Describir: Sequential Recurrent Encoders for Land Cover Mapping in the Brazilian Amazon using MODIS Imagery and Auxiliary Datasets