Gridded datasets are quite common in scientific computing as many disciplines produce large amounts of samples relying on regular spatial grid-structures that identify locations where measurements are taken. Very large gridded datasets with a lot of variables and measurements pose challenging problems on how to store, query, summarize, visualize and mine such datasets. One interesting and important task when analyzing such datasets is to find interesting contiguous regions, called interestingness hotspots, based on the domain expert's notion of interestingness which is captured in an interestingness function. There has been significant work in using spatio-temporal clustering for mining spatio-temporal datasets in literature. However, in this paper, we present a computational framework which uses an alternative nonclustering approach to obtain interestingness hotspots. We present a novel hotspot growing algorithm which grows interestingness hotspots from seed hotspots, and then post-processes the obtained hotspots to remove or merge overlapping hotspots. We claim that our approach is capable of identifying a much broader class of hotspots, which cannot be identified by traditional distance-based clustering algorithms. Our framework is evaluated in a case study for a very large 4-dimensional gridded air pollution dataset in which we find high correlation hotspots and low variance hotspots with respect to pollutants; moreover, we analyze the scalability of the proposed hotspot growing algorithm.