ROAD NETWORK ANALYSIS OF ALMATY: NETWORK SCIENCE APPROACH

ABSTRACT

Despite numerous studies linking network science and urban transport infrastructure, network science-based road network analysis remains underemphasized. In this paper, we aim to explore the structure of Almaty's road network and create methods to be applied to other cities with similar problems. Almaty was chosen due to the author's familiarity with the city and its infamous traffic jams and inefficient public transport system. Almaty traffic jams have become a huge problem and need a thorough study of its road network to establish the most significant nodes and crossroads. Through network science techniques, this paper discovers major structural characteristics and suggests potential solutions to decrease traffic and improve urban planning.

АННОТАЦИЯ

Несмотря на многочисленные исследования, связывающие сетевую науку с городской транспортной инфраструктурой, анализ дорожной сети на основе науки о сетях остается недостаточно изученным вопросом. В данной работе мы исследуем структуру дорожной сети Алматы и создаем методы, применимые к другим городам с аналогичными проблемами. Алматы был выбран по причине хорошего знакомства автора с городом, его печально известными дорожными пробками и неэффективной системой общественного транспорта. Пробки в Алматы являются огромной проблемой и требуют тщательного изучения дорожной сети для определения наиболее значимых узлов и перекрестков. С помощью методов науки о сетях данная работа выявляет основные структурные характеристики и предлагает потенциальные решения для уменьшения трафика и улучшения городского планирования.

Keywords: network science, road networks, Almaty, urban transportation, traffic analysis, centrality measures, spatial networks.

Ключевые слова: сетевая наука, дорожные сети, Алматы, городской транспорт, анализ дорожного движения, показатели центральности, пространственные сети.

1. Introduction

Road networks studies garnered a lot of attention in recent years, as well as studies utilizing network science methods to analyze various complex graph structures. And although there have been works exploring the intersection of these two, road structure analysis from the network science perspective is still an underexplored field. When looked through the lens of network science, road networks are conceptualized as spatial networks with nodes and edges representing intersections and edges, respectively. Understanding these networks and discovering their properties are essential not only for predicting traffic congestion and travel times, but also for assessing critical infrastructure resilience during emergencies and natural disasters [1].

In modern day, expanding cities are increasingly facing more and more traffic that are going beyond the expected limits of the road networks original designs. Network science provides powerful tools and methodologies for analyzing complex transportation systems, offering a different perspective  than conventional urban planning. It is especially useful for rapidly urbanizing cities like Almaty, in which street networks need to be able to adapt in a short time to keep up with shifting population trends and travel demands. By applying network science to studying Almaty's street network, we can not only better understand current traffic problems but can forecast future congestion hotspots before they are created. The research is a key contribution in closing this gap and bringing advanced tools to a major Central Asian metropolis, potentially offering solutions for similar urban hubs in emerging economies grappling with similar problems.

Almaty's transportation network (Figure 1) has a reputation in the eyes of both locals and tourists for frustrating traffic congestion and a lack of decent public transportation alternatives. These issues are just as bad as they read, and it is only through rigorous analytical methodologies, in our opinion, that the correct solutions can be discovered. Our analysis therefore examines the problem from the network science framework—structure allowing us to see where the most highly congested intersections are and how such key points function in the larger transportation network. These observations are most insightful when combined with the observation that there are streets in Almaty with disproportionately large volumes of traffic and in the process chronically creating bottlenecks within the city [2]. Aside from mere diagnosis of current traffic issues, the results here have great potential for informing future urban planning and allowing for more intelligent investments in infrastructure better suited for traffic flow.

