SIMILARITY AMONGMULTIPLE GEOGRAPHIC REPRESENTATIONS

A key ingredient of systems aiming to cope with multiple representations of geographic features is some method for assessing the correspondence and similarity of such representations. In other words, given two objects from two different data sources, one must be able to tell whether they model the same real world object and, in this case, measure their degree of similarity. This paper proposes an adaptation of the Equivalents Rectangles Method (ERM) to quantify the average distance between ambiguous cartographic representa-tions and uses the Cartographic Similarity Index (CSI) – an index based on areal distances – to evaluate how much a given geometric representation resembles another. To validate the proposal, a prototype system was implemented and experiments were conducted on two geographic databases from two different institutions responsible for mapping the city of Rio de Janeiro. These were first matched using feature names in order to independently establish object correspondence. Then, the ERM and CSI of 159 districts that make up the city were computed. Results show that 157 districts have an Adapted ERM lower than 100.00 m and a CSI of 70 % or greater. The method was thus able to detect 2 districts with significant dissimilarity, and these conflicts were later confirmed visually, indicating survey errors. In summary, while the proposed method is being used in a larger framework for ad hoc querying geographic data with multiple sources, it is also useful in other circumstances, such as in a preprocessing stage for data source integration or for assessment of data source quality.