Deep learning-based super-resolution (SR) methods have been widely used in natural images; however, their applications in satellite-derived sea surface temperature (SST) have not yet been fully discussed. Hence, it is necessary to analyze the validity of deep learning-based SR methods in SST reconstruction. In this study, an SR model, including multiscale feature extraction and multireceptive field mapping, was first proposed. Then, the proposed model and four other existing SR models were applied to SST reconstruction and analyzed. First, compared with the bicubic interpolation method, the SR models can improve the reconstruction accuracy. Compared with four other SR models, the proposed model can achieve the lowest mean squared error (MAE) in the East China Sea (ECS), in the northwest Pacific (NWP) and in the west Atlantic (WA), the second-lowest MAE in the southeast Pacific (SEP); the lowest root mean squared error (RMSE) in ECS and WA, the second-lowest RMSE in NWP and SEP. Additionally, ODRE model can acquire the highest or the second-highest peak single-to-noise ratio (PSNR) and structural similarity index (SSIM) in ECS, NWP and SEP. Moreover, the number of missing pixels and SST variety are two essential factors in the SR performance. The proposed multiscale feature extraction process can enhance the SR performance, especially for small regions and stable SST regions. Finally, while a deeper network can be helpful in achieving SR performance, the approach of simply adding more dilation convolutions may not enhance the reconstruction accuracy.

Cloud and cloud shadow detection can be considered as a classification process, in which the objective pixels are partitioned into cloud/cloud shadow or non-cloud/non-cloud shadow classes. However, some cloud pixels, especially the thin cloud pixels, can be considered as a mixture of reflectances of clouds and land objects. Fuzzy clustering may better characterize the status of one given pixel belonging to clouds or non-clouds. The fuzzy c-means method (FCM) was utilized in this study for cloud and cloud shadow detection. In addition, the “flood-fill” morphological transformation may misclassify some clear-sky areas surrounded by clouds as cloud shadows as a whole, so a modified cloud shadow index calculation was proposed. Moreover, a cloud and cloud shadow spatial matching strategy based on the projection direction and spatial coexistence was used to exclude some pseudo cloud shadows. Fewer predefined parameters and spectral bands are needed is one characteristic of the proposed method. In this study, 41 scenes with percentages of cloud cover from 4.99% to 97.63%, were utilized to validate of the FCM. The detected results demonstrate that the thick and thin clouds along with their associated cloud shadows can be precisely extracted by using the FCM. Compared with Fmask method, the FCM has relatively lower producer agreement rates, but it misclassifies as clouds fewer clear-sky pixels; compared with SVM method, the FCM can achieve better cloud detection accuracy. The results demonstrate that the FCM can attain a better balance between cloud pixel detection and non-cloud pixel exclusion.

Landsat and MODIS data have been widely utilized in many remote sensing applications, however, the trade-off between the spatial resolution and temporal frequency has limited their capacities in monitoring detailed spatio-temporal dynamics. Spatio-temporal fusion methods based on a linear model that considers the differences between fine- and coarse-spatial-resolution images as linear can effectively solve this trade-off problem, yet the existing linear fusion methods either regard the coefficients of the linear model as constants or have adopted regression methods to calculate the coefficients, both of which may introduce some errors in the fusion process. In this paper, we proposed an enhanced linear spatio-temporal fusion method (ELSTFM) to improve the data fusion accuracy. In the ELSTFM, it is not necessary to calculate the slope of the linear model, and the intercept, which can be deemed as the residual caused by systematic biases, is calculated based on spectral unmixing theory. Additionally, spectrally similar pixels in a given fine-spatial-resolution pixel’s neighborhood and their corresponding weights were used in the proposed method to mitigate block effects. Landsat-7/ETM+ and 8-day composite MODIS reflectance data covering two study sites with heterogeneous and homogenous landscapes were selected to validate the proposed method. Compared to three other typical spatio-temporal fusion methods visually and quantitatively, the predicted images obtained from ELSTFM could acquire better results for the two selected study sites. Furthermore, the resampling methods used to resample MODIS to the same spatial resolution of Landsat could slightly, but did not significantly influence the fusion accuracy, and the distributions of slopes of different bands for the two study sites could all be deemed as normal distributions with a mean value close to 1. The performance of ELSTFM depends on the accuracy of residual calculation at fine-resolution and large landscape changes may influence the fusion accuracy.

Dissolved oxygen (DO) is essential for assessing and monitoring the health of marine ecosystems. The phenomenon of ocean deoxygenation is widely recognized. Nevertheless, the limited availability of observations poses a challenge in achieving a comprehensive understanding of global ocean DO dynamics and trends. The study addresses the challenge of unevenly distributed Argo DO data by developing time–space–depth machine learning (TSD-ML), a novel machine learning-based model designed to enhance reconstruction accuracy in data-sparse regions. TSD-ML partitions Argo data into segments based on time, depth, and spatial dimensions, and conducts model training for each segment. This research contrasts the effectiveness of partitioned and non-partitioned modeling approaches using three distinct ML regression methods. The results reveal that TSD-ML significantly enhances reconstruction accuracy in areas with uneven DO data distribution, achieving a 30% reduction in root mean square error (RMSE) and a 20% decrease in mean absolute error (MAE). In addition, a comparison with WOA18 and GLODAPv2 ship survey data confirms the high accuracy of the reconstructions. Analysis of the reconstructed global ocean DO trends over the past two decades indicates an alarming expansion of anoxic zones.

Oceanic dissolved oxygen (DO) is crucial for oceanic material cycles and marine biological activities. However, obtaining subsurface DO values directly from satellite observations is limited due to the restricted observed depth. Therefore, it is essential to develop a connection between surface oceanic parameters and subsurface DO values. Machine learning (ML) methods can effectively grasp the complex relationship between input attributes and target variables, making them a valuable approach for estimating subsurface DO values based on surface oceanic parameters. In this study, the potential of ML methods for subsurface DO retrieval is analyzed. Among the selected ML methods, namely support vector regression (SVR), random forest (RF) regression, and extreme gradient boosting (XGBoosting) regression, the RF method generally demonstrates superior performance. As the depth increases, the accuracy of DO estimates tends to initially decrease, then gradually improve, with the poorest performance occurring at the depth of 600 dbar. The range of determination coefficients (R²) and root mean square error (RMSE) values based on the test dataset at different depths lies between 0.53 and 47.59 μmol/kg to 0.99 and 4.01 μmol/kg. In addition, compared to sea surface salinity (SSS) and sea surface chlorophyll-a (SCHL), sea surface temperature (SST) plays a more significant role in DO retrieval. Finally, compared to the pelagic interactions scheme for carbon and ecosystem studies (PISCES) model, the RF method achieves higher retrieval accuracies at depths above 700 dbar. In the deep ocean, the primary differences in DO values obtained from the RF method and the PISCES model-based method are noticeable in the vicinity of the equatorial region.