1.1 Literature Review

Network science has been used previously to analyze road networks as complex systems. Reza et al. [3] examined road networks in Porto, Portugal, applying network science indicators including degree distributions, clustering coefficients, and centrality measurements to reveal network structure vulnerabilities. They detected small-world characteristics in three Porto sub-sections and demonstrated the relationship between modularity values and network reachability. Expanding on this effort, Wang et al. [4] examined the influence of road network structures on ride-sharing accessibility in Atlanta and established higher degree centrality and reduced closeness centrality as being associated with improved service accessibility. This work extends network science applicability further by linking structural road network characteristics to functional outcomes such as transportation service efficiency. Another issue in applying network science to road networks is the computationally expensive nature of detailed inspection and simulations for which Pung et al. [5] gave an approach to road network simplification maintaining topological characteristics and de-densifying the network of elements with negligible effect on transportation functionality. Their approach was computationally viable upon testing over various cities and by keeping essentially the centrality distributions of the original network and consequently proving suitable for in-depth road network inspection of much larger networks in instances of limited compute resources.

A significant contribution to understanding road network vulnerability in cases of emergency comes from Santos et al. [6], who conducted a comprehensive analysis of 69 Japanese cities. Their study revealed critical relationships between city characteristics and network vulnerability, finding that cities with higher population and infrastructure investment tend to be more robust under random attacks, while being potentially more vulnerable to targeted disruptions due to their concentrated urban functions. Also, their analysis demonstrated that car dependency tends to make cities more vulnerable toward random attacks but less vulnerable toward targeted attacks, as it indicates a weaker concentration in urban functions. Contrary to cities with highly connected train networks  showed different vulnerability patterns. These findings are most applicable to the study of a city in a seismically active zone like Almaty because they show comparative data from cities that have been affected by a variety of natural disasters and infrastructural problems.

Major emphases in past studies on Almaty include optimizing the city’s public transit and transport systems. Kosherbay et al. [2] analyzed Almaty’s public transport network for its inefficient routes with the consequence of connectivity problems of individuals who live farther from the center. Their study advocates for the optimal transport routes and enhancing transport-related facilities to improve accessibility and traffic flow. Bramson et al. [7] analyzed human movement flow in the Greater Tokyo Area in multimodal transport networks. Their in-depth study based on socioeconomic information validates the huge contribution of a diversified transport network towards people’s daily lives and presents evidence-based grounds for improving the integration of different modes of transport in Almaty to improve quality of life. Ge et al. [8] examines the disturbance and robustness of public transport systems and emphasizes the importance of managing the information on disturbance and delays in transport networks. They talk about the sustainability and dependability of the network in terms of combinations of different transport networks. Their review’s results reveal similar problems in public transport with probable solutions from where Almaty could definitely benefit in the future.

Figure 1. Almaty’s road network visualized as a graph

2. Materials and Methods

Figure 2 illustrates the process we are using to analyze Almaty’s road network. We download the map of the city as a network from OpenStreetMap (OSM) through the OSMnx module [9], which uses the Overpass API to download city coordinates and query map data, and project them into a NetworkX format [10] for analysis. NetworkX module provides extensive road network information, including nodes and edges for road segments and intersections as illustrated in Figure 1 and dataset head shown in Table 1.

Having gathered and parsed our dataset we proceed with the network analysis, i.e., analysis of degree distribution, calculation of k-nearest neighbors (knn), and calculation of three different centrality measures. The final step is the identification and the analysis of the top nodes based on different measures: degree, closeness, and betweenness centrality.

Table 1.

Sample of Road Network Data

U - The starting node (intersection) ID of the road segment. V - The ending node (intersection) ID of the road segment. Key - A unique identifier for parallel edges between the same nodes. OsmID - The OpenStreetMap ID of the road segment. Lanes - The number of lanes in the road segment. Name - the name of the road. Length - The length of the road segment in meters.

