The recently developed Ice, Cloud, and Land Elevation Satellite 2 (ICESat-2), furnished with the Advanced Terrain Laser Altimeter System (ATLAS), delivers considerable benefits in providing accurate bathymetric data across extensive geographical regions. By integrating active lidar-derived reference seawater depth data with passive optical remote sensing imagery, efficient bathymetry mapping is facilitated. In recent times, machine learning models are frequently used to define the nonlinear connection between remote sensing spectral data and water depths, which consequently results in the creation of bathymetric maps. A salient model among these is the convolutional neural network (CNN), which effectively integrates contextual information concerning bathymetric points. However, current CNN models and other machine learning approaches mainly concentrate on recognizing mathematical relationships within the data to determine a water depth function and remote sensing spectral data, while oftentimes disregarding the physical light propagation process in seawater before reaching the seafloor. This study presents a physics-informed CNN (PI-CNN) model which incorporates radiative transfer-based data into the CNN structure. By including the shallow water double-band radiative transfer physical term (swdrtt), this model enhances seawater spectral features and also considers the context surroundings of bathymetric pixels. The effectiveness and reliability of our proposed PI-CNN model are verified using in situ data from St. Croix and St. Thomas, validating its correctness in generating bathymetric maps with a broad experimental R² accuracy exceeding 95% and remaining errors below 1.6 m. Preliminary results suggest that our PI-CNN model surpasses conventional methodologies.

Ocean color remote sensing technology has proven to be an indispensable tool for monitoring ocean conditions, as it has consistently provided critical data on global ocean optical properties, color, and biogeochemical parameters over several decades. With the rapid advancement of artificial intelligence, the integration of machine learning (ML) models into ocean color remote sensing has become a significant focus within the scientific community. This article provides a comprehensive review of the current status and challenges associated with ML models in ocean color remote sensing, assessing their applications in atmospheric correction, color inversion, carbon cycle analysis, and data reconstruction. This review highlights the advancements made in applying ML techniques, such as neural networks and deep learning, to improve data accuracy, enhance resolution, and enable more precise predictions of oceanic phenomena. Despite challenges such as model generalization and computational complexity, ML has significant potential for enhancing our understanding of marine ecosystems, facilitating real-time monitoring, and supporting global climate models.

The ocean absorbs a significant amount of carbon dioxide (CO₂) from the atmosphere, helping regulate Earth's climate. However, our knowledge of ocean CO₂ sink levels remains limited. This research focused on assessing daily changes in ocean CO₂ sink levels and air–sea CO₂ exchange, using a new technique. We used LiDAR technology, which provides continuous measurements during day and night, to estimate global ocean CO₂ absorption over 23 years. Our model successfully reproduced sea surface partial pressure of CO₂ data. The results suggest the total amount of CO₂ absorbed by oceans is higher at night than during the day. This difference arises from a combination of factors like temperatures, winds, photosynthesis, and respiration. Understanding these daily fluctuations can improve predictions of ocean CO₂ uptake. It may also help explain why current carbon budget calculations are not fully balanced—an issue scientists have grappled with. Overall, this pioneering study highlights the value of LiDAR's unique day–night ocean data coverage. The findings advance knowledge of ocean carbon cycles and their role in climate regulation. They underscore the need to incorporate day–night variability when assessing the ocean's carbon sink capacity.

