Reconstruct infrared sea surface temperature data based on an improved dincae method

Radiometers working in the infrared bands are easily affected by factors such as clouds and sea fog, which limits the spatial coverage of sea surface temperature (SST) data observed by remote sensing and affects the application of SST remote sensing data products. Data Interpolation Convolutional Auto-Encoder (DINCAE) is a data reconstruction method based on deep learning that extracts nonlinear relationships in data through a convolutional auto-encoder structure and reconstructs missing data using valid points available in the data. On the basis of DINCAE, we added the convolutional LSTM (ConvLSTM) to fully extract the time features in the data and improved the DINCAE method (T-DINCAE). The reconstruction error is calculated through cross validation and Argo buoy data. The results show that T-DINCAE has higher data reconstruction quality than DINCAE.