Because melt ponds have lower albedo than snow and ice, the ice albedo feedback process accelerates to increase the amount of solar absorption and sea ice melt. In this study, an automatic image analysis method is developed in order to detect open water, sea ice, and melt ponds using the forward looking camera images obtained from observations of the Arctic Ocean ice during summer and autumn. Comparing an automatic image analysis to visual observations of the images, we can detect melt ponds on the sea ice. First, this method performs filtering in the Fourier domain to smooth the brightness histogram of the image, and sea ice conditions are classified into three categories. The low pass filter (LPF) specifications for obtaining a high concordance rate were a cutoff frequency of 0.05 and a filter head of 51 using a Blackman window. Then, the concordance rate was 89.5% at one peak (for open water or sea ice only), 80.5% at two peaks (for open water and sea ice, or melt pond and sea ice), and 64.0% at three peaks (for open water, melt pond, and sea ice). Second, the surface conditions are classified into two types by using the brightness threshold at one peak, and making a relationship of the red and green histogram at two peaks. As a result, the concordance rate of two peaks achieved 88.2%. Finally, our image analysis method automatically enabled surface condition distinctions on cruise tracks in the Arctic Ocean.
View full abstract