Abstract
Ice concentration estimates are typically acquired from algorithms using passive microwave satellite data, and from image analysis charts, but these have limitations. Estimates acquired from passive microwave data have coarse spatial resolution, may have errors due to atmospheric contamination, and often perform poorly in marginal ice zones. Image analysis charts are not as precise, subject to analyst interpretation, and only available over specific geographic areas. We have implemented a U-net with synthetic aperture radar images as inputs and use ice concentration estimates retrieved from passive microwave data as training labels. The U-net, due to not being sensitive to patch size, is shown to be an improvement over previous work with convolutional neural networks that use fully connected layers at the output. Data augmentation and an L1 loss function were applied along with a novel training scheme that leverages curriculum learning. In this training scheme, the model is first trained with samples from open water and consolidated ice regions before incorporating samples from marginal ice regions. In a tenfold cross validation experiment, we achieve 3–4% mean absolute error comparing to estimates using passive microwave data and observe curriculum learning models having more stable training. Predictions on four with-held SAR scenes with difficult ice conditions were evaluated with image analysis charts. A mean absolute error of 7.18% is achieved, which is lower than errors associated with passive microwave data alone. Qualitative improvements in marginal ice zone estimates are achieved, while still preserving smooth consolidated ice regions, and openings in ice cover.
Highlights
S EA ice concentration is calculated as a numeric value between zero and one, defined as the total area of ice in a specified region divided by the total area of that region
Comparison of model predictions. (a)–(d) convolutional neural networks (CNNs) trained with patch size of 45 × 45, (e)–(h) U-net trained with L2 loss function, (i)–(l) U-net trained with L1 loss function, (m)–(p) U-net trained with L1 loss function with an enhanced data set, (q)–(t) U-net trained with L1 loss function with an enhanced dataset using curriculum learning
This is appropriate because we selected images for basic train that are of good visual agreement with the Synthetic aperture radar (SAR) images, whereas images for Hudson Strait evaluation were selected as images we wanted to improve on
Summary
S EA ice concentration is calculated as a numeric value between zero and one, defined as the total area of ice in a specified region divided by the total area of that region. Synthetic aperture radar (SAR) sensors and passive microwave sensors are more often used to monitor sea ice in arctic regions than optical sensors because they are less affected by cloud cover and do not depend on solar illumination Data provided from these sensors can be used to obtain ice concentration estimates. The present study builds upon the previous work of using deep learning models for ice concentration estimation. Passive microwave data has been used as training labels for a DenseNet model, a CNN with direct connections between all layers [11], to estimate ice concentration on the Gulf of St. Lawrence and Arctic Archipelago regions [5]. We follow a different approach to make multidimensional predictions that utilizes a U-net
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
