Spatial Data Cleaning

Abstract

There are always problems in spatial data, which makes spatial data cleaning the most important preparation for data mining. If spatial data are input without cleaning, the subsequent discovery may be unreliable and thus produce inaccurate output knowledge as well as erroneous decision-making results. This chapter discusses the problems that occur in spatial datasets and the various data cleaning techniques to remediate these problems. Spatial observation errors mainly include stochastic errors, systematic errors, and gross errors, such as incompleteness, inaccuracy, repetitiveness, inconsistency, and deformation in spatial datasets from multiple sources with heterogeneous characteristics. Classical stochastic error models are further categorized as indirect adjustment, condition adjustment, indirect adjustment with conditions, condition adjustment with parameters, and condition adjustment with conditions. All of these models are considered parameters as non-random variables in a generalized error model. By selecting weight for iteration, Li Deren made the assumption of two multi-dimensional alternates of Gauss-Markov when he established the distinction and reliability theory of adjustment. As a result, Li extended Baarda theory into multiple dimensions and realized the unification of robust estimation and least squares.