Applicability of the HBV model to a human-influenced catchment in northern China

With rapid socioeconomic development, domestic water demands have largely increased since the 1980s, particularly in arid areas (Liu et al. 2019). As a result, the observed stream flow has declined significantly in most rivers in dry climate regions (Wang et al. 2020). Meanwhile, both land use change induced by intensive human activities, and climate change in the context of global warming, have complicated runoff yield mechanisms and altered the hydrological cycle (Zhang et al. 2014; Sohoulande Djebou 2015). Environmental changes are challenging hydrological modeling and forecasting, as well as water resources management efforts (Guan et al. 2019; Liu et al. 2021a, 2021b). The mechanism of climate change and water use altering ephemeral rivers in (semi-)arid regions is, therefore, treated as one of the 23 Unsolved Problems in Hydrology (UPH) identified by the IAHS (International Association of Hydrology Sciences) community (Blöschl et al. 2019). Mathematical models can help to understand the impact mechanism of environmental change on the hydrological process under different land use scenarios (Bao et al. 2019; Song et al. 2020). However, previous studies indicated that hydrological models have a certain regional suitability due to differences in model structure, and runoff yield mechanisms (Rosero et al. 2009; Guan et al. 2019). It is, therefore, important to test a model's performance for a specific region before its application to support efforts such as flood control, water allocation, and drought relief.

As a conceptual hydrological model, the HBV (Hydrologiska Byrans VattenbalansavdeIning) model has been applied to hundreds of catchments all over the world (Bergström 1992; Engeland & Hisdal 2009; Dakhlaoui et al. 2012; Wu et al. 2017; Huang et al. 2019; Osuch et al. 2019), including the Huai River Basin in eastern China. The basin's runoff yield is mainly based on the mechanism of saturation excess (Li et al. 2012; Xu 2021). Zhao et al. (2007) applied the HBV model to the Guanzhai River, a tributary of the Huai River, and found that it simulated rainfall–runoff process well with the Nash–Sutcliffe model efficiency (NSE) coefficient over 0.75. Based on the discharge–stage relationship, the HBV model was used to determine the critical rainfall threshold of flash floods for the upper reaches of the Huai River (Lu & Tian 2015). Rainfall and snowfall are two types of precipitation in cold areas of northeastern China. The mechanism of runoff yield in cold regions is more complex than that in southern China because it needs to consider runoff generation from rainfall and snowmelt (Stewart 2010). Zhang et al. (2007) and Wu et al. (2017) tested the performance of the HBV model for discharge simulation of the Naoli River and the Mudan River in northeastern China, where communities are frequently threatened by both ice floods in the spring and storm floods in the summer and autumn. They found that the HBV model was effective in simulating discharge on both daily and monthly scales for these two river basins, with NSEs over 0.8 in most cases. China's coastal areas in the southeast are subject to high-intensity rainfall and frequent floods. The HBV model has also proved to be applicable to these areas according to the work by Hu et al. (2008) and Wang et al. (2014) .

Numerous studies indicate that the HBV model is applicable to catchments in different climatic zones (Zhao et al. 2007; Hu et al. 2008; Yang et al. 2010; Dakhlaoui et al. 2012; Wu et al. 2017). However, most case studies of the HBV model application employed a very consistent data series that was only slightly influenced by human activities (Vormoor et al. 2018; Jin et al. 2019). Due to the effects of intensive human activities, hydrological processes in arid or semi-arid regions have changed significantly (Sun et al. 2019; Yan et al. 2020; Guan et al. 2022). It is critical to evaluate the performance of the HBV model for widening its application. We took the Nanjuma River, located in a semi-arid region of China, as a case study to investigate hydrological changes of the river basin under a changing environment, and to evaluate the performance of the HBV model in different human-disturbance scenarios. The findings will provide strong support to improve capacity building of early warning, forecasting, rehearsing, and pre-scheming in flood control.

The Nanjuma River originates from Tiesuoya Mountain, flows through Yi County and Dingxing County and joins the Baigou River in Baigou Town, Gaobeidian, Hebei Province, forming the Daqing River, one of the major tributaries of the Hai River. The Nanjuma River Basin covers a drainage area of 2,157 km², bordered by the Cao River and Pu River in the south and the Beijuma River in the north. Influenced by the temperate continental monsoon, the climate of the basin is characterized by hot and rainy summers and autumns, and cold and dry winters and springs.

The Nanjuma River catchment is a highly developed agricultural area, where several large-scale irrigation regions exist, namely the Juyue irrigation region, Fanglaizhuo irrigation region, Caijiajing irrigation region, etc., with a total irrigation area of approximately 12,330 km². The irrigated area in the river basin reached its peak in the 1990s. As of 2021, 20 reservoirs have been constructed on the river and its tributaries for water supply and irrigation since the 1960s, including large and medium-sized reservoirs such as the Angezhuang Reservoir, Wanglong Reservoir, and Matou Reservoir, with a total capacity of 0.33 billion m³. Most of the reservoirs were reinforced and put into regular operation after the 1970s.

