The Juma River is a key tributary of the Daqing River system within the Haihe River Basin. It originates from the southern foothills of Qishan Mountain in Laiyuan County, Hebei Province. Its headwaters are characterized by the emergence of surface runoff from a cluster of underground springs. The river flows generally from northwest to southeast and bifurcates into two branches at Tiesuoya in Laishui County, Baoding City, Hebei Province. The southern branch, known as the South Juma River, continues to flow within Hebei Province, while the northern branch, referred to as the North Juma River, passes through Laishui County (Hebei), Fangshan District (Beijing), and Zhuozhou City (Hebei), eventually discharging into the Baiyangdian water system. The geographical location of the study area is shown in Figure 1.

The Juma River Basin exhibits hydrometeorological characteristics typical of rivers in northern monsoon regions. The average annual precipitation ranges from 550 to 650 mm, with approximately 70% falling during the flood season from June to September. The average annual temperature is between 11°C and 13°C. According to observational data from the Zijingguan Hydrological Station (1955–2019), the basin’s average annual runoff is approximately 1.56 billion cubic meters, with a significant proportion contributed by baseflow. Runoff distribution is highly uneven across seasons, with over 60% occurring in summer (June to August), while winter often experiences flow interruptions. Flood events in the basin are characterized by sharp rises and falls in discharge. The historical maximum peak flow has reached 10,000 m³/s. During the extraordinary flood event of July 2023 (“7·23” event), the runoff coefficient at Zhangfang Station reached 0.40, surpassing historical records and highlighting the basin’s hydrological sensitivity under extreme climatic conditions. Under the influence of global climate change, recent observations indicate a marked increase in the frequency of extreme precipitation events within the basin. From 2010 to 2020, the number of heavy rainfall days increased by 23% compared to the baseline period (1961–1990), exacerbating flood risk and intensifying pressure on flood management. This study focuses on the upstream region of the Juma River above the Zijingguan Hydrological Station. The area is located within a temperate continental monsoon climate zone and is characterized by distinct seasonal variation. Summers are typically hot and humid, whereas winters are cold and dry.

The fundamental data used in this study comprise two main categories: geographic information data and rainfall–runoff event data. The geographic information data primarily include the Digital Elevation Model (DEM), river network distribution, land use types, and soil texture. All datasets were obtained from the National Geomatics Center of China.¹ Standard pre-processing, including resampling of DEM and soil texture data to a consistent spatial resolution and reclassification of land use categories to match model input requirements, was performed to ensure compatibility with the CNFF model structure. Within the study basin, 14 rainfall stations and 2 hydrological stations are installed. Due to insufficient data availability at the Shimen Hydrological Station, this study selects the Zijingguan Hydrological Station as the forecast section. The Zijingguan (River) Station serves as the control point for the Zijingguan section, with a controlled drainage area of 1868.47 km². The basin’s topographic features and the spatial distribution of monitoring stations are illustrated in Figure 2. The hydrological data for the Zijingguan section are derived from observed records, which have been compiled and verified, ensuring high reliability. Hourly rainfall data from 14 rainfall stations—including Xieshan, Aihecun, Laiyuan, Chajin, Tuanyuancun, Huziyu, Shiziyu, Shimen, Dongtuanpu, and Wulonggou—were collected for the period 1955–2019. Additionally, flood event records for the Zijingguan section were compiled. From these, seven representative flood events were selected for model parameter calibration. These seven historical flood events were selected for model calibration to represent a diverse range of hydrological conditions, including varying flood magnitudes (from small to large peak discharges and runoff volumes), different seasons within the typical flood period of the basin, and distinct antecedent rainfall patterns. This selection aims to ensure robust parameterization and comprehensive model evaluation.

To evaluate the accuracy of flood forecasting, fur key indicators are employed: Relative Error of Runoff Depth (RER), Relative Error of Peak Discharge (REQ), Peak Time Error (TP), and the Nash–Sutcliffe Efficiency Coefficient (NSE) (Mathevet et al., 2006). The optimal values for RER, REQ, and TP are 0, while the optimal value for SNSE is 1. In model performance evaluation, simulation results are considered acceptable when the absolute values of RER and REQ are within 20%, TP is within ±2 h, and SNSE exceeds 0.6. The calculation formulas for these indicators are as follows:

Where Rs is the simulated runoff depth (mm); Ro is the observed runoff depth (mm); Qs is the simulated peak discharge (m³/s); Qo is the observed peak discharge (m³/s); Ts is the simulated time to peak (h); To is the observed time to peak (h); Qs(i) is the simulated discharge at time step i (m³/s); Qs(i) is the observed discharge at time step i (m³/s); Qo ¯ is the mean observed discharge (m³/s); and n is the length of the flood event time series.

