Finding Scientifically Interesting Places on Mars Using Region Discovery Algorithms

Mars is at the center of the solar system exploration efforts. Measurements collected by spacecrafts orbiting around the planet has been processed and organized into raster or pointbased datasets describing different aspects of Martian surface. Impressive knowledge about Mars was obtained by studying individual datasets, but additional information can be gained by data mining the fusion of disparate datasets. In this paper, we present a computational framework for region discovery geared toward finding scientifically interesting sites on Mars using multiple datasets. A novel family of density-based, representative-based, grid-based, and agglomerative clustering algorithms is proposed to find such sites. These algorithms search for contiguous regions maximizing a domain-expert-defined measure of interestingness. The proposed framework is used and evaluated in a case study in which we try to find sites in which extreme values of rampart impact craters co-locate with subsurface ice. Studying geological context of such sites provides a domain expert with an insight on a history of water on Mars. The case study features a mixture of raster and discrete datasets. Two approaches are explored to cope with this discrepancy. In the first approach, the raster is discretized and then regional co-location mining is applied. In the second approach, the point-based dataset is smoothed and transformed into a density function and then collocation regions are found from hotspots of the product of z-scores of the two densities. The paper also evaluates the potential of proposed framework for addressing scientific enquires in planetary sciences and other branches of geoscience.