A fully data-driven method for predicting Antarctic sea ice concentrations using temporal mixture analysis and an autoregressive model. Issue 2 (1st February 2017)

• Record Type: Journal Article
• Authors: Chi, Junhwa; Kim, Hyun-Cheol
• Is Part Of: Remote sensing letters. Volume 8:Issue 2(2017)
• Page Start: 106
• Page End: 115
• Publication Date: 2017-02-01
• Subjects: Remote sensing -- Periodicals; Remote sensing; Periodicals; 621.3678
• DOI: 10.1080/2150704X.2016.1234726
• Languages: English
• ISSNs: 2150-704X

Abstract

ABSTRACT: While sea ice dynamics have been gaining increased attention in climate change and global warming studies, remote sensing (RS) sensors, capable of detecting and characterizing detailed information on targets of interest, have come to play a critical role in acquiring image data over extended and inaccessible areas. Passive microwave sensors have been the most effective and consistent tool for characterizing daily sea ice cover at global scale. However, it is typically challenging to study temporally successive data acquired at high time frequencies, referred to as hypertemporal data. To address this issue, among various RS analysis techniques, temporal mixture analysis (TMA) approaches are often investigated for characterizing seasonal characteristics of environmental factors including sea ice concentration (SIC) in this study. The goal of the present study is to predict daily Antarctic SICs for 1 year through a combination of TMA results and time series analysis (TSA) without incorporation of environmental factors. First, temporally most significant sea ice signals, referred to as temporal endmembers (EMs), were found using signal processing algorithms, and then corresponding fractional abundances (FAs) associated with each EM were calculated using least squares solution. Using these FAs, subsequently, a single autoregressive (AR) model that typically fits all Antarctic SIC data for the period 1979–2013 was applied to predict SIC values for 2014. Daily SIC data reconstructed using the proposed method were qualitatively and quantitatively compared to those of using real FAs derived from a spectral unmixing method. It was found that AR model trained by the proposed method successfully predicts new FAs for 2014 and the FAs should be used to reconstruct resulting 2014 daily SIC images.