Uncertainty modeling for spatial data fusion and mining

Fusion and mining of uncertain heterogeneous spatial data in the cyber-physical space are challenging problems especially to deal in a coordinated way with both topological and geometrical uncertainties. This paper explores opportunities to meet these challenges by generalizing the Dynamic Logic of Phenomena (DLP) and the Neural Modeling Field (NMF) Theory for geo-spatial data. The main idea behind success of NMF and DLP in applications is matching the levels of uncertainty of the problem/model and the levels of uncertainty of the evaluation criterion used to identify the model with data. When a model becomes more certain then the evaluation criterion is also adjusted dynamically to match the adjusted model. This process mimics processes of the mind and natural evolution. This paper also outlines a generalization of k-nearest neighbors' algorithm in the DLP framework for geo-spatial structural data mining and fusion. The generalization is demonstrated on the problem of vector to raster conflation (VRC) of topologically and geometrically uncertain geo-spatial data. In this problem, multiple sources of data uncertainty include feature extraction from satellite imagery, vectorization of extracted features, and the whole process of generation of vector layers such as road and drainage networks. The derived conflation algorithm exploits generalized DLP with the lattice of models based on the hierarchy of topological and geometric uncertainties. The efficiency of this algorithm is shown on conflating satellite imagery with Tiger vector data.