Figure 2. Methodology figure

2.1. Degree Distribution

Degree distribution is a statistical characterization of how degrees (the number of each node's connections) are distributed in a network. It is the probability P(k) that a randomly selected node in the network has exactly k connections or edges. It provides insight into the structure of connectivity in the network, whether most nodes have an equal number of connections or there is great variability. There is a binomial distribution of k degrees in the network since each of n nodes is independently connected with the probability p. Mathematical expression of the binomial distribution, exactly k successes in n independent Bernoulli trials, each having a probability of p successes:

2.2. Centrality Measures

We will utilize degree centrality, closeness centrality and betweenness centrality.(Borgatti [11]]; Newman [12], Otte and Rousseau [13])

Closeness centrality is a metric in network science that measures the centrality or influence of a node in a network by its proximity to all other nodes. It tells us how quickly information or material can be sent from a point to all other points in the network. It is calculated by taking the average of the shortest paths from the point to all other points, then reciprocating it (e.g., turning a fraction upside down). If N is the number of nodes in the graph G, then the formula for the closeness centrality of a node v is as follows:

Betweenness Centrality is a measure in network science that quantifies a node's significance by its location as a go-between in the shortest paths linking other nodes in the network. It highlights nodes that are essential points of connection or bridges in the network topology. Betweenness centrality of a node v is the summation of the proportion for all-pairs shortest paths passing via v. For shortest paths  connecting nodes  and , and  is the number that node v is one, then the betweenness centrality mathematically is:

Avg Nearest Neighbor degree, also - knn (Xia et al. [14]) is a network science measure that calculates the average degree of the neighboring nodes of a node. The measure tells us how much the nodes tend to cluster with the same or different types of nodes. The neighboring average degree of a node v is the average degree of all the nodes that are in direct connection with v. Mathematically, it is defined as:

3. Results

3.1 Degree Distribution

Figure 3 shows Almaty's road network degree distribution in regular and log scales. As seen from the histogram, the maximum degree is 10, the minimum degree is one, and the mean degree is 5.01.

Figure 3. Degree distribution

3.2 Average Neighbour Degree

Calculating the average nearest neighbor degree helps to determine if a node tends to connect to others with many links. The results showed Almaty's road network to have an average nearest neighbor degree of 2.87. Figure 4a shows the box plot of the distribution of average neighbor degrees. The median is approximately 3, and the interquartile range (IQR) is approximately 2 to 4. The plot is slightly negatively skewed, with fewer nodes with very high average neighbor degrees than with low or moderate degrees. There are also some outliers on the lower end, corresponding to nodes with very low average neighbor degrees, including some with a degree close to zero..

The negative skew points to a concentration of nodes with relatively high average neighbor degrees, meaning many nodes are connected to neighbors with many connections. The presence of outliers on the lower end suggests some isolated or poorly connected nodes, which may represent areas with underdeveloped road infrastructure or bottlenecks in the network. Figure 4b is a line graph showing the frequency distribution of average neighbor degrees. Most notable is the high frequency of nodes with an average neighbor degree of 3, which vastly outnumber nodes of other degrees.

(a)                                                                               (b)

Figure 4. Average Neighbor Degree: (a) as a box plot, (b) as a line chart

3.3. Centrality Measures

Degree centrality values are in range from 0.000041 to 0.000412, the mean value is 0.000206, and the standard deviation is 0.000083. Histogram of the degree centrality shown in Figure 5a displays that the degree centrality distribution has more nodes with lower values.

(a)                                                 (b)                                                                 c)

Figure 5.Centrality Measures: (a) Degree, (b) Closeness, (c) Betweenness

The mean closeness centrality of Almaty's road network is 0.01372 (Table 2). Figure 6 shows the visualization of Almaty's road network by closeness centrality, with higher closeness values in lighter colors. The visualization shows that closeness centrality is uniformly distributed and becomes denser towards the center. Figure 5b shows the distribution of closeness centrality, which is more normally distributed, indicating a very even distribution of centrality values.

Table 2.

Centrality Measures Statistics

Betweenness centralities vary from 0.000000 to 0.114194 with a mean of 0.003010 and a standard deviation of 0.009822 (Table 2). Figure 5c shows that the distribution of betweenness centrality is highly skewed with most of the nodes having very low betweenness values.

The correlation among centrality measures (Table 3) reveals that degree centrality and closeness centrality have a correlation coefficient of 0.055326, degree centrality and betweenness centrality have a correlation coefficient of -0.023082, and closeness centrality and betweenness centrality have a correlation coefficient of 0.221728.

Table 3.

