Automatic High-Accuracy Sea Ice Monitoring in the Arctic Using MODIS Data

Abstract

High accuracy sea ice monitoring is of great significance in responding to global climate change and guiding safe navigation. The MODIS sensors carried by Aqua and Terra satellites have great potential in high-resolution monitoring of sea ice in polar regions. This study developed an automatic high-accuracy sea ice concentration (SIC) detection algorithm for the MODIS data. We used a machine learning algorithm that combines spectral and texture features to obtain high-precision sea ice recognition results. Then, according to the cloud characteristics in MODIS images and the full coverage of AMSR2 SICs, outlier removal and cloud area filling are carried out. Finally, the SIC result was calculated according to the distribution of ice types. We assessed the accuracy of our high-accuracy SIC product by comparing with the MOD29/MYD29 products and validating with the Landsat 8/9 images. We found that the deviation increased with the decrease of SIC value. The deviation are larger in the ice melting area and the ice edges area, but smaller in the area completely covered by ice. Compared with MOD29/MYD29 SICs, the SICs obtained in this study have higher accuracy, with an average RMSD of 7.64%. It can detect the details of relatively small leads, ice edges and fragmented ice areas. The high-accuracy SIC product we obtained is expected to provide long-term SIC records of high quality.