Improvement of lumped models in multiple inflows rivers by principal component analysis and reliability analysis

Abstract

Flood routing as an important part of flood management is a technique for predicting the flow in downstream of a river channel or reservoir. Lumped, semi-distributed and distributed models have been devised in this regard. The convex and Att-Kin models are capable of simulating floods in single branches, while in reality, rivers and channels are multiple inflows. The convex and modified Att-Kin models as the simplest lumped models in terms of the storage equation were developed based on an equivalent inflow for routing the multiple inflows rivers in the present study. The genetic algorithm, a quite robust algorithm, was used for parameter estimation of the extended models. The ability of the extended models in simulating the outflow hydrograph of multiple inflows systems was tested on two multiple inflows case studies. The results of extended models were compared with the equivalent Muskingum inflow model. Comparison of the extended models with the Muskingum model showed that the extended models with one parameter less than the Muskingum model could be suitable for use in flood routing of multiple inflows systems. The effect of inflow hydrographs at different time steps was investigated by the principal component analysis (PCA) and reliability analysis. The results showed that the outflow hydrograph of the case study was precisely simulated and predicted by the gene expression programming (GEP) and multilayer perceptron (MLP) models. The PCA and reliability analysis results were adopted for the lumped, GEP and MLP models. The outflow hydrograph was precisely simulated and predicted by the GEP and MLP models, while the precision of lumped models (extended convex, extended modified Att-Kin and Muskingum models) was not increased.