Spatial Pattern To Learn (SP2Learn).

Peter Bajcsy

Land-informatics workshop II, September 28-29, Urbana, 2007

This tutorial presents a framework for accurate estimation of geospatial models from sparse field measurements using image processing and machine learning. The motivation for our work is driven by the cost of field measurements and by the limitations of currently available physics-based modeling techniques. The goal is to improve our understanding of the underlying physical phenomena and increase the accuracy of geospatial models.

Our approach is to interpolate sparse field measurements, apply existing physics-based models, incorporate spatial constraints using image processing techniques, explore utilizing auxiliary raster measurements using machine learning, and perform optimization of all algorithmic parameters in supervised, as well as, in unsupervised manner.

The tutorial will illustrate the application of the framework to groundwater recharge and discharge (R/D) rate models. We use the physics-based R/D rate model that takes field measurements of hydrologic conductivity, water table level and bedrock elevation. We will explore the accuracy improvements when several image de-noising techniques with a decision tree machine learning technique are employed, and several remote sensing and terrestrial raster measurements are used, for example, slope, soil type and proximity to water bodies. The participants will be exposed to the analyses of spatially sparse geospatial measurements and utilization of image processing and data-driven modeling for these analyses.