Search

The AutoICE challenge, organized by multiple national and international agencies, seeks to advance the development of near-real-time sea ice products with improved spatial resolution, broader spatial and temporal coverage, and enhanced consistency. In this paper, we present a detailed description of our solutions and experimental results for the challenge. We have implemented an automated sea ice mapping pipeline based on a multi-task U-Net architecture, capable of predicting sea ice concentration (SIC), stage of development (SOD), and floe size (FLOE). The AI4Arctic dataset, which includes synthetic aperture radar (SAR) imagery, ancillary data, and ice-chart-derived label maps, is utilized for model training and evaluation. Among the submissions from over 30 teams worldwide, our team achieved the highest combined score of 86.3 %, as well as the highest scores on SIC (92.0 %) and SOD (88.6 %). Notably, the result analysis and ablation studies demonstrate that instead of model architecture design, a collection of strategies/techniques we employed led to substantial enhancement in accuracy, efficiency, and robustness within the realm of deep-learning-based sea ice mapping. Those techniques include input SAR variable downscaling, input feature selection, spatial–temporal encoding, and the choice of loss functions. By highlighting the various techniques employed and their impacts, we aim to underscore the scientific advancements achieved in our methodology. [ABSTRACT FROM AUTHOR]

The AutoIce challenge, organized by multiple national and international agencies, seeks to advance the development of near-real-time sea ice products with improved spatial resolution, broader spatial and temporal coverage, and enhanced consistency. In this paper, we present a detailed description of our solutions and experimental results for the challenge. We have implemented an automated sea ice mapping pipeline based on a multi-task U-Net architecture, capable of predicting sea ice concentration (SIC), stage of development (SOD), and floe size (FLOE) using Sentinel-1 SAR data. For model training and evaluation, we utilize the AI4Arctic dataset, which includes SAR imagery, corresponding passive microwave and auxiliary data, and ice chart-derived label maps. Among the submissions from over 30 teams worldwide, our team achieved the highest combined score of 86.3 %, as well as the highest scores on SIC (92.0 %) and SOD (88.6 %). Additionally, our result analysis showcases the effectiveness of various techniques, such as input SAR variable downscaling, spatial-temporal encoding, input feature selection, and loss function selection, in significantly improving the accuracy, efficiency, and robustness of deep learning-based sea ice mapping. [ABSTRACT FROM AUTHOR]

Bayesian neural networks (BNNs) have been demonstrated to be effective in accurate retrieval of sea ice concentration (SIC) from multisource data, while providing estimates of uncertainty, which are essential for downstream services. However, uncertainty obtained by BNNs is intrinsically uncalibrated, which indicates that it may not correlate well with model error. To address this issue, we investigate a new approach that combines an auxiliary (AUX) prediction interval (PI) estimator with the BNN-based SIC mean estimator to develop a well-calibrated SIC retrieval model that is both accurate and reliable. We adopt a training strategy called “uncertainty matching” to train the model, which ensures that the estimated uncertainties match the estimated PIs. We use a subset of AMSR2 brightness temperature data and ERA5 atmospheric data collected from 2014 to 2015 in the Baffin Bay area as input features of the model. Comparison between model inference and SIC labels obtained from the enhanced NASA Team (NT2) algorithm shows that the proposed approach is able to produce well-calibrated uncertainty with more accurate predictions in marginal ice zones (MIZs). [ABSTRACT FROM AUTHOR]

Copyright of Canadian Journal of Remote Sensing is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Accurate and timely forecasts of sea ice conditions are crucial for safe shipping operations in the Canadian Arctic and other ice-infested waters. Given the recent declining trend of Arctic sea ice extent in past decades, seasonal forecasts are often desired. In this study machine learning (ML) approaches are deployed to provide accurate seasonal forecasts based on ERA5 data as input. This study, unlike previous ML approaches in the sea ice forecasting domain, provides daily spatial maps of sea ice presence probability in the study domain for lead times up to 90 d using a novel spatiotemporal forecasting method based on sequence-to-sequence learning. The predictions are further used to predict freeze-up/breakup dates and show their capability to capture these events within a 7 d period at specific locations of interest to shipping operators and communities. The model is demonstrated in hindcasting mode to allow for evaluation of forecasted predication. However, the design allows for the approach to be used as a forecasting tool. The proposed method is capable of predicting sea ice presence probabilities with skill during the breakup season in comparison to both Climate Normal and sea ice concentration forecasts from a leading subseasonal-to-seasonal forecasting system. [ABSTRACT FROM AUTHOR]

