Retrieval of Melt Pond Fraction over Arctic Sea Ice during 2000–2019 Using an Ensemble-Based Deep Neural Network

• Published in: Remote Sensing
• Main Authors: Yifan Ding, Xiao Cheng, Jiping Liu, Fengming Hui, Zhenzhan Wang, Shengzhe Chen
• Format: Article in Journal/Newspaper
• Language: English
• Published: MDPI AG 2020
• Subjects: melt pond fraction; Arctic sea ice; deep neural network; Science; Q; Arctic; albedo; Sea ice
• Online Access: https://doi.org/10.3390/rs12172746; https://doaj.org/article/1849ea4c99aa469293a20ee80ffa60eb

The accurate knowledge of variations of melt ponds is important for understanding the Arctic energy budget due to its albedo–transmittance–melt feedback. In this study, we develop and validate a new method for retrieving melt pond fraction (MPF) over Arctic sea ice using all seven spectral bands of MODIS surface reflectance. We construct a robust ensemble-based deep neural network and use in-situ MPF observations collected from multiple sources as the target data to train the network. We examine the potential influence of using sea ice concentration (SIC) from different sources as additional target data (besides MPF) on the MPF retrieval. The results suggest that the inclusion of SIC has a minor impact on MPF retrieval. Based on this, we create a new MPF data from 2000 to 2019 (the longest data in our knowledge). The validation shows that our new MPF data is in good agreement with the observations. We further compare the new MPF dataset with the previously published MPF datasets. It is found that the evolution of the new MPF is similar to previous MPF data throughout the melting season, but the new MPF data is in relatively better agreement with the observations in terms of correlations and root mean squared errors (RMSE), and also has the smallest value in the first half of the melting season.