Workshop Program
Register now
Workshop Program

Suggestions for improving automatic ice charting with synthetic aperture data
by Dr. Anton Korosov

Abstract

Several Sentinel-1 image-based sea ice classification algorithms using a machine-learning-based model are proposed to support daily ice charting [Zakhvatkina et al., 2019; Park et al., 2020; Bouze et al., 2021]. The readily available ice charts from the operational ice services can reduce the amount of manual work in preparation of training data. A public ice chart is prepared through manual inspection of various sources of satellite imagery and other sources of data (Partington et al., 2003; Johannessen et al., 2006) and may contain operator-to-operator bias, such as inconsistent decisions against similar ice conditions. Sometimes the same ice floe is classified differently in two different ice charts, although it looks almost the same in the corresponding SAR backscattering images. Furthermore, the boundaries between different ice types in the ice chart are normally not as precise as those in the SAR image-based classification results. Therefore, the lower classification accuracies compared to those in previous studies (80% in Zakhvatkina et al., 2013; 91.7% in Liu et al., 2015; 87.2% in Aldenhoff et al., 2018), which used manually classified ice maps as training and validation reference, are expected. In addition, an overall bias may exist since the public ice charts are produced in the interest of marine safety (Karvonen et al., 2015). It appears feasible that usage of independent objective data from other types of satellite sensors, such as, e.g., CryoSat-2 or IceSat-2 altimeters, may help circumvent the problem of subjective and inconsistent judgments by ice experts. May proposed algorithms for ice type or ice concentration retrieval include calculation of grey level co-occurrence matrix and Haralick texture features in a sliding window [e.g. by Park et al., 2020; or review in Zakjvatkina et al., 2019]. Then a classifier (e.g. a support vector machine or a random forest classifier) is trained with the texture features on input and labels from the rasterized ice charts on output. Such multi-step algorithms have many hyper-parameters that are rarely tuned and are complex in nature. A new algorithm for ice-type classification using a convolutional neural network (CNN) was propsed recently [Boulze et al., 2021]. The CNN is also trained on reference ice charts produced by human experts. The hyper-parameters of the CNN (i.e., architecture, number of layers and neurons, etc) are thoroughly tested with automated procedures. The new algorithm is simpler to implement and tune and outperforms the existing random forest product for each ice type.

Poster

Open in a new tab


Topic : Theme 1: Cryosphere monitoring.
Reference : T1-E13

Back to the list of submissions
Previous submission · Next submission


Comments

Comments are only accessible to participants. If you are a participant, please log in to see them.


Hosts
Partners