Correlation of urban form with land use and their identification using GIS and ML

Abstract

Urban planning is very strongly associated with land use. However, the existing land use is not the outcome of only planning decisions but also of the morphology of the place. Urban form and function are coherently linked with each other, yet there are very few studies that actively sought the relationship between these two. This thesis tries to bridge this gap between Urban Form and Urban Land use. The aim of this thesis is to create a model that can predict Land use, using urban form as a base. The study is done on the city of Indore, using only secondary data. All the classification and study has been done at building level, using only vector datasets. The study can be divided into three stages: urban form identification, urban land use prediction and analysis of results. During Urban form identification, 17 urban form measures were calculated for each building and were reduced to 5 using PCA. These symbolised size, density, layout, complexity and regularity. These were used as inputs for unsupervised clustering done using k-means clustering. At K=14, where the objective of classification was achieved, the clustered were grouped into 7 urban form typologies using visual study. These were urban villages, unauthorised/regularised colonies, planned settlements, periphery of planned colonies, periphery of unplanned colonies, apartments, and special buildings. These were further analysed on basis of their cluster centres. This gave an output of Urban form map of Indore which was used as a base for land use prediction and identified the factors that affected the urban form. During land use prediction, 30 other measures were calculated which were taken from studying various land suitability studies and location choice models. 15 of these were compressed to 4 measures using PCA which reflected network, connectivity, Walkability and intersections. The land use prediction was done using Artificial Neural Networks (ANN). Sample data was taken from the detailed land use of ABD area of Indore. The sample itself had some inherent bias, hence the land use prediction analysis was done only on the ABD area. The calculated measures along with urban form measures were used to train model only on one urban form type of unauthorised/regularised colonies. The accuracy of the model for that urban form came to be 79.0%. Further the accuracy for the same model for other urban forms for same model was just 62.3% To analyse the interaction of urban form and land use, the ANN was run multiple times in multiple conditions. The earlier model when used to predict land use for buildings with different urban form just gave accuracy of 69.2%. Another case where ANN was run without taking any urban form measures as inputs further decreased the accuracy to just 58.5%. This proved that there is a strong relationship between urban form and land use. This proposed model of urban land use classification based on urban form can be hence used to predict land use for any city with ample sample data for the city.