Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Hierarchical Information-sharing Convolutional Neural Network for the Prediction of Arctic Sea Ice Concentration and Velocity

Younghyun Koo, Maryam Rahnemoonfar

TL;DR

This study proposes a novel multitask fully conventional network architecture named hierarchical information-sharing U-net (HIS-Unet) to predict daily SIC and SIV, which outperforms other statistical approaches, sea ice physical models, and neural networks without such information-sharing units.

Abstract

Forecasting sea ice concentration (SIC) and sea ice velocity (SIV) in the Arctic Ocean is of great significance as the Arctic environment has been changed by the recent warming climate. Given that physical sea ice models require high computational costs with complex parameterization, deep learning techniques can effectively replace the physical model and improve the performance of sea ice prediction. This study proposes a novel multi-task fully conventional network architecture named hierarchical information-sharing U-net (HIS-Unet) to predict daily SIC and SIV. Instead of learning SIC and SIV separately at each branch, we allow the SIC and SIV layers to share their information and assist each other's prediction through the weighting attention modules (WAMs). Consequently, our HIS-Unet outperforms other statistical approaches, sea ice physical models, and neural networks without such information-sharing units. The improvement of HIS-Unet is more significant to when and where SIC changes seasonally, which implies that the information sharing between SIC and SIV through WAMs helps learn the dynamic changes of SIC and SIV. The weight values of the WAMs imply that SIC information plays a more critical role in SIV prediction, compared to that of SIV information in SIC prediction, and information sharing is more active in marginal ice zones (e.g., East Greenland and Hudson/Baffin Bays) than in the central Arctic.

Hierarchical Information-sharing Convolutional Neural Network for the Prediction of Arctic Sea Ice Concentration and Velocity

TL;DR

This study proposes a novel multitask fully conventional network architecture named hierarchical information-sharing U-net (HIS-Unet) to predict daily SIC and SIV, which outperforms other statistical approaches, sea ice physical models, and neural networks without such information-sharing units.

Abstract

Forecasting sea ice concentration (SIC) and sea ice velocity (SIV) in the Arctic Ocean is of great significance as the Arctic environment has been changed by the recent warming climate. Given that physical sea ice models require high computational costs with complex parameterization, deep learning techniques can effectively replace the physical model and improve the performance of sea ice prediction. This study proposes a novel multi-task fully conventional network architecture named hierarchical information-sharing U-net (HIS-Unet) to predict daily SIC and SIV. Instead of learning SIC and SIV separately at each branch, we allow the SIC and SIV layers to share their information and assist each other's prediction through the weighting attention modules (WAMs). Consequently, our HIS-Unet outperforms other statistical approaches, sea ice physical models, and neural networks without such information-sharing units. The improvement of HIS-Unet is more significant to when and where SIC changes seasonally, which implies that the information sharing between SIC and SIV through WAMs helps learn the dynamic changes of SIC and SIV. The weight values of the WAMs imply that SIC information plays a more critical role in SIV prediction, compared to that of SIV information in SIC prediction, and information sharing is more active in marginal ice zones (e.g., East Greenland and Hudson/Baffin Bays) than in the central Arctic.
Paper Structure (21 sections, 13 equations, 8 figures, 3 tables)

This paper contains 21 sections, 13 equations, 8 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Dynamic sea ice models
  4. Deep learning for sea ice prediction
  5. Multi-task neural network
  6. Data
  7. Sea ice velocity
  8. Sea ice concentration
  9. ERA5 climate reanalysis
  10. Sea ice physics model
  11. Method
  12. Hierarchical information-sharing U-net
  13. Baseline models
  14. Training and testing models
  15. Evaluation metrics
  16. ...and 6 more sections

Figures (8)

  • Figure 1: Architecture of multi-task CNN models tested in this study: (a) Hierarchical information-sharing U-net (HIS-Unet), (b) early-branched U-net (EB-Unet), and (c) late-branched U-net (LB-Unet).
  • Figure 2: Schematic diagram of the (a) Weighting attention module (WAM), (b) Channel attention module, and (c) Spatial attention module.
  • Figure 3: Comparison of monthly accuracy of the models in 2022: (a) R of SIC, (b) R of SIV, (c) RMSE of SIC, and (d) RMSE of SIV
  • Figure 4: The difference of R and RMSE between HIS-Unet and EB-Unet: (a) R of SIC, (b) RMSE of SIC, (c) R of SIV, and (d) RMSE of SIV. All pixel values are the average of all days of 2022 when the SIC and SIV values of the pixels are valid.
  • Figure 5: Six Arctic subregions defined in this study: Central Arctic (CA), Chukchi/Beaufort Seas (CBS), Laptev/East Siberian Seas (LESS), Kara/Barents Seas (KBS), East Greenland (EG), Hudson/Baffin Bays (HBB)
  • ...and 3 more figures