A method to extend categorical triple collocation (CTC) to five data sets, three of which must have conditionally independent errors, is presented. The method is shown to enable comparison of three passive microwave sea ice concentration data sets, and can easily be extended to for use with more data sets. The method is evaluated in the Gulf of Saint Lawrence, on the east coast of Canada, during the freeze-up period. It is found that the ice concentration data set from the European space agency sea ice climate change initiative (ESA-SICCI) and the NASA Team 2 (NT2) algorithms have higher balanced accuracy than that from the Artist sea ice (ASI) algorithm. [ABSTRACT FROM AUTHOR]

Synthetic aperture radar (SAR) sea ice imagery is a promising source of data for sea ice data assimilation. Classification of SAR sea ice imagery into ice and water is of particular relevance due to its relationship with ice concentration, a key variable in sea ice data assimilation systems. With increasing volumes of SAR data, automated methods to carry out these classifications are of particular importance. Although several automated approaches have been proposed, none look at the impact of including an estimate of uncertainty of the model parameters and input features on the classification output. This article uses an established database of SAR image features to train a multilayer perceptron (MLP) neural network to classify pixel locations as either ice, water, or unknown. The classification accuracies are benchmarked using a recently developed logistic regression approach for the same database. The two methods are found to be comparable. The MLP approach is then enhanced to allow uncertainty to be estimated at each pixel location. Following methods proposed in the deep learning community, two kinds of uncertainty are considered. The first, epistemic uncertainty, is that due to uncertainty in the MLP weights. The second kind of uncertainty, aleatoric uncertainty, is that which cannot be explained by the model, and is therefore associated with the input data. It is found that including these uncertainties in the MLP models reduces their accuracies slightly, but also reduces misclassification rates. This is of particular importance for data assimilation applications, where misclassifications could severely degrade the analysis. [ABSTRACT FROM AUTHOR]

Accurate estimates of sharp features in the sea ice cover, such as leads and ridges, are critical for shipping activities, ice operations and weather forecasting. These sharp features can be difficult to preserve in data fusion and data assimilation due to the spatial correlations in the background error covariance matrices. In this article, a set of data fusion and data assimilation experiments are carried out comparing two objective functions, one with a conventional l2-norm and one that imposes an additional l1-norm on the derivative of the ice thickness state estimate. The latter is motivated by analysis of high resolution ice thickness observations derived from an airborne electromagnetic sensor demonstrating the sparsity of the ice thickness in the derivative domain. Data fusion and data assimilation experiments (using a 1 D toy sea-ice model) are carried out over a wide range of background and observation error correlation length scales. Results show the superiority of using an l1–l2 regularisation framework. For the data fusion experiments it was found when both background and observation error correlation length scales are zero, the ice thickness root mean squared error for the l1–l2 method was 0.16 m as compared to 0.20 m for the l2 method. The differences between the methods were greater when the background error correlation length scale was relatively short (approximately five times the analysis grid spacing), and were not significant for larger background error correlation length scales (e.g. 10 times the analysis grid spacing). For data assimilation experiments it was found that openings in the ice cover were captured better with the l1–l2 regularisation, with reduced errors in ice thickness, concentration and velocity. In addition, the ice thickness derivatives in the analyses were found to be more sparse when the l1–l2 method was used and are closer to the those from the true model run. [ABSTRACT FROM AUTHOR]

