The Effect of ACRC on the Results of Cartographic Classification Depending on Spatial Autocorrelation

Abstract

This paper examines a recently published method for cartographic classification of spatial data, called Autocorrelation based Regioclassification (ACRC). The main benefit of ACRC is its self-adaptive weighting of attribute values and spatial proximity within the classification process. The according weight is calculated on the basis of the degree of spatial autocorrelation factually present in the data. Hence this fundamental property of spatial dependency is explicitly considered. Any arbitrary component in the classification process can be avoided due to the strict statistical approach. As demonstrated in this paper, ACRC results in visually less complex choropleth maps compared to standard classification algorithms, whereas the amount of complexity reduction depends on the degree of spatial autocorrelation present within the data set. As a trade-off, the goodness of variance fit (GVF) of the classification is slightly reduced. To help the user to estimate the visual and statistical effect of the ACRC method, we suggest a statistical measure expressing the ratio of visual complexity and GVF. In an introductory section we shortly summarize the framework of the ACRC method and the major challenges of classifying with spatial data in general. Within this context we further extend the argument for an explicit consideration of spatial dependencies in (cartographic) spatial data classification. After a brief presentation of the method itself we examine the effect on the classification result. For this, the ACRC is applied to three sample data sets, exhibiting different degrees of spatial autocorrelation. On the basis of these results the self-adaptiveness as well as the general applicability of the method are demonstrated