GeoComputation 2000 HomeConference ProgrammeAlphabetical List of AuthorsPaper

A Semi-Automatic Method to Construct Territorial Partitions


Key words: Spatial Aggregation, Territorial Pertinence, Territorial Partition, Agricultural Flows, Cutting Out Algorithms, Genetic Algorithms, Hierarchical Clustering, Exploratory Spatial Data Analysis.

This research problem is based on the need of the Regional Service of the French Agricultural Statistical Institute (SCESS) to adjust and then use new territorial partitions.  These partitions are needed to provide better statistical information about "communal " agriculture. This paper is presented at the end of the research (1996-1999).

Our work is concerned with the statistical or computational methods used to construct the territorial partitions. The partitions are based on geographical entities, such as Fench communes or cantons, or, more generally, aggregates of these basic entities. The aim is to obtain a optimal spatial partitions dealing with three geographical scales: communes, aggregates of communes, and links between them (agricultural flows).

We compute a "territorial pertinence" associated with every local aggregate and with the whole partition (distribution of aggregates pertinences). This index evaluates the quality of the partitions and permits the "best one" to be found. Other statistical indicators evaluate the accuracy of the aggregates and completeness of information. They can be used to make the results of the cutting out algorithm relative. The classification of communes within spatial aggregates is directed using these indices.

According to our operational partners, the classification of communes occurs in two stages that mix quantitative and qualitative evaluations. First, we apply and compare two different aggregation methods: hierarchical classification and "cutting out" using genetical algorithms. These two methods produce different "optimal" partitions. Second, the user chooses the best partition, with respect to the distributions of the statistical indices of quality. Using his tpractical knowledge of agriculture and the evolution of the statistical indices, the user can thenpermutate communes between some aggregates to improve their chosen optimal partition.  This is done using an exploratory data analysis tool developed within the XLISP-STAT statistical environment.

This approach has been tested on the French department of Isere, and will be generalized in the Rhone-Alps region, and hopefully, to all of France in the long term.