Rainfall-Runoff Modeling: Comparison of Two Approaches with Different Data Requirements

Abstract

Among several hydrological models developed over the years, the most widely used technique for estimating direct runoff depth from storm rainfall i.e., the United States Department of Agriculture (USDA) Soil Conservation Service’s (SCS) Curve Number (CN) method was adopted in the present study. In addition, the Muskingum method, which continues to be popular for routing of runoff in river network, was used in the developed model to route surface runoffs from different subbasin outlet points up to the outlet point of the catchment. SCS CN method in combination with Muskingum routing technique, however, required a detailed knowledge of several important properties of the watershed, namely, soil type, land use, antecedent soil water conditions, and channel information, which may not be readily available. Due to this complexity of semi-distributed conceptual approach (SCS CN method) and non-linearity involved in rainfall-runoff modeling, researchers also attempted another less data requiring approach for runoff prediction, i.e., the neural network approach, which is inherently suited to problems that are mathematically difficult to describe. The purpose of this study was to compare the rainfall-runoff modeling performance of semi-distributed conceptual SCS CN method (in combination with Muskingum routing technique) with that of empirical ANN technique. The models were coded in C language and to make them user friendly, a Graphical User Interface (GUI) was also developed in Visual Basic 6.0. The developed models were tested for Kangsabati catchment, situated in the western part of West Bengal, India. Monsoon data of 1996 to 1999 were used for calibration of the models whereas they were validated for another four years (1987, 1989, 1990, and 1993) monsoon data. Modeling efficiency (ME) and coefficient of residual mass (CRM) were used as performance indicators. Results indicated that for Kangsabati catchment, the empirical runoff prediction approach (ANN technique), in spite of requiring much less data, predicted daily runoff values more accurately than semi-distributed conceptual runoff prediction approach (SCS CN method).