We recently developed a method to improve the NASA GISS ModelE Earth System Model, a global model capable of making projections of the Earth system on climatic (decade to century) timescales. Our method uses best statistical practices to reduce and quantify uncertainties, and machine learning to speed the process. Here we present new work refining this methodology to make it more accurate, efficient, and useful. We also present applications of this work, including its proposed use to guide development of new NASA satellite platforms, as well as extending our projections to metrics that matter most to non-scientists, for example the prevalence of droughts, heat waves, and extreme rainfall.