Subsurface density (SD) is a crucial dynamic environment parameter reflecting a 3-D ocean process and stratification, with significant implications for the physical, chemical, and biological processes of the ocean environment. Thus, accurate SD retrieval is essential for studying dynamic processes in the ocean interior. However, complete spatiotemporally accurate SD retrieval remains a challenge in terms of the equation of state and physical methods. This study proposes a novel multiscale mixed residual transformer (MMRT) neural network method to compensate for the inadequacy of the existing methods in dealing with spatiotemporal nonlinear processes and dependence. Considering the spatial correlation and temporal dependence of dynamic processes within the ocean, the MMRT addresses temporal dependence by fully using the transformer’s processing of time-series data and spatial correlation by compensating for deficiencies in spatial feature information through multiscale mixed residuals. The MMRT model was compared with the existing random forest (RF) and recurrent neural network (RNN) methods. The MMRT model achieves the best accuracy with an average determination coefficient (${R}^{2}$) of 0.988 and an average root mean square error (RMSE) of 0.050 kg/m3 for all layers. The MMRT model not only outperforms the RF and RNN methods regarding reliability and generalization ability when estimating global ocean SD from remote sensing data but also has a more interpretable encoding process. The MMRT model offers a new method for directly estimating SD using multisource satellite observations, providing significant technical support for future remote sensing super-resolution and prediction of subsurface parameters.

3-D 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 network (CGKDN) model based on the ocean knowledge-driven model. The model introduced K-means clustering for the partitions of ocean processes, knowledge distillation (KD) 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 root mean square error (RMSE) (°C) across time-series and vertical levels of CGKDN/Institute of Atmospheric Physics (IAP)/OCEAN5 ocean analysis-reanalysis (ORAS5)/deep ocean remote sensing (DORS) ocean subsurface temperature (OST) is 0.590/0.598/0.690/0.723, and the average RMSE (PSU) of CGKDN/IAP/ORAS5 ocean subsurface salinity (OSS) 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.

Remote sensing study of the carbon cycle in coastal marine systems using machine learning methods has received significant attention recently. The partial pressure of carbon dioxide (CO2) in seawater (pCO2w) is a crucial parameter for quantifying the air-sea carbon dioxide exchange. However, previous studies did not consider the effect of sea surface roughness (SSR) onpCO2wcaused by wind, waves, and other ocean dynamics. In this study, for the first time, we used SSR data derived from Synthetic Aperture Radar (SAR), with sea surface temperature (SST), chlorophyll-a concentration (Chl-a), sea surface salinity (SSS) conventional remote sensing data to predict thepCO2wdata along the North American East Coast from 2015 to 2021 using the Cubist algorithm. Results show that the semi-analytic algorithm, Cubist, performs best among 20 statistical and machine learning models. Moreover, compared with the control experiment without the SSR data, after adding SSR as an independent variable, the final Cubist model’s coefficient of determination (R2) increased from 0.88 to 0.95, and the root mean square error reduced from 21.75 to 14.79 μatm. Our results showed significant improvement over the previous study (R2= 0.8), proving the applicability of applying SSR data in retrieving high spatial resolution carbonate system parameters in the future, especially for coastal regions where wind and wave dynamics are more variable.

Accurate and detailed ocean interior salinity data are fundamental to the studies of hydrological cycle, ocean current, and global climate change. This study adopted multisource satellite-based sea surface parameters and Argo float data for downscaling the subsurface salinity with higher spatial resolution based on the Light Gradient Boosting Machine (LightGBM) and the Convolutional Neural Networks (CNN) algorithms. Moreover, we utilized the sparse EN4 profiles gridded salinity data with quarter degree (EN4-GOSQ) for the accuracy validation of downscaling results. The model performance was well measured by the determination coefficient (R2), root mean square error (RMSE), and relative RMSE (RRMSE). The R2and RRMSE indicated the advanced LightGBM model is more preferable for downscaling global subsurface salinity than the CNN model.

Accurately retrieving ocean interior parameters from remote sensing observations is essential for ocean and climate studies because direct observations are sparse and costly. Furthermore, high-resolution structure of seawater properties is critical for understanding the oceanic processes and changes on multiple scales. Here, we designed a new method based on a deep neural network to retrieve subsurface temperature anomaly (STA) and subsurface salinity anomaly (SSA) in the Pacific Ocean at high (1/4°) and super (1/12°) horizontal resolution. We utilized multisource satellite-observed sea surface data (e.g., sea level, temperature, salinity, and wind vector) as inputs. The results revealed that our model retrieved the high- and super-resolution STA/SSA with high accuracy, and the model was reliable in a wide range of depths (near surface to 4000 m) and times (all months in 2014). Regarding the high-resolution STA (SSA) estimation, the average coefficient of determination ($R^{2}$) was 0.984 (0.966), and the average root-mean-squared error (RMSE) was 0.068 °C (0.016 psu). For the super-resolution STA, the average$R^{2}$was 0.988 and RMSE was 0.093 °C. Here, we established an effective technique that improved the resolution and accuracy of estimating the ocean interior parameters from satellite observation. The new technique provides some new insights into oceanic observation and dynamics.

Sea surface temperature (SST) prediction has raised considerable attention in various ocean-related fields. However, these methods were only limited to the time-sequence prediction of some isolated points, and their spatial linkage was not considered. Furthermore, these studies only predict the temperature of sea surface, but the subsurface temperature in the inner ocean is much more important. In this letter, we propose a model of multilayer convolutional long- and short-term memory (M-convLSTM) to predict 3-D ocean temperature, comprising convolutional neural networks (CNNs), long- and short-term memory (LSTM), and multiple layer stacking to consider the horizontal and vertical temperature variations from sea surface to subsurface to be about 2000 m below. Global marine environment observation data (ARGO) are used to conduct the prediction of 3-D ocean temperature in this letter, and the results demonstrate the overall good accuracy of forecast and ARGO data.

