Detection of Antarctic Surface Meltwater Using Sentinel-2 Remote Sensing Images via U-Net With Attention Blocks: A Case Study Over the Amery Ice Shelf
Lihang Niu,Xueyuan Tang,Shuhu Yang,Yun Zhang,Lei Zheng,Lijuan Wang
DOI: https://doi.org/10.1109/tgrs.2023.3275076
IF: 8.2
2023-01-01
IEEE Transactions on Geoscience and Remote Sensing
Abstract:Surface meltwater critically impacts the Antarctic mass balance and global sea level rise. Quantifying the extent of surface meltwater in Antarctica on a large scale is a challenging task. Traditional methods, such as thresholding, have many limitations. We used a deep learning method, the U-Net with attention blocks, to automatically extract surface meltwater from Sentinel-2 images. We inserted attention mechanism blocks into U-Net to assign different weights to all pixels and channels to utilize the high resolution and multiple channels of Sentinel-2 images. The model was used to map surface water bodies in Sentinel-2 images, and the average accuracy reached 0.9969 on the test dataset. In East Antarctica, the Amery ice shelf (AIS) exhibits the largest surface meltwater area. Studying surface meltwater dynamics on the AIS is useful for understanding the East Antarctic mass balance and demonstrating the model performance. We analyzed the classification results for surface water bodies on the AIS from January 2017 to 2022. Spatially, 96% of surface water bodies are concentrated inland of the AIS from 70° S to 73° S and account for 93% of the region 20 km from the coastline of the AIS. Temporally, the water body area varies considerably in different years, with a maximum in 2017 (932.54 km $^{2}$ ) and a minimum in 2021 (58.34 km $^{2}$ ). The spatial distribution of surface water bodies on the AIS is controlled by the firn air content (FAC), katabatic winds, bare rocks, and blue ice. The interannual variability is associated with complex climate factors, including temperature, surface net solar radiation, snowfall, and snowmelt, among which temperature and snowfall show a strong correlation.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geochemistry & geophysics