Rainfall-runoff modeling using computational intelligence techniques

Abstract

Rainfall and corresponding Runoff estimation are substantially dependent on various geographic, climatic, and biotic features of the catchment or basin under study and these factors often induce a linear, non-linear or highly complex relation between rainfall and runoff. The few of key factors include precipitation, percolation, infiltration, evaporation, stream flow, and air temperature. Plenty of Rainfall-Runoff (RR) regression models are available, each one distinguished by a varying level of complexity and data requirement. Most of the time due to complex relationship between rainfall and runoff the traditional models (SCN-CN, MISDc, GA, CN4GA) with regression equations don't resembles the correct scene of rainfall-runoff connection. Computational Intelligence (CI) approaches play a key role in modeling those complex tie-ups between rainfall and runoff. The rainfall-runoff process was modeled using a mamdani Fuzzy Inference System (FIS) implemented within a layered design of Artificial Neural Network (ANN) and was applied to a small area of Koshi basin in Bihar, using 12 year's (1980–1992) observed records of daily rainfall, soil moisture and runoff. A comparison was also made between proposed models and existing soft computing models. The proposed computational intelligence model proves significantly better than existing soft computing models in terms of performance.