Search

Algorithms designed for ice–water classification of synthetic aperture radar (SAR) sea ice imagery produce only binary (ice and water) output typically using manually labeled samples for assessment. This is limiting because only a small subset of labeled samples are used, which, given the nonstationary nature of the ice and water classes, will likely not reflect the full scene. To address this, we implement a binary ice–water classification in a more informative manner considering the uncertainty associated with each pixel in the scene. To accomplish this, we have implemented a Bayesian convolutional neural network (CNN) with variational inference to produce both aleatoric (data-based) and epistemic (model-based) uncertainty. This valuable information provides feedback as to regions that have pixels more likely to be misclassified and provides improved scene interpretation. Testing was performed on a set of 21 RADARSAT-2 dual-polarization SAR scenes covering a region in the Beaufort Sea captured regularly from April to December. The model is validated by demonstrating: 1) a positive correlation between misclassification rate and model uncertainty and 2) a higher uncertainty during the melt and freeze-up transition periods, which are more challenging to classify. By incorporating the iterative region growing with semantics (IRGS) segmentation algorithm and an uncertainty value-based thresholding algorithm, the Bayesian CNN classification outputs are improved significantly via both numerical analysis and visual inspection.

In recent years, the adoption of deep learning (DL) techniques for predicting sea ice concentration (SIC) given both passive microwave (PM) data and reanalysis data has seen a growing interest. For use in downstream services, these SIC estimates should be accompanied by uncertainty estimates. To provide these estimates, we utilize a heteroscedastic Bayesian neural network (HBNN), which can estimate both model (epistemic) and data (aleatoric) uncertainty. We use both PM and atmospheric data as our input features and demonstrate that both are needed for accurate SIC estimates. Results show that, over an annual cycle, the months of melt onset, such as April, May, and June, produce the highest uncertainties relative to other months, with total (epistemic + aleatoric) uncertainties of approximately 20%, while areas in the marginal ice zone contributed highest total uncertainty of 25% spatially. When considering an average over the test year, the level of uncertainty due to the data (aleatoric) is consistent with other studies, at 10%–15%. The advantage of our approach is that the uncertainties are specific to the data instance, and both model and data uncertainties are estimated.

AbstractThe Cumberland Sound Beluga is a threatened population of belugas and the assessment of the population is done by a manual review of aerial surveys. The time-consuming and labor-intensive nature of this job motivates the need for a computer automated process to monitor beluga populations. In this paper, we investigate convolutional neural networks to detect whether a section of an aerial survey image contains a beluga. We use data from the 2014 and 2017 aerial surveys of the Cumberland Sound, conducted by the Fisheries and Oceans Canada to simulate two scenarios: (1) when one annotates part of a survey and uses it to train a pipeline to annotate the remainder and (2) when one uses annotations from a survey to train a pipeline to annotate another survey from another time period. We experimented with a number of different architectures and found that an ensemble of 10 CNN models that leverage Squeeze-Excitation and Residual blocks performed best. We evaluated scenarios (1) and (2) by training on the 2014 and 2017 surveys, respectively. In both scenarios, the performance on (1) is higher than (2) due to the uncontrolled variables in the scenes, such as weather and surface conditions.

AbstractIn this study, sea ice edge derived from three passive microwave (PM) algorithms, ARTIST sea ice (ASI), enhanced NASA Team 2 (NT2), and Bootstrap (BT), are compared to those derived from the daily Canadian Ice Service charts over a primarily seasonal ice zone in the eastern Canadian Arctic for 2013–2021. To determine the ice edge error, we introduced an edge-length-based displacement measure called the edge displacement error (EDE), a dimensionless measurement obtained by dividing the weighted average Hausdorff distance by the ice edge length. We found that the ASI algorithm has the highest EDE on average, while the BT algorithm has the lowest one. In October (the beginning of the freeze-up period), the ice edge exhibits significant meandering, and the EDE is less sensitive to changes in the charted area. In the freeze-up period, the PM algorithms have the highest mean EDE value relative to other months due to the appearance of thin ice. A greater range of EDE values was observed in April than in other months. Throughout this region, the wind speed varies the most in April and May, whereas in April, the air temperature fluctuates more than in the other months.

Abstract. A method to automatically combine binary ice/water information from synthetic aperture radar (SAR) sea ice images with the Advanced Microwave Scanning Radiometer-EOS (AMSR-E) daily ice concentration product is proposed for the purpose of generating sea ice concentration estimates with improved detail and accuracy. First, each pixel in the SAR image is labeled as ice or water using the MAp-Guided Ice Classification (MAGIC) SAR image classification system. Second, the labeled pixels are modeled as a Bernoulli process and combined with the AMSR-E ice concentration data in a Bayesian framework to generate improved ice concentration estimation. Visually interpreted ice/water extent and sea ice image analyses from the Canadian Ice Service (CIS) are used as comparison data. The combination of SAR ice/water labeled pixels with the AMSR-E ice concentration is shown to improve the ice concentration estimates, especially at the ice edge where substantial improvements are observed. Although the present study uses ice/water information from SAR, the method is general and could be used with other sources of ice/water remote sensed data.Résumé. Une méthode pour combiner automatiquement les informations binaires de glace/eau des images de radar à ouverture synthétique «synthetic aperture radar» (SAR) de la glace de mer avec le produit de la concentration quotidienne des glaces de «Advanced Microwave Scanning Radiometer-EOS» (AMSR-E), est proposée dans le but de produire des estimations de concentration de glace de mer en améliorant les détails et la précision. Tout d’abord, chaque pixel de l’image SAR est étiqueté comme de la glace ou de l’eau en utilisant le système de classification de l’image SAR «MAp-Guided Ice Classification» (MAGIC). Deuxièmement, les pixels étiquetés sont modélisés comme un processus de Bernoulli et combinés avec les données de concentration de glace AMSR-E dans un cadre Bayésien pour générer une meilleure estimation de la concentration de glace. L’étendue de la glace/eau interprétée visuellement et les analyses de l’image de la glace de mer du Service canadien des glaces «Canadian Ice Service» (CIS) sont utilisées comme données de comparaison. La combinaison des pixels SAR étiquetés glace/eau avec la concentration de glace AMSR-E améliore les estimations de concentration de glace, en particulier à la lisière de glace où des améliorations substantielles sont observées. Bien que la présente étude utilise les informations de glace/eau de SAR, la méthode est générale et peut être utilisée avec d’autres sources de données de télédétection de glace/eau.