A Wavelet and Neuro-Fuzzy Conjunction Model to Forecast Air Temperature Variations at Coastal Sites

The present paper investigates performance of neuro-fuzzy (NF) and wavelet-neuro-fuzzy (WNF)conjunction models in short- and long-term forecasts of air temperature. A NF and two WNF models were developed and validated using daily air temperature data collected at two coastal stations in Iran, namely, Ahwaz and Izeh. The comparison of ANFIS and WANFIS models indicated that the conjunction models performed better than the single ANFIS model especially in forecasting weekly and monthly air temperatures. The coefficient of determination (R2), the root mean square error (RMSE) and adjusted coefficient of efficiency (E1) were used as comparison criteria. For the Ahwaz and Izeh stations, the WANFIS model increased the accuracy of single ANFIS model by 23-13% (Ahwaz) and 21-8% (Izeh) with respect to RMSE in forecasting one-week and one-month ahead maximum air temperatures, respectively