Sea Ice Monitoring with CFOSAT Scatterometer Measurements Using Random Forest Classifier

Xiaochun Zhai,Fangli Dou,Zhixiong Wang,Na Xu,Rui Xu,Zhaojun Zheng,Xingying Zhang

doi:10.3390/rs13224686

Abstract

The Ku-band scatterometer called CSCAT onboard the Chinese–French Oceanography Satellite (CFOSAT) is the first spaceborne rotating fan-beam scatterometer (RFSCAT). This paper performs sea ice monitoring with the CSCAT backscatter measurements in polar areas. The CSCAT measurements have the characteristics of diverse incidence and azimuth angles and separation between open water and sea ice. Hence, five microwave feature parameters, which show different sensitivity to ice or water, are defined and derived from the CSCAT measurements firstly. Then the random forest classifier is selected for sea ice monitoring because of its high overall accuracy of 99.66% and 93.31% in the Arctic and Antarctic, respectively. The difference of features ranked by importance in different seasons and regions shows that the combination of these parameters is effective in discriminating sea ice from water under various conditions. The performance of the algorithm is validated against the sea ice edge data from the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF) on a global scale in a period from 1 January 2019 to 10 May 2021. The mean sea ice area differences between CSCAT and OSI SAF product in the Arctic and Antarctic are 0.2673 million km2 and −0.4446 million km2, respectively, and the sea ice area relative errors of CSCAT are less than 10% except for summer season in both poles. However, the overall sea ice area derived from CSCAT is lower than the OSI SAF sea ice area in summer. This may be because the CSCAT is trained by radiometer sea ice concentration data while the radiometer measurement of sea ice is significantly affected by melting in the summer season. In conclusion, this research verifies the capability of CSCAT in monitoring polar sea ice using a machine learning-aided random forest classifier. This presented work can give guidance to sea ice monitoring with radar backscatter measurements from other spaceborne scatterometers, particular for the recently launched FY-3E scatterometer (called WindRad).

Highlights

Licensee MDPI, Basel, Switzerland.Polar sea ice, as an important input to the global climate model and a sensitive indicator of climate change, has always been paid attention by climate researchers [1,2].Sea ice forms a new interface between the upper ocean and the lower atmosphere, which can change the radiation and energy balance of the ocean surface and isolate the heat exchange and water vapor exchange between the ocean and the atmosphere by blocking the wind field’s momentum input to the ocean [3]
The accuracies of sea ice and water are 0.94, 0.98, the random forest classifier is selected for sea ice monitoring in this study
The difference in feature important rank in the different seasons implies that the combination of these parameters is effective in discriminating sea ice from water under various conditions

Summary

Introduction

As an important input to the global climate model and a sensitive indicator of climate change, has always been paid attention by climate researchers [1,2]. Sea ice forms a new interface between the upper ocean and the lower atmosphere, which can change the radiation and energy balance of the ocean surface and isolate the heat exchange and water vapor exchange between the ocean and the atmosphere by blocking the wind field’s momentum input to the ocean [3]. Sea ice modulates the atmosphere–ocean interaction by influencing the exchange of matter and energy, ocean surface albedo, ocean temperature, and salinity circulation, and has a significant impact on global climate [6]

Objectives

Methods

Results

Conclusion