Monitoring the sea ice cover is important for both climate studies and ice operations, such as shipping. It is challenging to validate even basic essential variables, such as the sea ice extent, due to a lack of appropriate validation data. Instead of focusing on validation, this paper looks at the use of categorical triple collocation (CTC) for the task of quantitatively comparing three colocated data sets. CTC has been developed and used in earlier studies to rank binary data sets. In this paper, we extend earlier studies and bring in recent results from the binary classification community to estimate the class imbalance (the relative proportion of each class, ice or water). We then use this class imbalance to obtain quantitative estimates of the proportion correct of ice (sensitivity) and the proportion correct of water (specificity). The methodology is first tested using toy data, after which three data sets from the Gulf of Saint Lawrence, on the east coast of Canada, are used. These data sets are from an ice–ocean model, a passive microwave sea ice concentration retrieval, and a sea ice concentration retrieval from synthetic aperture radar (SAR). By looking at both the sensitivity and the specificity, it is found that the passive microwave data have difficulty in recognizing ice during freeze-up, but they perform well at obtaining the correct water observations. This distinction cannot be made by ranking the data sets. The CTC method is compared with, and found to be complementary to, a validation using ice/water states from the interactive multisensor snow and ice mapping system. [ABSTRACT FROM AUTHOR]

Sea ice prediction centres are moving toward the assimilation of ice thickness observations under the simplifying assumption that the observation errors are uncorrelated. The assumption of uncorrelated observation errors is attractive because the errors can be represented by a diagonal observation error covariance matrix, which is inexpensive to invert. In this paper a set of idealized experiments are carried out to investigate the impact of this assumption on sea ice analyses. A background error covariance matrix is generated using a 1D toy model for sea, i.e. forced with idealized models of the ocean and atmosphere. Analysis error covariance matrices are then calculated using this matrix for both correlated and uncorrelated observation error covariance matrices, . The results indicate when the true is correlated, using a diagonal approximation results in an analysis that is overconfident at the large scales, in that the analysis error standard deviation at the large scales is underestimated. It is also shown that for the largest observation error correlation length scale tested, 150 km, the analysis error standard deviation for ice thickness is reduced by 10.8% relative to the background error standard deviation when has the correct correlation length scale of 150 km, whereas when a diagonal approximation to is used in combination with an inflation factor, the reduction is to 6.3%. [ABSTRACT FROM AUTHOR]

In this study, a convolutional neural network (CNN) is used to estimate sea ice concentration using synthetic aperture radar (SAR) scenes acquired during freeze-up in the Gulf of St. Lawrence on the east coast of Canada. The ice concentration estimates from the CNN are compared to those from a neural network (multi-layer perceptron or MLP) that uses hand-crafted features as input and a single layer of hidden nodes. The CNN is found to be less sensitive to pixel level details than the MLP and produces ice concentration that is less noisy and in closer agreement with that from image analysis charts. This is due to the multi-layer (deep) structure of the CNN, which enables abstract image features to be learned. The CNN ice concentration is also compared with ice concentration estimated from passive microwave brightness temperature data using the ARTIST sea ice (ASI) algorithm. The bias and RMS of the difference between the ice concentration from the CNN and that from image analysis charts is reduced as compared to that from either the MLP or ASI algorithm. Additional results demonstrate the impact of varying the input patch size, varying the number of CNN layers, and including the incidence angle as an additional input. [ABSTRACT FROM AUTHOR]

A new tool has been developed to calculate dynamic, state-specific tie points, to aid in the assimilation of various types of satellite data into Environment and Climate Change Canada's Regional Ice Ocean Prediction System. These tie points are referred to as characteristic values (CVs). In this study, CVs are calculated for RadarSat-2 ScanSAR-Wide-A HH-HV backscatter data from October 2010 to September 2011. In this collection, the mean, standard deviation, and percentile distribution of backscatter at locations and times identified as being either ice or open water are represented over different relevant categories affecting the signal. The resulting water CVs are compared with modeled backscatter values, and are in close agreement at midrange wind speeds (5-14 m s-1), where wind slicks are not present. When compared against previously reported values, the ice CVs correspond best for ice conditions with fairly uniform backscatter distributions, such as the Arctic during the spring. When the ice and water CVs are compared to each other, the best cases for the assimilation of RadarSat-2 data are evident. In these cases, the CV distributions at a given incidence angle and wind speed are well separated from each other, such as in the far range (40°-50°) at midrange wind speeds. This set of CVs will be used for an initial assimilation of binary ice and open water retrievals. Future work will include a more complex treatment of ice CVs to address mixed ice types, and the application of CVs to other types of satellite data, including those from passive microwave sensors. [ABSTRACT FROM AUTHOR]

