Particle Filter Networks for Ei NiñO-Southern Prediction

The El Nino-Southern Oscillation (ENSO) is the strongest interannual change signal on Earth, which has a significant impact on global weather and climate anomalies. However, the level of ENSO forecast has almost stagnated in the past 20 years, and it can no longer meet the major needs of national disaster prevention and mitigation. No matter the traditional assimilation set Kalman filter target observation method, or the subsequent use of CNN and Transformers attempts have their corresponding shortcomings. In this paper, a model based on particle filter mixed empirical mode and radial basis function neural network (ENSOPFN) is proposed. ESNOPFN combines empirical mode decomposition (EMD), radial basis neural network (RBFMN), incremental kernel space independent vector prediction, particle filter and weighted combination of the model architecture. It has been tested that it can predict ENSO robustly over a period of 1 to 12 months, and has improved the prediction accuracy. According to the analysis of the future El Nino event forecast, we find that it is more suitable to deal with the nonlinear activities caused by the earth activities, and the forecast for the next 12 months and 24 months has also achieved good results, and provides new inspiration and reference for the supplement and optimization of other models.