

0.5 < NSE \leq 0.65

and

\pm 15 % < PBIAS < \pm 25 %

, the simulation results are considered acceptable; when

0.65 < NSE \leq 0.75

and

\pm 10 % < PBIAS < \pm 15 %

, the simulation results are considered good; and when

0.75 < NSE \leq 1.0

and

PBIAS \leq \pm 10 %

, the simulation results are considered excellent [59]. These evaluation metrics are used for the main analysis, with other indicator parameters used for auxiliary analysis. At the same time, we also analyze the performance of the SHUD model on a daily scale, warm and cold seasons, annual scale, and different regions and stations to comprehensively evaluate the model’s simulation effects.

3. Results and Discussion

3.1. Streamflow Simulation Results

3.1.1. Daily/Monthly Scale Streamflow Simulation Performance

Figure 3 and Figure 4 display the performance of the daily/monthly scale streamflow simulated at the Tangnaihai hydrological station during the calibration and validation periods, utilizing the optimal parameter set obtained from automated parameter optimization. By comparing the observed and simulated flow duration curves (FDC), the model’s predictive capability across various flow ranges can be assessed, with the x-axis of the FDC curves presented on a logarithmic scale for a more intuitive representation of high-flow conditions.

From the daily scale FDC curve in Figure 3a, it is observed that the simulated results have a lower degree of agreement with the actual observed FDC curves at both low- and high-flow extremes, exhibiting an underestimation of peak and valley flows. Nonetheless, the simulated outcomes adequately reflect the overall trend of the observed streamflow FDC curves.

Figure 3b illustrates that the scatter plot fit line of the simulation closely approximates the 45° line, indicating a strong correlation between the simulated and observed results. However, the slope of the fit line is 0.8, with the daily scale observation–simulation fit line equation being y = 0.8x + 75.4, suggesting that the model’s simulated values are slightly lower than the observed values, with this underestimation becoming increasingly apparent as the observed flow increases.

The monthly FDC curve in Figure 4a demonstrates improved agreement between simulated and observed values in both low- and high-flow extremes, indicating that the processing of monthly scale data may mitigate the stochastic fluctuations in streamflow data, bringing the simulated and observed data closer. The fit line equation in Figure 4b is y = 0.863x + 36.2, with a slope increase compared to the daily scale but still below 1, indicating that the monthly scale simulated values are still slightly lower than the observed values, though the underestimation is reduced compared to the daily scale.

Figure 3c and Figure 4c present the hydrographs of simulated and observed streamflow. The simulated results are fundamentally consistent with the observed data in terms of trends, indicating that the model can effectively reflect the characteristics of observed streamflow changes throughout the simulation period. The model’s ability to capture the general trend of streamflow fluctuations, with the timing of peak flow occurrences aligning well with the observed results, suggests that the SHUD model can adeptly capture the variability of streamflow. The simulated results are fairly consistent with observed outcomes during both high- and low-flow periods, indicating that the SHUD model can accurately simulate the alternating patterns of low- and high-flow seasons within the streamflow process.

Synthesizing the daily and monthly scale simulation results, the SHUD model adequately captures the trends and amplitudes of streamflow changes during both the calibration and validation periods. Although there is a certain degree of underestimation at high-flow conditions, the model performs well overall, providing reliable simulation results for the study of hydrological processes in the Yellow River source region.

3.1.2. Daily/Monthly Scale Streamflow Simulation Performance Statistics

A statistical analysis was conducted on the daily streamflow simulation performance at the Tangnaihai hydrological station from 2006 to 2018, encompassing five key metrics: R², KGE, NSE, PBIAS, and RMSE (see Table 2). The analysis revealed that the model effectively simulated the streamflow trends in the majority of the years, with 77% of the years having an NSE exceeding 0.65, 70% of the years with a KGE greater than 0.7, and an average R² of 0.79.

The R² values fluctuated between 0.65 and 0.89 from 2006 to 2018, with an average of 0.79, indicating that the SHUD model accurately captured the overall trend of streamflow on a daily scale. In years with superior performance, such as 2009 and 2014, the R² values reached 0.89 and 0.87, respectively, demonstrating a robust trend-fitting capability. However, the R² values for 2015 and 2006 were comparatively low at 0.65 and 0.71, respectively, suggesting that the model’s trend-capturing ability was weaker in those years. The KGE values exhibited a broad range of variation, from negative values to 0.9, reflecting significant differences in the model’s comprehensive performance across different years.

The KGE value for 2014 was 0.9, indicating an exceptional comprehensive performance in terms of bias, variance, and correlation for that year. In contrast, the KGE values for 2006 and 2016 were comparatively low at 0.31 and −0.25, respectively, with the negative value in 2016 indicating a severe deficiency in the model’s predictive capability for that year.

The NSE ranged from −1.9 to 0.82, with an average of 0.47. The NSE values for most years fell between 0.5 and 0.8, indicating that the model could reasonably simulate the fluctuations of the observed data in those years. However, the lowest NSE value in 2016 was −1.9, demonstrating a poor fit to the data fluctuations for that year.

Further analysis of the errors revealed systematic biases in certain years, with the majority of years having negative PBIAS values, indicating a general trend of underestimation by the model. The PBIAS value for 2016 was as high as 55.1%, indicating a systemic overestimation of streamflow by the model in that year. The RMSE ranged from 177.44 to 449.95 m³ s⁻¹, with an average of 272.66 m³ s⁻¹, and in 70% of the years, the RMSE was below 300 m³ s⁻¹, demonstrating a high level of predictive accuracy.

In years with superior performance, the metric values were balanced and high, indicating that the model could comprehensively capture the overall trend of streamflow while maintaining good predictive accuracy and minimal systematic bias. In contrast, in years with inferior performance, such as 2016 and 2006, the model exhibited poor performance in metrics like R², KGE, and NSE, indicating deficiencies in capturing the overall trend and fluctuations of streamflow. The issues in 2016 were more severe, characterized by overestimation bias (extremely high PBIAS) and poor predictive accuracy (high RMSE), whereas in 2006, the overestimation bias (positive PBIAS) was lower, and predictive accuracy (RMSE) was relatively better.

Overall, the model demonstrated high equilibrium in years with superior performance, with excellent performance across all metrics, indicating that the model could effectively capture the trends, fluctuations, and predictive accuracy of streamflow. In years with inferior performance, the model’s performance exhibited significant variability, with individual metrics potentially revealing distinct issues that may have interacted with one another, leading to overall subpar model performance.

The simulation results in 2008 during the calibration period and in 2014 during the validation period were notably outstanding. Figure 5 illustrates the hydrographs for these two years, showcasing a high degree of congruence between the simulated and observed hydrological processes, further corroborating the model’s exceptional performance during these years.

In 2008, the model demonstrated commendable R² (0.83) and KGE (0.81) values, indicating an excellent performance in capturing the overall trend and a comprehensive representation. Compared to 2014, although the KGE value was slightly lower, it still denoted a satisfactory performance in trend capture and bias control. An NSE of 0.76 signifies the model’s efficacy in simulating data fluctuations with a strong predictive capability. A PBIAS of −3.2% suggests a minor underestimation of streamflow by the model, with a negligible bias. The RMSE at 180.71 m³ s⁻¹ indicates a well-performing predictive accuracy within a reasonable margin of error.

In 2014, the model exhibited an outstanding R² value of 0.87, reflecting the precise capture of the overall streamflow trend. A KGE value reaching 0.90 denotes an exceptional performance in terms of bias, variance, and correlation. An NSE of 0.86 illustrates the model’s proficiency in fitting the fluctuations of observed data; the high values of KGE and NSE indicate the model’s ability to not only capture the general trend but also accurately simulate the variability of streamflow. A PBIAS of −7.4% indicates a minor systematic bias towards the underestimation of observed values; the low RMSE (177.44 m³ s⁻¹) signifies a high level of predictive precision.

By considering these metrics holistically, a more comprehensive understanding of the model’s performance can be achieved, identifying its strengths and weaknesses across different years. The superior performance in the favorable years (2014 and 2008) is primarily attributed to the model’s accurate capture of overall trends, provision of high-precision predictions, and maintenance of minor systematic biases. Conversely, the subpar performance in less favorable years (2016 and 2006) is due to significant deficiencies in trend capture, predictive capability, and accuracy, coupled with substantial systematic biases.

The current literature has noted similar issues when utilizing the WRF-Hydro model for simulating monthly streamflow scales in the Yellow River source region from 2009 to 2018 [16]. The discrepancies between observed and simulated values across different years may be the result of a confluence of climatic factors, intrinsic model limitations, and uncertainties in precipitation input data.

To investigate the relationship between streamflow simulation performance and meteorological factors, Figure 6 presents an integrated view of the dynamic changes in temperature, precipitation, and streamflow depth at the Tangnaihai hydrological station in the Yellow River source region from 2006 to 2018. Streamflow depth refers to the depth of water layer obtained by evenly distributing the total streamflow volume over a certain cross-section during the calculation period across the catchment area upstream of that cross-section. Figure 6a focuses on monthly scale data, while Figure 6b provides an annual scale perspective. On the monthly scale, Figure 6a reveals a significant similarity in the trend of changes between streamflow and precipitation, while the synchronicity between temperature and streamflow is relatively weaker. This trend is also reflected in the annual scale data in Figure 6b. Further literature analysis indicates that the impact of precipitation and temperature on streamflow accounts for 64.2% and 25.93%, respectively [60], emphasizing that the sensitivity of streamflow to precipitation is significantly higher than its sensitivity to temperature. This is also consistently supported by other studies [27,61,62,63,64]. When calculating the annual streamflow ratio, we found that the observed streamflow ratio in the Yellow River source region during the study period was 0.28, while the simulated streamflow ratio was 0.25, indicating a certain degree of difference between the two. Especially in 2016, this difference was more pronounced, with the observed annual streamflow ratio being 0.20 and the simulated streamflow ratio being 0.25. This inconsistency between observations and simulations may indicate a significant deviation between the meteorological data input to the model and the actual meteorological conditions.

