Geospatial data are usually featured with three-dimensions in space and fast evolvement with time. The delivery of a large volume of geospatial data requires a well built network environment. This fact also challenges the online visualization. In order to accelerate data transmission and rendering, progressive methods have been applied in this paper. A k-d tree structure is incorporated to construct the multi-resolution data model thus facilitating progressive transmission process. We implement the framework in a flexible disturbed system. Based on the experiment on dust simulation output, this manner is demonstrated to be an effective way to delivery geospatial data.