Summer sea ice concentration in the Chukchi Sea derived from AMSR2 and NWP data with machine learning approach

Abstract

Arctic sea ice concentration (SIC) is a primary information for the prediction of climate change and the development of sea route in polar oceans. Passive microwave (PM) sensors have provided SIC of the Arctic Ocean since the 1970s. The SIC retrieval algorithms for PM observations could produce inaccurate SIC in summer due to ice surface melting and/or atmospheric effects. In this study, we developed summer SIC estimation models for Advanced Microwave Scanning Radiometer-2 (AMSR2) observations in the Chukchi Sea by using numerical weather prediction (NWP) data from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim reanalysis and rule-based machine learning approaches―Decision Tree (DT) and Random Forest (RF). We computed 42,480 values (samples) of SIC from KOrea Multi-Purpose SATellite-5 (KOMPSAT-5) synthetic aperture radar (SAR) images acquired in the Chukchi Sea during summer (JulySeptember) from 2015 to 2017. Eighty percent of the KOMPSAT-5 SIC values were used as training dataset for the development of SIC estimation models and the remaining values were used as test dataset. The brightness temperatures measured at each channel of AMSR2 and their combinations, and the atmospheric parameters (atmospheric water vapor, wind speed, sea level pressure, 2-m temperature, and 925 hPa-temperature) predicted by the ERA-Interim reanalysis were used as input variables for the SIC estimation models. The RF model produced more accurate SICs than the DT model. The SICs estimated from the RF model showed the value of root mean square error (RMSE) less than 9% compared to the KOMPSAT-5 SAR SICs.