Interrelationship of sea ice-salinity-chlorophyll a changes using multi-satellite based on machine learning analysis in Ross Sea, Antarctica

As insolation increases during the Antarctic summer, sea ice melts, and the iron in the sea ice is released into the ocean Iron leads to phytoplankton blooms at sea ice margins and polynya in Antarctica, a high nutrient and low chlorophyll region. However, it is difficult to investigate the direct effect of sea ice on chlorophyll a changes because the observed regions are physically difficult to access. Therefore, this study intends to investigate the sea ice effect on chlorophyll a variation by using the decreasing salinity when sea ice melts in the Ross Sea, Antarctica. Using Random Forest, an ensemble bagging tree method of machine learning, the relationship was analyzed by 11 variables, including physical variables (sea surface temperature, photosynthetically available radiation, atmospheric temperature, wind, and salinity) as input data and chlorophyll a data observed from satellites as output data. As a result, the square of the correlation coefficient (R²) of the test data set was 0.97, and the root mean square error (RMSE) was 0.41 mg m^-3, showing high accuracy. In addition, the importance of salinity was identified by calculating the variable importance in the model. These results provide the importance of salinity in predicted future chlorophyll a changes in the Antarctic Ocean due to climate change.