FORMULATION AND VALIDATION OF HYBRID CONCEPTUAL MODELS FOR RUNOFF GENERATION

Abstract

A number of well-established conceptual and physically based modelling approaches are available for the purpose of simulation of rainfall-runoff process of various catchments. In the development of such models, the runoff is generated based on infiltration excess runoff generation concept for separating excess rainfall that generates runoff from the uniformly distributed rainfall over the catchment. Although such models are used throughout the world, one school of thought attributes the failure of some of these models to their inability to reproduce the dynamic variation of the saturated areas within the catchment, particularly in the catchments located in the humid climatic zones. The non-linear nature of catchment response to storm events could be attributed to the dynamic variation in the accumulation and horizontal movement of water in the upper layers of the soil. Accordingly, the catchments produce runoff based on the saturation excess runoff concept which considers that the runoff from any point of the catchment is generated for the incident rainfall at that point only when the soil tension water capacity requirement at that point is fully satisfied by the incident rainfall. Also, the runoff is generated for a given rainfall only from that fraction of the area of the catchment wherein the soil tension water capacity requirement is fully satisfied. Based on the Dunne’s concept of soil moisture replenishment, depletion and redistribution mechanism, many models have been developed. Notable among them is the Xinanjiang model, which is taken as the base model in the present investigation. The Xinanjiang model represents the dynamic variation of the saturated areas through a conceptual distribution function for reproducing the catchment response with a smaller number of quasi-physically meaningful parameters for large scale catchments in humid climatic zones. While catchments of humid climate zones may follow the saturation excess runoff generation mechanism, the catchments located in dry and average climate zones may still follow the infiltration excess runoff generation mechanism. Accordingly, Hu et al. (2005) applied the concept of combined, i.e., saturation excess and infiltration excess runoff generation mechanisms, for runoff generation of three catchments of China and they showed that the combined mechanisms of runoff generation is able to reproduce the observed runoff closely for humid and semi-humid catchments. A careful study of the interpretation of Hu et al. (2005) about the concept of Horton infiltration capacity leads to the inference that it is the lumped representation of the point variability of the infiltration capacity of the pervious area of the catchments at any time during the rainfall process. This interpretation enables one to consider the Horton infiltration capacity used in the infiltration excess runoff generation mechanism takes care of the point variability of the infiltration capacity rate ii | P a g e throughout pervious area of the catchment. Therefore, there is a necessity for studying runoff generation of the catchment based only on Horton’s runoff generation mechanism based on its interpretation given by Hu et al., (2005) for rainfall-runoff modelling. The present study uses this interpretation for runoff generation. Besides the study also uses the combined mechanisms of runoff generation using SCS-CN method and as well as saturation excess method based on Zhao et al. (1992) approach. Therefore, in the present study, the following modifications in runoff generation mechanism of the Xinanjiang model have been proposed i) Incorporation of Soil Conservation Curve Number (SCS-CN) formulation for surface runoff generation to take care of infiltration excess runoff generation mechanism, which is ubiquitous in most of the catchments and missing in Xinanjiang model. In the proposed formulation, the spatial soil moisture capacity (WM) is considered as the function of the parameter S (maximum retention potential of soil in SCS-CN method) as proposed by Lin et al., (2014). Therefore, WM could be evaluated from average curve number of the watershed. For further computation, the parameter S is visualized as current soil water retention capacity and updated on daily basis as the difference of WM and W (which is nothing but the current soil moisture deficit of the soil), i.e. when the value of W becomes zero then S is equal to WM. Also when W reaches WM (state of saturation in soil water store zone) S is equal to zero or SCSCN equal to 100, thus simulation of saturation excess runoff mechanism. In this way, the value of S is updated at each computational time step using the soil moisture updation procedure of Xinanjiang model. Under this model, the surface runoff is generated by SCS-CN method then remaining rainfall is infiltrates and add to the soil moisture and other components of total runoff are generated in the same way as in the original Xinanjiang