Finding pattern in Spanish cities

30 Aug 2024

Take home message

• Classification of cities may be possible based on land use, spatial configuration and socioeconomic data

• Cities with more than 100000 do not show characteristic of monofuncionality, such characteristic may be found in smaller cities

Classification of cities

Cities with more than 100.000 inhabitants can be classify in 5 different groups based on land use, spatial configuration and socioeconomic fetures. This groups are:

• Bid and diverse cities

• Small and monofunctional cities

• Embrionaries cities

• Transition

• Large and costal cities

Technical knowledge

• Processing spatial data iteratively using Python (Geopandas).

• Cleaning data in Python (pandas, regex).

• Preprocesing data in Python: scaling features using MinMax, StandardScaler, and PowerTransfomer methods.

• Exploratory data analysis (Outliers, features distributions).

• Assessing multiple combinations of parameters to find optimal solution for factor analysis.

• Evaluating the best combination of parameters to find optimal solution for hierarchical clustering using Silhouette as evaluation metric.

Objetive

Find groups of Spanish cities with more than 100,000 inhabitants based on land use, spatial configuration of patches, and socioeconomic data.

Figure 1. Flowchard describing the complete process to obtain cluster of cities.

Classification of urban land use

Using the Cadastral Classifier plugin in QGIS we downloaded the spatial and attribute data for Spanish cities with more than 100.000 inhabitants and classified in 12 different land uses, after further agregation and ommiting clases of low interest we ended up with nine land use classes. Finally the percetaje of each land use relative to total area was obteined for each city.

Cleaning data

Socioeconomic data was obtained from different governmental sources for each city. Using python a new dataset with organized data for land use, socioeconomic data, and spatial configuration was created. All different source of data were join using sql operations in pandas. Data from socioeconomic source was process from different files to created a new data file with the counties needed

Exploratory data analysis

Features were scaled using three different methods, MinMax, StandardScaler, and PowerTransformation of the sklearn library in Python. Additionaly, a function applied to the features allowed to find outliers using the criterio of the 1.5 Interquartil range.

Factor Analysis

Factor analysis applied to reduce the dimensionality and get new variables that could explained most of the variance of the data, also this new variables could be interpreted to facilitate the understanding of structure. Iterative factor analysis were made using the PSYCH library in R, a function was created to test the combination of different parameters to find the optimal solution for the Factor analysis (number of factors, rotation, minres). Solution with best structure meaning high explanation, low communality.

Cluster analysis

From the new factors different solutions were tested evaluation for different datasets. We tested removing few outliers, and differets number of clusters, linkage. We obtained aproximately 2000 combination of cluster. Many soluctions with high values of silhouette were meaningless since distribution of sample in the different cluster completly uneven. One of the optimal solution was chosen based on the interpretability of the clusters.