Accurate estimation of ocean's interior salinity information based on surface remote sensing data is quite significant for understanding complex dynamic processes in the ocean. This study adopts two kinds of ensemble learning algorithm, Random Forest (RF) and Gradient Boosting Decision Tree (GBDT) to estimate the subsurface salinity anomaly (SSA) in the upper 2000 m of the global ocean from multisource satellite-based sea surface parameters. The model performance is measured by R-square (R2) and normalized root-mean-square error (NRMSE). The results indicate the RF and GBDT models are both well suitable for retrieving SSA in the global ocean's interior and RF model outperforms GBDT model; the models accuracy generally decreased with the depth below 500 m. This study is helpful in understanding subsurface and deeper ocean environment response to recent global warming.

The objective of this study is to identify, catalog, and relate surface phenomena observed in SAR imagery of the Delaware Bay with concurrent environmental conditions (tides, winds, currents, river discharge, etc). To accomplish this objective, we examined a sequence of RADARSAT-1 SAR containing approximately 176 standard mode scenes collected over the Delaware Bay and adjacent coastal waters between May 1997 and February 2004. In this paper we report on 58 standard mode-2 scenes collected during ascending orbits. Unique to this RADARSAT image time sequence is the contemporaneous set of in situ meteorological and oceanographic observations that were collected from shipboard platforms and fixed stations throughout the Delaware Bay during the image acquisition period.

To gain insight into the effects of currents in tidal channels on SAR imaging, a series of hydrodynamic and SAR simulations were conducted using an idealized tidal channel. Preliminary results from the hydrodynamic and SAR modeling reveal the following: (1) peak modulations lie over mid-channel during flood and over the channel banks during ebb, (2) there is little, if any, change in these positions during respective tidal phases; only during transition is peak migration evident, and (3) there are distinct relationships between NRCS and along-channel current velocity and between NRCS and tide height.

In this study we examine and comparatively analyze synthetic aperture radar and near infrared Thematic Mapper signatures observed in the restricted tidal channel of a shallow lagoonal estuary. Both imaging systems show coherent structures, identified as longitudinal convergence zones, that appear as bright and dark bands aligned parallel with the axis of the tidal channel in question. This result indicates that Doppler shifting is not the mechanism responsible for the observed SAR signatures. Furthermore, modulation of relative intensity shows similar trends in location and magnitude of peak intensity with respect to tidal phase. Background variances from the unperturbed regions also show similarities between the two sensor systems. Finally, TM imagery, in the context employed here, is limited by the solar illumination cycle with the optimal period from May through August for the northern latitude site examined here.

Seasonal coastal upwelling in the U.S. Mid-Atlantic coastal ocean normally occurs during the summer months because of generally alongshore southerly wind episodes. Southerly winds force an offshore surface Ekman flow over the inner continental shelf. Colder and nutrient-rich waters from below upwell toward the surface replacing offshore-flowing surface waters. Synthetic Aperture Radar (SAR) observations from the European Remote Sensing (ERS) satellite ERS-2 before and after upwelling-favorable wind episodes in early summer 1996 along the New Jersey coast are presented. Lower backscatter conditions appearing in the SAR imagery after the onset of upwelling demonstrate the influence of the upwelling regime on the sea surface roughness. Satellite sea surface temperature (SST) observations and in-situ sea temperature vertical profiles confirm upwelling conditions. Three key mechanisms are suggested to explain the lower radar returns observed under upwelling conditions, an increase in the atmospheric marine boundary layer stability, an increase in the viscosity of surface waters, and the presence of biogenic surfactants in the upwelling region.

Empirical orthogonal functions (EOFs) are used to study a time-series of IR images from the Delaware Bay region, to determine the dominant patterns of sea surface temperature (SST) variance. Thirty-two months of AVHRR imagery, from February 1992 to September 1994, were used in the study. The SST images were averaged into monthly means for the analysis. EOF analysis decomposed the time series into its component parts. For Delaware Bay, the seasonal solar heating cycle dominated the results, accounting for 95.3% of the total variance. Higher modes in this area are interpreted as the result of spring heating and fall cooling of shallow river and bay waters ahead of offshore waters. The EOF results provide independent satellite derived evidence for lateral variability in the circulation of the estuary, which had previously been found in in-situ studies. The lateral variability in the bay circulation is a result of Delaware Bay's drowned river valley bathymetry.

The idea of shape matching is applied to track edge features automatically in a pair of AVHRR IR satellite images. The centroid and radius weighted mean are chosen as shape-specific points of the edges in two sequential images. Through the correspondence of these two shape-specific points, the whole edge's properties of rotation, translation, and scaling could be obtained. Also, a better correspondence in the second image is chosen according to the similarity comparison. After that, the total velocities of the points in the pattern can be computed. Velocity components in normal and tangential directions for certain points on the edge are also computed through a simple trial-and-error procedure and vector decomposition.>

A method for quantitatively measuring ocean surface movement, using sequential 10.8- mu m-band AVHRR images, is presented. An ordered statistical edge detection algorithm is used to select ocean thermal pattern features by detecting and mapping gradients and at the same time discriminating between the water surface, land, and clouds. Use of edge detection to select features in this manner reduces the need to perform preprocess screening and masking to remove clouds and land. A constrained correlation based feature recognition scheme is then used to find the best match to the pattern feature in a subsequent image. Surface displacement direction and distance are calculated for each selected point with average period velocity being computed based on elapsed time.>