The watershed hydrological model is a significant tool for hydrological simulation. At present, hydrological models are mainly divided into conceptual models and physically based distributed models. Conceptual hydrological models generalize runoff generation and runoff routing processes based on the mass conservation and experimental simulation or empirical function relationship. The conceptual model is simple and easy to be established, but the parameters are difficult to be determined by the theoretical formula. Generally, it needs to be calibrated according to the observed flow [1
], which limits the application of the conceptual model in ungauged areas. For the physically-based distributed hydrological models, their parameters can be estimated by catchment characteristics and meteorological conditions, which is able to fully consider the influence of soil, land use types and terrain characteristics, and better describe the spatial variability of watershed. However, due to the problems of over parameterization and scale, the use of complex models does not necessarily lead to better hydrological simulation results at watershed outlet [3
]. In view of the advantages and disadvantages of conceptual hydrological models and physically based distributed hydrological models, Robinson et al. [8
] suggested reconsidering the use of simpler approaches and found connections between physically based and conceptual models. Later on, Koren et al. [4
] investigated an approach that combined the features of a widely used the conceptual SAC–SMA model and physically based kinematic routing model in the development and parameterization of a modeling system referred to as HL-RMS. A modeling framework in which lumped, semi-distributed and fully distributed approaches could be constructed and tested was provided in their study. The procedures for deriving a priori parameters as well as the computational efficiency of the HL-RMS have been supported by the Distributed Model Intercomparison Project (DMIP) [9
]. Moore et al. (2006) [11
] developed a conceptual-physical area-wide model (e.g., the G2G model) supported by terrain and soil property data, and demonstrated the potential value of distributed conceptual-physical models for flood warning and for identifying flood prone locations.
The Xin’anjiang model is a popular conceptual hydrological model in China. Since it was first developed in the 1970s, it has been widely used in humid and semi-humid areas, and has been employed into China’s national flood forecasting system [12
]. However, the determination of sensitive parameters is a significant problem for the model. Generally, it is difficult to estimate these parameters by geographical information so that these parameters have to be calibrated through historical observation especially runoff routing parameters. For instance, the river network recession constant Cs
] of the lag-and-route method represents the retention and storage capacity of river network and quantifies the flood peak attenuation. It has been widely recognized that the simulated flood events by the Xin’anjiang model are extremely sensitive to this parameter. Generally, the value of Cs
related to catchment drainage area, river network topology, overland slope and flow velocity, etc. Although many studies have applied to the estimation this parameter values, they have some limitations. For example, Zhao [15
] deduced the equation of Cs
with the change of the reference river length and flow velocity from the analysis of physical mechanism, but the method still needs to rely on the observed hydrological data. Later, some scholars utilized a statistical method to study the relationship between Cs
and geomorphic characteristics. For example, Xu et al. [16
] calibrated the parameters of 13 watersheds by observed data in Huangshan area, China, and obtained the empirical relationship between Cs
and drainage area. Hu et al. [17
] extracted 10 underlying surface factors of 29 watersheds in Southern Anhui province, China, and then established the regression relationship between Cs
and the underlying surface characteristics. However, these methods can only achieve good simulation results for catchments with similar hydrological properties as studied catchments. Therefore, it is difficult to apply these methods to other catchments. Besides, the parameter Lag
, another important parameter controlling the river network routing in the Xin’anjiang model and influencing flood propagation and peak time, is also not easy to estimate due to lack of clear physical meaning. What is more, the Muskingum method [18
] employed by the Xin’anjiang assumes that the channel is a virtual channel so that its parameters cannot be estimated by channel characteristics and hydraulic properties. Specifically, it is generally known that the time constant Ke
of Muskingum method is identical to concentration time interval of the sub catchment, and another parameter x still needs to be calibrated by observation. Overall, it is challenging to apply the Xin’anjiang model in ungauged area.