Accuracy Grading: Based on the ‘Specifications for Hydrological Forecasting Information’ (GB/T 22482–2008), flood forecast accuracy is categorized to standardize performance evaluation. For key indicators such as the Relative Error of Peak Discharge (REQ) and Relative Error of Runoff Depth (RER), a forecast is typically classified as follows:

For the Peak Time Error (TP), a forecast within ±2 h of the observed peak is generally considered to meet high-grade accuracy requirements (i.e., Class A). This framework is now introduced early in our methodology and is used consistently to interpret model performance in the Results and Discussion sections.

Taking the upper reaches of the North Juma River Basin as the research object, the runoff generation, runoff concentration, evaporation, and river channel evolution patterns of each model are shown in the following table (see Table 1), the use of a ‘Three-layer Evaporation’ module in the Xin’anjiang model versus a ‘Daily Average Evaporation’ in the other two reflects a key structural difference in how soil moisture dynamics are accounted for. Conversely, employing the ‘Dynamic Muskingum Method’ for river routing in all three models ensures consistency in the confluence process, allowing our comparison to focus primarily on the differences in the runoff generation modules.

The parameters of each of the three hydrological models were optimized using a multi-event joint calibration approach, which involved simultaneously utilizing seven representative flood events (see Table 2) to identify a single, robust parameter set that best captures the overall rainfall-runoff characteristics of the upper Juma River basin. The calibration process was conducted by manually adjusting the parameters to maximize the Nash–Sutcliffe efficiency (NSE) and minimize errors in runoff depth and peak discharge across all seven flood events. The final optimized parameter values for each model are presented in Tables 3–5. All the calibrated parameter values fall within the physically reasonable range for similar climatic regions, confirming the reliability of the calibration results. The key parameters of each model are described in the following sections.

The simulation results of seven selected flood events using three hydrological models—Xinanjiang (XAJ), Vertical Mixed Runoff Generation (HHC), and Dahuofang (DHF)—were evaluated based on four key performance indicators defined by Equations (1–4) described in Section 2.3. Table 6 provides a comprehensive summary of this comparative analysis. For each flood event and each model, the table presents the Nash-Sutcliffe Efficiency (NSE), the Relative Error in Runoff Depth (RER, in %), the Relative Error in Peak Discharge (REQ, in %), and the Peak Timing Error (TP, in hours). Furthermore, based on the accuracy criteria outlined in our methodology, a “Pass/Fail” assessment is provided for each indicator to quickly evaluate whether the simulation meets the operational forecasting standards. A “Pass” indicates that the result falls within the acceptable range (e.g., |REQ| ≤ 20% and |TP| ≤ 2 h for Class A). The following subsections provide a detailed analysis and discussion of these results, assessing the overall performance, stability, and applicability of each model for flood forecasting in the Upper Juma River Basin.

Selected flood events are presented to illustrate the model performance under different flood types. The event on August 17, 1955 (19550817) is used as an example of a large flood with a double peak; the event on August 8, 1963 (19630808) represents an extraordinary flood with a double peak; the event on July 21, 2012 (20120721) demonstrates a large flood with a single peak; and the event on July 20, 2016 (20160720) serves as an example of a small flood with a double peak. The observed and simulated hydrographs for these events are shown in Figure 3.

Overall, the simulation performance was better for large flood events and poorer for small floods. As flood magnitude increased, the simulation accuracy for single-peak floods improved. However, for the flood event on July 20, 2016, all three models exhibited a delayed peak time, which is consistent with the findings of Zhai et al. (2020), who reported that under varying rainfall intensities—from moderate rain to extremely heavy rain—the average flood peak appeared earlier and the passing rate of the determinacy coefficient improved, indicating a significant enhancement in simulation performance. Nonetheless, under moderate rainfall conditions, the CNFF model still exhibited a certain degree of delay in simulating the peak time. Among the three models, the Xin’anjiang model showed superior simulation results compared to the Vertical Mixed Runoff Generation Model and the Dahuofang Model. In summary, the Xin’anjiang model demonstrates higher simulation accuracy and is more suitable for application in the upper reaches of the Juma River compared to the other two models.

