Classifying patterns of land cover

July 2010

Comparison  between two maps of land use/land cover (LULC) is a fundamental task in remote sensing and geospatial data analysis with application to change detection, validation of models, and accuracy assessment. In the paper presented at the 2010 IEEE International Geoscience and Remote Sensing Symposium Stepinski proposes to classify a collection of land use/land cover maps sharing common set of categories on the basis of their patterns using information-theoretic definition of similarity, based on the concept of mutual information.  A pattern is a specific composition of LULC categories and their spatial arrangement in a given landscape; it represents a higher level abstraction of landscape than a single LULC category. For example, a LULC map of a city serves as means for visual assessment of spatial relations between its constituent categories, but it also defines a pattern - characteristic fingerprint of this particular city in terms of LULC. A collection of different cities can be grouped into classes on the basis of similarities between their patterns. 

Discovering spatio-social motifs of electoral support

June 2010

Association  analysis provides a natural, data-centric framework for the discovery of patterns of explanatory variables that are linked to a certain outcome. In the  paper presented at the 1st International Conference on Computing for Geospatial Research & Application   Stepinski, Salazar, and Ding demonstrate how such a framework can be applied for political analysis, using an expository example of discovering different spatio-social motifs of support for Barack Obama in the 2008 presidential election. Election results and thirteen different socio-economic explanatory variables, tabulated at the county level, are used as an input for calculating a collection of discriminative patterns having disproportionately large support within the counties won by Obama. These patterns are synthesized into a small number of larger socio-economics motifs using a novel pattern similarity measure that outputs a concise summary readily interpretable in terms of political analysis. The method discovers two major Obama constituencies that different in their socio-economic makeup and in their geographical distributions. 

Geographical distribution of crater depths on Mars 

March 2010

A global map of crater depth/diameter (d/D) ratio is important for studies that use crater morphology for determining the rate and amount of surface degradation (including degradation due to the presence of subsurface ice), or to the studies of target strengths. In a paper presented in the 41st Lunar and Planetary Science Conference Stepinski reported on the geographical distribution of crater depths derived from an automatic estimate of the depths for 75,919 craters located over the entire surface of Mars. The paper discusses the implications of the results to location of the Martian cryosphere and identifies locations of the deepest craters on Mars.

More extensive valley networks on Mars

November 2009

In a paper published in the Journal of Geophysical Research — Planets Luo and Stepinski used an innovative computer program to produce a new and more detailed global map of the valley networks on Mars. The findings indicate the networks are more than twice as extensive (2.3 times longer in total length) than had been previously depicted in the only other planet-wide, manually drawn map of the valleys. Further, regions that are most densely dissected by the valley networks roughly form a belt around the planet between the equator and mid-southern latitudes, consistent with a past climate scenario that included precipitation and the presence of an ocean covering a large portion of Mars’ northern hemisphere.

Machine learning expedites planetary mapping

October 2009

The algorithm developed by Stepinski and Bagaria and published in the IEEE Geoscience and Remote Sensing Letters uses a novel combination of techniques to yield maps having informational esthetics similar to manually drawn maps. In the first stage the algorithm classifies small patches of terrain into functionally predefined terrain type labels using empirical knowledge. This mimics the action of an analyst who first looks into smallest components of landscape. The second stage of the algorithm uses machine learning (clustering) and image processing (segmentation) techniques to calculate regionally-gathered ensembles of terrain type labels and uses them to output terrain classes thus performing a division of landscape reminiscent of traditional geologic units.

About Me

Tomasz Stepinski is a staff scientist at the Lunar and Planetary Institute in Houston, Texas, USA. More ...

Things to come

Catalog of sub-km size craters on Mars will be constructed from high-resolution images and using machine learning-based algorithm. The method is designed to produce "million craters" global catalogs of sub-km craters on Mars. Such catalogs will be utilized for deriving high spatial resolution and high temporal precision stratigraphy on regional or even planetary scale.   

Geographical distribution of craters depths on the Moon  will be calculated using new high resolution global lunar topography based on the LOLA mesurements. 

Data mining and engineering models for assessing wind damage risk to houses in the Harris County is in development in collaboration with a group at the Rice University.  The models are validated using ground truth data collected in the aftermath of hurricane Ike. The ultimate goal is to empower individual citizens to make more informed evacuation decision choices.

We are working on a framework for discovery of associations between environmental factors and the spatial distribution of biodiversity across the contiguous United States. A pressing problem in biodiversity studies is to develop the optimal strategy for protecting the species given limited resources. In order to design such a strategy it is necessary to understand associations between environmental factors and the spatial distribution of biodiversity. In this context we will apply our method to discover existence of different environments which associate with the high levels of biodiversity.