Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts

Manveer Tamber, K. Andrea Scott, Leif Toudal Pedersen

    Research output: Contribution to journalJournal articleResearchpeer-review

    106 Downloads (Pure)

    Abstract

    Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from SAR in an automated manner. Often this is 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 0.921 to 0.979. 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.
    Original languageEnglish
    JournalIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
    Volume15
    Pages (from-to)1502-1513
    Number of pages12
    ISSN1939-1404
    DOIs
    Publication statusPublished - 2022

    Keywords

    • Convolutional neural network (CNN)
    • Ice Concentration
    • Synthetic Aperture Radar (SAR) data

    Fingerprint

    Dive into the research topics of 'Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts'. Together they form a unique fingerprint.

    Cite this