The U.S. Geological Survey’s Land Change Monitoring, Assessment, and Projection (LCMAP) initiative characterizes changes in land cover, use, and condition with the goal of producing land change information that improves understanding of the earth system and provides insight into the impacts of land change on society. For LCMAP, all available high-quality data from the Landsat archive is used in a time series approach to detect land surface change. Annual thematic land cover maps are produced by classifying time series models. In this paper, we describe optimization of the classification method used to derive the thematic land cover product. We investigated the influences of auxiliary data, sample size, and training from different sources such as the U.S. Geological Survey’s Land Cover Trends project and National Land Cover Database (NLCD 2001 and NLCD 2011). Results were evaluated and validated based on independent data from the training dataset. We found that refining auxiliary data effectively reduced artifacts in the thematic land cover map that are related to data availability (i.e., SLC-off). The classification accuracy and stability were improved considerably by using a total of 20 million training pixels with a minimum of 600,000 and a maximum of 8 million training pixels per class. Finally, the NLCD 2001 training data delivered the best classification accuracy. Comparing to the original LCMAP classification strategy (Trends training data, 20,000 samples), the optimized classification strategy considerably improved the annual land cover map accuracy.