The SHUD model, while limited in capturing flood peaks in areas like the Yellow River source with complex terrains and sparse data, offers valuable insights during flood seasons. Its inaccuracies mainly stem from precipitation data inaccuracy and an optimization focus that compromises extreme flow simulation. Despite these, the model is effective for analyzing trends, supporting flood assessments, and water resource management, though it requires improvements for precise flood peak predictions.

To enhance the SHUD model’s performance, we recommend improving input data accuracy with higher-resolution satellite and radar data and expanding ground-based observations. Additionally, multi-objective calibration methods should be used to better represent hydrological processes. The model’s applicability should be tailored to the research objectives, focusing on flood-specific metrics for simulation, long-term runoff trends, and annual runoff totals. These efforts will address current limitations and prepare the model for future flood season applications.

Table 3 reveals the daily/monthly scale streamflow simulation results at the Tangnaihai station during the calibration and validation periods. The negative PBIAS, with an absolute value less than 10%, indicates a slight underestimation trend of simulated streamflow compared to the observed values. This trend is consistent with the phenomenon shown in Figure 3b and Figure 4b, where the slope of simulated streamflow is slightly lower than that of observed streamflow. Additionally, NSE values are all above 0.7, and KGE values are basically higher than 0.8, further confirming the high degree of agreement between simulated and observed streamflow in overall trends. Nevertheless, the daily scale RMSE values are 261.6 m³ s⁻¹ and 297.06 m³ s⁻¹, respectively, reflecting a certain degree of average error between simulated and observed values, especially due to significant deviations in some data points leading to higher RMSE values. However, the daily scale R² values are 0.74 and 0.72, respectively, showing a high degree of consistency and similar trends between observed and simulated streamflow. There is a slight decline in model performance from the calibration period to the validation period, but according to the evaluation grade classification standards of [59], the SHUD model’s monthly scale streamflow simulation effect at the Tangnaihai hydrological station is rated as “excellent”, while the daily scale performance is slightly inferior, which is consistent with findings in the existing literature [47,48,49,50,51,52]. Appendix A Table A1 summarizes some hydrological model evaluation indicators applied in the Yellow River source region. Except for a few very small indicators, the daily scale R² and NSE ranges are 0.73 to 0.84 and 0.56 to 0.93, respectively, and the monthly scale R² and NSE ranges are 0.63 to 0.91 and 0.55 to 0.91. The SHUD model’s daily/monthly scale R² and NSE values are all within the aforementioned ranges, showing competitive performance compared to other models. The comprehensive evaluation of five performance indicators further confirms the effectiveness of the SHUD model in effectively capturing key characteristics of observed streamflow processes. Compared with the SWAT model, which has the best performance in the Yellow River source region, the SHUD model’s performance is highly comparable, demonstrating the potential and reliability of the SHUD model in the field of hydrological simulation.

3.1.3. Sub-Basins Streamflow Simulation Performance Statistics

To assess the model’s performance across various watersheds, the Yellow River source region was segmented into three sub-basins, as depicted in Figure 1. Table 4 presents the evaluative metric for the daily/monthly scale SHUD model streamflow simulations at hydrological stations within each sub-basin. Upon analysis of these metrics, it was found that the Maqu station demonstrated the most distinguished simulation outcomes, followed by the Tangnaihai and Jungong stations. Conversely, the Jimai station, situated in the upper reaches of the Yellow River source region, exhibited relatively poorer simulation results. These findings align with the performance outcomes of selected hydrological models at various stations within the Yellow River source region as compiled in Appendix A Table A1 [22,28,65]. During the period from 2006 to 2018, the absolute values of PBIAS for the simulated daily/monthly scale streamflow at the Maqu, Jungong, and Tangnaihai stations were all less than 10%, with NSE values exceeding 0.65/0.75. These results indicate that the SHUD model’s simulation outcomes at multiple hydrological stations are quite reliable.

Figure 7 illustrates the hydrographs and scatter plots for three distinct sub-basins (the source area–Jimai, Jimai–Maqu, and Maqu–Tangnaihai) from 2006 to 2018, demonstrating the model’s satisfactory performance in simulating the daily scale distribution characteristics of the controlled station flows within each sub-basin. The simulated streamflow was largely consistent with the observed streamflow over most periods, effectively reflecting the seasonality and interannual variability of water resources. However, certain biases were observed in the simulation of peak and valley flows. For the source area–Jimai sub-basin (Figure 7b), there was a significant discrepancy between the model’s simulated flows and the observed flows. The slope of the linear regression equation (0.542) indicates an overall underestimation of the simulated flows, with some underestimation in the simulation of peak flows, yet the overall trend remained consistent. The slopes of the linear regression equations for the Jimai–Maqu and Maqu–Tangnaihai sub-basins were 0.896 and 0.914, respectively, both approaching 1, indicating a high degree of congruence between the simulated and observed streamflows.

In terms of daily scale streamflow simulation, the SHUD model’s performance ranking across the three sub-basins was Maqu–Tangnaihai, Jimai–Maqu, and source area–Jimai. This suggests that the SHUD model’s simulation performance in the seasonal frozen soil areas of the Yellow River source region is more ideal, with relatively larger biases in the perennial frozen soil areas. This indicates that the SHUD model’s simulation performance in the middle and lower sub-basins of the Yellow River source region is quite satisfactory, with simulated flows at hydrological stations closely matching the observed values. The model is capable of effectively reflecting the daily scale distribution characteristics of streamflow and the spatial streamflow changes from upstream to downstream.

Despite the SHUD model demonstrating certain deviations in simulating peak and nadir streamflow values, particularly at the upstream hydrological stations where performance lags slightly behind other stations, it still exhibits a commendable congruence with the observed sequences on daily and monthly scales. This is especially pronounced in the prediction of overall trends and cyclic characteristics, enabling the SHUD model to effectively forecast the streamflow processes in the Yellow River source region. The calibrated SHUD model is not only capable of capturing the precipitation–streamflow relationship within the basin but also accurately reproducing the timing of peak streamflow events and the characteristics of dry and wet season variations. This indicates that the SHUD model possesses a high degree of applicability and reliability in streamflow simulation within the Yellow River source region, providing significant scientific underpinnings for water resource management and basin planning. During the simulation process, the SHUD model adeptly reflects seasonal and interannual variations, which is crucial for understanding and predicting the hydrological dynamics of the Yellow River source region. Although there are some limitations, the overall performance of the SHUD model remains satisfactory, particularly in simulating the long-term trends and principal characteristics of streamflow.

3.1.4. High/Low Streamflow Simulation Performance Statistics

The flow duration curve (FDC) is not only effective in reflecting the goodness of fit across various flow ranges but also demonstrates the model’s capability to reproduce the frequency of observed streamflow. The FDC captures the streamflow characteristics of a watershed from low- to high-flow states comprehensively [59,66,67]. To visually present the effectiveness of the SHUD model in simulating different flow ranges of measured streamflow, the flows corresponding to the 5%, 50%, and 95% probabilities, denoted as Q₅, Q₅₀, and Q₉₅, were determined using the FDC. By analyzing the ratios of high and low flows relative to the median flow, Q₅/Q₅₀ and Q₉₅/Q₅₀, we gained an in-depth understanding of the model’s performance in simulating high and low flows.

The main stem hydrological stations established from upstream to downstream along the Yellow River—Jimai, Maqu, Jungong, and Tangnaihai—exhibit increasing streamflow volumes, with corresponding Q₅, Q₅₀, and Q₉₅ values also increasing sequentially. The ratios Q₅/Q₅₀ and Q₉₅/Q₅₀ provide a perspective on the performance of high and low flows from the overall streamflow distribution. Observational data indicate that the Tangnaihai station has the largest Q₅/Q₅₀ and Q₉₅/Q₅₀ ratios, while the Jungong and Jimai stations have the smallest. In contrast, the simulated data show that the Jungong station has the largest Q₅/Q₅₀ and Q₉₅/Q₅₀ ratios, with the Tangnaihai station having the smallest. By comparing the observed and simulated streamflow volumes corresponding to Q₅ and Q₉₅ in Table 5, it was found that the Jimai station underestimated high flows and overestimated low flows during simulation; Maqu and Jungong stations overestimated both high and low flows; while Tangnaihai station underestimated these two flows. Zhang et al. [24] also found that the peak streamflow at the Jimai, Maqu, and Tangnaihai stations was underestimated when applying the SPHY model to simulate streamflow in the Yellow River source region, which may be related to the CMFD data underestimating peak precipitation. The performance in simulating high and low flows at different stations shows significant differences, emphasizing the necessity for a detailed assessment of model performance to more accurately simulate and understand streamflow dynamics.

3.1.5. Intra-Annual Distribution of Streamflow Simulation

Figure 8a further illustrates that the simulated multi-year average monthly streamflow levels and trends closely align with the observed values. With the exception of the Jimai hydrological station, the SHUD model performs better in simulating streamflow from May to July and September to December, particularly in capturing peak flows. This indicates that the SHUD model is effective in simulating monthly-scale streamflow in the middle and lower reaches of the Yellow River source region.

Figure 8a further indicates that the simulated multi-year average monthly streamflow levels and trends are close to the observed values. With the exception of the Jimai station, the SHUD model performs more admirably in simulating the streamflow processes from May to July and September to December, particularly in the simulation of peak flows. This suggests that the SHUD model exhibits satisfactory performance in the monthly scale streamflow simulation of the middle and lower sub-basins of the Yellow River source region.

Zhang et al. [21] utilized data from 23 meteorological stations surrounding the Yellow River source region to drive the SWAT model and analyzed the observed and simulated streamflow at the Tangnaihai station. They found that initial simulations and adjustments to the snowmelt process parameters, as well as the incorporation of elevation zones, yielded promising results. However, in terms of intra-annual distribution, these models tended to underestimate the streamflow volume from January to April and failed to capture the temporary decline in streamflow in August. By concurrently altering snowpack parameters, employing elevation zones, and adjusting groundwater parameters, the simulation and observation discrepancies for January to April were reduced, and the bimodal peak characteristic within the year was successfully simulated. The observed bimodal phenomenon may be the result of the interaction between precipitation and temperature. However, the multi-year average monthly observed streamflow at the Jimai, Maqu, Jungong, and Tangnaihai stations all exhibit a bimodal pattern, where an increase in precipitation leads to a rise in streamflow volume, while increased evapotranspiration due to higher temperatures leads to a decrease in streamflow volume. This bimodal pattern was not fully reflected in the simulation results of this study.