Correlation between Centrality Measures

Figure 6. Almaty’s closeness centrality visualized

4. Discussion

The network analysis reveals several critical problems of Almaty's road infrastructure, specifically highlighting structural vulnerabilities that may contribute to the city's persistent traffic problems. The most striking finding emerges from the betweenness centrality analysis that is highly skewed, visualized in Figure 5c, which reveals a severe concentration of critical bridge nodes along Al-Farabi Avenue (Figure 7). Specifically, seven of the top ten nodes with highest betweenness centrality are clustered along a single section of Al-Farabi Avenue, spanning from its merging point with Sain Avenue to the intersection with Shashkina street. This concentration of critical nodes along this single pathway clearly indicate the part of the road network to suffer most from the congestion in the case of equally spread traffic flow as seen in morning and evening rush hours. Such a concentration is an example of how a disruption of one high-density corridor potentially could have cascading effects throughout the neighboring parts of the network ultimately causing more traffic jams.

The degree distribution analysis (maximum degree: 10, minimum: 1, mean: 5.01) signals a moderately connected network structure. However, the bigger picture is more complex when considered alongside the average nearest neighbor degree (mean: 2.87) and its negatively skewed distribution. The relatively low average nearest neighbor degree suggests that even high-degree nodes tend to connect to intersections with fewer connections, potentially creating bottlenecks where traffic from well -connected areas flows into less-connected regions.

Figure 7. Almaty’s map with top betweenness centrality nodes marked

The centrality correlation analysis shown in Table 3 reveals concerning patterns, such as weak positive correlation between degree and closeness centrality (0.055) which in turn, combined with the weak negative correlation between degree and betweenness centrality (-0.023), might suggest that the network's most connected nodes aren't necessarily its most critical bridges. This structural characteristic differs from more optimally designed networks where high-degree nodes typically serve as efficient traffic distributors. The moderate positive correlation between closeness and betweenness centrality (0.222) further emphasizes the reliance of the network on a small number of central pathways, making it sensitive to congestion and disruption.

The analysis of closeness centrality (mean: 0.013772 from Table 2, and visualized in Figure 6) uncovered intriguing patterns in the network's overall accessibility. The top nodes by closeness centrality cluster around major arteries - particularly intersections on Ryskulov Avenue (with Tlendieva, Kokoray, and Moskvina streets), near Rayimbek Avenue (intersecting with Sain Street), and at Zhubanov-Altynsarin junction. These nodes, being most centrally positioned with shortest average path lengths to all other nodes, should theoretically facilitate efficient traffic distribution.

Another weak point of Almaty’s road network comes from average neighbor degree distribution (Figure 4) which shows the tendency of a node to connect to other nodes with many connections. So, having the high frequency of nodes with the same average neighbor degree of 3, combined with the presence of outliers having very low neighbor degrees, suggests presence of isolated or poorly connected areas within the network. In this case even well-positioned intersections may not effectively distribute traffic due to insufficient connectivity in surrounding areas.

5. Conclusions

This study uncovers significant structural vulnerabilities that contribute to the city's traffic challenges. The most crucial finding is the severe concentration of critical bridge nodes along Al-Farabi Avenue, with seven of the top ten betweenness centrality nodes located along a single section of this one street.

Our analysis identified several key structural issues:

1. Over-reliance on Al-Farabi Avenue as a primary east-west connector

2. Shortage of alternative routes in major traffic corridors

3. Poor integration between high-degree nodes and critical bridge nodes

4. Uneven distribution of network connectivity, with some areas particularly separated

5. Suboptimal placement of high-closeness centrality nodes relative to traffic flow needs

The static nature of our network analysis presents certain limitations in capturing the full complexity of Almaty's transportation challenges. Real-world traffic patterns are dynamic, and tend to vary significantly throughout the day and across different seasons. Hence, to fully get and provide reasonable suggestions for urban planners for further development of Almaty, understanding these temporal variations, real-time traffic data and historical patterns are fundamental.

