Web based Geographic Information Systems (WebGIS) are graphics intensive applications that can overload networks causing delivery delays leading to user dissatisfaction and lost productivity. This work proposes a novel approach to address timely delivery of WebGIS content over the Internet by employing intelligent caching and pre-fetching of images based on a probabilistic model. We demonstrate that such a model can be efficiently created based upon data mined from the log files of WebGIS applications and that such models are effective at eliminating or reducing the waiting times for subsequent WebGIS requests.