Furthermore, the simulated streamflow values during the ice period from winter to spring of the following year are lower than expected, possibly due to insufficient understanding of the distribution of groundwater in high mountain areas. The aquifer thickness set in the model may be inaccurate, leading to a discrepancy between the simulated groundwater storage and the natural conditions of the basin, thereby causing a deviation between the simulated and observed streamflow values.

Although the SHUD model can effectively simulate the monthly scale streamflow variation processes and reflect the seasonal distribution characteristics of the flow at the control stations of each sub-basin, there are still certain biases in simulating the bimodal peak characteristics, the streamflow during the ice-bound period from January to April in plateau rivers, and the streamflow in August. Further improvement and optimization of the model, especially considering factors such as the distribution of groundwater in high mountain areas, are expected to significantly enhance the model’s accuracy in simulating the streamflow processes of the basin.

By delineating the seasonal demarcations of the Yellow River source region, the period from May to September is designated as the warm season, while October to April of the following year is categorized as the cold season. Based on this delineation, this study conducted a comparative seasonal analysis of the SHUD model’s performance in simulating streamflow within the Yellow River source region between 2006 and 2018. Figure 8b presents a visual representation of the simulated error percentages for each station within the three sub-basins during the warm and cold seasons, reflecting the discrepancies between the SHUD model’s predicted streamflow and actual observed values.

The Source Area–Jimai Sub-basin: Within this sub-basin, the relative simulation errors for the warm and cold seasons are 74.24% and −39.37%, respectively. The cold season’s simulation error is significantly negative, while the warm season error manifests as a higher positive value, indicating a substantial underestimation by the SHUD model during the cold season and a notable overestimation during the warm season.

Jimai–Maqu Sub-basin: At the Maqu station, the relative errors for the warm and cold seasons are 21.21% and −38.31%, respectively. The warm season error is a minor positive value, suggesting that the model’s predictions slightly exceed actual observations, while the cold season indicates an underestimation. This reveals that the model’s performance during the warm season is markedly superior to that of the cold season.

Maqu–Jungong–Tangnaihai Sub-basin: In this area, the relative errors for the warm season at the Jungong and Tangnaihai stations are 12.93% and 13.50%, respectively, while in the cold season, they significantly decrease to −41.69% and −39.61%, respectively. This indicates a pronounced underestimation at both stations during the cold season, with a slight overestimation during the warm season, particularly highlighting the more pronounced underestimation during the cold season and the model’s significantly better predictive capacity during the warm season.

Overall, the SHUD model generally exhibits a trend of underestimating streamflow during the cold season, evident from the negative values of the blue bars in most sub-basins in Figure 8b. In contrast, during the warm season, the model tends to overestimate streamflow volumes, albeit with significant differences in overestimation errors between the source area–Jimai sub-basin and the other two sub-basins. From a holistic basin perspective, the model’s predictions are generally larger during the warm season and smaller during the cold season. Among all monitored stations, the Maqu station demonstrates the most accurate simulation performance during the cold season, while the Jungong station stands out during the warm season. Specifically, the SHUD model’s relative error for streamflow in the middle and lower reaches of the Yellow River source region during the warm season ranges from 12.93% to 21.21%, while the cold season error ranges from −38.31% to −41.69%. These data indicate that after calibration, the SHUD model’s simulation effectiveness during the warm season in the middle and lower reaches of the Yellow River source region is significantly better than during the cold season, but there is still room for improvement, especially in enhancing the accuracy of cold season streamflow simulation.

The SHUD model’s superior performance during the warm season over the cold season may be attributed to the stability of meteorological conditions, the simplification of surface processes, and the high accessibility and accuracy of hydrological and meteorological data during the warm season. During the warm season, precipitation primarily occurs as rain, while the cold season involves more complex hydrological processes such as snowfall, snowmelt, and soil freezing and thawing, increasing the uncertainty of model simulation [68]. Furthermore, the meteorological and hydrological observation conditions during the warm season are more favorable, with key observational data being more abundant and accurate. In contrast, the observation conditions during the cold season are more adverse, with hydrological and meteorological data being less accessible and having larger observational errors, affecting the model’s simulation, calibration, and validation.

During the warm season, surface processes are relatively straightforward, while in the cold season, the freezing–thawing cycles of the surface and soil significantly impact the migration and storage of water. The freezing and thawing processes of permafrost directly and complexly affect hydrological processes by altering soil moisture, regulating evaporation, and influencing surface streamflow. Zheng et al. [20] used the Noah LSM model to investigate the impact of soil moisture content and soil freezing–thawing status on streamflow in the Yellow River source region, finding that considering soil freezing–thawing would overestimate streamflow from May to September, while disregarding soil freezing–thawing would overestimate winter streamflow and underestimate streamflow from April to July.

The SHUD model includes a simplified permafrost module capable of characterizing partial permafrost processes during simulation. This module records surface and subsurface temperatures and calculates the proportion of permafrost freezing based on accumulated temperature, adjusting the moisture state of the surface and subsurface, thereby affecting parameters and variables related to permafrost hydrological processes. These adjustments alter the rate of water infiltration, taking into account the impact of permafrost on hydrological processes to some extent. However, due to the relative simplicity of the permafrost module in the SHUD model, it may fail to fully capture the impact of soil freezing processes on streamflow during the cold season, leading to an overestimation; during the warm season, it may inaccurately simulate the melting process and the release of moisture from the soil, leading to an underestimation of streamflow.

3.2. Uncertainty Analysis

The accuracy of hydrological model simulations is influenced by a multitude of uncertainties, including those related to model inputs, calibration data, model structure, parameters, calibration strategies, and initial conditions. These uncertainties, while independent, are interwoven and collectively impact model outputs [69,70].

To mitigate uncertainties associated with initial conditions, data from 2000 to 2005 were utilized as a spin-up period to stabilize the model. Nonetheless, parameter uncertainty significantly affects simulation outcomes. Hydrological models, particularly those based on physical principles, encompass numerous parameters and assumptions that cannot fully and precisely describe the characteristics of a watershed, and the optimized parameter sets also harbor uncertainties. The phenomenon of “different parameters leading to similar outcomes” is inevitable. Uncertainty in model structure stems from the simplification and assumptions made regarding natural hydrological processes. Although the SHUD model is applicable in the Yellow River source region, its simulation results cannot fully replicate the observed values.

Data quality, model structure, parameters, and calibration strategies jointly influence the streamflow simulation at Jimai station, with the impact of model structure being particularly significant. The permafrost module in the SHUD model is relatively simplistic, limiting its comprehensive representation of the freeze–thaw processes of permafrost. Jimai station, located in a zone of perennial permafrost, experiences significant impacts of permafrost on water infiltration, evaporation, and soil water storage, especially in the simulation of streamflow in high mountain areas [13,14,71,72]. The SHUD model shows deficiencies in simulating streamflow during winter and spring seasons and the freeze–thaw period, particularly for short time series simulations. The current optimization strategy primarily targets the outlet of the source region, Tangnaihai station; incorporating streamflow at Jimai station into the optimization objectives may improve its simulation effectiveness. However, the multitude of model parameters and the complexity of parameter tuning present challenges, and optimization methods still face significant hurdles. Future research should consider comprehensive indicators or high spatiotemporal resolution observational variables as optimization targets.

The disparity in the simulation of high and low streamflow stems from calibration strategies, data quality, model structure, and parameter settings. In automated optimization targeting NSE as the objective, the simulation of high and low streamflow may be compromised. In extreme value studies, adjusting the weights of the objective function or employing indicators focused on extremes can improve simulation outcomes. The uncertainty of precipitation input is also a critical factor. Streamflow in the Yellow River source region is sensitive to precipitation [27], and the accuracy of precipitation data directly affects the simulation results [73,74]. Precipitation errors may arise from observation, data processing, and interpolation errors. Due to the low density of data points, the representativeness of station observations and reanalysis precipitation data in complex terrain areas is insufficient, leading to uncertainties in streamflow simulation [75,76,77].

Figure 9 displays the regression analysis of CMFD precipitation data compared to data from nine meteorological stations in the Yellow River source region on daily, monthly, and annual scales. On a daily scale, CMFD data underestimate precipitation, with a weak correlation; on a monthly and annual scale, the correlation is strong, especially on a monthly scale, where CMFD estimates are close to meteorological station data. While CMFD data perform well on a monthly and annual scale, there is an issue of underestimation on a daily scale. This may be one of the reasons for the differences in streamflow simulation capabilities. Flow indicators such as average flow and high/low streamflow depend on precipitation volume, but the specific influencing factors on the simulation capabilities of high and low streamflow are numerous and cannot fully account for simulation errors [65,78,79,80].

This study highlights the application of the SHUD model in the Yellow River source region, a cold and high-altitude area with extensive permafrost and complex snowmelt dynamics. The findings demonstrate the model’s capability to capture spatial and temporal variations in streamflow under extreme climatic and hydrological conditions, providing valuable insights into its applicability in cold-region environments. However, uncertainties in simulating cold season processes reveal the need for targeted improvements. Refinements such as integrating a detailed permafrost module or adopting multi-objective optimization strategies could further enhance the model’s accuracy and reliability in permafrost-dominated regions.

