By Athanasios Strantzalis (Greece)
This research deals with a fundamental issue of Geo-information science, that of cartographic generalization and more specifically with the generalization of raster categorical data. Map generalization is the process that reduces and simplifies the details of the map; is performed mainly for visualization purposes, when a map is converted from a fine to a coarse scale / resolution.
However, during the generalization process the information and quality of the map are also reduced and errors are introduced. The challenge is to apply generalization in a way that optimizes the generalization objectives. Four generalization criteria were used, regarding the size and shape of the map entities, as well as the maintenance of the semantics and the proportion of the categorical classes. In order to obtain a generalized map that minimizes the weighted sum of the individual generalization criteria, the Simulated Annealing (SA) optimization algorithm was used.
SA is an iterative optimization algorithm that mimics the technique of the annealing process in metallurgy. In every step of the iterative procedure, a map entity is randomly selected and reclassified, while the effects of that reclassification in the map are evaluated, with the use of the defined generalization objective function. Based on the evaluation results, it is decided whether the proposed change is accepted or rejected. The iteration continues until a maximum number of iterations is reached, or the result becomes stable.
The developed algorithm was applied to two 120x120 pixels sections of the SoilGrids1km product, which is a global soil product produced by ISRIC – World Soil Information. The two selected areas were chosen in order to demonstrate behavior of the generalization at different levels of map complexity. The generalization took place at three threshold values of 5, 17 and 65 pixels for the minimum mapable unit.
The performance of the SA was evaluated based on the required computational time and on the goodness of the generalization. Additionally, the results of the developed generalization algorithm were compared with the results of a commercial generalization methodology. It was shown that when the parameters of Simulated Annealing are configured properly, the developed algorithm can outperform the commercial methodology, in terms of value of the objective function.
Keywords: Cartographic Generalization; Soil Map Generalization; Taxonomic Distance; Simulated Annealing Optimization; SoilGrids1km.