The study demonstrates that the flood simulation results of the Xin’anjiang model show a high degree of agreement with observed data in the upper reaches of the Juma River, confirming its applicability in the semi-arid to semi-humid mountainous regions of northern China. This finding, which confirms the applicability of the Xin’anjiang model in this basin, is consistent with the results of Che (2024). The upper Juma River basin is characterized by a temperate continental monsoon climate, with a flood season that is concentrated between June and September. Precipitation predominantly occurs in the form of heavy rainfall. The region has shallow soil layers with limited water storage capacity, a deep groundwater table, and a low contribution from baseflow. The Xin’anjiang model is based on the saturation-excess runoff generation mechanism, wherein rainfall generates surface runoff only after the soil becomes saturated (Lu, 2021; Jin et al., 2022; Ren-Jun, 1992). This model utilizes a “runoff yield capacity curve” to simulate the soil saturation process, effectively capturing runoff generation following rapid soil saturation. Given the shallow soil and low water retention capacity of the Juma River basin, even moderate rainfall can quickly saturate the soil, triggering saturation-excess runoff. This characteristic enables the Xin’anjiang model to accurately capture the key processes of rainfall-runoff transformation, thereby improving flood forecasting accuracy in the region.

The Vertical Mixed Runoff Generation Model focuses on simulating vertical water exchange between the soil and groundwater, making it more suitable for regions with significant baseflow contributions. However, the Juma River basin has a deep groundwater table and a negligible baseflow contribution (<10%). The model’s emphasis on vertical processes results in underestimation of surface runoff. Moreover, it typically assumes sufficient soil infiltration capacity and neglects the infiltration-excess runoff mechanism. During intense rainfall events, this leads to an underestimation of runoff volumes. For example, during the 20,120,721 flood event, characterized by high rainfall intensity, the model significantly underestimated peak discharge compared to observed values, indicating its inability to accurately simulate flood responses under high-intensity rainfall. This shortcoming demonstrates that the model is not suitable for the Juma River basin or similar mountainous regions in northern China, particularly under extreme precipitation conditions.

The Dahuofang Model is centered on the infiltration-excess mechanism, where surface runoff is generated only when rainfall intensity exceeds soil infiltration capacity. This mechanism performs well in regions with thick soils and high water retention capacity. However, in the Juma River basin, soil often becomes saturated after continuous rainfall, meaning that even moderate rainfall can produce runoff via the saturation-excess mechanism. The Dahuofang Model fails to represent this process. When the soil is unsaturated, the model assumes continuous infiltration, overlooking the threshold effects of saturation-excess runoff. Consequently, the model underperforms during initial rainfall and prolonged rain events. For example, under pre-event drought conditions, the model may allocate all early rainfall to infiltration, delaying or underestimating the flood peak. Furthermore, the model is highly sensitive to infiltration-related parameters, requiring accurate calibration of initial infiltration rate, infiltration decay coefficients, and others. However, infiltration capacity in the Juma River basin is heavily influenced by antecedent soil moisture conditions (e.g., low water absorption in dry soils), complicating parameter calibration and undermining the reliability of flood forecasting.

In different climatic regions—particularly under short-duration, high-intensity rainfall—both saturation-excess and infiltration-excess mechanisms may coexist within small watersheds. Since a single runoff generation mechanism cannot fully capture the diverse rainfall-runoff responses of various basins, it is difficult to establish an accurate quantitative relationship using a single model type (Zhai et al., 2020). The geographic and climatic conditions of the Juma River basin may cause saturation-excess runoff to occur rapidly at the onset of rainfall, While infiltration-excess runoff can also occur during localized downpours—conceptually, for example, when hourly rainfall exceeds 30 mm and surpasses the soil’s infiltration capacity even if the soil is not saturated—this threshold is used illustratively here unless specific regional studies inform it. The generally good performance of the Xin’anjiang model (XAJ), particularly in capturing the initial rising limb and peak timing for many events, suggests that saturation-excess runoff, driven by rising water tables in shallow soils, is a dominant mechanism in the Juma basin. However, occasional underestimations during very intense, short-duration rainfall within larger events might hint at challenges in fully capturing localized infiltration-excess contributions, suggesting a potential co-existence or temporal shift in dominant mechanisms under specific storm characteristics. Therefore, a model capable of handling both mechanisms simultaneously is needed to accommodate the spatial and temporal variability of rainfall in the Juma River basin, particularly under conditions such as early drought followed by intense, concentrated rainfall.