4. Conclusions

This study employed the SHUD distributed hydrological model to conduct a comprehensive simulation of streamflow in the Yellow River source region from 2006 to 2018 and assessed the model’s performance across different seasons, scales, and sub-basins, as well as its associated uncertainties. The following are the primary conclusions:

(1): The SHUD model demonstrates a fundamental alignment with the observed hydrograph, with NSE and R² values exceeding 0.7 on daily and monthly scales, KGE values generally greater than 0.8, and PBIAS controlled within $\pm 10 %$ . According to the current evaluation metrics, the model achieves an excellent level in simulating monthly streamflow in the Yellow River source region, with daily streamflow simulation outcomes comparable to the existing studies. The model effectively reflects the distribution characteristics of streamflow on daily and monthly scales, particularly excelling in capturing the spatial variations from upstream to downstream, thereby validating its applicability in the Yellow River source region and its capability to simulate hydrological processes with high spatiotemporal resolution.
(2): The SHUD model’s simulation performance in the Yellow River source region exhibits distinct seasonal and regional variations. The warm season simulation outcomes are significantly superior to those of the cold season, especially in the middle and lower reaches. On daily, monthly, and warm season scales, the model’s performance ranking across the three sub-basins is as follows: Maqu–Tangnaihai, Jimai–Maqu, and the source area–Jimai sub-basin, showing an increasing trend from upstream to downstream. This may be associated with stable meteorological conditions during the warm season, simplified surface processes, and high-quality hydrological and meteorological data. The uncertainty in cold season simulation increases, primarily due to the influence of complex hydrological processes such as snowfall, snowmelt, and permafrost freeze–thaw cycles. In extreme flow simulations, the Jimai station underestimates high flows and overestimates low flows; Maqu and Jungong stations overestimate both high and low flows; while Tangnaihai station underestimates these two types of flows.
(3): The uncertainties in the application of the SHUD model in the Yellow River source region primarily stem from input data, model structure, parameter settings, and calibration strategies. The model’s simplified treatment of soil freeze–thaw processes limits its precision in simulating streamflow during the winter and spring seasons. Furthermore, the CMA-ES automated parameter optimization, which targets NSE as the optimization goal, may not adequately consider the simulation of high and low streamflow.
(4): Constructed based on the principles of conservation of mass, energy, and momentum, the SHUD model possesses the advantage of coupling with other physical processes. Future research can enhance the model’s hydrological simulation capabilities in regions with high-altitude, perennial permafrost, and seasonal frost distribution by introducing a more refined permafrost module. This study not only improves the simulation accuracy of streamflow changes in the Yellow River source region but also provides significant scientific evidence for the region to address climate change.

In summary, this study highlights the successful application of the SHUD model in the Yellow River source region, bridging a critical gap in hydrological modeling for cold, high-altitude regions. The model demonstrated its capability to simulate streamflow with high accuracy, effectively capturing spatial and temporal variations despite the existing uncertainties and areas requiring refinement. Beyond validating the performance of the model, this study provided novel insights into the impacts of permafrost and freeze–thaw dynamics on hydrological processes, significantly advancing our understanding of streamflow behavior in cold-region basins. Furthermore, this study established a comprehensive methodological framework tailored to address region-specific hydrological challenges, such as snowmelt and freeze–thaw cycles. This framework not only supports hydrological forecasting but also offers strong scientific backing for water resource management, climate adaptation, and wetland conservation efforts in the Yellow River source region. Looking ahead, continuous model refinements, including advanced parameter optimization and structural enhancements, are anticipated to further improve simulation accuracy and reliability. These efforts will empower the SHUD model to better address the evolving climatic and hydrological challenges faced by high-altitude regions, reinforcing its role as a robust tool for climate change adaptation and sustainable water management.

5. Discussion

(1): Application Prospects for High-Altitude Wetland Research: The purpose of this study in evaluating the SHUD model is to lay the foundation for future high-altitude wetland research. The 12 traditional hydrological models mentioned in Appendix A have limitations in dealing with wetland-specific hydrological processes, such as snow and ice accumulation and ablation, as well as permafrost freezing and thawing, and thus cannot meet the needs of wetland research. The SHUD model, with its comprehensive process representation and high spatial resolution, provides a new perspective for simulating wetland hydrological processes, especially in terms of simulating the interaction between surface water and groundwater and the contribution of snowmelt to flow, showing potential application value.
(2): Long-term and Short-term Applications of the Model: According to the evaluation results of the SHUD model, we believe that the model is not only suitable for long-term hydrological trend research but also for short-term flood event research. For example, in the flood research conducted by [81] in North America, the SHUD model was able to capture the hydrological dynamics of flood events. Additionally, the applicability of the SHUD model extends to the study of hydrological process changes in the Qinghai–Tibet Plateau under future climate change conditions, especially against the backdrop of warming and increased extreme precipitation. The focus of this study’s evaluation was on long-term water flow processes, hence the NSE, which can reflect the average state and time series, was chosen as the calibration parameter. This provides a scientific basis for the model’s applicability on different time scales.
(3): Adjustment and Optimization of Model Parameters: To serve different research purposes, the SHUD model needs to adjust calibration parameters based on different research objectives. In long-term trend research, NSE, as a key indicator of model performance, can well reflect the model’s simulation capability of the overall hydrological cycle. However, in short-term flood event research, more attention may need to be paid to the model’s response to extreme events, thus requiring the introduction of other evaluation indicators, such as the accuracy of peak flow and the deviation of peak time. By adjusting model parameters, the SHUD model can better adapt to different research needs and improve the model’s applicability and prediction accuracy.
(4): Limitations of the Model and Future Improvement Directions: Although the SHUD model has shown good performance in this study, there are some limitations, especially in simulating hydrological processes in cold seasons and extreme flood events. Future research can further improve the model’s performance by introducing more refined surface process modules, improving the representation of permafrost and snow processes, and optimizing parameter calibration strategies. In addition, with the development of remote sensing and big data technologies, future research can consider integrating more real-time monitoring data and high-resolution remote sensing products to improve the quality of model input data and the model’s predictive capability.

Author Contributions

Conceptualization, C.W., H.C. and X.M.; Methodology, T.B., C.W. and H.C.; Software, T.B.; Validation, T.B.; Data curation, Z.L.; Writing—review & editing, T.B., C.W., H.C., D.S., Y.C. and C.Z.; Funding acquisition, H.C., X.M. and Z.L. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Science and Technology Program of Gansu Province (Grant Nos. 24RCKB009, 23JRRA654, 23JRRA609, 22JR5RA048) and the West Light Foundation for Western Cross Team of the Chinese Academy of Sciences (xbzg-zdsys-202215).

Data Availability Statement

Publicly available datasets were analyzed in the study. The Land Use data can be found at the website: https://data.mint.isi.edu/files/raw-data/land-use/USGS_LCI/. The Soil Classification data can be found at the website: http://www.fao.org/soils-portal/soil-survey/soil-maps-and-databases/harmonized-world-soil-database-v12/en/. The CMFD meteorological data can be found at the website: http://data.tpdc.ac.cn/zh-hans/data/8028b944-daaa-4511-8769-965612652c49/. The Stations Precipitation data can be found at the website: https://m.data.cma.cn/data/cdcdetail/dataCode/SURF_CLI_CHN_MUL_DAY_V3.0.html, accessed on 24 October 2024. The streamflow data were requested from the relevant department due to the restriction policy of the data provider.

Conflicts of Interest

The authors declare no conflicts of interest.

Appendix A

Table A1. Performance of various hydrological models in streamflow simulation in the Yellow River source region.

Model	Simulation Period	Temporal Resolution	R² Calibration/Validation	NSE Calibration/Validation	Streamflow Simulation Performance	Notes
SWAT [12]	1975–1990	Monthly	0.77–0.85/0.78–0.86	0.63–0.85/0.57–0.83	Overestimates peak in some years and underestimates in others; underestimates low flow.	The default parameters with three types of snowmelt algorithms consistently underestimated the streamflow. After parameter adjustment, the degree of deviation was reduced, leading to an improvement in simulation performance.
SWAT [21]	1961–1990	Monthly	-	0.91/0.89	Overestimates peak in some years and underestimates in others; underestimates low flow.	In terms of intra-annual distribution: the simulation underestimated the streamflow during February and March; apart from the adjustment of groundwater parameters, the remaining simulation scenarios did not exhibit the characteristic transient decrease in streamflow observed in August.
SWAT [28]	2006–2017	Monthly	0.89/0.91	0.87/0.86	Most stations underestimate peak and overestimate low flow in some years.	Modeling Capability at Various Stations: The streamflow simulations at Jun Gong, Tangnaihai, and Maqu stations were closest to the observed values; followed by Tangke; then Mentang and Jimai; and finally Ruoergai and Dashui stations.
GBEHM [13]	1981–2000	Daily	0.84/0.73	0.77/0.67	Generally overestimates peak and overestimates low flow in most years.	The simulation performance at Maqu station is similar to that at Tangnaihai station; whereas the performance at Jimai station is comparatively poor. Overestimation of low flow values: this may be due to errors in soil depth. The simulation utilized the HWSD dataset, which entails uncertainties in soil depth.
VIC [82]	1961–1990	Daily	-	0.91/0.93	Generally underestimates peak in most years and underestimates low flow in some years, with a general underestimation of low flow.	The NSE was greater than 0.90 during both the calibration and validation periods, with the corresponding Er less than 3%.
GXAJ [83]	2014.4–2014.12	Daily	-	0.897/0.807	Underestimates peak for the vast majority of the time, with significant overestimation in a few instances; overestimates low flow.	Possible reasons include: (1) the gridded precipitation dataset from rain gauge stations may not adequately capture the spatiotemporal variations in precipitation. (2) The average daily streamflow volume during the calibration period is smaller than that during the validation period. Parameters calibrated under wet conditions may not sufficiently represent the hydrological characteristics of dry years, affecting the simulation performance.
WaSiM [22]	1980–2014	Daily	-	0.89	Overall, underestimates peak, with a few instances of overestimation; generally underestimates low flow, with occasional overestimation.	The simulation performance at the stations is as follows: Tangnaihai, Jun Gong, and Maqu stations are the top performers, followed by Jimai, and finally Mentang. At Jun Gong and Maqu stations, the simulated streamflow during the spring months (March, April, and May) is less than the observed data, while this is not evident at the Mentang station.
HEQM [65]	1976–2016	Daily	-	0.88/0.81	Generally underestimates annual peak and low flow in most years.	Maqu and Tangnaihai stations exhibit superior simulation performance, followed by Jimai station.
SPHY [24]	1990–2015	Monthly/Daily	-	0.76/0.74	Overall, underestimates peak; underestimates low flow in some years and overestimates in others.	The peak streamflow at Jimai, Maqu, and Tangnaihai stations was underestimated, possibly due to the CMFD’s underestimation of precipitation peaks.
WRF-Hydro [16]	2009–2018	Daily	-	0.56–0.67/0.15–0.94	Overestimates peak and underestimates low flow in most years.	Errors in flood peaks during the flood season were relatively large, possibly because the CMFD data overestimate the concentration of regional precipitation; simulation performance was generally superior in wet years compared to dry years.
VIC [84]	1977–1987	Monthly/Daily	-	0.80–0.81/0.74–0.85	Generally underestimates peak in most years.	The significant deviation in the peak streamflow is likely due to errors in extreme precipitation within the meteorological data. The modeling performance at the stations, in order of proficiency, is Tangnaihai and Maqu, followed by Tangke and Jimai.
Noah& Noah_wo-FT [20]	1984–2009	Monthly	0.87/0.63	0.87/0.60	Without considering soil freeze–thaw: peak was underestimated in most years, low flow was occasionally underestimated. With consideration of soil freeze–thaw: peak was overestimated in some instances and underestimated in others, and the low flow was overestimated.	Intra-annual distribution: Accounting for soil freeze–thaw overestimates the streamflow from May to September; disregarding soil freeze–thaw overestimates the streamflow during the winter months (October to February) and underestimates the streamflow from April to July.
CLM5.0 [19]	2007–2016	Monthly	-	0.55	Generally overestimates peak and underestimates low flow in most years.	Hydrological models, when calibrated with observational data, exhibit satisfactory performance. Surface models, which incorporate more complex mechanisms to simulate the interaction between surface energy and the water cycle, do not perform as well in streamflow simulation as hydrological models.
CREST-Snow [25]	2004–2014	Daily	-	0.66–0.86/0.40–0.8	In a few years, peak was generally overestimated, while it was underestimated in most years; overall, low flow was overestimated.	Models with different precipitation inputs yield varying simulation outcomes.

