Melt ponds on Arctic sea ice determined from MODIS satellite data using an artificial neural network

Melt ponds on sea ice strongly reduce the surface albedo and accelerate the decay of Arctic sea ice. Due to different spectral properties of snow, ice, and water, the fractional coverage of these distinct surface types can be derived from multispectral sensors like MODIS using a spectral unmixing algorithm. The unmixing was implemented using a multilayer perceptron (MLP) to reduce computational costs. Arctic-wide melt pond fractions and sea ice concentrations are derived from the level 3 MODIS surface reflectance product. The validation of the MODIS melt pond data set was conducted with aerial photos from the MELTEX campaign 2008 in the Beaufort Sea, data sets from the National Snow and Ice Data Center (NSIDC) for 2000 and 2001 from four sites spread over the entire Arctic, and with ship observations from the trans-Arctic HOTRAX cruise in 2005. The root-mean-square errors (RMSE) range from 3.8 % for the comparison with HOTRAX data, over 10.7 % for the comparison with NSIDC data, to 10.3 % and 11.4 % for the comparison with MELTEX data, with correlations coefficients ranging from R2 = 0.28 to R2 = 0.45. The mean annual cycle of the melt pond fraction for the entire Arctic shows a strong increase in June, reaching a maximum of 15 % by the end of June. The zonal mean of melt pond fractions indicates a dependence of the temporal development of melt ponds from the geographical latitude, and has its maximum in mid-July in latitudes between 80° and 88° N. Furthermore, the MODIS results are used to estimate the influence of melt ponds on retrievals of sea ice concentrations from passive microwave data. Results from a case study comparing sea ice concentrations from ASI-, NASA Team 2-, and Bootstrap-algorithms with MODIS sea ice concentrations indicate an underestimation of around 40 % for sea ice concentrations retrieved with microwave algorithms.