Comparison of Short-Term Prediction Effects of the Sea Surface Temperature and Salinity Based on BP and RBF Neural Network

In order to make an accurate short-term prediction of Sea Surface Temperature (SST) and Sea Surface Salinity (SSS) data, the short-term prediction results of Back Propagation (BP) and Radial Basis Function (RBF) neural network methods are compared and analyzed by using the SST and SSS data obtained by multi-station marine observation buoys. Firstly, when the prediction days are fixed to 5 days, the Mean Squared Errors (MSE) of the prediction results of different training days are compared, and then determine that the MSE with 20 days of observation data as the training set is the smallest. Then, taking the SST and SSS data of the first 20 days in January，April, July and October 2009 obtained by the observation buoy of PAPA station as the training set, BP and RBF neural networks are trained respectively. Thereafter, the trained two neural network models are applied to the prediction of SST and SSS data from the 21st to 25th days of each month. The results show that both BP and RBF neural network can effectively predict the seasonal changes of SST and SSS data. But the comparative experiments of different prediction days show that the overall prediction effect of RBF neural network is better than BP neural network. Finally，through the prediction experiment of multisite data, it is verified that the RBF neural network model has strong applicability and higher accuracy, and can become a powerful tool for short-term prediction of SST and SSS data.