Table A2. Calibration parameter ranges and optimal values.

Variable Name	Variable Meaning	Minimum Value of Variable	Maximum Value of Variable	Minimum Value of Parameter Adjustment Coefficient	Maximum Value of Parameter Adjustment Coefficient	Optimized Adjustment Coefficient Value	Mean Value of Parameter
GEOL_KSATH	Horizontal conductivity of ground water (ms⁻¹)	0.67	5.43	0.0001	10	1.72	3.23
GEOL_KSATV	Vertical conductivity of ground water (ms⁻¹)	0.07	0.54	0.0001	10	0.84	0.16
GEOL_KMACSATH	Horizontal conductivity of macropore (ms⁻¹)	67.74	543.3	0.0001	1	1	188.1
GEOL_MACVF	Vertical macropore areal fraction (m²m⁻²)	0.01	0.01	0	10	4.05	0.04
GEOL_THETAS	Porosity, saturated soil moisture (m³m⁻³)	0.36	0.49	0.5	1.5	0.5	0.22
GEOL_THETAR	Residual soil moisture (m³m⁻³)	0.01	0.01	0	10	1.31	0.01
GEOL_DMAC	Macropore depth (m)	1	1	0	3	0	0
SOIL_KINF	Vertical conductivity of top soil (ms⁻¹)	0.17	1.38	0.0001	10	10	4.92
SOIL_KMACSATV	Vertical conductivity of soil macropore (ms⁻¹)	1.67	138.47	0.0001	10	0.27	13.5
SOIL_ALPHA	$α$ , van Genuchten soil parameter (m⁻¹)	6.91	10	1	10	1	4.07
SOIL_BETA	$β$ , van Genuchten soil parameter (-)	1.47	10	0.8	5	4.56	5.52
ET_ETP	Transpiration	0.5	0.5	0.5	2	0.5	0.5
RIV_KH	Conductivity of river bed (ms⁻¹)	0.1	0.1	0.001	1000	3.16	0.316
RIV_BEDTHICK	Depth of river cross section (m)	0.1	0.1	0.001	10	0.001	0.0001
AQ_DEPTH+	Thickness of aquifer (m)	20	20	−1	20	−1	19
FZN_SURFMAX	Maximum Temperature of the Permafrost Surface Layer (℃)	10	10	1	15	10	10
FZN_SURFMIN	Minimum Temperature of the Permafrost Surface Layer (℃)	−10	−10	−10	0	−10	−10
FZN_SURFDAY	Number of Freezing Days of the Permafrost Surface Layer (d)	7	7	1	15	7	7
FZN_SUBMAX	Maximum Temperature of the Permafrost Subsurface Layer (℃)	10	10	1	15	10	10
FZN_SUBMIN	Minimum Temperature of the Permafrost Subsurface Layer (℃)	−10	−10	−30	0	−10	−10
FZN_SUBDAY	Number of Freezing Days of the Permafrost Subsurface Layer (d)	30	30	1	60	30	30

The data presented in columns 3 and 4 illustrate the range of physical values for the parameters themselves. For instance, the range for the horizontal hydraulic conductivity parameter GEOL_KSATH is between 0.67 and 5.43 ms⁻¹. The subsequent columns (5, 6, and 7) present the dimensionless adjustment coefficients for the aforementioned parameters. To illustrate, in the initial column, the adjustment coefficients span a range of 0.0001 to 10, with a value of 1.72 signifying that the optimal parameter value is 1.72 times the original physical parameter value. Column 8 displays the mean value across the entire watershed after parameter adjustment. For example, a value of 3.23 in the first row indicates that following calibration, the spatial average of the horizontal hydraulic conductivity of the groundwater layer is 3.23 ms⁻¹. The optimal adjustment coefficients are determined based on a comparative analysis of model simulation results and observed data. (+) in the table represents the adjusted parameter values on the basis of the initial data, such as the original aquifer thickness set at 20 m, so the aquifer thickness used in the calibrated model is 20 m − 1 m = 19 m. No (+) represents multiplication of the adjusted parameter values based on the initial data.

Table A3. The performance metrics and their equations used to evaluate the model performance.

Metrics	Calculation Formula	Meaning
R²	$R^{2} = \frac{{[\sum_{i = 1}^{n} (O_{i} - \bar{O}) (S_{i} - \bar{S})]}^{2}}{\sum_{i = 1}^{n} {(O_{i} - \bar{O})}^{2} \sum_{i = 1}^{n} {(S_{i} - \bar{S})}^{2}}$	The percentage of variance in the observed data that is explained by the model [85].
KGE	$KGE = 1 - \sqrt{{(ρ - 1)}^{2} + {(α - 1)}^{2} + {(β - 1)}^{2}}$	Goodness-of-fit measures provide an analysis of the relative importance of different components (correlation, bias, and variance) in hydrological simulation [59].
NSE	$NSE = 1 - \frac{\sum_{i = 1}^{n} {(S_{i} - O_{i})}^{2}}{\sum_{i = 1}^{n} {(O_{i} - \bar{O})}^{2}}$	Evaluate the overall trend fit between the simulated results and observed data in time series [86] quantify the relative magnitude of residual variance (noise) compared to the variance of the observed data [57,59].
PBIAS (%)	$PBIAS = \frac{\sum_{i = 1}^{n} (S_{i} - O_{i})}{\sum_{i = 1}^{n} O_{i}} \times 100$	The bias of the evaluated data is expressed as a percentage. It measures the average trend of whether the simulated values are larger or smaller than the observed values. Negative values indicate underestimation by the model, while positive values indicate overestimation [86].
RMSE (m³ s⁻¹)	$RMSE = \sqrt{\frac{1}{n} \sum_{i = 1}^{n} {(S_{i} - O_{i})}^{2}}$	Reflects the deviation of simulated values from observed values; sensitive to both extreme large and small errors [87].

n represents the number of streamflow data points;

S_{i}

and

O_{i}

are the simulated streamflow and observed streamflow at time i, respectively; and

\bar{S}

and

\bar{O}

denote the mean values of simulated and observed streamflow.

ρ

is the linear regression coefficient between the simulated values

S_{i}

and observed values

O_{i}

;

