Abstract
Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from synthetic aperture radar (SAR) in an automated manner. This is often done using ice charts as training data. However, in these charts, an ice concentration label is given to a large region, which may not have a spatially uniform sea ice concentration distribution at the prediction scale of the CNN. This leads to representativity errors, which can be more pronounced at intermediate sea ice concentrations. In this study, we first investigate ways to perturb the ice chart labels to obtain improved predictions to account for the label uncertainty for intermediate ice concentrations. We then propose a method to augment the ice chart data by rescaling the information in the SAR imagery. The method is found to lead to improved accuracy in comparison to using the ice chart labels alone, with accuracy improving from 0.919 to 0.966. The sea ice concentration maps with the augmented labels also have much finer detail than the other approaches evaluated. These details are visually in agreement with expected sea ice concentration from the SAR data.
Highlights
Sea ice concentration is defined as the fraction of a given portion of the ocean surface that is covered by sea ice
The model trained with the binary cross-entropy loss function will be referred to as the binary cross entropy (BCE) model, the model trained with the mean absolute error loss function will be referred to as the mean-absolute error (MAE) model, the model trained with the mean squared error loss function will be referred to as the mean-squared error (MSE) model, and the model trained with the mean-split loss function will be referred to as the MS model
In this study we have developed a novel method to improve the representation of fine scale details in Convolutional neural networks (CNNs) predictions of sea ice concentration from synthetic aperture radar (SAR) imagery when ice charts are used to provide the training labels
Summary
Sea ice concentration is defined as the fraction of a given portion of the ocean surface that is covered by sea ice. It is considered an essential climate variable by the World Meteorological Organization (WMO) due to the role it plays in climate, and in moderating the heat and momentum transfer at the ocean-ice and ocean-atmosphere interfaces It is a key variable in operational ice monitoring, as it is an impediment for ship traffic at high latitudes. The main source of remote sensing data used for operational ice monitoring is synthetic aperture radar (SAR) imagery. These data are acquired at the low frequencies of the microwave spectrum and are insensitive to atmospheric moisture. Due to the complexity of the interaction of the SAR signal and the sea ice or ice/snow/ocean system, and the imaging geometry, there is not a straightforward mapping between the signal received by the sensor (backscatter) and the surface properties
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
