AI Method For Ocean Eddy Identification And Parameter Inversion Based On Profile Dat

Oceanic eddies have a large number,wide distribution,high energy content,and strong coercion.They are ideal carriers for studying material circulation,energy cascades,and sphere coupling,and are of great significance to human activities and marine science.Based on the dynamic sea level anomaly and temperature anomaly characteristics of eddies,traditional oceanographic eddy detection methods usually rely on experts to manually set thresholds or adjust parameters,which cannot guarantee accuracy and cannot be applied to all ocean fields.At the same time,due to the low resolution of satellite altimeters,nearly 90% of the sea level dynamic anomalies caused by oceanic eddies cannot be observed.This paper proposes a deep neural network model to identify oceanic eddies and invert the corresponding eddy attribute information by combining vertical profile data and altimeter-calibrated ocean temperature data.The research results are of great significant to both human activities and the field of ocean science.The research content of this paper is as follows:(1)Construct an oceanic eddy profile dataset,including Expendable Bathythermograph(XBT)data and Array for Real-time Geostrophic Oceanographic(Argo)data.In addition,we also created a sea level anomaly(SLA)dataset observed by satellite altimeters as the ground truth for the final eddy identification.(2)Construct a deep neural network model EDTR with a self-attention mechanism,which is suitable for two different vertical profile data.We compared the performance of neural networks such as Res Net34,Mobile Net V2 and Efficient V1 on classification results,and finally selected the EDTR model as the classification model.The EDTR model analyzes the correlation between the vertical profile data through the self-attention mechanism,and classifies the XBT and Argo data through information fusion.Finally,the model achieved accuracy of 98.22% and 99.02% on the two vertical profile data,respectively.(3)We used the EDTR model to reclassify eddies that were not recognized by the altimeter,and then identified more new eddies.To verify the authenticity of these newly identified eddies,we performed distribution pattern verification and sea surface temperature anomaly(SSTA)independent verification.Furthermore,we invert properties such as radius,amplitude,and energy of the newly identified eddies using different methods of machine learning.