Global warming has made the Arctic increasingly available for marine operations and created a demand for reliable operational sea ice forecasts to increase safety. Because ocean-ice numerical models are highly computationally intensive, relatively lightweight ML-based methods may be more efficient for sea ice forecasting. Many studies have exploited different deep learning models alongside classical approaches for predicting sea ice concentration in the Arctic. However, only a few focus on daily operational forecasts and consider the real-time availability of data needed for marine operations. In this article, we aim to close this gap and investigate the performance of the U-Net model trained in two regimes for predicting sea ice for up to the next 10 days. We show that this deep learning model can outperform simple baselines by a significant margin, and we can improve the model’s quality by using additional weather data and training on multiple regions to ensure its generalization abilities. As a practical outcome, we build a fast and flexible tool that produces operational sea ice forecasts in the Barents Sea, the Labrador Sea, and the Laptev Sea regions.

The eddy‐resolving (1/12°) global ocean reanalysis GLORYS12 is assessed against 14 years (2002–2015) of independent hydrographic observations collected at 59.5°N in the North Atlantic. Two multi‐observations statistical analyses, ISAS‐15 and ARMOR‐3D, and an eddy‐permitting (1/4°) reanalysis, GLORYS025, also contribute to this comparison. The mean thermohaline structure along the 59.5°N section revealed by the observations is well‐reproduced by GLORYS12, except for the overflow waters. A good agreement with observations is found for linear trends over the whole period, exhibiting a dipole‐like pattern with a cooling/freshening in the main thermocline and a warming/salinization below. However, localized discrepancies with observations suggest the need for improvement in the reanalysis system (especially in the overflows representation and the consistency between the forcing and data assimilation system) and the deep observational array. The reanalysis reliably represents the ocean heat content in the upper 700‐m layer but shows significant differences with observations between 700 and 2,000 m. The meridional volume and heat transports across the 59.5°N section are compared for years when ADCP observations were available. The reanalysis does not reproduce the variability observed in the western boundary current but agrees well with observed transports in the other parts of the section. The reanalysis reproduces the major mesoscale eddy features that contribute to the meridional transport and provides the large‐scale context of their location. The analysis of time correlation at all measurement points demonstrated that GLORYS12 is the most accurate among the analyzed datasets used in this study to represent and explain the observed ocean characteristics and variability along that section.

Polar mesocyclones (MCs) are small marine atmospheric vortices. The class of intense MCs, called polar lows, are accompanied by extremely strong surface winds and heat fluxes and thus largely influencing deep ocean water formation in the polar regions. Accurate detection of polar mesocyclones in high-resolution satellite data, while challenging, is a time-consuming task, when performed manually. Existing algorithms for the automatic detection of polar mesocyclones are based on the conventional analysis of patterns of cloudiness and they involve different empirically defined thresholds of geophysical variables. As a result, various detection methods typically reveal very different results when applied to a single dataset. We develop a conceptually novel approach for the detection of MCs based on the use of deep convolutional neural networks (DCNNs). As a first step, we demonstrate that DCNN model is capable of performing binary classification of 500 × 500 km patches of satellite images regarding MC patterns presence in it. The training dataset is based on the reference database of MCs manually tracked in the Southern Hemisphere from satellite mosaics. We use a subset of this database with MC diameters falling in the range of 200–400 km. This dataset is further used for testing several different DCNN setups, specifically, DCNN built “from scratch”, DCNN based on VGG16 pre-trained weights also engaging the Transfer Learning technique, and DCNN based on VGG16 with Fine Tuning technique. Each of these networks is further applied to both infrared (IR) and a combination of infrared and water vapor (IR + WV) satellite imagery. The best skills (97% in terms of the binary classification accuracy score) is achieved with the model that averages the estimates of the ensemble of different DCNNs. The algorithm can be further extended to the automatic identification and tracking numerical scheme and applied to other atmospheric phenomena that are characterized by a distinct signature in satellite imagery.

We investigate in this paper the sensitivity of the representation of the Denmark Strait overflow produced by a regional z-coordinate configuration of NEMO (version 3.6) to the horizontal and vertical grid resolutions and to various numerical and physical parameters. Three different horizontal resolutions, 1∕12, 1∕36, and $\mathrm{1}/\mathrm{60}{}^{\circ }$ , are respectively used with 46, 75, 150, and 300 vertical levels. In the given numerical set-up, the increase in the vertical resolution did not bring improvement at eddy-permitting resolution ($\mathrm{1}/\mathrm{12}{}^{\circ }$ ). We find a greater dilution of the overflow as the number of vertical level increases, and the worst solution is the one with 300 vertical levels. It is found that when the local slope of the grid is weaker than the slope of the topography the result is a more diluted vein. Such a grid enhances the dilution of the plume in the ambient fluid and produces its thickening. Although the greater number of levels allows for a better resolution of the ageostrophic Ekman flow in the bottom layer, the final result also depends on how the local grid slope matches the topographic slope. We also find that for a fixed number of levels, the representation of the overflow is improved when horizontal resolution is increased to 1∕36 and $\mathrm{1}/\mathrm{60}{}^{\circ }$ , with the most drastic improvements being obtained with 150 levels. With such a number of vertical levels, the enhanced vertical mixing associated with the step-like representation of the topography remains limited to a thin bottom layer representing a minor portion of the overflow. Two major additional players contribute to the sinking of the overflow: the breaking of the overflow into boluses of dense water which contribute to spreading the overflow waters along the Greenland shelf and within the Irminger Basin, and the resolved vertical shear that results from the resolution of the bottom Ekman boundary layer dynamics. This improves the accuracy of the calculation of the entrainment by the turbulent kinetic energy mixing scheme (as it depends on the local shear) and improves the properties of the overflow waters such that they more favourably compare with observations. At 300 vertical levels the dilution is again increased for all horizontal resolutions. The impact on the overflow representation of many other numerical parameters was tested (momentum advection scheme, lateral friction, bottom boundary layer parameterization, closure parameterization, etc.), but none had a significant impact on the overflow representation.

Total Cloud Cover (TCC) retrieval from ground-based optical imagery is a problem that has been tackled by several generations of researchers. The number of human-designed algorithms for the estimation of TCC grows every year. However, there has been no considerable progress in terms of quality, mostly due to the lack of systematic approach to the design of the algorithms, to the assessment of their generalization ability, and to the assessment of the TCC retrieval quality. In this study, we discuss the optimization nature of data-driven schemes for TCC retrieval. In order to compare the algorithms, we propose a framework for the assessment of the algorithms’ characteristics. We present several new algorithms that are based on deep learning techniques: A model for outliers filtering, and a few models for TCC retrieval from all-sky imagery. For training and assessment of data-driven algorithms of this study, we present the Dataset of All-Sky Imagery over the Ocean (DASIO) containing over one million all-sky optical images of the visible sky dome taken in various regions of the world ocean. The research campaigns that contributed to the DASIO collection took place in the Atlantic ocean, the Indian ocean, the Red and Mediterranean seas, and the Arctic ocean. Optical imagery collected during these missions are accompanied by standard meteorological observations of cloudiness characteristics made by experienced observers. We assess the generalization ability of the presented models in several scenarios that differ in terms of the regions selected for the train and test subsets. As a result, we demonstrate that our models based on convolutional neural networks deliver a superior quality compared to all previously published approaches. As a key result, we demonstrate a considerable drop in the ability to generalize the training data in the case of a strong covariate shift between the training and test subsets of imagery which may occur in the case of region-aware subsampling.