Abstract:
This study compares three different approaches to continuous-time modelling of the daily rainfall-runoff response of the Dedtalai basin, India. The main objective is to compare the modelled response of the watershed using a Conceptual Rainfall-Runoff (CRR) model, a Data-Based Mechanistic (DBM) model, and an Artificial Neural Network (ANN) model. The Dedtalai watershed is a large semiarid basin (6,705 km2) with ephemeral rivers and it is located within the Tapi basin (65,145 km2). Daily forcing climate and discharge data are available from 1990 to 1998, although with very limited spatial coverage. The CRR model is based on a deterministic model chosen subjectively based on modeller experience. The DBM modelling philosophy identifies a rainfall-runoff transfer function using only the input-output data, with no prescribed conceptual structure. The physical interpretation of the transfer function, in terms of conceptual stores, is compared with the simplified hydrological representations of the applied CRR models. The ANN model is a three layer back propagation ANN, with observed climate and streamflow data as inputs. The models are identified and validated using two periods each of about four years. The estimate fl ows are compared visually and using least-squares objective functions. The ANN and DBM models performed best in validation, with NSE values of 0.95 and 0.64 compared to 0.41 for the CRR model. However the CRR model has wider applicability to simulation because it does not need observed flow inputs. This paper concludes by providing discussion and guidance about how the different approaches can be used in a complementary manner for modelling rainfall-runoff response in large semi-arid areas.
