Search for: All records

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. 

Abstract Marine heatwaves (MHWs)—extreme ocean temperature anomalies—are increasing in frequency and intensity globally, yet their vertical structure and drivers remain underexplored in the Caribbean Sea (CS), a region of critical ecological and socioeconomic importance. Using high resolution climate model simulations and reanalysis fields, we show that MHW intensity in the CS peaks in the subsurface, where vertical gradients of temperature are stronger, and that MHWs are associated with the passage of Rossby waves. Projections under the RCP8.5 scenario reveal that while the ocean continues to warm to the end of the century, subsurface MHW intensity weakens. This decline is associated with a weaker thermocline, reduced ocean currents and eddy kinetic energy, and less energetic Rossby waves. Reducing uncertainties in climate projections will be essential to improve our understanding of how marine heat extremes may evolve and affect Caribbean ecosystems. 

Key Points Salinity variability in the Virgin Islands basin (VIB) is primarily driven by horizontal salinity advection Three paths dominate the import of waters into the VIB: two paths linked with Amazon plume waters and the third with Atlantic inflow Mesoscale features steer low salinity waters from the southeastern Caribbean toward the VIB 

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. 

Abstract Earth’s energy imbalance (EEI) is a fundamental metric of global Earth system change, quantifying the cumulative impact of natural and anthropogenic radiative forcings and feedback. To date, the most precise measurements of EEI change are obtained through radiometric observations at the top of the atmosphere (TOA), while the quantification of EEI absolute magnitude is facilitated through heat inventory analysis, where ~ 90% of heat uptake manifests as an increase in ocean heat content (OHC). Various international groups provide OHC datasets derived from in situ and satellite observations, as well as from reanalyses ingesting many available observations. The WCRP formed the GEWEX-EEI Assessment Working Group to better understand discrepancies, uncertainties and reconcile current knowledge of EEI magnitude, variability and trends. Here, 21 OHC datasets and ocean heat uptake (OHU) rates are intercompared, providing OHU estimates ranging between 0.40 ± 0.12 and 0.96 ± 0.08 W m⁻²(2005–2019), a spread that is slightly reduced when unequal ocean sampling is accounted for, and that is largely attributable to differing source data, mapping methods and quality control procedures. The rate of increase in OHU varies substantially between − 0.03 ± 0.13 (reanalysis product) and 1.1 ± 0.6 W m⁻² dec⁻¹(satellite product). Products that either more regularly observe (satellites) or fill in situ data-sparse regions based on additional physical knowledge (some reanalysis and hybrid products) tend to track radiometric EEI variability better than purely in situ-based OHC products. This paper also examines zonal trends in TOA radiative fluxes and the impact of data gaps on trend estimates. The GEWEX-EEI community aims to refine their assessment studies, to forge a path toward best practices, e.g., in uncertainty quantification, and to formulate recommendations for future activities. 

Abstract Atmospheric rivers (ARs) result in precipitation over land and ocean. Rainfall on the ocean can generate a buoyant layer of freshwater that impacts exchanges between the surface and the mixed layer. These “fresh lenses” are important for weather and climate because they may impact the ocean stratification at all time scales. Here we use in situ ocean data, collocated with AR events, and a one-dimensional configuration of a general circulation model, to investigate the impact of AR precipitation on surface ocean salinity in the California Current System (CCS) on seasonal and event-based time scales. We find that at coastal and onshore locations the CCS freshens through the rainy season due to AR events, and years with higher AR activity are associated with a stronger freshening signal. On shorter time scales, model simulations suggest that events characteristic of CCS ARs can produce salinity changes that are detectable by ocean instruments (≥0.01 psu). Here, the surface salinity change depends linearly on rain rate and inversely on wind speed. Higher wind speeds ( U > 8 m s −1 ) induce mixing, distributing freshwater inputs to depths greater than 20 m. Lower wind speeds ( U ≤ 8 m s −1 ) allow freshwater lenses to remain at the surface. Results suggest that local precipitation is important in setting the freshwater seasonal cycle of the CCS and that the formation of freshwater lenses should be considered for identifying impacts of atmospheric variability on the upper ocean in the CCS on weather event time scales. Significance Statement Atmospheric rivers produce large amounts of rainfall. The purpose of this study is to understand how this rain impacts the surface ocean in the California Current System on seasonal and event time scales. Our results show that a greater precipitation over the rainy season leads to a larger decrease in salinity over time. On shorter time scales, these atmospheric river precipitation events commonly produce a surface salinity response that is detectable by ocean instruments. This salinity response depends on the amount of rainfall and the wind speed. In general, higher wind speeds will cause the freshwater input from rain to mix deeper, while lower wind speeds will have reduced mixing, allowing a layer of freshwater to persist at the surface. 

Abstract Since the mid-2000s, the Argo oceanographic observational network has provided near-real-time four-dimensional data for the global ocean for the first time in history. Internet (i.e., the “web”) applications that handle the more than two million Argo profiles of ocean temperature, salinity, and pressure are an active area of development. This paper introduces a new and efficient interactive Argo data visualization and delivery web application named Argovis that is built on a classic three-tier design consisting of a front end, back end, and database. Together these components allow users to navigate 4D data on a world map of Argo floats, with the option to select a custom region, depth range, and time period. Argovis’s back end sends data to users in a simple format, and the front end quickly renders web-quality figures. More advanced applications query Argovis from other programming environments, such as Python, R, and MATLAB. Our Argovis architecture allows expert data users to build their own functionality for specific applications, such as the creation of spatially gridded data for a given time and advanced time–frequency analysis for a space–time selection. Argovis is aimed to both scientists and the public, with tutorials and examples available on the website, describing how to use the Argovis data delivery system—for example, how to plot profiles in a region over time or to monitor profile metadata.