Blanca Arellano (p), Josep Roca

The separation between the countryside and the city, from rural and urban areas, has been one of the central themes of the literature on urban and territorial studies. The seminal work of Kingsley Davis in the 1950s and 1960s introduced a wide and fruitful debate which, however, has not yet concluded in a rigorous definition that allows for comparative studies at the national and subnational levels of a scientific nature. In particular, the United Nations (UN) definition of urban and rural population is overly linked to political and administrative factors that make it difficult to use data adequately to understand the human settlement structure of different countries.
The present paper seeks to define a more rigorous methodology for the identification of rural and urban areas. For this purpose it uses the night lights supplied by the SNPP satellite, and more specifically by the VIIRS sensor for the determination of the urbanization gradient, and by means of the same construct a more realistic indicator than the statistics provided by the UN. The arrival of electrification to nearly every corner of the planet is certainly the first and most meaningful indicator of artificialisation of land. In this sense, this paper proposes a new methodology designed to identify highly impacted (urbanized) landscapes worldwide based on the analysis of satellite imagery of night-time lights. The developed methodology allows comparing the degree of urbanization of the different countries and regions, surpassing the dual approach that has traditionally been used. This paper enables us to identify the different typologies of urbanized areas (villages, cities and metropolitan areas), as well as “rural”, “rurban”, “periurban” and “central” landscapes.

Katarzyna Kopczewska (p)

There are plenty of cluster-based measures of geographical and sectoral concentration. All based on two-dimensional table by regions and sectors, refer to empirical or theoretical distributions as the baselines. One can mention here sectoral concentration measures for regions: with uniform distribution (Relative H, Theil’s H, Shannon’s H, Ogive, Refined Diversification), with empirical distribution (National Averages, Relative Diversity, Hachman, Hallet, Kullback-Leibler Divergence, Krugman Dissimilarity, Lilien), with transformed empirical distribution (Gini, Relative Specialisation), with no distribution (Herfindahl, Absolute Diversity) and geographical concentration measures for industries with uniform distribution (Relative H, Theil’s H, Shannon’s H, Kullback-Leibler Divergence), with empirical distribution (Krugman Concentration, Bruhlart & Traeger, Agglomeration V, Clustering Bergstrand index), with transformed empirical distribution (Gini, Locational Gini, Ellison-Glaeser, Maurell-Sedillot), with no distribution (Moran I for LQ). There are also overall concentration measures for whole economy (Theil total and Geographic concentration index) and measures for single “cell” /for sector in region/ as Location Quotient. Selection of the measure in most of the existing applied studies is quite random, what weakens the conclusions.
Kopczewska et al. (2017) show the correlation and sensitivity of measures for the examples of real and generated data, reserving that all conclusions are approximated and require deeper studies. They indicate few group of measures, which behave similarly and should be treated as substitutes, not complements.
This paper develops the above mentioned study by Monte Carlo simulation for data generated from different distributions. It’s first goal is to determine the correlation between measures based on simulation, and not only point estimates, what can support the division of measures into homogenous groups. It’s second goal is to study the sensitivity of measures for extreme values (in sector and region) to assess its vulnerability and real range of values possible to obtain (aside from theoretical limits, which are often hardly obtainable).
Technically, the matrix k x n, for n regions and k sectors, will be filled with data generated from different distributions. Initially, for every cell, on the basis of uniform distribution, the parameters of normal distribution will be drawn. Secondly, the random numbers from given normal distribution will be assigned. This double randomization is expected to create different (also extreme) patterns of two-dimensional table by regions and sectors. For every kxn matrix generated, the set of geographical and sectoral concentration measures is calculated. Study is to detect correlations and sensitivity of the cluster-based measures and support the recommendation of the analytical toolbox.

Peter Van Den Besselaar, Ali Khalili (p), Klaas Andries de Graaf

see extended abstract