GIS and Transport
By reviewing ‘Accident and traffic analysis using GIS’
Introduction
With urbanization and globalization playing a critical role, traffic and transportation are increasingly important in the growth of cities, as development is based on the connectivity and accessibility (Shafabakhsh, Famili and Bahadori, 2017) However, it can also be an obstruct of sustainable economic growth in terms of traffic accidents due to the loss of property as well as injuries and death. Since GIS is specialized in performing complex spatial analysis, it is can be largely applied in analyzing problems of accident and traffic (LeGates, 2005)
In finding articles on this topic, ’GIS For Transportation’ column on ‘Esri’ website was looked through to get an overall knowledge of in what area and how GIS can be applied. Then keywords such as ‘GIS’, ‘Transport’ and ‘Traffic’ were put in to the search box on the website of ‘Journal of Transport Geography’. Among a number of articles, the article selected to be reviewed, is a research wherein the accident and traffic of Highway in Coimbatore District of India were analyzed using GIS. It was selected as it contained the important tool ‘Kernel Density Estimation’ to evaluating the probability of accidents and has been implemented into practice. Moreover, it is of value to see how and to what extent GIS is applied in developing countries.
This essay will discuss the implementation and impact of GIS in transport by reviewing the article ‘Accident and traffic analysis using GIS’, written by Selvasofia and Arulraj in 2016. First, the summary of the research will be given, then the strength and weaknesses of it will be analysed. Finally, suggestions will be proposed.
Summary of the article
The research question is to identify the hotspots and predict the probability zones on national highways in Coimbatore District of India, using Kernel and Weighted Sun Overlay analysis in GIS.
The method of the research can be divided into 3 parts. First in the collection of data. Spatial data, such as location, boundary extend and road network, is collected and added into database, with the accident site mapped on. Non-spatial data, including date, time, type of the accidents, vehicles involved as well as special conditions, such as licensed or non-licensed and drunk driving, is collected from the police station and added as attributes. Moreover, the locations of bus stops, hospitals, ,police stations, markets, schools, clubs and hotels are also added.
Secondly, accident hotspots are identified and the probability zones are classified from high to low, by calculating the density of accident hotspots with Kernel Density Estimation.
Thirdly, Buffer Analysis is used to identify the bus stops within 500m from the accident hotspots, which is aimed at minimizing the probability of congestion.
Finally, the Weighted Sum Overlay Analysis is performed and the accident locations are ranked according to the probability, with weights assigned to factors, such as the accident hotspot level, the travel time to the nearest hospital, the jurisdiction of police stations and so on.
There are three major findings. Firstly, eight bus stations are found within 500m from the accident hotspot and are suggested to be removed. Then it is found that, as the highways extend from the centre of the city, the probability of accidents is likely to reduce. Moreover, Periyanayakampalayam, Sarcarsamakulam, Thondamuthur, Madukkari and sulur are the locations wherein most accidents happened. With respect to this finding, alternatives and measures are offered to local authorities.
Comments on strength and weaknesses
Strength
The strength of this research will be analysed from the factors considered, methodologies and effect.
As analyzing the probability of accidents, factors such as the jurisdiction of police stations, the travel time to the nearest hospital, markets, schools, clubs, hotels, have been taken into consideration. Thus, the results can be more reasonable and reliable.
Apart from it, Kernel Density Estimation (KDE), Buffer Analysis and Weighted Sum Overlay Analysis are used. KDE is the most used spatial data analysis to determine the risk spread of the accidents (Satria and Castro, 2016). Since the point pattern may have a density at any location not just where it is displayed, KDE is more accurate comparing to point density calculation and it allows point data to be analysed in a consistent form (Shafabakhsh, Famili and Bahadori, 2017). Weighted Sum Overlay provides the ability to weight and combine multiple factors to perform an integrated analysis (ArcGIS for Desktop, 2016). In this case, the prediction of probability maybe more realistic and reasonable, with different values assigned to different factors. In addition, it is more favourable than Weighted Overlay Analysis, as it can maintain the attribute resolution without rescaling back to the evaluation scale (Selvasofia and Arulraj, 2016).
