Premium
Simulation of rainfall and runoff time series using robust machine learning *
Author(s) -
Alizadeh Amir,
Rajabi Ahmad,
Shabanlou Saeid,
Yaghoubi Behrouz,
Yosefvand Fariborz
Publication year - 2021
Publication title -
irrigation and drainage
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.421
H-Index - 38
eISSN - 1531-0361
pISSN - 1531-0353
DOI - 10.1002/ird.2518
Subject(s) - autocorrelation , precipitation , series (stratigraphy) , surface runoff , extreme learning machine , wavelet , time series , computer science , moving average , environmental science , artificial intelligence , statistics , machine learning , meteorology , mathematics , geology , artificial neural network , geography , paleontology , ecology , biology
In this paper, the precipitation and runoff time‐series data of the Shaharchay River basin from 2000 to 2017 are simulated by a modern hybrid artificial intelligence technique. In order to develop the mentioned artificial intelligence model, the extreme learning machine (ELM), differential evolution and wavelet transform are combined and the SAELM and WASAELM hybrid models are provided. Initially, the most effective lags of the time‐series data are distinguished using an autocorrelation function. Using the lags, seven artificial intelligence models are defined for each of the SAELM and WSAELM models. To simulate precipitation and runoff, the sym and coif mother wavelets are chosen as the optimal ones, respectively. For the best model, the values of R2, the scatter index and the Nash–Sutcliffe efficiency coefficient (NSC) for simulating precipitation yielded 0.967, 0.208 and 0.965, respectively. Furthermore, a sensitivity analysis shows that the lags (t‐1), (t‐2) and (t‐12) are regarded as the most effective input lags. Ultimately, an uncertainty analysis is carried out for the superior model that the performance of this model in simulating precipitation and runoff is over‐ and underestimated, respectively.