Incorporating non-uniformity and non-linearity of hydrologic and catchment characteristics in rainfall-runoff modeling using conceptual, data-driven, and hybrid techniques

Abstract

The rainfall-runoff (RR) process in a catchment is non-uniform, complex, dynamic, and non-linear in nature. Although a number of advanced conceptual and data-driven techniques have been proposed in the past, the accurate estimation of daily runoff still remains a challenging task. A majority of conceptual models proposed so far suffer from the assumptions of linearity during their modeling. In this paper, novel hybrid approaches are proposed that are capable of exploiting the strength of both conceptual and data-driven techniques in RR modeling. A conceptual technique is first used to generate sub-basins' runoff hydrographs in upstream reaches and then data-driven techniques are employed for routing them to the outlet of the catchment. The hybrid models' performances are compared with standalone conceptual and data-driven models by employing the daily rainfall, runoff, and temperature data derived from the Kentucky River basin, USA. The results show that the proposed hybrid models, which do not assume the RR process to be a linear process to simulate the flow, outperform their individual counterparts. It is concluded that in order to achieve improved accuracy in RR modeling, the real-life process needs to be represented as accurately as possible in the modeling effort rather than making simplified assumptions.