Rainfall–Runoff Modeling Based on Genetic Programming

Abstract

AbstractThe rainfall–runoff process is highly nonlinear, time varying, spatially distributed, and not easily described by simple models. Various models have been developed to simulate this process, including lumped conceptual models, distributed physically based models, and empirical black‐box models. Either conceptual or distributed physically based models require a significant amount of data for calibration and validation, whereas in most cases, it is difficult to collect all the data necessary with sufficient accuracy for such models. Traditional black‐box models such as artificial neural networks provide means to ease the data demands for model calibration and validation; however, the information they provide add little insights for interpretation of the underlying process.Using data‐driven techniques such asgenetic programming(GP), one can attempt to model the rainfall–runoff process based on available hydrometeorology data. Genetic programming can also be used in combination with conceptual models to discover physically interpretable models or equations describing the physical processes. After a detailed review of the conventional applications of genetic programming in rainfall–runoff modeling, this article introduces a novel scheme of conceptual rainfall–runoff modeling based on GP. A tropical case study is provided as a prototype for illustration.