Integration of Observational Data and Behavioral Models for Spatio-Temporal Interpolation —Application to Reconstructing Long-Term Land Use and Land Cover Changes

Abstract—Spatio-temporal interpolation to generate voxel-field data in a space-time domain from observational data is indispensable to many spatiotemporal reconstruction and visualization of dynamic spatial phenomena. However, only very primitive interpolation methods such as nearest neighbor interpolation based on a Voronoi diagram are proposed for nominal or “class variable” data such as land use or land cover data. In interpolating nominal data with these primitive methods, we cannot make use of knowledge on spatial or temporal patterns or behavior of the object. The authors propose a spatiotemporal interpolation scheme for generating a voxel-field of nominal data under the framework of optimization of likelihood. The likelihood is computed from the fitness to both observational data and expected patterns/behavior described by a behavioral model or rules specific to the object. Any model which provides likelihood or probability to a given spatio-temporal pattern can be used in this framework. For the optimization of likelihood, a geneticalgorithm (GA) was combined with the Hill-Climbing (HC) method to increase the efficiency and reliability of optimization. Through some experiments, it is demonstrated that the GA/HC based interpolation method can generate voxelfields which fit both the observational data and to the knowledge of its behavior and that the reliability of interpolation can be evaluated quantitatively in terms of the maximal likelihood. Finally, the method is applied to the reconstruction of long term land cover changes from BC 7500 to present.