In terms of its effect, suggestions based on the analysis are given, such as removing the bus stops near the hot spots, improving the infrastructure, increasing the lanes, providing necessary notice boards and so on (Selvasofia and Arulraj, 2016). What’s more, it seems to be useful in predicting the traffic pattern and the probability of accidents in reality, which can help decision-making in urban management policy formulation.
Weakness
However, limits also exist, in terms of the inaccuracy of data, subjectivity in assigning weights and ignorance of factors such as time, infrastructure conditions, neighbouring roads and so on.
Since the police office may not be able to collect the information of all the accidents, the data in the analysis collected is likely to be inaccurate. As the success of identifying the probability of the accident relies on the accurate and reliable data, the reliability of the research can be limited.
Apart from it, subjectivity is unavoidably involved in performing the Weighted Sum Overlay Analysis. Weights are assigned differently and values are given individually, depending on circumstances and personal preferences.
What’s more, factors such as time and infrastructure conditions, are not taken into account, therefore, the results can be deficient. Probability of accidents can vary in time. For example, accidents are more likely to happen during the peak-hour and night due to the high flow and poor visibility. Apart from it, locations with bad road conditions, less lanes and high flow probably should be taken into account.
Moreover, in performing the Weighted Sum Overlay, the values given to the inputs, as an important information, are now showed and clarified in the articles, which may cuase difficulty in understanding for readers.
Questions
With respect to the questions I have got in reviewing the article, they can be described as below. What is Kernel Density Estimation? Why do medium probable zones have more hotspots than high probable zones? What is the difference between Weighted Overlay Analysis and Weighted Sum Overlay Analysis? Why the author prefers the latter? Then I managed to figure them out through looking up the explanation in the “Esri Support GIS Dictionary’ and performing them in ArcGIS with the methods given on ‘ArcGIS Pro’ and ‘ArcGIS for Desktop’ websites. Moreover, reading more articles with similar methodologies also helps me understand it. Nevertheless, I still do not understand why the medical information is added into the consideration of prediction of probability, since there is no relationship between them.
Suggestions and conclusion
Suggestions will be given from two aspects. One is based on its limits. Regards of the insufficient accident information, it can be improved by using GIS to monitor the roads and restore the accident sites with locations. By this approach, data can be more reliable and timely, thus prediction can be more accurate. In terms of the subjectivity in giving weights and values, although it is unavoidable, effort can be made to reduce the effect. For instance, Moran’s I statistic, which is used to analyse the spatial autocorrelation, probably can be utilized to evaluate the spatial dependence of the accident location (Satria and Castro, 2016). Getis-Ord Gi* can help to find the dependence of spatially distributed variables, which is highly likely to contribute to hotspots identification (ArcGIS Pro, 2017). Moreover, Regression Analysis can be used in assessing the relationship between different variables (Aghajani, et al., 2017). Importantly, Bayesian Estimation is highlighted by many scholars as a complement to Kernel Density Estimation, for it includes data from the past experience, expert opinions and prior belief (Dereli and Erdogan, 2017)
The other aspect is about the further development of the prediction of accident probability. First, network analysis can be used to determine the most appropriate locations and numbers of hospitals, police stations and necessary facilities. With network cost taken into consideration, the relocation may allow these facilities to respond quickly according different accident sties. Apart from it, real-time traffic monitoring may be performed with this model to reduce the probability of accidents. That is to say, as locations with high flow to be identified through monitoring, control over the numbers of vehicles can be taken, if it is highly likely to cause traffic accidents. Moreover, if accidents occur, it may also provide alternatives to drivers who are intended to enter the area.
In all, through monitoring, planning and managing the transportation using GIS, a smarter and safer infrastructure system can be built to reduce the probability of accidents, enhance the quality of daily life and advance the development of economy.