MODIS data (MODIS Surface Reflectance 8-Day L3 Global 500m SIN Grid V005) of frequency bands 1, 3 und 4 are use in an artificial neural network to obtain the melt pond cover fraction on Arctic sea ice.
The method uses the fact that for surface types melt ponds, sea ice, snow, and open water different reflectance values are measured in the above-mentioned MODIS frequency bands. An artificial neural network has been developed. The approach of Tschudi et al. (2008) (see references) has been used to obtain a training data set of typical reflectances for selected regions and typical steps of melt pond cover development. This data set was subsequently used to train the neural network. After evaluation of the training results it has been applied to MODIS reflectances of bands 1, 3, 4 projected into a 500 m grid-cell size polar-stereographic grid to classify above-mentioned surface types.
The surface type distribution obtained is analysed and converted into a 12.5 km x 12.5 km grid-cell size product. The melt pond fraction is weighted with the sea ice concentration to obtain the relative melt pond fraction on sea ice. See section references for more information.
The data set offered comprises, on the one hand, the full set of melt pond fraction, its standard deviation, the open water fraction and the number of usable 500 m grid cells per 12.5 km grid cell. Those 12.5 km grid cells with less than 10% usable 500 m grid cells or more than 85% open water fraction are flagged. On the other hand, we offer in addition melt pond fraction, its standard deviation and the open water fraction for almost clear-sky areas, i.e. 12.5 km grid cells with more than 90% usable 500 m grid cells; areas with more than 85% open water fraction are again flagged.
This is version v02 of this data set. It differs from version v01 by a bias correction of the melt pond cover fraction and the open water fraction which were biased by 8% and 3%, respectively in the old version.
Last data update: August 27, 2015.