To better understand the unprecedented rate, extent, and intensity of change that the land system is currently experiencing, land change science requires remote sensing classification products that are both highly accurate and spatial-temporally consistent. The current generation of land cover products does not meet these requirements, resulting in substantial bias and inaccuracies in the detection, monitoring, and quantification of land cover change. To address this problem, this thesis develops an improved classifier for multi-temporal land-cover mapping that combines advancements in machine learning with domain knowledge from remote sensing. Improvement of the technique makes it tractable to obtain highly accurate and spatial-temporally consistent land cover maps automatically and without post-processing. Application of the proposed classifier shows significant improvement upon the state-of-the-art in multi-temporal land-cover mapping. The accuracy of change detection, accuracy at individual dates, and temporal consistency among land-cover change trajectories are all improved over those of results produced by competitive techniques in literature. Altogether, this work substantially advances the capability of multi-temporal land-cover mapping through better use of existing geographic information and improved statistical decision making. Futures studies that employ this classifier can more accurately and consistently explain the places, periods, and types of land-cover change occurring on our Earth.