According to the observations of hydro-meteorological variables, the average annual temperature is approximately 8.7 °C. The average annual precipitation is less than 500 mm, with 75% of the falls in the flood season from June to September. Meanwhile, the annual precipitation shows high inter-annual variability, with a ratio of the maximum value to the minimum value exceeding 3.0.

The HBV model, developed by the Swedish Meteorological and Hydrological Institute (SMHI), is often used in discharge simulation, flood forecasting, and assessment of environmental change (Deckers et al. 2010; Chen et al. 2012; Tian et al. 2013). The model can simulate stream flow for a small-scale catchment as a lumped hydrological model. It also can be applied to distributed hydrological simulation for a large-scale river basin by dividing the catchment into sub-basins using the DEM (digital elevation model). The HBV model consists of the following four kernel modules: snow melting, soil moisture calculation, hydrological response, and flow routing. Compared with other hydrological models, the HBV model has the advantages of a simple structure, relatively few parameters, a physically based mechanism of runoff yield, and ease of application (Liu et al. 2021a, 2021b). The model inputs include daily precipitation, air temperature, and potential evaporation (often substituted by the observed E601 evaporation). The model outputs include the simulated daily rainfall–runoff process of each sub-basin, and the simulated discharge at the outlet hydrometric station of the study basin.

Influenced by changes in climate and human activities, data series of streamflow may vary with time while presenting a certain tendency. It is of importance to detect the variation trend of hydrological series for understanding the impact of human activities. The Mann–Kendall (M–K) nonparametric trend test (M–K method) is a commonly used method to detect the variation trend of time series (Mann 1945; Wang et al. 2013). Based on an assumption of stability of a time series, the method does not require the samples to follow a certain statistical distribution. The defined M–K statistic approximately follows a standard normal distribution with a positive/negative value indicating an increasing/decreasing trend. A significant variation trend is detectable when the absolute M–K statistic is greater than 1.96 at the significance level of 0.05.

According to the variation in annual runoff, the data series could be roughly divided into three phases: 1961–1979, 1980–1999, and 2000–2017. The first phase can be characterized as a natural alternation of wet and dry in the annual runoff, where the runoff varies from 400 to 100 mm. Runoff in the second phase shows a more considerable disparity and variability, ranging from 500 to 20 mm. The third phase has a steady low stream flow with an annual runoff under 30 mm in all years and zero flow in several years.

Previous studies show that the Hai River Basin has been increasingly influenced by human activities since the late 1970s (Bao et al. 2012; Yan et al. 2020). The Hai River Basin could be treated as a quasi-natural state before 1980 as it was only slightly influenced by human disturbance. After 1980, the recorded runoff gauged at key hydrometric stations on the Hai River decreased by more than 50% relative to the previous period while the precipitation approximately reduced by 10% (Wang et al. 2020). The sensitivity analysis indicated that a 10% change in precipitation will lead to an 18–25% change in the runoff the for Hai River basin (Wang et al. 2017), which is much lower than the observed runoff reduction, illustrating a huge human influence on stream flow in the 1980s and beyond.

To evaluate the performance of a hydrological model, a data series with good consistency and longer time span covering at least 7–8 consecutive years is needed (Guan et al. 2019). Previous analysis shows that the consistency of the recorded discharge series of the Nanjuma River Basin from 1961 to 2017 was destroyed due to environmental changes. We, therefore, calibrated the HBV model based on the three segment phases identified in Section 2.1. For the data series of each segment phase, the first year is taken as the warm-up period to minimize the effect of the initial state condition, which is usually given based on expert knowledge. The data in the last 3–5 years are usually used to validate the model. The rest of the data series is used for model calibration. The calibration and validation periods of each segment data series are given in Table 1.

We took the NSE as the objective function. The model parameters under different human influences were calibrated and are given in Table 2. The model performance on discharge simulation is statistically given in Table 3.

Table 2 shows that the calibrated model parameters in the first natural phase are close to those in the second period in which human activities moderately influenced hydrological processes. However, there is a large difference between parameters in the natural phase and those in the third phase when intensive human activities strongly influenced the stream flow.

