Research On Sea Surface Temperature Prediction Based On Spatiotemporal Characteristics Analysi

Sea Surface Temperature(SST)is an important geophysical parameter in the study of global climate change.On the one hand,the prediction of sea surface temperature change can create favorable conditions for the exploitation of Marine living resources,mineral resources and tourism resources.On the other hand,accurate prediction of sea surface temperature change can prevent many dangerous events,such as global warming,continuous sea level rise,El Nino phenomenon and so on.Therefore,this thesis proposes the HDC-Bi GRU-AT prediction model,which extracts temporal and spatial features,and the S-L-HBMSA prediction model,which extracts temporal and spatial features and integrates various feature elements,in order to further improve the prediction accuracy.(1)Most of the existing methods consider the time characteristics of sea surface temperature,but ignore the spatial characteristics.Even if some methods extract the time and space characteristics at the same time,there is insufficient extraction of spatial characteristics.In order to improve the prediction accuracy of sea surface temperature,a hybrid hollow convolution and bidirectional gated cyclic network model(HDC-Bi GRU-AT)based on attention mechanism with encoding and decoding structure is proposed.In the model coding stage,the mixed cavity convolution can extract the spatial characteristics of SST,and the bidirectional gated cyclic neural network can capture the temporal characteristics of SST.By adding the attention mechanism,a higher weight coefficient is assigned to the important information,so as to realize the information encoding,which can improve the prediction accuracy of the model in the decoding stage.Using OISST V2.0 high resolution daily mean sea surface temperature data provided by NOAA,the sea surface temperature data in the East China Sea and the South China Sea were intercepted for research.The model input 30 consecutive days of historical observation data,output the next 7 days of SST predicted value.Compared with some existing deep learning methods,the error index values of the proposed prediction model are all lower than those of the existing methods in the prediction task with only sea surface temperature data as input,which fully verifies the feasibility and effectiveness of the proposed method.(2)Sea surface temperature can be affected by external factors,such as Shortwave Radiation(SWR)from the sun and Longwave Radiation(LWR)from the atmosphere and the ground.Based on the prediction model of HDC-Bi GRU-AT,a prediction model of sea surface temperature(S-L-HBMSA)with multi-feature elements is proposed.In this prediction model,SWR,LWR and sea surface temperature data were fused as initial inputs.Then,the attention mechanism in the HDC-Bi GRU-AT hybrid prediction model was replaced by a multi-head self-attention mechanism.The multi-head attention mechanism was able to calculate the weight of different mapping results,process information at different levels in the input sequence,and calculate the global attention.The prediction accuracy of sea surface temperature is further improved.