Resumen

ROUSSEAU-FIGUEROA✝, Pablo Andrés et al. The Influence of Edge Effect on Precipitation Forecast Using DWT Diadic, MODWT, ANN y ANFIS. Tecnol. cienc. agua [online]. 2016, vol.7, n.3, pp.93-113. ISSN 2007-2422.

The influence of the edge effect by using Neuro-fuzzy Wavelet hybrid models to forecast monthly time series of precipitation one month in advance is analyze in this study. In order to implement the models, data from climatologic station, located at Emilio Lopez Zamora dam in the city of Ensenada northwestern Baja California, Mexico, was used. In particular we explore four methods: a) Discrete Wavelet Transform using the Mallat algorithm (DWT); b) Maximal Overlap Discrete Wavelet Transform (MODWT); c) Feed Forward Back Propagation (FFBP), and, d) Adaptive-Network-based Fuzzy Inference System (ANFIS). Two preprocessing approaches currently used in the literature to predict climatic series of hydrological variables were applied. In the first approach, four methods for performing convolution transform are used and discussed their relationship with the phenomenon of edge effect. The results show that the hybrid model used has a significant influence to improve network training for prediction purposes, however the opposite happens at forecasting step due to edge effect. Meanwhile, in the second approach, we found that reconstruction of time series must be done using wavelet coefficients of five year scales period, but the results show that there is a significant noise component in the signal. Finally, it was found that the easier and recommended method to be used for this type of time series is self ANFIS.

Palabras llave : Precipitation forecast; discrete wavelet transform; adaptive-network-based fuzzy inference system (ANFIS); Mallat algorithm; edge effect.