There was not a single pattern that the city employed to grow uniformly. It was the opposite actually, the city's structure had to undergo different eras through different eras, such as the Soviet Union and currently the Independent Kazakhstan. The lack of the natural river or lake at its core makes this city rather unusual compared to most other cities. The mountainous topography of the environment also restricts the potential network growth, namely for the southern part of the city.

Future research must be directed towards integrating the network analysis with the broader urban contexts, including land use patterns, points of interest, and multi-modal transport networks. Special emphasis must be placed on the nodal points such as business districts, schools, and residential concentrations, as those are the main generators of everyday traffic. Further, as Almaty is a fast-growing city both in population and size, analysis of the city in totality poses a certain challenge. Instead, an effort to delineate the different components of Almaty and analyzing each of those components in isolation would result in more accurate findings and implementable solutions.

These findings present just one perspective on examining Almaty's traffic problems from a network science perspective. Although they represent just one component of a complex puzzle, when combined with temporal data and spatial contexts, these findings can create extensive solutions to Almaty's urban transportation problems in future studies. The ultimate goal should be a more effective and resilient road network that better serves the needs of the population of this fast-growing city.

References:

1. F. Ahmadzai, K. M. L. Rao, and S. Ulfat, ‘Assessment and modelling of urban road networks using Integrated Graph of Natural Road Network (a GIS-based approach)’, Journal of Urban Management, vol. 8, 11 2018.
2. K. Kosherbay, A. Mussagaliyeva, G. Nyussupova, and J. Strobl, ‘ANALYSIS OF THE STATE OF PUBLIC TRANSPORT IN ALMATY’, Geojournal of Tourism and Geosites, vol. 45, pp. 1534–1542, 2022.
3. S. Reza, M. C. Ferreira, J. J. M. Machado, and J. M. R. S. Tavares, ‘Road networks structure analysis: A preliminary network science-based approach’, Annals of Mathematics and Artificial Intelligence, vol. 92, pp. 215–234, 1 2024.
4. M. Wang, Z. Chen, L. Mu, and X. Zhang, ‘Road network structure and ride-sharing accessibility: A network science perspective’, Computers, Environment and Urban Systems, vol. 80, 3 2020.
5. J. Pung, R. M. D’Souza, D. Ghosal, and M. Zhang, ‘A road network simplification algorithm that preserves topological properties’, Applied Network Science, vol. 7, 12 2022.
6. J. Santos, N. D. Safitri, M. Safira, V. Varghese, and M. Chikaraishi, ‘Road network vulnerability and city-level characteristics: A nationwide comparative analysis of Japanese cities’, Environment and Planning B: Urban Analytics and City Science, vol. 48, p. 239980832199931, 03 2021.
7. A. Bramson, M. Hori, B. Zha, and H. Inamoto, ‘Social and economic flows across multimodal transportation networks in the Greater Tokyo Area’, Applied Network Science, vol. 5, 12 2020.
8. L. Ge, S. Voß, and L. Xie, ‘Robustness and disturbances in public transport’, Public Transport, vol. 14, pp. 191–261, 3 2022.
9. G. Boeing, ‘Modeling and Analyzing Urban Networks and Amenities with OSMnx. Working paper’, 2024.
10. A. Hagberg, P. Swart, and D. Chult, ‘Exploring Network Structure, Dynamics, and Function Using NetworkX’, Proceedings of the 7th Python in Science Conference, 01 2008.
11. S. P. Borgatti, ‘Centrality and network flow’, Social Networks, vol. 27, no. 1, pp. 55–71, 2005.
12. M. Newman, Networks, 2nd ed. Oxford, UK: Oxford University Press, 2018.
13. E. Otte and R. Rousseau, ‘Social Network Analysis: A Powerful Strategy, also for the Information Sciences’, Journal of Information Science, vol. 28, pp. 441–453, 12 2002.
14. S.-Y. Xia, Z.-Y. Xiong, Y.-G. Luo, L.-M. Dong, and G.-H. Zhang, ‘Location difference of multiple distances based k-Nearest Neighbors algorithm’, Knowledge-Based Systems, vol. 90, 10 2015.