Date of Award

5-11-2024

Document Type

Thesis

School

School of Civil Engineering

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Dr.R.Selvakumar

Keywords

Urbanization, Region Generation, Centroid Shift, Landscape Metrics, Prediction, Modelling, Support Vector, Random Forest, Drivers

Abstract

Urban populations have grown substantially in the modern era and the information age as people seek better living prospects. Currently, more than half of the world's population resides in the cities. Such growth must be monitored to ensure that people enjoy the boons of urban spaces, not the banes. Recognising the significance of urban studies, several researchers have attempted to analyse and predict urban dynamics using remote sensing, Geographic Information Systems, and machine learning methods. However, certain limitations exist, especially in improving the CAMarkov model's prediction performance. In addition, studies have suggested that growth characteristics, such as pattern, intensity, and direction, must be included as driving factors in prediction models. These hypotheses served as the basis of the present study. The research also emphasised the necessity for regional urban growth studies beyond its corporation limit.

Salem, a non-metropolitan city, was earmarked as a smart city because of its economic importance and was chosen as the region of interest. During the preliminary field visit, it was noticed that there was a considerable floating population from neighbouring towns, such as Omalur, Sankari, Rasipuram, and Vazhapadi. This prompted the conduct of a regional analysis that included these towns. Land Use/Land Cover (LULC) was the preliminary data required for this analysis. However, owing to the lack of a high-resolution LULC dataset, this study began by creating the LULC dataset using pan-sharpened Landsat images. Using Landsat data from 2020, the performances of three machine-learning algorithms the Random Forest Algorithm (RFA), Maximum Likelihood Classifier (MLC), and Support Vector Machine (SVM) were evaluated. The results showed that the SVM output had the highest accuracy of 0.945; subsequently, the LULC data were prepared for 2001 and 2011. The LULC classes include agricultural, built-up, others (barren, fallow land), and restricted (water bodies, mining, and forest). The created dataset yielded overall accuracies of 91.4%, 92.3%, and 95.2% for 2001, 2011, and 2020, respectively, with kappa statistics of 0.884, 0.896, and 0.935. These results demonstrate the reliability of the LULC maps for further research. Subsequently, changes in LULC were examined, revealing a consistent uptrend in urban expansion, characterised by the conversion of agricultural land to barren land and its subsequent development into built-up areas. The time-series analysis of changes in LULC revealed a steady increase in urban growth, characterized by converting agricultural land to barren land and eventually transforming it into builtup areas. This growth pattern is expected to continue and accelerate as the city expands and transforms into a smart city.

The subsequent phase involved comparing the prediction performance of the CA-Markov model using two methods: 1) without driver variables and 2) with driver variables. The model output without drivers exhibited a kappa statistic of approximately 0.75. Although the value was statistically significant, differences in the area measures were noted. The predicted built-up area exceeded the actual value by 111 square kilometres. These results indicate the need to incorporate drivers. Subsequently, the study measures urban growth characteristics, such as direction, pattern, and intensity, and incorporates them as drivers in a limited data scenario. The objectives of the directional study were to (1) delineate the limits of urban growth for Salem and its neighbouring towns, (2) conduct directional analysis within these limits, (3) use centroid shift analysis to explore the macro-level direction of urban growth using centroid shift analysis, and (4) use landscape metrics to quantify the dynamics of urban cover changes in Salem and neighbouring towns. The major outcomes of the study are that the study provided evidence of interconnectedness between the city and towns and the that the region is still in urbanization phase and has potential for planned development. In addition, the analysis of the built-up extent across the three study periods revealed that, contrary to the current corporation de jure boundary of 9 km, the actual extent of the city was 12 km from the city centre. This emphasises the importance of upgrading city boundaries to enable proactive urban planning.

The study then mapped four different types of growth patterns for 2001 2011 and 2011 2020: infill, sprawl, scattered, and ribbon. The study revealed a significant increase in built-up areas between 2011 and 2020, with ribbon development emerging as the most common type of urban growth. According to the findings, the main drivers of urbanisation are population increase and economic development. Additionally, the findings demonstrate that urban growth is taking place rapidly on the city's periphery, encroaching on agricultural land and potentially impacting the local economy. The study also found that neighbouring towns, namely Omalur, Rasipuram, Sankari, and Vazhapadi, influence the urban growth patterns of Salem. These changes are attributed to the dynamic interaction between population growth, accessibility, agricultural land, and urban planning considerations.

The Urban Expansion Differentiation Index (UEDI) was used to identify the dominant pattern for each growth. The studies found that sprawl is dominant pattern, particularly on the fringes of cities, resulting in encroachment of agricultural land, particularly around Salem. This trend could negatively influence agricultural productivity and the local economy. The Urban Expansion Intensity Index (UEII) was then used to determine the intensity of growth, which was classified into five categories: very high, high, medium, low, and very low. Between 2001 and 2011, high-and very high-intensity values were primarily found in the core areas of Salem and Omalur.

However, these high-intensity values exhibited a dispersal pattern during the subsequent decade (2011-2020), becoming more common in the city and town cores and suburban areas. The objective of this study is to improve the model by incorporating these patterns. Consequently, a future growth region was created around pre-existing built-up areas by considering both pattern and intensity. A novel method was developed and a region was generated using this new algorithm within an arcpy environment. In the final phase, the CA Markov model was executed using the ultimate drivers selected using the Cramer's V index. The drivers include pattern, intensity, and direction based on growth characteristics. The model performed well, with a kappa value of 0.9 and a 9.01% difference between the actual and predicted built-up areas. Urban growth was then predicted for 2030 2100 at 10-year intervals. The predictions indicate that growth will be at its highest over the next three decades before experiencing a notable decline after 2070.

Additionally, a prediction was made based on a simulated scenario to comprehend unidirectional growth by hypothetically limiting the growth to Salem. The output was compared with the bidirectional Business as Usual (BAU) scenario. Unidirectional analyses suggest a potential merger between Salem and Omalur by 2100, whereas bidirectional outcomes anticipate conurbation by 2050. These findings demonstrate the importance of bidirectional analysis, which accounts for growth disparities, in offering a comprehensive perspective on regional studies. In summary, this study provides insight into the urban dynamics of Salem and its peripherals, and presents a novel approach to model future growth using the CA-Markov model. Thus, blazing a trail for forthcoming researchers in the field of urban studies.

Download

Share

COinS