Preprints
https://doi.org/10.5194/tc-2021-102
https://doi.org/10.5194/tc-2021-102

  15 Apr 2021

15 Apr 2021

Review status: this preprint is currently under review for the journal TC.

Improving Surface Melt Estimation over Antarctica Using Deep Learning: A Proof-of-Concept over the Larsen Ice Shelf

Zhongyang Hu1, Peter Kuipers Munneke1, Stef Lhermitte2, Maaike Izeboud2, and Michiel van den Broeke1 Zhongyang Hu et al.
  • 1Institute for Marine and Atmospheric research Utrecht, Utrecht University, Utrecht, The Netherlands
  • 2Department of Geoscience & Remote Sensing, Delft University of Technology, Delft, The Netherlands

Abstract. Accurately estimating surface melt volume of the Antarctic Ice Sheet is challenging, and has hitherto relied on climate modelling, or on observations from satellite remote sensing. Each of these methods has its limitations, especially in regions with high surface melt. This study aims to demonstrate the potential of improving surface melt simulations by deploying a deep learning model. A deep-learning-based framework has been developed to correct surface melt from the regional atmospheric climate model version 2.3p2 (RACMO2), using meteorological observations from automatic weather stations (AWSs), and surface albedo from satellite imagery. The framework includes three steps: (1) training a deep multilayer perceptron (MLP) model using AWS observations; (2) correcting moderate resolution imaging spectroradiometer (MODIS) albedo observations, and (3) using these two to correct the RACMO2 surface melt simulations. Using observations from three AWSs at the Larsen B and C Ice Shelves, Antarctica, cross-validation shows a high accuracy (root mean square error = 0.95 mm w.e. per day, mean absolute error = 0.42 mm w.e. per day, and coefficient of determination = 0.95). Moreover, the deep MLP model outperforms conventional machine learning models (e.g., random forest regression, XGBoost) and a shallow MLP model. When applying the trained deep MLP model over the entire Larsen Ice Shelf, the resulting, corrected RACMO2 surface melt shows a better correlation with the AWS observations for two out of three AWSs. However, for one location (AWS 18) the deep MLP model does not show improved agreement with AWS observations, likely due to the heterogeneous drivers of melt within the corresponding coarse resolution model pixels. Our study demonstrates the opportunity to improve surface melt simulations using deep learning combined with satellite albedo observations. On the other hand, more work is required to refine the method, especially for complicated and heterogeneous terrains.

Zhongyang Hu et al.

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on tc-2021-102', Anonymous Referee #1, 30 Jun 2021
  • RC2: 'Comment on tc-2021-102', Anonymous Referee #2, 20 Sep 2021

Zhongyang Hu et al.

Zhongyang Hu et al.

Viewed

Total article views: 687 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
506 171 10 687 6 7
  • HTML: 506
  • PDF: 171
  • XML: 10
  • Total: 687
  • BibTeX: 6
  • EndNote: 7
Views and downloads (calculated since 15 Apr 2021)
Cumulative views and downloads (calculated since 15 Apr 2021)

Viewed (geographical distribution)

Total article views: 690 (including HTML, PDF, and XML) Thereof 690 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 21 Sep 2021
Download
Short summary
Antarctica is shrinking, and part of the mass loss is caused by higher temperatures leading to more snowmelt. We use computer models to estimate the amount of melt, but this can be inaccurate exactly in areas with most melt. It's because the model cannot account for small, darker areas like rocks or darker ice. Therefore, we trained a computer using artificial intelligence. Satellite images that show these darker areas were provided to the computer, and it computed an improved estimate of melt.