With the development of Digital Elevation Model (DEM) technology, more geographic data (river channel topology, slope and underlying surface characteristics) can be obtained. Some runoff routing schemes with physical basis of routing methods have been developed based on DEM for the Xin’anjiang model over the years. For example, Yao [19
] proposed the distributed Grid-Xinanjiang model, a distributed conceptual-physical model. It successfully applied the grid-based model to two Chinese catchments for the simulation of hydrologic processes and demonstrated that the model has the capability of providing valuable spatial distributions of hydrologic variables. A physically based approach using geographically based information for a priori parameter estimates was presented as well, and it was proved that the proposed method is effective [20
]. However, the parameters in the model are more complex and the scale of model parameters needs further study. In addition, to improve the flood prediction capability of the Xinanjiang model in ungauged nested catchments, Yao [21
] used the concept of geomorphic unit hydrograph (GIUH) with physical basis to improve the Xin’anjiang model, and established the Xin’anjiang (XAJ)-GIUH model. However, it still uses the Muskingum method for channel routing, so its parameter must be calibrated in ungauged area.
The purpose of this paper is to determine a simple, efficient and reasonable routing approach by combining the conceptual and distributed model feature so that the model parameters can be estimated by geographical information. The isochrone method [22
] was chosen in this paper to replace the lag-and-route component of the Xinanjiang model. It has the advantage of requiring a limited number of parameters whereas the necessary geomorphologic data can be obtained from DEM. Furthermore, we used the grid-to-grid Muskingum–Cunge–Todini (MCT) method [25
] for channel routing. The MCT method was selected to improve the channel routing module of the Xin’anjiang model for the following reasons: First, it can guarantee the mass conservation in the calculation process. Second, its parameters can be determined according to the river channel geometric characteristics and hydraulic properties. Third, MCT method can be easily programmed and have high computation efficiency.
This paper is organized as follows: the methodology of the new catchment scheme is proposed in the Section 2
; a concise description of the research catchment is given, and data sources and model comparison methods are proposed in Section 3
; In Section 4
, the improved model and Xin’anjiang model are applied to the study catchment, and the simulation results of the two models are compared and discussed; additionally, the model’s ability to predict hydrologic processes at interior points without model recalibration is evaluated. The last section gives a summary and more discussions on the limitations and future directions for improvements.
This paper presents a new runoff routing scheme and coupled it with the Xin’anjiang model, in which the isochrone method is employed for overland routing and grid-to-grid MCT method is employed for main channel routing. What’s more, based on the geographic information of the watershed, we establish a set of parameters estimation methods for the new runoff routing scheme. Compared with the original Xin’anjiang model, the advantage of the improved model is that it can predict the flow for internal stations, and its routing parameters can be determined according to geographic information. In this scheme, the length of routing path from each grid flow to the main river channel is determined by DEM based on D8 principle. The flow velocity of overland is mainly determined by land cover types and average slope. Hence, we can determine the lag time of each grid droplet arriving at the main channel. The lag time histogram is employed to determine the percentage of the overland flow arrived at the river channel for each sub catchment at each time step. After using MCT method instead of Muskingum method, the parameters of the model can also be estimated according to river order, channel geometry and channel gradient. This makes the runoff routing scheme established in this paper can be used only with DEM and a multiplier on overland Manning’s roughness . The strengths of our new routing scheme have been demonstrated through its applications to the Tunxi catchment with four interior stations. This approach is a powerful tool for linking the hydrologic response of catchments to their geomorphologic characteristics, which can provide an insight to the hydrologic behavior of ungauged catchments.
Although tests performed in this study with observed data have shown the potential of the new routing scheme model, two additional issues need further research to improve the prediction accuracy of the new runoff routing model.
(1) The problems of steep rising and falling limbs hydrographs and early peak time caused by the traditional isochrone method, due to constant flow velocity and lack of the consideration of river network routing, can be significantly reduced by the new routing scheme. In the future study, other flow velocity formula such as Manning’s formula can be considered to replace the overland velocity formula in this paper for overland routing, so that the average overland velocity changes with time. In addition, the regulation and storage capacity of river network still needs to be further addressed.
(2) With the changes of underlying surface conditions affected by human activities and climate change, land cover types are constantly evolving. Therefore, in the case of land cover and land use change, the fixed model parameter values are not always applicable. Improving the quality of input data (precipitation, land cover, soil type and channel basic information, etc.) is beneficial to increase the accuracy of flood simulation and forecasting. It is anticipated that a new efficient methodology that utilizes appropriate survey information, topographic maps and remote sensing data will be employed in further applications of the model.