ON THE ROLE OF QUALITY RELATIONS IN PATCH-BASED SPATIAL DATA UPDATING Arie CROITORU and Yerahmiel DOYTSHER Technion – Israel Institute of Technology Department of Civil Engineering, Division of Geodetic Engineering Haifa 32000, Israel ariec@tx.technion.ac.il , doytsher@geodesy.technion.ac.il KEY WORDS: Updating, Accuracy, Collocation, Transformation, Distortion field model. ABSTRACT As spatial data becomes a central component in a variety of applications, the demand for up-to-date data is on the rise. In order to shorten the updating cycle time local updating is preferred, in which patches of up-to-date data must be incorporated into the existing data set. Although this can be done by using a global transformation model or a rubber-sheeting scheme, it is argued that in the case of patch-based updating the accuracy relations and its spatial variations must be considered. This requires adopting a field model for the various distortions in the data sets, as well as the implementation of proper computational tools. It is suggested that collocation can not only be used as such a tool, but that it may also encompass additional advantages, such as the ability to estimate the distortions in one data set based on the distortions in another data set. An example to the field model and the estimation of unknown signals in a data set are also presented. 1. INTRODUCTION The demand for up-to-date spatial data has long been self-evident. Nowadays, As up-to-date spatial data are becoming a fundamental component in a variety of engineering, analysis and management operations and as this data is becoming readily available to a growing community of users, the requirement for up-to-date data is on the rise. In order to facilitate this requirement an updating process is employed. This process can be carried out either on a global scale, where the entire data set is replaced by a new up-to-date data, or on a local scale, where distinct areas in the existing data set are updated. Due to various drawbacks of the global scale updating process, such as its long duration and the considerable resources required for its implementation, a local updating process, during which only patches in the existing data are updated, is frequently preferred. Both the global and the local updating schemes share several fundamental processing steps that are required for their success. These steps usually include extracting up-to-date data, detecting and classifying changes, and finally, incorporating the up-to-date data with the existing data. Up-to-date data extraction usually consists of processing up-to-date aerial photographs or a re-mapping of the interest area using a variety of surveying methods. Change detection and classification, which is at the heart of the updating process, consist of a comparison between the up-to-date and the existing data, resulting in a designation of areas or objects that were affected by change. The final step of incorporating the up-to-date data with the existing data usually consists of transforming the up-to-date into the existing data set using a variety of transformations. Most of the research effort in recent years was dedicated to the first two steps of the updating process, yet little attention was given so far to the problem of incorporating the existing data set and the up-to-date data (in the case of a local updating process, this includes incorporating several patches of up-to-date data). The data incorporation problem is commonly solved by employing various geometric transformations. These transformations are realized by mathematical models with various degrees of freedom, ranging from a rigid- body transformation with three degrees of freedom up to an affine transformation with six degrees of freedom, or a projective transformation with eight degrees of freedom (Fagan and Soehngen, 1987). The transformation process begins with the measurement of homologous points in both data sets. If redundant points were identified the transformation parameters may then be estimated using the well known least squares adjustment technique, during which weights may be assigned to each measurement (Greenfeld, 1997 a ) ; (Greenfeld, 1997 b ). In the case of control points weights may be assigned by the rank of each point in the control network hierarchy (Greenfeld,