Quantified research on the Arctic Ocean carbon system is poorly understood, limited by the scarce available data. Measuring the associated phytoplankton responses to air–sea CO₂ fluxes is challenging using traditional satellite passive ocean color measurements due to low solar elevation angles. We constructed a feedforward neural network light detection and ranging (LiDAR; FNN-LID) method to assess the Arctic diurnal partial pressure of carbon dioxide (pCO₂) and formed a dataset of long-time-series variations in diurnal air–sea CO₂ fluxes from 2001 to 2020; this study represents the first time spaceborne LiDAR data were employed in research on the Arctic air–sea carbon cycle, thus providing enlarged data coverage and diurnal pCO₂ variations. Although some models replace Arctic winter Chl-a with the climatological average or interpolated Chl-a values, applying these statistical Chl-a values results in potential errors in the gap-filled wintertime pCO₂ maps. The CALIPSO measurements obtained through active LiDAR sensing are not limited by solar radiation and can thus provide ‘fill-in’ data in the late autumn to early spring seasons, when ocean color sensors cannot record data; thus, we constructed the first complete record of polar pCO₂. We obtained Arctic FFN-LID-fitted in situ measurements with an overall mean R² of 0.75 and an average RMSE of 24.59 µatm and filled the wintertime observational gaps, thereby indicating that surface water pCO₂ is higher in winter than in summer. The Arctic Ocean net CO₂ sink has seasonal sources from some continental shelves. The growth rate of Arctic seawater pCO₂ is becoming larger and more remarkable in sectors with significant sea ice retreat. The combination of sea surface partial pressure and wind speed impacts the diurnal carbon air–sea flux variability, which results in important differences between the Pacific and Atlantic Arctic Ocean. Our results show that the diurnal carbon sink is larger than the nocturnal carbon sink in the Atlantic Arctic Ocean, while the diurnal carbon sink is smaller than the nocturnal carbon sink in the Pacific Arctic Ocean.

Seawater pH is a direct proxy of ocean acidification, and monitoring the global pH distribution and long-term series changes is critical to understanding the changes and responses of the marine ecology and environment under climate change. Owing to the lack of sufficient global-scale pH data and the complex relationship between seawater pH and related environmental variables, generating time-series products of satellite-derived global sea surface pH poses a great challenge. In this study, we solved the problem of the lack of sufficient data for pH algorithm development by using the massive underway sea surface carbon dioxide partial pressure (pCO₂) dataset to structure a large data volume of near in situ pH based on carbonate calculation between underway pCO₂ and calculated total alkalinity from sea surface salinity and relevant parameters. The remote sensing inversion model of pH was then constructed through this massive pH training dataset and machine learning methods. After several tests of machine learning methods and groups of input parameters, we chose the random forest model with longitude, latitude, sea surface temperature (SST), chlorophyll a (Chla), and Mixed layer depth (MLD) as model inputs with the best performance of correlation coefficient (R² = 0.96) and root mean squared error (RMSE = 0.008) in the training set and R² = 0.83 (RMSE = 0.017) in the testing set. The sensitivity analysis of the error variation induced by the uncertainty of SST and Chla (SST ≤ ±0.5 °C and Chla ≤ ±20%; RMSE_SST ≤ 0.011 and RMSE_Chla ≤ 0.009) indicated that our sea surface pH model had good robustness. Monthly average global sea surface pH products from 2004 to 2019 with a spatial resolution of 0.25° × 0.25° were produced based on the satellite-derived SST and Chla products and modeled MLD dataset. The pH model and products were validated using another independent station-measured pH dataset from the Global Ocean Data Analysis Project (GLODAP), showing good performance. With the time-series pH products, refined interannual variability and seasonal variability were presented, and trends of pH decline were found globally. Our study provides a new method of directly using remote sensing to invert pH instead of indirect calculation based on the construction of massive underway calculated pH data, which would be made useful by comparing it with satellite-derived pCO₂ products to understand the carbonate system change and the ocean ecological environments responding to the global change.

The size and control mechanism of the Southern Ocean’s carbon fluxes remain highly uncertain due to sparse winter observations. Here, we integrate satellite light detection and ranging (LIDAR) measurements with machine learning to assess the Southern Ocean air-sea CO ₂ fluxes between 2007 and 2020. We reveal that CO ₂ outgassing south of 50°S was underestimated by up to 40% in previous studies. While the midlatitude Southern Ocean (30° to 50°S) strengthens as a carbon sink, the high-latitude region (50° to 90°S) shows Southern Annular Mode (SAM)–modulated alternation between uptake and outgassing. The air-sea CO ₂ partial pressure difference (Δ p CO ₂ ) increasingly dominates flux variability over wind-driven transfer velocity. We propose a framework involving three latitudinal loops with differing p CO ₂ controls: (i) Antarctic (salinity/sea ice), (ii) polar front (atmospheric CO ₂ /chlorophyll), and (iii) subpolar (sea surface temperature/CO ₂ ). The findings underscore the winter processes’ critical role and necessitate year-round observations to understand Southern Ocean’s global carbon cycle impact.