Anovel method based on the curvelet transform and active contour method to automatically detect the ice edge in Synthetic Aperture Radar (SAR) imagery is proposed. The method utilizes the location of high curvelet coefficients to determine regions in the image likely to contain the ice edge. Using an ice edge from passive microwave sea ice concentration for initialization, these regions are then joined using the active contour method to obtain the final ice edge. The method is evaluated on four dual polarization SAR scenes of the Labrador sea. Through comparison of the ice edge with that from image analysis charts, it is demonstrated that the proposed method can detect the ice edge effectively in SAR images. This is particularly relevant when the marginal ice zone is diffuse or the ice is thin, and using the definition of ice edge from the passive microwave ice concentration would underestimate the ice edge location. It is expected that the method may be useful for operations in marginal ice zones, such as offshore drilling, where a high resolution estimate of the ice edge location is required. It could also be useful as a first guess for an ice analyst, or for the assimilation of SAR data. [ABSTRACT FROM AUTHOR]

In this paper, the assimilation of binary observations calculated from synthetic aperture radar (SAR) images of sea ice is investigated. Ice and water observations are obtained from a set of SAR images by thresholding ice and water probabilities calculated using a supervised maximum likelihood estimator (MLE). These ice and water observations are then assimilated in combination with ice concentration from passive microwave imagery for the purpose of estimating sea ice concentration. Due to the fact that the observations are binary, consisting of zeros and ones, while the state vector is a continuous variable (ice concentration), the forward model used to map the state vector to the observation space requires special consideration. Both linear and non-linear forward models were investigated. In both cases, the assimilation of SAR data was able to produce ice concentration analyses in closer agreement with image analysis charts than when assimilating passive microwave data only. When both passive microwave and SAR data are assimilated, the bias between the ice concentration analyses and the ice concentration from ice charts is 19.78%, as compared to 26.72% when only passive microwave data are assimilated. The method presented here for the assimilation of SAR data could be applied to other binary observations, such as ice/water information from visual/infrared sensors. [ABSTRACT FROM AUTHOR]

Copyright of Atmosphere -- Ocean (Taylor & Francis Ltd) is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

In this paper, sea-ice thickness values are calculated along the Labrador coast using data from two sensors representative of those available for operational data assimilation. Data from the first sensor, the Moderate Resolution Imaging Spectroradiometer (MODIS), are used to calculate the ice thickness using a heat balance equation. Relationships between the MODIS ice thickness and polarization ratio from the Advanced Microwave Scanning Radiometer-Earth Observing System (AMSR-E) are used to calculate the thickness of thin ice (less than 0.2 m) from the AMSR-E data. This is done for each frequency on the AMSR-E sensor in the range of 6.9-36.5 GHz. Through comparison with data from ice charts, it is found that the errors are lowest for thickness values calculated from low-frequency AMSR-E data. The accuracies of the ice thickness from MODIS, AMSR-E, operational ice charts, and two moored upward looking sonars are further assessed using the triple collocation method. It is found that the error associated with ice thickness from AMSR-E is the lowest and the error associated with ice thickness from MODIS is the highest. While the MODIS data represent the small-scale variability of the sea-ice thickness better than the AMSR-E data, the MODIS data can produce spurious values of ice thickness due to unmasked clouds. To use ice thickness from MODIS in an automated algorithm, quality control would need to be applied to the MODIS data to remove unmasked clouds which lead to spurious values of thick ice. The errors calculated for the ice thickness from AMSR-E, which are calculated based on a relationship calibrated with MODIS ice thickness from a clear-sky day, indicate that these data would be useful for operational data assimilation. [ABSTRACT FROM AUTHOR]

Abstract: In this paper a method to assimilate data from a visual/infrared (VIS/IR) sensor for the purpose of estimating sea ice concentration is investigated. A novel forward model is presented which allows the reflectances and brightness temperatures from the VIS/IR sensor to be assimilated. The VIS/IR data are assimilated in combination with ice concentration retrievals from a passive microwave sensor using 3D variational data assimilation into a 5km×5km grid of sea ice concentrations. Results are compared with those from an experiment which assimilates only the ice concentration retrievals. It is found that the VIS/IR data have a positive impact as ice on the sea ice state through comparison with independent data. The proportion of points correctly identified asice or open water increased in each region studied by including the AVHRR data, with the largest increase being from 0.921 to 0.935. This positive impact is due to the higher resolution of the VIS/IR data as compared with the passive microwave data, that improves the representation of open water in bays and coastal polynyas. There is however, a degradation in the accuracy of the sea ice state in some regions of the analysis from assimilating the VIS/IR data. Possible ways of improving the impact from VIS/IR data are discussed. [Copyright &y& Elsevier]