α

is the ratio of the standard deviations of the simulated values

S_{i}

to the observed values

O_{i}

; and

β

is the ratio of the mean values of the simulated values

S_{i}

to the observed values

O_{i}

References

Tang, Q.H.; Liu, X.C.; Li, Z.; Yun, X.; Zhang, X.; Yu, Q.; Li, J.; Zhang, Y.; Cui, H.; Sun, S.; et al. Integrated Water Systems Model for Terrestrial Water Cycle Simulation. Adv. Earth Sci. 2019, 34, 115–123. (In Chinese) [Google Scholar]
Shu, L.; Chen, H.; Meng, X.; Chang, Y.; Hu, L.; Wang, W.; Shu, L.; Yu, X.; Duffy, C.; Yao, Y.; et al. A review of integrated surface-subsurface numerical hydrological models. Sci. China Earth Sci. 2024, 67, 1459–1479. (In Chinese) [Google Scholar] [CrossRef]
Maxwell, R.M.; Putti, M.; Meyerhoff, S.; Delfs, J.O.; Ferguson, I.M.; Ivanov, V.; Kim, J.; Kolditz, O.; Kollet, S.J.; Kumar, M.; et al. Surface-subsurface model intercomparison: A first set of benchmark results to diagnose integrated hydrology and feedbacks. Water Resour. Res. 2014, 50, 1531–1549. [Google Scholar] [CrossRef]
Peel, M.C.; McMahon, T.A. Historical development of rainfall-runoff modeling. Wiley Interdiscip. Rev. Water 2020, 7, e1471. [Google Scholar] [CrossRef]
Shu, L.; Ullrich, P.A.; Duffy, C.J. Simulator for Hydrologic Unstructured Domains (SHUD v1.0): Numerical modeling of watershed hydrology with the finite volume method. Geosci. Model Dev. 2020, 13, 2743–2762. [Google Scholar] [CrossRef]
Shu, L.; Li, X.; Chang, Y.; Meng, X.; Chen, H.; Qi, Y.; Wang, H.; Li, Z.; Lyu, S. Advancing understanding of lake–watershed hydrology: A fully coupled numerical model illustrated by Qinghai Lake. Hydrol. Earth Syst. Sci. 2024, 28, 1477–1491. [Google Scholar] [CrossRef]
Shu, L.; Ullrich, P.; Meng, X.; Duffy, C.; Chen, H.; Li, Z. rSHUD v2.0: Advancing the Simulator for Hydrologic Unstructured Domains and unstructured hydrological modeling in the R environment. Geosci. Model Dev. 2024, 17, 497–527. [Google Scholar] [CrossRef]
Meng, X.; Chen, H.; Li, Z.; Zhao, L.; Zhou, B.; Lv, S.; Deng, M.; Liu, Y.; Li, G. Review of climate change and its environmental influence on the Three-River Regions. Plateau Meteorol. 2020, 39, 1133–1143. (In Chinese) [Google Scholar]
Wang, D.; Tian, S.; Jiang, S.; Dong, X.; Li, Z.; Zhang, L. Research progress of the evolution of runoff in the source area of the Yellow River. Yellow River 2020, 42, 90–95. (In Chinese) [Google Scholar]
Nan, T.; Chen, J.; Ding, Z.; Li, W.; Chen, H. Deep learning-based multi-source precipitation merging for the Tibetan Plateau. Sci. China Earth Sci. 2023, 66, 852–870. (In Chinese) [Google Scholar] [CrossRef]
Zhong, D.; Dong, Z.; Fu, G.; Bian, J.; Kong, F.; Wang, W.; Zhao, Y. Trend and change points of streamflow in the Yellow River and their attributions. J. Water Clim. Chang. 2021, 12, 136–151. [Google Scholar] [CrossRef]
Zhang, X.; Srinivasan, R.; Debele, B.; Hao, F. Runoff simulation of the headwaters of the Yellow River using The SWAT model with three snowmelt algorithms. JAWRA J. Am. Water Resour. Assoc. 2008, 44, 48–61. [Google Scholar] [CrossRef]
Qin, Y.; Yang, D.; Gao, B.; Wang, T.; Chen, J.; Chen, Y.; Wang, Y.; Zheng, G. Impacts of climate warming on the frozen ground and eco-hydrology in the Yellow River source region, China. Sci. Total Environ. 2017, 605, 830–841. [Google Scholar] [CrossRef] [PubMed]
Yang, J.; Wang, T.; Yang, D.; Yang, Y. Insights into runoff changes in the source region of Yellow River under frozen ground degradation. J. Hydrol. 2023, 617, 128892. [Google Scholar] [CrossRef]
Li, G.; Meng, X.; Blyth, E.; Chen, H.; Shu, L.; Li, Z.; Zhao, L.; Ma, Y. Impact of fully coupled hydrology-atmosphere processes on atmosphere conditions: Investigating the performance of the WRF-Hydro model in the Three River Source Region on the Tibetan Plateau, China. Water 2021, 13, 3409. [Google Scholar] [CrossRef]
Chen, Y.; Wen, J.; Yang, C.; Long, Y.; Li, G.; Jia, H.; Liu, Z. Applicability of Different Precipitation Products and the WRF-Hydro Model over the Source Region of the Yellow River. Chin. J. Atmos. Sci. 2024, 48, 923–937. (In Chinese) [Google Scholar]
Sheng, M.; Lei, H.; Jiao, Y.; Yang, D. Evaluation of the runoff and river routing schemes in the community land model of the Yellow River Basin. J. Adv. Model. Earth Syst. 2017, 9, 2993–3018. [Google Scholar] [CrossRef]
Zheng, D.; Van Der Velde, R.; Su, Z.; Wen, J.; Wang, X. Assessment of Noah land surface model with various runoff parameterizations over a Tibetan river. J. Geophys. Res. Atmos. 2017, 122, 1488–1504. [Google Scholar] [CrossRef]
Deng, M.; Meng, X.; Lu, Y.; Shu, L.; Li, Z.; Zhao, L.; Chen, H.; Shang, L.; Sheng, D.; Ao, X. Impact of climatic and vegetation dynamic change on runoff over the Three Rivers Source Region based on the Community Land Model. Clim. Dyn. 2023, 61, 1193–1208. [Google Scholar] [CrossRef]
Zheng, D.; van der Velde, R.; Su, Z.; Wen, J.; Wang, X.; Yang, K. Impact of soil freeze-thaw mechanism on the runoff dynamics of two Tibetan rivers. J. Hydrol. 2018, 563, 382–394. [Google Scholar] [CrossRef]
Zhang, Y.; Su, F.; Hao, Z.; Xu, C.; Yu, Z.; Wang, L.; Tong, K. Impact of projected climate change on the hydrology in the headwaters of the Yellow River basin. Hydrol. Process. 2015, 29, 4367–4558. [Google Scholar] [CrossRef]
Sun, A.; Yu, Z.; Zhou, J.; Acharya, K.; Ju, Q.; Xing, R.; Huang, D.; Wen, L. Quantified hydrological responses to permafrost degradation in the headwaters of the Yellow River (HWYR) in High Asia. Sci. Total Environ. 2020, 712, 135632. [Google Scholar] [CrossRef] [PubMed]
Nkiaka, E.; Nawaz, N.R.; Lovett, J.C. Evaluating Global Reanalysis Datasets as Input for Hydrological Modelling in the Sudano-Sahel Region. Hydrology 2017, 4, 13. [Google Scholar] [CrossRef]
Zhang, T.; Li, D.; Lu, X. Response of runoff components to climate change in the source-region of the Yellow River on the Tibetan plateau. Hydrol. Process. 2022, 36, e14633. [Google Scholar] [CrossRef]
Hong, Z.; Han, Z.; Li, X.; Long, D.; Tang, G.; Wang, J. Generation of an improved precipitation dataset from multisource information over the Tibetan Plateau. J. Hydrometeorol. 2021, 22, 1275–1295. [Google Scholar]
Zhang, M.; Zhu, K.; Lu, F.; Dai, Y.; Song, X.; Yu, S. Analysis of changes of hydrological elements and driving factors in the source area of the Yellow River. Yellow River 2023, 45, 30–37. (In Chinese) [Google Scholar]
Xu, R.; Qiu, D.; Gao, P.; Wu, C.; Mu, X.; Ismail, M. Prediction of streamflow based on the long-term response of streamflow to climatic factors in the source region of the Yellow River. J. Hydrol. Reg. Stud. 2024, 52, 101681. [Google Scholar] [CrossRef]
Zhao, G.; Tian, S.; Jing, Y.; Cao, Y.; Liang, S.; Han, B.; Cheng, X.; Liu, B. Establishing a quantitative assessment methodology framework of water conservation based on the water balance method under spatiotemporal and different discontinuous ecosystem scales. J. Environ. Manag. 2023, 346, 119006. [Google Scholar] [CrossRef]
Huang, Q.; Qin, G.; Zhang, Y.; Tang, Q.; Liu, C.; Xia, J.; Chiew, F.H.S.; Post, D. Using remote sensing data-based hydrological model calibrations for predicting runoff in ungauged or poorly gauged catchments. Water Resour. Res. 2020, 56, e2020WR028205. [Google Scholar] [CrossRef]
Yang, J.; Huang, M.; Zhai, P. Performance of the CRA-40/Land, CMFD, and ERA-Interim datasets in reflecting changes in surface air temperature over the Tibetan Plateau. J. Meteorol. Res. 2021, 35, 663–672. [Google Scholar] [CrossRef]
Wen, T.; Guo, Y.; Dong, S.; Dong, Y.; Lai, X. Assessment of CRU, ERA5, CMFD grid precipitation data for the Tibetan Plateau from 1979 to 2017. Arid Zone Res. 2022, 39, 684–697. (In Chinese) [Google Scholar]
Wu, P.; Liu, Y.-J.; Wang, J.; Ding, Y.-H. Revisiting the variations of precipitation and water vapour budget over the Tibetan Plateau. Adv. Clim. Chang. Res. 2023, 14, 77–84. [Google Scholar] [CrossRef]
Gil, Y.; Garijo, D.; Khider, D.; Knoblock, C.A.; Ratnakar, V.; Osorio, M.; Vargas, H.; Pham, M.; Pujara, J.; Shbita, B.; et al. Artificial Intelligence for Modeling Complex Systems: Taming the Complexity of Expert Models to Improve Decision Making. ACM Trans. Interact. Intell. Syst. 2021, 11, 11:1–11:49. [Google Scholar] [CrossRef]
Shu, L.; Chang, Y.; Wang, J.; Chen, H.; Li, Z.; Zhao, L.; Meng, X. A Brief Review of Numerical Distributed Hydrological Model SHUD. Adv. Earth Sci. 2022, 37, 680–691. (In Chinese) [Google Scholar]
Chang, Y.; Li, X.; Shu, L.; Ji, H. Comprehensive Hydrological Analysis of the Buha River Watershed with High-Resolution SHUD Modeling. Water 2024, 16, 2015. [Google Scholar] [CrossRef]
Wang, D.; Liu, Y.; Kumar, M. Using Nested Discretization for a Detailed yet Computationally Efficient Simulation of Local Hydrology in a Distributed Hydrologic Model. Sci. Rep. 2018, 8, 5785. [Google Scholar] [CrossRef]
Hansen, N. The CMA Evolution Strategy: A Comparing Review. In Towards a New Evolutionary Computation: Advances in the Estimation of Distribution Algorithms; Lozano, J.A., Larrañaga, P., Inza, I., Bengoetxea, E., Eds.; Studies in Fuzziness and Soft Computing; Springer: Berlin/Heidelberg, Germany, 2006; pp. 75–102. [Google Scholar]
Zhang, Y.; Liu, L.; Bai, W. Grassland Degradation in the Source Region of the Yellow River. Acta Geogr. Sin. 2006, 61, 3–14. (In Chinese) [Google Scholar]
Chen, L.; Liu, C. Influence of Climate and Land-Cover Change on Runoff of the Source Regions of Yellow River. China Environ. Sci. 2007, 27, 559–565. (In Chinese) [Google Scholar]
Chang, G.; Li, L.; Zhu, X. Changes and Influencing Factors of Surface Water Resources in the Source Region of the Yellow River. J. Geogr. Sci. 2007, 62, 312–320. (In Chinese) [Google Scholar]
Mo, X.; Liu, S.; Hu, S. Co-Evolution of Climate-Vegetation-Hydrology and Its Mechanisms in the Source Region of Yellow River. Acta Geogr. Sin. 2022, 77, 1730–1744. (In Chinese) [Google Scholar]
