TIME SERIES DROUGHT FORECASTING USING RBF WAVELET-GMDH MODEL WITH STANDARDIZED PRECIPITATION INDEX
Loading...
Date
Journal Title
Journal ISSN
Volume Title
Publisher
Department of Statistics, Nasarawa State University Keffi
DOI
Abstract
rought forecasting is a significant aspect of time series forecasting among researchers using numerous available models to achieve the desired goal.
This paper, therefore, proposes an advanced method to improve the drought forecasting capability of the existing models.
The propose model is the Radial Basis Function (RBF) Wavelet-Group Method of Data Handling (W-GMDH) developed by integrating Discrete Wavelet Transform (DWT) with GMDH model using the Standardized Precipitation Index (SPI) drought data for forecasting to assess the effectiveness of the model. We used four SPI data series (SPI3, SPI6, SPI9 and SPI12) made up of 600 of SPI data from January 1968 to December 2015. This was divided into 80% for training and 20% for testing. The results of the proposed model were compared with that of the W-ARIMA and RBF GMDH models with the Root Mean Square Error (RMSE), Mean Average Error (MAE) and coefficient of correlation (R) as the performance measures. The outcome of the proposed RBF W-GMDH model for SPI3 data RMSE = 0.2210, MAE = 0.1879 and R = 0.9495; SPI6 data RMSE = 0.1976, MAE = 0.1745 and R = 0.9644; SPI9 data RMSE = 0.4076, MAE = 0.3947 and R = 0.9526 and SPI12 data RMSE = 0.3016, MAE = 0.2339 and R = 0.9420 which outpowered the results of the W-ARIMA and W- GMDH, indicating that the proposed method can achieve the best forecasting performance in terms of accuracy for each of the SPI data series. The key role played by the DWT is to smooth the analysis of SPI data obtained after the wavelet decomposition used to decompose the SPI data into different number of components to minimize the forecasting error. In all the results, the proposed model has a minimum error indicating its superiority over the other models.
