Retrieval of subsurface dissolved oxygen from surface oceanic parameters based on machine learning.

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.

Tree-based approach for exploring marine spatial patterns with raster datasets.

From multiple raster datasets to spatial association patterns, the data-mining technique is divided into three subtasks, i.e., raster dataset pretreatment, mining algorithm design, and spatial pattern exploration from the mining results. Comparison with the former two subtasks reveals that the latter remains unresolved. Confronted with the interrelated marine environmental parameters, we propose a Tree-based Approach for eXploring Marine Spatial Patterns with multiple raster datasets called TAXMarSP, which includes two models. One is the Tree-based Cascading Organization Model (TCOM), and the other is the Spatial Neighborhood-based CAlculation Model (SNCAM). TCOM designs the "Spatial node→Pattern node" from top to bottom layers to store the table-formatted frequent patterns. Together with TCOM, SNCAM considers the spatial neighborhood contributions to calculate the pattern-matching degree between the specified marine parameters and the table-formatted frequent patterns and then explores the marine spatial patterns. Using the prevalent quantification Apriori algorithm and a real remote sensing dataset from January 1998 to December 2014, a successful application of TAXMarSP to marine spatial patterns in the Pacific Ocean is described, and the obtained marine spatial patterns present not only the well-known but also new patterns to Earth scientists.