Table 3 indicates that (1) the HBV model performs better on monthly discharge simulation than on daily discharge simulation. The NSEs on monthly scale are much higher than those on the daily scale for all three phases. (2) The HBV model works the best on discharge simulation in the first phase when the catchment was in the quasi-natural state. Both daily-scale NSEs in the calibration and validation periods exceed 0.6, while monthly-scale NSEs in both periods are 0.88 and 0.71, respectively. The relative errors of simulations in the calibration and validation periods are quite small, −3.6 and 2.7%, respectively, representing a good performance of the HBV model in the Nanjuma River. (3) The HBV model is acceptable for discharge simulation with daily-scale NSEs over 0.45 and monthly-scale NSEs over 0.55 when the catchment was moderately influenced by human activities, but the RE of simulations is larger than those in the natural state. (4) The HBV model has a mediocre performance for simulating hydrological processes on both daily and monthly scales when the catchment was strongly disturbed by intensive human activities. In the period 2000–2017, NSEs on both daily and monthly scales are quite small and even below zero while REs are beyond 15%. It is also found that the HBV model overestimates discharge when the catchment was moderately or strongly influenced by human activities.

Figure 4 shows that the hydrographs of observed and simulated runoff fit nicely on daily and monthly scales, which means the HBV model performs well for discharge simulation in the natural phase. The HBV model may overestimate or underestimate peak discharge while low flow is generally underestimated with the exception of that in several dry months. In addition, the absolute simulation error of high discharge is higher than that of low discharge, but the RE of high-flow simulation is relatively low with comparison with that of low-flow simulation. However, an NSE is sensitive to simulation of high flow. A good simulation of high flow can effectively increase the NSE.

The Nanjuma River Basin is in the East Asian Monsoon climate zone. Most of the high flows in the basin result from torrential rain in flood seasons, which often occurs abruptly and is highly heterogeneous in distribution in space. The density of rain gauges can influence model performance to some extent (Tegegne et al. 2017). Although the selected five meteorological stations are evenly distributed in the study basin and can monitor most of the large-scale rainfall information all over the catchment, it may miss local rainstorm information in some cases. It is, therefore, essential to increase the density of the hydrological monitoring network and acquire more information based on remote sensing, radar monitoring, etc., to improve the accuracy of peak discharge simulation in semi-arid regions.

The low flow in dry seasons mainly consists of subsurface flow and baseflow, which are generated from unsaturated and saturated soil zones. In this study, the NSE was taken as the objective function to calibrate the HBV model. Previous studies show that the accuracy of high-flow simulation can affect the NSE value to some extent (Guan et al. 2019). The NSE will probably be high when high flows are well simulated, while the NSE is not sensitive to changes in the accuracy of low-flow simulation. It is essential to select a suitable objective function if we take low flow as a focus.

Regional climate, topography, as well as human activities can also influence model performance for flow simulation. Numerous studies indicated that hydrological models perform better for the humid catchments than for the arid catchments, while hydrological modeling for plain catchment is a greater challenge than for mountain catchment (Yang et al. 2000; Wang et al. 2006; Vilaseca et al. 2021). In addition, intensive human activities make the mechanism of runoff yield more complex and thus change the hydrological regime (Liu et al. 2021a, 2021b). In this study, due to the influence of human activities, the HBV has been increasingly losing its capacity to capture real features of hydrological processes (Figure 7). It is urgent to enhance the research on mechanisms of human impact on hydrology to improve hydrological modeling and forecasting in a changing environment.

Environmental changes, particularly intensive human activities, are challenging the hydrological modeling. It may be helpful to strengthen the capacity of the HBV model in hydrological modeling by improving the model structure, selecting suitable objective functions, and using data from multiple sources. However, it is critical to quantify impacts of human activity and incorporate the quantified impacts in further studies.

In this study, taking the Nanjuma River Basin, a human-disturbed catchment, as a study case, we analyzed changes in the hydrological regime of the river basin under a changing environment, and investigated performance of the HBV model in different situations of human disturbances. Results can be concluded as follows:

The recorded streamflow decreased significantly under the influence of climate change and human activities. Characteristics of runoff variation and human activities in the river basin suggest the data series could be divided into three phases: the quasi-natural period from 1961 to 1979, the moderate human-disturbance period from 1980 to 1999, and the intensive human-disturbance period from 2000 to 2017.

The HBV model has a good performance in simulating hydrological processes on daily and monthly scales when the Nanjuma River catchment was in a natural state. The NSEs in both the calibration and validation periods are above 0.63 and the RE of simulation is less than 5%. Relatively, the model performs better in monthly discharge simulation than in daily discharge modeling.

The HBV model is acceptable for discharge simulation when the catchment is moderately influenced by human activities although the RE of simulation is greater than that in the natural state. The HBV model almost totally lost its capacity to capture hydrological features when the catchment was intensively regulated. It is, therefore, of great importance to study hydrological modeling under a changing environment in further study for supporting operational flow forecasting and water resources management.

This research was financially supported by the National Natural Science Foundation of China (Grant nos 41830863, 52121006, and 92047301) and the National Key Research and Development Programs of China (2021YFC3200201).

Yu.W. did formal analysis and wrote the original draft; Yanj.W. and Z.B. conceptualized and prepared methodology; J.J. and Yan W. did data curation and software analysis; Q.J. and X.D. led discussions and gave suggestions; C.L. and G.S. collected data and edited the article.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.