@Article{GangSaNordBevi:1996:PrSeSu,
               author = "Gang, Li Wei and Sa, Leonardo Deane Abreu and Nordemann, Daniel 
                         Jean Roger and Bevilaqua, Rute Maria",
                title = "Predictions of Sea Surface Temperature in Tropical Ocean using 
                         neural networks",
              journal = "Bulletin of the American Meteorological Society",
                 year = "1996",
               volume = "68",
               number = "1",
                pages = "23--33",
             keywords = "METEOROLOGIA, REDES NEURAIS, SUPERFICIE DO MAR, TEMPERATURA.",
             abstract = "A review of researches on the relationship between the tropical 
                         ocean sea surface temperatures (SST)and rainfall anomalies in 
                         Northeast Brazil was introduced. In this work, two neural network 
                         models are implemented to reconstruct and predict the time series 
                         of the SST in two regions: the tropical Atlantic ocean (Wright 
                         index, from 1854 to 1985) and the tropical Pacific ocean (regions 
                         Nino 1-2: 0 N-10 S, 270 E-280 E and Nino4: 5 N-5 S, 160 E-150 E, 
                         from 1950 to 1995). The selected neural networks include 
                         Backpropagation Neural Network (BNN) and Time Delay Neural Network 
                         (TDNN). Both were implemented in the neural network stimulator 
                         SNNS. For the Wright index, the trained Backpropagation Neural 
                         Network successfully predicted the index of the following four 
                         months with the relative errors from 1.40 to 3.34. For SST in Nino 
                         1-2 and Nino4 regions, the Time Delay Neural Network was used for 
                         reconstruction and prediction. Comparing with the next six month 
                         observations and predictions, all of them are located within the 
                         predicted error bars. These results show that neural network 
                         methods may be used, within certain limits, for prediction and 
                         evaluation of predictability of time series measured from 
                         phenomena influenced by complex climatic and geophysical 
                         processes, like SST.",
                 issn = "0003-0007",
                label = "7907",
           targetfile = "1996_gang.pdf",
        urlaccessdate = "05 maio 2024"
}