Search for: All records

Abstract The Southern Ocean is a region of intense air–sea exchange that plays a critical role for ocean circulation, global carbon cycling, and climate. Subsurface chlorophyll‐a maxima, annually recurrent features throughout the Southern Ocean, may increase the energy flux to higher trophic levels and facilitate downward carbon export. It is important that model parameterizations appropriately represent the chlorophyll vertical structure in the Southern Ocean. Using BGC‐Argo chlorophyll profiles and the Biogeochemical Southern Ocean State Estimate (B‐SOSE), we investigate the sensitivity of chlorophyll vertical structure to model parameters. Based on the sensitivity analysis results, we estimate optimized parameters, which efficiently improve the model consistency with observations. We characterize chlorophyll vertical structure in terms of Empirical Orthogonal Functions and define metrics to compare model results and observations in a series of parameter perturbation experiments. We show that chlorophyll magnitudes are likely to respond quasi‐symmetrically to perturbations in the analyzed parameters, while depth and thickness of the subsurface chlorophyll maximum show an asymmetric response. Perturbing the phytoplankton growth tends to generate more symmetric responses than perturbations in the grazing rate. We identify parameters that affect chlorophyll magnitude, subsurface chlorophyll or both and discuss insights into the processes that determine chlorophyll vertical structure in B‐SOSE. We highlight turbulence, differences in phytoplankton traits, and grazing parameterizations as key areas for improvement in models of the Southern Ocean. 

Abstract Recent advances in explainable artificial intelligence (XAI) methods show promise for understanding predictions made by machine learning (ML) models. XAI explains how the input features are relevant or important for the model predictions. We train linear regression (LR) and convolutional neural network (CNN) models to make 1-day predictions of sea ice velocity in the Arctic from inputs of present-day wind velocity and previous-day ice velocity and concentration. We apply XAI methods to the CNN and compare explanations to variance explained by LR. We confirm the feasibility of using a novel XAI method [i.e., global layerwise relevance propagation (LRP)] to understand ML model predictions of sea ice motion by comparing it to established techniques. We investigate a suite of linear, perturbation-based, and propagation-based XAI methods in both local and global forms. Outputs from different explainability methods are generally consistent in showing that wind speed is the input feature with the highest contribution to ML predictions of ice motion, and we discuss inconsistencies in the spatial variability of the explanations. Additionally, we show that the CNN relies on both linear and nonlinear relationships between the inputs and uses nonlocal information to make predictions. LRP shows that wind speed over land is highly relevant for predicting ice motion offshore. This provides a framework to show how knowledge of environmental variables (i.e., wind) on land could be useful for predicting other properties (i.e., sea ice velocity) elsewhere. Significance StatementExplainable artificial intelligence (XAI) is useful for understanding predictions made by machine learning models. Our research establishes trustability in a novel implementation of an explainable AI method known as layerwise relevance propagation for Earth science applications. To do this, we provide a comparative evaluation of a suite of explainable AI methods applied to machine learning models that make 1-day predictions of Arctic sea ice velocity. We use explainable AI outputs to understand how the input features are used by the machine learning to predict ice motion. Additionally, we show that a convolutional neural network uses nonlinear and nonlocal information in making its predictions. We take advantage of the nonlocality to investigate the extent to which knowledge of wind on land is useful for predicting sea ice velocity elsewhere. 

Southern Ocean air–sea fluxes are a critical component of the climate system but are historically undersampled due to the remoteness of the region. While much focus has been placed on interannual flux variability, it has become increasingly clear that high-frequency fluctuations, driven by processes like storms and (sub-)mesoscale eddies, play a nonnegligible role in longer-term changes. Therefore, collecting high-resolution in situ flux observations is crucial to better understand the dynamics operating at these scales, as well as their larger-scale impacts. Technological advancements, including the development of new uncrewed surface vehicles, provide the opportunity to increase sampling at small scales. However, determining where and when to deploy such vehicles is not trivial. This study, conceived by the Air–Sea Fluxes working group of the Southern Ocean Observing System, aims to characterize the statistics of high-frequency air–sea flux variability. Using statistical analyses of atmospheric reanalysis data, numerical model output, and mooring observations, we show that there are regional and seasonal variations in the magnitude and sign of storm- and eddy-driven air–sea flux anomalies, which can help guide the planning of field campaigns and deployment of uncrewed surface vehicles in the Southern Ocean. 