In this paper a method to directly assimilate brightness temperatures from the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) to produce ice concentration analyses within a three-dimensional variational data assimilation system is investigated. To assimilate the brightness temperatures a simple radiative transfer model is used as the forward model that maps the state vector to the observation space. This allows brightness temperatures to be modeled for all channels as a function of the total ice concentration, surface wind speed, sea surface temperature, ice temperature, vertically integrated water vapor, and vertically integrated cloud liquid water. The brightness temperatures estimated by the radiative transfer model are sensitive to the specified values for the sea ice emissivity. In this paper, two methods of specifying the sea ice emissivity are compared. The first uses a constant value for each polarization and frequency, while the second uses a simple emissivity parameterization. The emissivity parameterization is found to significantly improve the fit to the observations, reducing both the bias and the standard deviation. Results from the assimilation of brightness temperatures are compared with those from assimilating a retrieved ice concentration in the context of initializing a coupled ice-ocean model for an area along the east coast of Canada. It is found that with the emissivity parameterization the assimilation of brightness temperatures produces ice concentration analyses that are in slightly better agreement with operational ice charts than when assimilating an ice concentration retrieval, with the most significant improvements during the melt season. [ABSTRACT FROM AUTHOR]

Sea ice observations through satellite imaging have led to advancements in environmental research, ship navigation, and ice hazard forecasting in cold regions. Machine learning and, recently, deep learning techniques are being explored by various researchers to process vast amounts of Synthetic Aperture Radar (SAR) data for detecting potential hazards in navigational routes. Detection of hazards such as sea ice floes in Marginal Ice Zones (MIZs) is quite challenging as the floes are often embedded in a multiscale ice cover composed of ice filaments and eddies in addition to floes. This study proposes a segmentation model tailored for detecting ice floes in SAR images. The model exploits the advantages of both convolutional neural networks and convolutional conditional random field (Conv-CRF) in a combined manner. The residual UNET (RES-UNET) computes expressive features to generate coarse segmentation maps while the Conv-CRF exploits the spatial co-occurrence pairwise potentials along with the RES-UNET unary/segmentation maps to generate final predictions. The whole pipeline is trained end-to-end using a dual loss function. This dual loss function is composed of a weighted average of binary cross entropy and soft dice loss. The comparison of experimental results with the conventional segmentation networks such as UNET, DeepLabV3, and FCN-8 demonstrates the effectiveness of the proposed architecture. [ABSTRACT FROM AUTHOR]

The Canadian Arctic Archipelago (CAA) presents unique challenges to the determination of melt onset (MO) using remote sensing data. High spatial resolution data is required to discern melt onset among the islands and narrow waterways of the region. Current passive microwave retrievals use daily averaged 19 GHz and 37 GHz data from the multi-channel microwave radiometer (SMMR) and/or the special sensor microwave/imager (SSM/I). The development of a new passive microwave melt onset method capable of using higher resolution data is desirable. The new passive microwave melt onset method described here, named the Dynamic Threshold Variability Method (DTVM), uses higher resolution data from the 37 GHz vertically-polarized channel from the advanced microwave scanning radiometers (AMSR-E and AMSR-2). The DTVM MO detection methodology differs from previously presented passive microwave Arctic MO methods in that it does not use a fixed threshold of a brightness temperature parameter. Instead, the DTVM determines MO dates based on the distribution of dates corresponding to the exceedance of a range of brightness temperature variability thresholds. The method also uses swath data instead of daily averaged brightness temperatures, which is found to lead to improved melt detection. Two current passive microwave MO methods are compared and evaluated for applicability in the CAA alongside the DTVM. The DTVM provides MO dates at a higher spatial resolution than earlier methods in addition to higher correlation with MO dates from surface air temperature (SAT) reanalyses. It is found that, for some years, MO dates in the CAA exhibit a latitudinal dependence, while in other years the MO dates in the CAA are relatively uniform across the domain. [ABSTRACT FROM AUTHOR]