model. The proposed SCS-CN inspired Xinanjiang model has been named as XIN-CN model. ii) The proposed DVIC model is the modified form of the Hu et al. (2005) model. As the Hu et al. (2005) considered both the runoff generation mechanism i.e. saturation excess and infiltration excess runoff generation mechanism simultaneously, using both the distribution curve i.e. the distribution curve of tension water capacity and distribution curve of infiltration capacity. It is however seen that the Hu et al. (2005) model does not perform well as it was expected and also, the Hu et al. (2005) model is very complex in its runoff generation process as it uses six steps for generating surface runoff as well as ground water runoff. Therefore, a simplified and more iii | P a g e realistic hybrid conceptual model is proposed in this study. The proposed DVIC model considers only the distribution of infiltration capacity curve for surface runoff generation, which uses only two steps for surface runoff generation and ground water runoff is generated when soil moisture exceeds the field capacity of the soil moisture. As FmΔt is the function of point soil infiltration capacity (F´Δt) and the value of F´Δt is variable in nature because it varies from 0 to F´mΔt therefore, the proposed model (DVIC) shows its variability in terms of infiltration capacity distribution curve. Also the average time interval infiltration capacity FmΔt, itself changes in each time interval (or daily) therefore, the proposed model (DVIC) is dynamic in nature, therefore, the model has been named as Dynamic Variable Infiltration Capacity (DVIC) model. The performance of both the proposed hybrid XIN-CN and DVIC models and four existing variants of the Xinanjiang model viz. Zhao (1992), Nirupama (1996), Hu et al. (2005) and Lin et al. (2014) have been evaluated using observed data from 20 watersheds of different size and shape situated in different climatic zones of India. Available observed hydrological data have been split into two groups, data in one group has been used to calibrate parameters of the model, and data in other group have been used to validate the performance of the calibrated model. The performance of the models has been assessed using the statistical indices NSE, R2, SE and RE (as %) as well as on the basis of visual assessment of hydrographs. To evaluate performance of selected models, the watershed selected for this study have been grouped into three categories as wet, average and dry based on average value of runoff coefficient. Accordingly, the watershed having a runoff coefficient more than 0.65 has been classified as a wet watershed, the watershed having a runoff coefficient between 0.36 and 0.65, has been classified as average watershed and the watershed having a runoff coefficient less than or equal to 0.35, has been classified as a dry watershed (Gan et al., 1997) representing humid, average and dry climatic conditions respectively. Analysis of results obtained reveals that the Xinanjiang model and its other variants studied herein performs relatively poorly in estimating the discharge in catchments located in average and dry climatic zones which are mostly dominated by the infiltration excess runoff generation mechanism compared to those in humid zones, which are primarily dominated by the saturation excess runoff generation mechanism. This inference clearly indicates the inadequacy of the runoff generation mechanism adopted in the Xinanjiang model. The proposed hybrid conceptual models (XIN-CN and DVIC) can account for both infiltration excess as well as saturation excess runoff generation mechanisms based on watershed soil water status thus making them amenable iv | P a g e for use in all categories of catchments. Comparative evaluation of results obtained using proposed and existing four existing versions of the Xinanjiang model on hydrological data of 20 watersheds located in different climatic zones of India clearly indicate better performance of proposed models. The observed peak runoff is better simulated by proposed models. Better results in terms of close visual match between observed and model computed discharge obtained using DVIC and XIN-CN and high value on NSE both during calibration and validation periods indicate that the adoption and amalgamation of Hortonian runoff generation mechanism is very much need along with saturation excess mechanism to improve performance of the model for all catchments (i.e. in all climatic zones) in the present study. The overall performance ranking based on statistical evaluation indicators of the proposed models (DVIC and XIN-CN) and existing versions of the Xinanjiang model is indicated below XIN-CN > DVIC > ZHAO (1992) > NIRUPAMA (1996) > LIN (2014) > HU ET AL. (2005) The proposed models have simple structure and can simulate both infiltration excess and saturation excess runoff generation mechanisms based on catchment wetness status and can be used as a flexible tool for rainfall runoff modeling in all categories of catchments.