Croitoru & Doytsher Joint Workshop on Multi-Scale Representations of Spatial Data Ottawa, Canada, July, 7 th -8 th , 2002 1 RANDOM FIELDS AND (MARKED) POINT PROCESSES: A PRACTICAL COMPARISON OF TWO STOCHASTIC MODELS Arie CROITORU and Yerach DOYTSHER Technion – Israel Institute of Technology Faculty of Civil Engineering, Division of Geodetic Engineering Technion City, Haifa 32000, Israel ariec@tx.technion.ac.il, doytsher@geodesy.technion.ac.il Key words: vector data quality, random field, marked point process ABSTARCT Reliable detailed information on the positional accuracy of spatial data is essential to end-users. In order to accommodate this two essential tools, namely reporting tools (“Metadata”) and modeling tools (a mathematical framework describing the positional accuracy) have been developed in recent years. One such modeling tool is the random field stochastic model, in which the discrepancies between two data sets are treated as a random process (a "signal") over a two or three-dimensional domain. The characteristics of a random field may be attained using well-known geostatistcal estimators, such as the variogram and the correlelogram. The foremost advantage of this approach is that the errors are considered as a spatial phenomenon, in which correlations are accounted for. This contribution discusses the link between the random field model and two other stochastic tools, Least Squares Collocation (LSC), and point process analysis. It is argued that this linkage is required due to the inability of the random field model to carry out signal filtering or prediction, and its lack of sensitivity to the spatial distribution of the data points in the domain that is being evaluated. Each of these tools is described, and its usage in the context of spatial data is discussed. INTRODUCTION Recent advances in information technology as well as in spatial data capturing and processing techniques provide a wealth of spatial information. Such information can be easily and quickly downloaded from various data providers via the internet or other broad-band communication facilities, and its usage is almost immediate as it can be downloaded in a variety of commonly used formats. This so-called "plug-and-play" approach is new to spatial data providers, as well as to data end users ("users"). Unlike the situation where data is collected by an organization according to its particular specifications and for its exclusive usage, data is no longer “tailor-made” and direct communication between the data provider and the user is no longer assured. This may result in data misuse, forcing the user to check the applicability of the data. Data providers may also be affected by this new situation, as there is no guaranty of proper usage of the data, and as the proficiency of the user is unknown. This may expose data providers to undesirable liability and other legal complications. It is possible to resolve some of these difficulties by providing additional descriptive information regarding the data set at hand. Such additional information may be delivered to the user by adding metadata to the data set, in which various aspects of the data (such as data quality information, spatial reference information, or temporal information) are detailed. Yet for many applications the available