Suwandana, E.; Kawamura, K.; Sakuno, Y.; Kustiyanto, E. Thematic Information Content Assessment of the ASTER GDEM: A Case Study of Watershed Delineation in West Java, Indonesia. Remote Sens. Lett. 2011, 3, 423–432. [Google Scholar] [CrossRef]
Broxton, P.D.; Zeng, X.; Sulla-Menashe, D.; Troch, P.A. A Global Land Cover Climatology Using MODIS Data. J. Appl. Meteorol. Climatol. 2014, 53, 1593–1605. [Google Scholar] [CrossRef]
Fischer, G.; Nachtergaele, F.; Prieler, S.; Van Velthuizen, H.; Verelst, L.; Wiberg, D. Global Agro-Ecological Zones Assessment for Agriculture (GAEZ 2008); IIASA: Laxenburg, Austria; FAO: Rome, Italy, 2008; Volume 10. [Google Scholar]
Yang, K.; He, J.; Tang, W.J.; Qin, J.; Cheng, C.K. On Downward Shortwave and Longwave Radiations over High Altitude Regions: Observation and Modeling in the Tibetan Plateau. Agric. For. Meteorol. 2010, 150, 38–46. [Google Scholar] [CrossRef]
He, J.; Yang, K.; Tang, W.J.; Lu, H.; Qin, J.; Chen, Y.Y.; Li, X. The First High-Resolution Meteorological Forcing Dataset for Land Process Studies over China. Sci. Data 2020, 7, 25. [Google Scholar] [CrossRef] [PubMed]
Eckhardt, K.; Arnold, J.G. Automatic Calibration of a Distributed Catchment Model. J. Hydrol. 2001, 251, 103–109. [Google Scholar] [CrossRef]
Fontaine, T.A.; Cruickshank, T.S.; Arnold, J.G.; Hotchkiss, R.H. Development of a Snowfall–Snowmelt Routine for Mountainous Terrain for the Soil Water Assessment Tool (SWAT). J. Hydrol. 2002, 262, 209–223. [Google Scholar] [CrossRef]
Tripathi, M.P.; Panda, R.K.; Raghuwanshi, N.S. Identification and Prioritisation of Critical Sub-Watersheds for Soil Conservation Management Using the SWAT Model. Biosyst. Eng. 2003, 85, 365–379. [Google Scholar] [CrossRef]
Fernandez, G.P.; Chescheir, G.M.; Skaggs, R.W.; Amatya, D.M. Development and Testing of Watershed-Scale Models for Poorly Drained Soils. Trans. ASAE 2005, 48, 639–652. [Google Scholar] [CrossRef]
Singh, J.; Knapp, H.V.; Arnold, J.G.; Demissie, M. Hydrological Modeling of the Iroquois River Watershed Using HSPF and SWAT 1. JAWRA J. Am. Water Resour. Assoc. 2005, 41, 343–360. [Google Scholar] [CrossRef]
Van Liew, M.W.; Veith, T.L.; Bosch, D.D.; Arnold, J.G. Suitability of SWAT for the Conservation Effects Assessment Project: Comparison on USDA Agricultural Research Service Watersheds. J. Hydrol. Eng. 2007, 12, 173–189. [Google Scholar] [CrossRef]
Chai, T.; Draxler, R.R. Root Mean Square Error (RMSE) or Mean Absolute Error (MAE)?—Arguments Against Avoiding RMSE in the Literature. Geosci. Model Dev. 2014, 7, 1247–1250. [Google Scholar] [CrossRef]
Olden, J.D.; Poff, N.L. Redundancy and the Choice of Hydrologic Indices for Characterizing Streamflow Regimes. River Res. Appl. 2003, 19, 101–121. [Google Scholar] [CrossRef]
Legates, D.R.; McCabe, G.J. Evaluating the use of “goodness-of-fit” measures in hydrologic and hydroclimatic model validation. Water Resour. Res. 1999, 35, 233–241. [Google Scholar] [CrossRef]
Knoben, W.J.M.; Freer, J.E.; Woods, R.A. Inherent benchmark or not? Comparing Nash–Sutcliffe and Kling–Gupta efficiency scores. Hydrol. Earth Syst. Sci. 2019, 23, 4323–4331. [Google Scholar] [CrossRef]
Gupta, H.V.; Kling, H.; Yilmaz, K.K.; Martinez, G.F. Decomposition of the mean squared error and NSE performance criteria: Implications for improving hydrological modelling. J. Hydrol. 2009, 377, 80–91. [Google Scholar] [CrossRef]
Mizukami, N.; Rakovec, O.; Newman, A.J.; Clark, M.P.; Wood, A.W.; Gupta, H.V.; Kumar, R. On the choice of calibration metrics for “high-flow” estimation using hydrologic models. Hydrol. Earth Syst. Sci. 2019, 23, 2601–2614. [Google Scholar] [CrossRef]
Moriasi, D.; Arnold, J.; Van Liew, M.; Bingner, R.D.; Harmel, R.D.; Veith, T. Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations. Trans. ASABE 2007, 50, 885–900. [Google Scholar] [CrossRef]
Wu, J.; Zheng, H.; Xi, Y. SWAT-based runoff simulation and runoff responses to climate change in the headwaters of the Yellow River, China. Atmosphere 2019, 10, 509. [Google Scholar] [CrossRef]
Lan, Y.; Zhao, G.; Zhang, Y.; Wen, J.; Liu, J.; Hu, X. Response of runoff in the source region of the Yellow River to climate warming. Quat. Int. 2010, 226, 60–65. [Google Scholar] [CrossRef]
Yuan, F.; Berndtsson, R.; Zhang, L.; Bertacchi Uvo, C.; Hao, Z.C.; Wang, X.P.; Yasuda, H. Hydro climatic trend and periodicity for the source region of the Yellow River. J. Hydrol. Eng. 2015, 20, 05015003. [Google Scholar] [CrossRef]
Xie, J.; Xu, Y.-P.; Guo, Y.; Wang, Y. Detecting the dominant contributions of runoff variance across the source region of the Yellow River using a new decomposition framework. Hydrol. Res. 2021, 52, 1015–1032. [Google Scholar] [CrossRef]
Wang, X.; Li, H.; Chen, R.; Liu, J.; Liu, G.; Han, C. Runoff evolution characteristics and driving factors of Yellow River above Lanzhou Station from 1956 to 2020 under changing environment. Adv. Earth Sci. 2022, 37, 726–741. (In Chinese) [Google Scholar]
Wang, W.; Zhang, Y.; Geng, X.; Tang, Q. Impact classification of future land use and climate changes on flow regimes in the Yellow River Source Region, China. J. Geophys. Res. Atmos. 2021, 126, e2020JD034064. [Google Scholar] [CrossRef]
Yilmaz, K.K.; Gupta, H.V.; Wagener, T. A process-based diagnostic approach to model evaluation: Application to the NWS distributed hydrologic model. Water Resour. Res. 2008, 44, 2007WR006716. [Google Scholar] [CrossRef]
Simolo, C.; Brunetti, M.; Maugeri, M.; Nanni, T. Improving estimation of missing values in daily precipitation series by a probability density function-preserving approach. Int. J. Climatol. 2010, 30, 1564–1576. [Google Scholar] [CrossRef]
Ran, S.; Wang, X.; Luo, Y. Predicting climate change and its impact on runoff in snow-ice basin with multi-climate models. Arid Land Geogr. 2021, 44, 807–818. (In Chinese) [Google Scholar]
Shi, X.; Wood, A.W.; Lettenmaier, D.P. How essential is hydrologic model calibration to seasonal streamflow forecasting? J. Hydrometeorol. 2008, 9, 1350–1363. [Google Scholar] [CrossRef]
Moges, E.; Demissie, Y.; Larsen, L.; Yassin, F. Sources of hydrological model uncertainties and advances in their analysis. Water 2020, 13, 28. [Google Scholar] [CrossRef]
Wang, T.; Yang, H.; Yang, D.; Qin, Y.; Wang, Y. Quantifying the streamflow response to frozen ground degradation in the source region of the Yellow River within the Budyko framework. J. Hydrol. 2018, 558, 301–313. [Google Scholar] [CrossRef]
Ma, Q.; Jin, H.-J.; Bense, V.F.; Luo, D.-L.; Marchenko, S.S.; Harris, S.A.; Lan, Y.-C. Impacts of degrading permafrost on streamflow in the source area of Yellow River on the Qinghai-Tibet Plateau, China. Adv. Clim. Chang. Res. 2019, 10, 225–239. [Google Scholar] [CrossRef]
Nijssen, B.; Lettenmaier, D.P. Effect of precipitation sampling error on simulated hydrological fluxes and states: Anticipating the global precipitation measurement satellites. J. Geophys. Res. Atmos. 2004, 109, 2003JD003497. [Google Scholar] [CrossRef]
Su, F.; Hong, Y.; Lettenmaier, D.P. Evaluation of TRMM Multisatellite Precipitation Analysis (TMPA) and its utility in hydrologic prediction in the La Plata Basin. J. Hydrometeorol. 2008, 9, 622–640. [Google Scholar] [CrossRef]
Sun, R.; Yuan, H.; Liu, X.; Jiang, X. Evaluation of the latest satellite–gauge precipitation products and their hydrologic applications over the Huaihe River Basin. J. Hydrol. 2016, 536, 302–319. [Google Scholar] [CrossRef]
Bárdossy, A.; Anwar, F. Why do our rainfall–runoff models keep underestimating the peak flows? Hydrol. Earth Syst. Sci. 2023, 27, 1987–2000. [Google Scholar] [CrossRef]
Yong, B.; Zhang, J.Y.; Wang, G.Q. Key scientific issues of hydrological forecast in the headwater area of Yellow River. Adv. Water Sci. 2023, 34, 159–171. (In Chinese) [Google Scholar]
Tisseuil, C.; Vrac, M.; Lek, S.; Wade, A.J. Statistical downscaling of river flows. J. Hydrol. 2010, 385, 279–291. [Google Scholar] [CrossRef]
Zhang, Y.; Chiew, F.H.S.; Li, M.; Post, D. Predicting runoff signatures using regression and hydrological modeling approaches. Water Resour. Res. 2018, 54, 7859–7878. [Google Scholar] [CrossRef]
Penas, F.J.; Barquin, J.; Alvarez, C. A comparison of modeling techniques to predict hydrological indices in ungauged rivers. Limnetica 2018, 37, 145–158. [Google Scholar]
Shu, L. Impacts of Urbanization and Climate Change on the Hydrological Cycle: A Study in Modern and Ancient Land Use Change; The Pennsylvania State University: University Park, PA, USA, 2017. [Google Scholar]
Jin, J.; Wang, G.; Zhang, J.; Yang, Q.; Liu, C.; Liu, Y.; Bao, Z.; He, R. Impacts of climate change on hydrology in the Yellow River source region, China. J. Water Clim. Chang. 2020, 11, 916–930. [Google Scholar] [CrossRef]
Yuan, F.; Wang, B.; Shi, C.; Cui, W.; Zhao, C.; Liu, Y.; Ren, L.; Zhang, L.; Zhu, Y.; Chen, T.; et al. Evaluation of hydrological utility of IMERG Final run V05 and TMPA 3B42V7 satellite precipitation products in the Yellow River source region, China. J. Hydrol. 2018, 567, 696–711. [Google Scholar] [CrossRef]
Cuo, L.; Zhang, Y.; Gao, Y.; Hao, Z.; Cairang, L. The impacts of climate change and land cover/use transition on the hydrology in the upper Yellow River Basin, China. J. Hydrol. 2013, 502, 37–52. [Google Scholar] [CrossRef]
Coffey, M.E.; Workman, S.R.; Taraba, J.L.; Fogle, A.W. Statistical procedures for evaluating daily and monthly hydrologic model predictions. Trans. ASAE 2004, 47, 59–68. [Google Scholar] [CrossRef]
Nash, J.E.; Sutcliffe, J.V. River flow forecasting through conceptual models Part I—A discussion of principles. J. Hydrol. 1970, 10, 282–290. [Google Scholar] [CrossRef]
Odusanya, A.E.; Mehdi, B.; Schürz, C.; Oke, A.O.; Awokola, O.S.; Awomeso, J.A.; Adejuwon, J.O.; Schulz, K. Multi-site calibration and validation of SWAT with satellite-based evapotranspiration in a data-sparse catchment in Southwestern Nigeria. Hydrol. Earth Syst. Sci. 2019, 23, 1113–1144. [Google Scholar] [CrossRef]

