Sea surface temperature (SST) significantly affects the processes of air–sea interactions, and, thus, forms an important indicator of climate changes. In SST predictions, the approach of artificial neural networks (ANNs) is data-driven, unlike that of the numerical models, which are physics-based. In this letter, Operational SST and Ice Analysis (OSTIA) data set were used for training ANN models and verifying prediction results. Typically, time series SST data of a certain period were directly used for ANN’s training. To reduce the prediction errors caused by SST variations, the authors propose to separate SST time series data into climatological monthly mean and monthly anomaly data sets and construct two neural network models. The combination of these two models gives the final SST prediction results. This method was used for 12-month lead time SST prediction in the South China Sea. The average bias and standard deviation between the predicted SST and OSTIA SST are −0.16 °C and 0.37 °C, respectively. The percentage of SST difference between the predicted SST and OSTIA SST, within ±0.5 °C and ±1 °C, is 71.24% and 95.22%, respectively. The results indicate that the proposed training method gives good prediction accuracy.

Sea surface temperature (SST) in the China Seas has shown an enhanced response in the accelerated global warming period and the hiatus period, causing local climate changes and affecting the health of coastal marine ecological systems. Therefore, SST distribution prediction in this area, especially seasonal and yearly predictions, could provide information to help understand and assess the future consequences of SST changes. The past few years have witnessed the applications and achievements of neural network technology in SST prediction. Due to the diversity of SST features in the China Seas, long-term and high-spatial-resolution prediction remains a crucial challenge. In this study, we adopted long short-term memory (LSTM)-based deep neural networks for 12-month lead time SST prediction from 2015 to 2018 at a 0.05° spatial resolution. Considering the sub-regional differences in the SST features of the study area, we applied self-organizing feature maps (SOM) to classify the SST data first, and then used the classification results as additional inputs for model training and validation. We selected nine models differing in structure and initial parameters for ensemble to overcome the high variance in the output. The statistics of four years’ SST difference between the predicted SST and Operational SST and Ice Analysis (OSTIA) data shows the average root mean square error (RMSE) is 0.5 °C for a one-month lead time and is 0.66 °C for a 12-month lead time. The southeast of the study area shows the highest predictable accuracy, with an RMSE less than 0.4 °C for a 12-month prediction lead time. The results indicate that our model is feasible and provides accurate long-term and high-spatial-resolution SST prediction. The experiments prove that introducing appropriate class labels as auxiliary information can improve the prediction accuracy, and integrating models with different structures and parameters can increase the stability of the prediction results.

Due to the application demand, users have higher expectations for the accuracy and resolution of sea surface temperature (SST) products. Recent advances in deep learning show great advantages in exploiting massive ocean datasets, and provides opportunities for investigating regional SST predictions in an efficiency approach. However, for deep learning-based SST prediction to be adopted by users, the output must be accurate. This paper investigates the 7-day SST prediction over the China seas and their adjacent waters at a 0.05° spatial resolution. To improve the prediction’s accuracy, we designed a deep learning model combining the three-dimensional convolution and long short-term memory under multi-input multi-output strategy. The Operational SST and Sea Ice Analysis (OSTIA) SST anomaly was used as training data. To test the model prediction ability, we verified the predicted results with the Sub-seasonal to Seasonal (S2S) prediction data from 2015 to 2019. Validation of the predicted SSTs using the OSTIA test datasets show that the root-mean-square error increases from 0.27°C to 0.53°C during the 1- to 7-day lead time, with predictability decreases from southeast to northwest in the study area. Furthermore, the comparison of predicted SST and S2S data with Argo shows that our model is slightly more accurate, which can achieve -0.08°C bias, with a standard deviation of 0.35°C for a 1-day lead time and -0.07°C bias, with a standard deviation of 0.59°C for a 7-day lead time. The results indicate that the proposed deep learning model is accurate and can be applied in regional daily SST prediction.

Sea surface temperature (SST) is one of the most important factors related to the ocean and the climate. In studying the domains of eddies, fronts, and current systems, high-resolution SST data are required. However, the passive microwave radiometer achieves a higher spatial coverage but lower resolution, while the thermal infrared radiometer has a lower spatial coverage but higher resolution. In this paper, in order to improve the performance of the super-resolution SST images derived from microwave SST data, we propose a transformer-based SST reconstruction model comprising the transformer block and the residual block, rather than purely convolutional approaches. The outputs of the transformer model are then compared with those of the other three deep learning super-resolution models, and the transformer model obtains lower root-mean-squared error (RMSE), mean bias (Bias), and robust standard deviation (RSD) values than the other three models, as well as higher entropy and definition, making it the better performing model of all those compared.

In recent years, coral reef ecosystems have been affected by global climate change and human factors, resulting in frequent coral bleaching events. A severe coral bleaching event occurred in the northwest of Hainan Island, South China Sea, in 2020. In this study, we used the CoralTemp sea surface temperature (SST) and Sentinel-2B imagery to detect the coral bleaching event. From 31 May to 3 October, the average SST of the study area was 31.01 °C, which is higher than the local bleaching warning threshold value of 30.33 °C. In the difference images of 26 July and 4 September, a wide range of coral bleaching was found. According to the temporal variation in single band reflectance, the development process of bleaching is consistent with the changes in coral bleaching thermal alerts. The results show that the thermal stress level is an effective parameter for early warning of large-scale coral bleaching. High-resolution difference images can be used to detect the extent of coral bleaching. The combination of the two methods can provide better support for coral protection and research.

The primary measure of the quality of sea surface temperature (SST) fields obtained from satellite-borne infrared sensors has been the bias and variance of matchups with co-located in-situ values. Because such matchups tend to be widely separated, these bias and variance estimates are not necessarily a good measure of small scale (several pixels) gradients in these fields because one of the primary contributors to the uncertainty in satellite retrievals is atmospheric contamination, which tends to have large spatial scales compared with the pixel separation of infrared sensors. Hence, there is not a good measure to use in selecting SST fields appropriate for the study of submesoscale processes and, in particular, of processes associated with near-surface fronts, both of which have recently seen a rapid increase in interest. In this study, two methods are examined to address this problem, one based on spectra of the SST data and the other on their variograms. To evaluate the methods, instrument noise was estimated in Level-2 Visible-Infrared Imager-Radiometer Suite (VIIRS) and Advanced Very High Resolution Radiometer (AVHRR) SST fields of the Sargasso Sea. The two methods provided very nearly identical results for AVHRR: along-scan values of approximately 0.18 K for both day and night and along-track values of 0.21 K for day and night. By contrast, the instrument noise estimated for VIIRS varied by method, scan geometry and day-night. Specifically, daytime, along-scan (along-track), spectral estimates were found to be approximately 0.05 K (0.08 K) and the corresponding nighttime values of 0.02 K (0.03 K). Daytime estimates based on the variogram were found to be 0.08 K (0.10 K) with the corresponding nighttime values of 0.04 K (0.06 K). Taken together, AVHRR instrument noise is significantly larger than VIIRS instrument noise, along-track noise is larger than along-scan noise and daytime levels are higher than nighttime levels. Given the similarity of results and the less stringent preprocessing requirements, the variogram is the preferred method, although there is a suggestion that this approach overestimates the noise for high quality data in dynamically quiet regions. Finally, simulations of the impact of noise on the determination of SST gradients show that on average the gradient magnitude for typical ocean gradients will be accurately estimated with VIIRS but substantially overestimated with AVHRR.