Abstract The Southern Ocean is rich in highly dynamic mesoscale eddies and substantially modulates global biogeochemical cycles. However, the overall surface and subsurface effects of eddies on the Southern Ocean biogeochemistry have not been quantified observationally at a large scale. Here, we co‐locate eddies, identified in the Meta3.2DT satellite altimeter‐based product, with biogeochemical Argo floats to determine the effects of eddies on the dissolved inorganic carbon (DIC), nitrate, and dissolved oxygen concentrations in the upper 1,500 m of the ice‐free Southern Ocean, as well as the eddy effects on the carbon fluxes in this region. DIC and nitrate concentrations are lower in anticyclonic eddies (AEs) and increased in cyclonic eddies (CEs), while dissolved oxygen anomalies switch signs above (CEs: positive, AEs: negative) and below the mixed layer (CEs: negative, AEs: positive). We attribute these anomalies primarily to eddy pumping (isopycnal heave), as well as eddy trapping for oxygen. Maximum anomalies in all tracers occur at greater depths in the subduction zone north of the Antarctic Circumpolar Current (ACC) compared to the upwelling region in the ACC, reflecting differences in background vertical structures. Eddy effects on air–sea exchange have significant seasonal variability, with additional outgassing in CEs in fall (physical process) and additional oceanic uptake in AEs and CEs in spring (biological and physical process). Integrated over the Southern Ocean, AEs contribute 0.01 Pg C (7 ) to the Southern Ocean carbon uptake, and CEs offset this by 0.01 Pg C (2 ). These findings underscore the importance of considering eddy impacts in observing networks and climate models. 

Abstract West Antarctic Ice Sheet mass loss is a major source of uncertainty in sea level projections. The primary driver of this melting is oceanic heat from Circumpolar Deep Water originating offshore in the Antarctic Circumpolar Current. Yet, in assessing melt variability, open ocean processes have received considerably less attention than those governing cross-shelf exchange. Here, we use Lagrangian particle release experiments in an ocean model to investigate the pathways by which Circumpolar Deep Water moves toward the continental shelf across the Pacific sector of the Southern Ocean. We show that Ross Gyre expansion, linked to wind and sea ice variability, increases poleward heat transport along the gyre’s eastern limb and the relative fraction of transport toward the Amundsen Sea. Ross Gyre variability, therefore, influences oceanic heat supply toward the West Antarctic continental slope. Understanding remote controls on basal melt is necessary to predict the ice sheet response to anthropogenic forcing. 

Open-ocean polynyas formed over the Maud Rise, in the Weddell Sea, during the winters of 2016–2017. Such polynyas are rare events in the Southern Ocean and are associated with deep convection, affecting regional carbon and heat budgets. Using an ocean state estimate, we found that during 2017, early sea ice melting occurred in response to enhanced vertical mixing of heat, which was accompanied by mixing of salt. The melting sea ice compensated for the vertically mixed salt, resulting in a net buoyancy gain. An additional salt input was then necessary to destabilize the upper ocean. This came from a hitherto unexplored polynya-formation mechanism: an Ekman transport of salt across a jet girdling the northern flank of the Maud Rise. Such transport was driven by intensified eastward surface stresses during 2015–2018. Our results illustrate how highly localized interactions between wind, ocean flow and topography can trigger polynya formation in the open Southern Ocean. 

Modeled water-mass changes in the North Pacific thermocline, both in the subsurface and at the surface, reveal the impact of the competition between anthropogenic aerosols (AAs) and greenhouse gases (GHGs) over the past 6 decades. The AA effect overwhelms the GHG effect during 1950–1985 in driving salinity changes on density surfaces, while after 1985 the GHG effect dominates. These subsurface water-mass changes are traced back to changes at the surface, of which ~70% stems from the migration of density surface outcrops, equatorward due to regional cooling by AAs and subsequent poleward due to warming by GHGs. Ocean subduction connects these surface outcrop changes to the main thermocline. Both observations and models reveal this transition in climate forcing around 1985 and highlight the important role of AA climate forcing on our oceans’ water masses. 

Physics-based simulations of Arctic sea ice are highly complex, involving transport between different phases, length scales, and time scales. Resultantly, numerical simulations of sea ice dynamics have a high computational cost and model uncertainty. We employ data-driven machine learning (ML) to make predictions of sea ice motion. The ML models are built to predict present-day sea ice velocity given present-day wind velocity and previous-day sea ice concentration and velocity. Models are trained using reanalysis winds and satellite-derived sea ice properties. We compare the predictions of three different models: persistence (PS), linear regression (LR), and a convolutional neural network (CNN). We quantify the spatiotemporal variability of the correlation between observations and the statistical model predictions. Additionally, we analyze model performance in comparison to variability in properties related to ice motion (wind velocity, ice velocity, ice concentration, distance from coast, bathymetric depth) to understand the processes related to decreases in model performance. Results indicate that a CNN makes skillful predictions of daily sea ice velocity with a correlation up to 0.81 between predicted and observed sea ice velocity, while the LR and PS implementations exhibit correlations of 0.78 and 0.69, respectively. The correlation varies spatially and seasonally: lower values occur in shallow coastal regions and during times of minimum sea ice extent. LR parameter analysis indicates that wind velocity plays the largest role in predicting sea ice velocity on 1-day time scales, particularly in the central Arctic. Regions where wind velocity has the largest LR parameter are regions where the CNN has higher predictive skill than the LR.