A wavelet-SARIMA-ANN hybrid model for precipitation forecasting

Journal of Water and Land Development

Jan Adamowski McGill University, Faculty of Agricultural and Environmental Sciences, Department of Bioresource Engineering, Quebec, Canada, H9X 3V9EmailOther articles by this author:De Gruyter OnlineGoogle Scholar¹  Kazimierz Adamowski University of Ottawa, Department of Civil Engineering, CanadaEmailOther articles by this author:De Gruyter OnlineGoogle Scholar²  Yagob Dinpashoh University of Tabriz, Department of Water Engineering, IranEmailOther articles by this author:De Gruyter OnlineGoogle Scholar³  Maryam Shafaei University of Tabriz, Department of Water Engineering, IranEmailOther articles by this author:De Gruyter OnlineGoogle Scholar³  Ahmad Fakheri-Fard University of Tabriz, Department of Water Engineering, IranEmailOther articles by this author:De Gruyter OnlineGoogle Scholar³ 

[1] McGill University, Faculty of Agricultural and Environmental Sciences, Department of Bioresource Engineering, Quebec, Canada, H9X 3V9;University of Ottawa, Department of Civil Engineering, Canada;University of Tabriz, Department of Water Engineering, Iran

关键词: Keywords: artificial neural network (ANN);    precipitation forecasting;    seasonal auto regressive integrated moving average (SARIMA);    water resources management;    wavelet;   

DOI  :  10.1515/jwld-2016-0003

学科分类：农业科学（综合）

来源: Instytut Technologiczno-Przyrodniczego / Institute of Technology and Life Sciences

【 摘 要 】

Given its importance in water resources management, particularly in terms of minimizing flood or drought hazards, precipitation forecasting has seen a wide variety of approaches tested. As monthly precipitation time series have nonlinear features and multiple time scales, wavelet, seasonal auto regressive integrated moving average (SARIMA) and hybrid artificial neural network (ANN) methods were tested for their ability to accurately predict monthly precipitation. A 40-year (1970–2009) precipitation time series from Iran’s Nahavand meteorological station (34°12’N lat., 48°22’E long.) was decomposed into one low frequency subseries and several high frequency sub-series by wavelet transform. The low frequency sub-series were predicted with a SARIMA model, while high frequency subseries were predicted with an ANN. Finally, the predicted subseries were reconstructed to predict the precipitation of future single months. Comparing model-generated values with observed data, the wavelet-SARIMA-ANN model was seen to outperform wavelet-ANN and wavelet-SARIMA models in terms of precipitation forecasting accuracy.

【 授权许可】

Unknown