The reconstruction of the ocean’s 3D thermal structure is essential to the study of ocean interior processes and global climate change. Satellite remote sensing technology can collect large-scale, high-resolution ocean observation data, but only at the surface layer. Based on empirical statistical and artificial intelligence models, deep ocean remote sensing techniques allow us to retrieve and reconstruct the 3D ocean temperature structure by combining surface remote sensing observations with in situ float observations. This study proposed a new deep learning method, Convolutional Long Short-Term Memory (ConvLSTM) neural networks, which combines multisource remote sensing observations and Argo gridded data to reconstruct and produce a new long-time-series global ocean subsurface temperature (ST) dataset for the upper 2000 m from 1993 to 2020, which is named the Deep Ocean Remote Sensing (DORS) product. The data-driven ConvLSTM model can learn the spatiotemporal features of ocean observation data, significantly improves the model’s robustness and generalization ability, and outperforms the LighGBM model for the data reconstruction. The validation results show our DORS dataset has high accuracy with an average R² and RMSE of 0.99/0.34 °C compared to the Argo gridded dataset, and the average R² and NRMSE validated by the EN4-Profile dataset over the time series are 0.94/0.05 °C. Furthermore, the ST structure between DORS and Argo has good consistency in the 3D spatial morphology and distribution pattern, indicating that the DORS dataset has high quality and strong reliability, and well fills the pre-Argo data gaps. We effectively track the global ocean warming in the upper 2000 m from 1993 to 2020 based on the DORS dataset, and we further examine and understand the spatial patterns, evolution trends, and vertical characteristics of global ST changes. From 1993 to 2020, the average global ocean temperature warming trend is 0.063 °C/decade for the upper 2000 m. The 3D temperature trends revealed significant spatial heterogeneity across different ocean basins. Since 2005, the warming signal has become more significant in the subsurface and deeper ocean. From a remote sensing standpoint, the DORS product can provide new and robust data support for ocean interior process and climate change studies.

Global ocean heat content (OHC) is generally estimated using gridded, model and reanalysis data; its change is crucial to understanding climate anomalies and ocean warming phenomena. However, Argo gridded data have short temporal coverage (from 2005 to the present), inhibiting understanding of long-term OHC variabilities at decadal to multidecadal scales. In this study, we utilized multisource remote sensing and Argo gridded data based on the long short-term memory (LSTM) neural network method, which considers long temporal dependence to reconstruct a new long time-series OHC dataset (1993–2020) and fill the pre-Argo data gaps. Moreover, we adopted a new machine learning method, i.e., the Light Gradient Boosting Machine (LightGBM), and applied the well-known Random Forests (RFs) method for comparison. The model performance was measured using determination coefficients (R²) and root-mean-square error (RMSE). The results showed that LSTM can effectively improve the OHC prediction accuracy compared with the LightGBM and RFs methods, especially in long-term and deep-sea predictions. The LSTM-estimated result also outperformed the Ocean Projection and Extension neural Network (OPEN) dataset, with an R² of 0.9590 and an RMSE of 4.45 × 10¹⁹ in general in the upper 2000 m for 28 years (1993–2020). The new reconstructed dataset (named OPEN-LSTM) correlated reasonably well with other validated products, showing consistency with similar time-series trends and spatial patterns. The spatiotemporal error distribution between the OPEN-LSTM and IAP datasets was smaller on the global scale, especially in the Atlantic, Southern and Pacific Oceans. The relative error for OPEN-LSTM was the smallest for all ocean basins compared with Argo gridded data. The average global warming trends are 3.26 × 10⁸ J/m²/decade for the pre-Argo (1993–2004) period and 2.67 × 10⁸ J/m²/decade for the time-series (1993–2020) period. This study demonstrates the advantages of LSTM in the time-series reconstruction of OHC, and provides a new dataset for a deeper understanding of ocean and climate events.