Figure 1. Distribution of the Yellow River source region, river system, and the geographic locations of observation stations.

Figure 2. The unstructured SHUD coarse/fine mesh for the Yellow River source region generated by the rSHUD tool.

Figure 3. Flow duration curves (a), scatter plot (b), and hydrograph processes (c) of daily observed and simulated streamflow at the Tangnaihai hydrological Station.

Figure 4. Flow duration curves (a), scatter plot (b) and hydrograph processes (c) of monthly observed and simulated streamflow at the Tangnaihai hydrological Station.

Figure 5. Hydrographs and scatter plots of daily observed and simulated streamflow at the Tangnaihai hydrological station for 2008 (a,b) and 2014 (c,d).

Figure 6. Monthly scale (a) and annual scale (b) temperature, precipitation, and observed and simulated streamflow at Tangnaihai hydrological station from 2006 to 2018, with temperature and precipitation as the annual averages from CMFD.

Figure 7. Hydrographs (a,c,e) and scatter plots (b,d,f) of daily observed and simulated streamflow at hydrological stations in the Yellow River source region: (a,b) Jimai station, (c,d) Maqu station, and (e,f) Jungong station.

Figure 8. Monthly average values of observed and simulated streamflow (a) and error percentage for simulated streamflow during warm and cold seasons (b) at four hydrologic stations in the Yellow River source region.

Figure 9. Comparison of precipitation on daily (a), monthly (b), and annual (c) scales between meteorological stations and the CMFD in the Yellow River source region.

Table 1. Data description.

Data	Source	Data Specification	Function
Watershed Boundary	DEM	Vector polygon	Model Setup and Parameterization
River Network	DEM	Vector polylines
Elevation	ASTER GDEM	Raster, 1 arc-second
Land Use	LCI MODIS-basedGlobal Land Cover Climatology	500 m resolution
Soil Classification	HWSD (v2.01)	1 km resolution
Meteorological Data	CMFD	0.1 deg, 3 h interval	Model Driving Data/Meteorological Factors Analysis
Stations Precipitation	SURF_CLI_CHN _MUL_DAY_V3.0	Daily precipitation data from 9 meteorological stations in China: Maduo, Dari, Maqin, Jiuzhi, Hongyuan, Ruoergai, Maqu, Henan, and Xinghai.	Precipitation Data Evaluation
Observation Streamflow	Hydrologic station records in the Yellow River source region (2006∼2018)	Station	Model Calibration and Validation
		Jimai (45,019 km²)
		Maqu (86,048 km²)
		Jungong (98,414 km²)
		Tangnaihai (121,972 km²)

Table 2. Model performance statistics of daily streamflow for each year at the Tangnaihai hydrological station.

Year	R²	KGE	NSE	PBIAS (%)	RMSE (m³ s⁻¹)
2006	0.71	0.31	−0.03	14.5	257.9
2007	0.84	0.7	0.69	10.5	269.52
2008	0.83	0.81	0.76	−3.2	180.71
2009	0.89	0.77	0.79	−20.6	257.93
2010	0.71	0.66	0.65	−21.1	324.39
2011	0.87	0.83	0.82	−8.4	202.02
2012	0.86	0.55	0.67	−31.3	449.95
2013	0.72	0.73	0.67	−19	281.76
2014	0.87	0.9	0.86	−7.4	177.44
2015	0.65	0.8	0.61	−4.6	221.79
2016	0.71	−0.25	−1.9	55.1	337.5
2017	0.83	0.87	0.8	5.9	197.07
2018	0.78	0.73	0.73	−19.1	386.63

Table 3. Model performance statistics for calibration and validation periods at the Tangnaihai hydrological station.

Metrics	Calibration Period 2006–2010		Validation Period 2011–2018		Study Period 2006–2018
Metrics	Daily Scale	Monthly Scale	Daily Scale	Monthly Scale	Daily Scale	Monthly Scale
R²	0.74	0.80	0.72	0.79	0.73	0.79
KGE	0.84	0.85	0.79	0.84	0.82	0.86
NSE	0.70	0.76	0.71	0.77	0.71	0.77
PBIAS (%)	−6.3	−6.5	−9.0	−9.1	−8.0	−8.1
RMSE (m³ s⁻¹)	261.1	221.7	297.1	246.6	284.1	238.0

Table 4. Evaluation of streamflow simulation on daily and monthly scales at four hydrologic stations.

Station	Daily Scale from 2006 to 2018					Monthly Scale from 2006 to 2018
Station	R²	KGE	NSE	PBIAS (%)	RMSE (m³ s⁻¹)	R²	KGE	NSE	PBIAS (%)	RMSE (m³ s⁻¹)
Jimai	0.6	0.45	0.08	25.8	116.43	0.69	0.41	0.13	25.6	104.34
Maqu	0.74	0.85	0.73	−3.4	199.81	0.8	0.89	0.78	−3.5	163.81
Jungong	0.71	0.8	0.69	−9.8	247.44	0.78	0.84	0.75	−9.9	205.24
Tangnaihai	0.73	0.82	0.71	−8	284.12	0.79	0.86	0.77	−8.1	237.96

Table 5. Observed and simulated streamflow characteristics in the Yellow River Source Region from 2006 to 2018.

Station	Observed (2006∼2018)					Simulated (2006∼2018)
Station	Q₅ m³ s⁻¹	Q₅₀ m³ s⁻¹	Q₉₅ m³ s⁻¹	Q₅/ Q₅₀	Q₉₅/ Q₅₀	Q₅ m³ s⁻¹	Q₅₀ m³ s⁻¹	Q₉₅ m³ s⁻¹	Q₅/ Q₅₀	Q₉₅/ Q₅₀
Jimai	479.17	145.83	2.31	3.29	0.01	398.15	105.32	28.94	3.76	0.28
Maqu	1098.38	346.06	19.68	3.17	0.06	1255.79	303.24	89.12	4.14	0.29
Jungong	1234.95	393.52	30.09	3.14	0.08	1496.53	359.95	137.73	4.16	0.38
Tangnaihai	1750	439.81	153.94	3.98	0.35	1479.17	474.54	35.88	3.12	0.07

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Bu, T.; Wang, C.; Chen, H.; Meng, X.; Li, Z.; Chen, Y.; Sheng, D.; Zhao, C. An Investigation into the Applicability of the SHUD Model for Streamflow Simulation Based on CMFD Meteorological Data in the Yellow River Source Region. Water 2024, 16, 3583. https://doi.org/10.3390/w16243583

AMA Style

Bu T, Wang C, Chen H, Meng X, Li Z, Chen Y, Sheng D, Zhao C. An Investigation into the Applicability of the SHUD Model for Streamflow Simulation Based on CMFD Meteorological Data in the Yellow River Source Region. Water. 2024; 16(24):3583. https://doi.org/10.3390/w16243583

Chicago/Turabian Style

Bu, Tingwei, Chan Wang, Hao Chen, Xianhong Meng, Zhaoguo Li, Yaling Chen, Danrui Sheng, and Chen Zhao. 2024. "An Investigation into the Applicability of the SHUD Model for Streamflow Simulation Based on CMFD Meteorological Data in the Yellow River Source Region" Water 16, no. 24: 3583. https://doi.org/10.3390/w16243583

APA Style

Bu, T., Wang, C., Chen, H., Meng, X., Li, Z., Chen, Y., Sheng, D., & Zhao, C. (2024). An Investigation into the Applicability of the SHUD Model for Streamflow Simulation Based on CMFD Meteorological Data in the Yellow River Source Region. Water, 16(24), 3583. https://doi.org/10.3390/w16243583

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu