Among popular regression models in geoscience, deep learning gains increasing attention but depends on advanced computational resources. The deep stack and tremendous amount of parameters in deep learning models greatly increases the complexity and challenges of applying on geospatial datasets. This session will discuss the latest progresses on advanced cyberinfrastructure for automating and accelerating deep learning on large-scale remote-sensed or field-observed geospatial datasets including satellite imageries, aerial images, ground sensor network, model simulations, reanalysis, radar data, surveyed tables, etc. We will discuss how to bring community experiences together and collaborate on building, integrating, and using advanced geospatial cyberinfrastructure to address big data-related questions raised in solving fundamental geoscience problems, e.g., crop land cover monitoring, crop yield prediction, drought forecasting, air quality forecasting, earthquake warning, hurricane forecasting, etc.