Three-dimensional Ocean temperature and salinity data are the basis for studying ocean dynamic processes and warming. Satellite remote sensing observations on the ocean surface are abundant and full-coverage, while in-situ observations in the ocean interior are very sparse and unevenly distributed. Currently, the remote sensing inversion models of temperature and salinity in the ocean interior are unable to learn both global and local detail information, and modeling layer-by-layer blocks the connection between vertical depth levels, resulting in poor accuracy. In this study, we proposed a novel Clustering-Guided and Knowledge-Distillation Networks (CGKDN) model based on the ocean knowledge-driven. The model introduced K-means clustering for the partitions of ocean processes, knowledge distillation fusing global and local detail information, and adaptive depth gradient loss linking the vertical depth dimension, which enhanced the interpretability and accuracy of the model. Comparison of the reconstructions with the existing major publicly available datasets through the validation of 10% EN4 in-situ profile observations from 2001 to 2020 reveals that the reconstructions are more accurate. Concretely, the average RMSE (°C) across time series and vertical levels of CGKDN/IAP/ORAS5/DORS ocean subsurface temperature is 0.590/0.598/0.690/0.723, and the average RMSE (PSU) of CGKDN/IAP/ORAS5 ocean subsurface salinity is 0.101/0.103/0.106, respectively. Furthermore, the downscaled quarter-degree reconstructions present more mesoscale detail signals, consistent with the ARMOR3D data. This study not only improves the estimation accuracy of subsurface temperature and salinity, but also serves the study of ocean interior dynamic processes and variabilities, and provides valuable references for reconstructing other ocean subsurface physical variables.

Satellite remote sensing plays a fundamental role in observing oceanic processes by providing large-scale, long-term, and continuous measurements. With the increasing availability of multisource satellite data, challenges such as data gaps, complex environmental conditions, and the limitations of conventional retrieval methods have become more evident. In recent years, artificial intelligence (AI) has emerged as a practical and effective approach to address these issues. This article reviews the development of AI techniques in satellite ocean remote sensing, focusing on three main application areas: parameter retrieval, data reconstruction, and image-based ocean phenomenon detection. For geophysical variable retrieval, AI models such as convolutional neural networks (CNNs) and Transformer architectures have improved the accuracy of ocean waves, sea surface, salinity, wind, and ocean color estimates, especially under extreme or noisy conditions. In the field of data reconstruction, AI methods enable the completion of missing data in both surface and subsurface ocean layers, offering finer spatial–temporal resolution and better consistency than traditional interpolation approaches. For image interpretation, deep learning (DL) models have been applied to detect and segment dynamic ocean features such as mesoscale eddies, internal waves, sea ice, and tropical cyclones (TCs), achieving high efficiency and precision. This article also highlights the integration of AI with physical knowledge, the use of multisource fusion, and the trend toward near real-time (NRT) applications. These developments indicate that AI will play an increasingly important role in future satellite-based ocean observation and environmental monitoring.

The recognition of sea ice is of great significance for reflecting climate change and ensuring the safety of ship navigation. Recently, many deep learning based methods have been proposed and applied to segment and recognize sea ice regions. However, the diverse scales of sea ice areas, the zigzag and fine edge contours, and the difficulty in distinguishing different categories of sea ice challenges to existing sea ice recognition methods. In this paper, a global-local transformer network, called SeaIceNet, is proposed for sea ice recognition in optical remote sensing images. In SeaIceNet, a dual global-attention head is proposed to capture global information. On this basis, a global-local feature fusion mechanism is designed to fuse global structural correlation features and local spatial detail features. Furthermore, a detail-guided decoder is developed to retain more high-resolution detail information during feature reconstruction for improving the performance of sea ice recognition. Extensive experiments on several sea ice datasets demonstrated that the proposed SeaIceNet has better performance than existing methods in multiple evaluation indicators. Moreover, it excels in addressing challenges associated with sea ice recognition in optical remote sensing images, including the difficulty in accurately identifying irregular frozen ponds in complex environments, the broken and unclear boundaries between sea and thin ice that hinder precise segmentation, and the loss of high-resolution spatial details during model learning that complicates refinement.