The aim of this work is to develop an automated approach for sea ice mapping and ice concentration retrieval using visible/infrared measurements from geostationary platforms. This study is a part of the cryosphere algorithm development activities for the future GOES-R ABI sensor. SEVIRI/METEOSAT data was used as prototype. The developed algorithm is designed to be completely autonomous where only GOES-R images will be used as inputs to the algorithm. The algorithm final products will be daily ice charts and ice concentration maps. In order to achieve reliable and accurate final products, two key issues have been addressed. Firstly, it is necessary to have a strong cloud detection tool to make sure that only cloud-free pixels are presented to the ice mapping algorithm. A variable threshold approach has been proposed in this study to take into account the varying sun-satellite geometry. Secondly, reflectances of all the Caspian Sea pixels were simulated for all the possible sun-satellite geometries and for both water (ice free) and pure ice pixels at different frequencies. A neural network based approach was used to generate the simulated reflectances. Two exhaustive samples of water and ice pixels have been selected using MODIS images to train and validate the network. A tie-point approach has then been used to calculate ice fraction. The produced maps were validated using MODIS images and IMS product. The obtained results showed an acceptable agreement with IMS charts in both space and time. This implies that, in addition to their unique capabilities in weather-related applications, GOES-R images might also be beneficial in sea ice mapping and classification.