Where are the most dangerous road locations in Hong Kong?

An investigation of traffic crash hot spots and hot zones

April 1, 2020

Introduction

Traffic casualties and fatalities are some negative social impacts of the transport sector. Promoting road safety requires a thorough understanding of the spatial characteristics of traffic crashes, hence enabling us to devise effective countermeasures. In a lot of countries from different parts of the world, there are strong initiatives to achieve the "Vision Zero" objective. This study offers a new spatial analytical method, called "Hotzone methodology", to identify dangerous road sections. The newly developed add-in for detecting these hotzones will help policymakers to implement effective corrective measures.

An overview of traffic crashes in Hong Kong

Trend of traffic crashes in Hong Kong (2000-2018)

As of 2018, there were 15,935 road traffic crashes and 19,637 casualties. 107 people were killed. Among the 15,935 crashes, 0.7 percent were fatal, 10.6 percent were causing serious injury and 88.8 percent were causing slight injury.

The number of traffic crashes fluctuates moderately, ranging from 14,300 as the minimum, to 16,200 as the maximum. The figure has increased by around 6.6 percent.

Severity of traffic crashes

The trend of slight traffic crashes is noteworthy. Between 2000 and 2018, there is a 18.4 percent increase. This requires further investigation and bold incentives to tackle the increasing trend.

The trends of fatal and serious traffic crashes are noticeably declining. There is a 34 percent and 40.7 percent decrease in fatal and serious traffic crashes respectively.

Types of road users

It is also important to examine the vulnerable road user groups, including pedestrians and cyclists. When comparing the figures of 2010 and 2018, the share of pedestrians involved in traffic crashes has decreased from 20.4 percent to 16.8 percent. As for cyclists, the share has decreased from 9.2 percent to 8.8 percent.

Characteristics of traffic crashes in Hong Kong (2017)

Traffic crash locations in 2017

There were 13,806 geovalidated traffic crashes in 2017.

Spatial clustering of traffic crashes

The heat map shows the clustering pattern of traffic crashes in Hong Kong in 2017. "Blue" refers to low density of traffic crashes. "Red" and "yellow" indicate moderate density and high density respectively. There are several observations:

Traffic crashes are highly clustered in Kowloon and the northern part of Hong Kong Island (Admiralty and Wan Chai).
Multiple new towns (including Tuen Mun, Yuen Long, Tsuen Wan, Sha Tin, Tai Po and Sheung Shui) are observed with a moderate to high level of clustering of traffic crashes. Tsuen Wan and Yuen Long are having the highest density of traffic crashes.
Highways and major trunk roads are also observed with a moderate level of traffic crash density.

How do we usually analyse the spatial pattern of traffic crashes?

Junction Blacksites identified by the Transport Department (2018 Q4)

The hotspot analysis is typically used to detect dangerous road locations. The method of "Junction Blacksites" used by the Transport Department is in line with the traditional "hotspot" analysis of traffic crashes. The hotspots are typically located at road junctions (i.e intersections).
The definition of a Junction Blacksite used by the Transport Department is either (i) there are 6 or more pedestrian injury accidents in the past 1 year; or (ii) there are 9 or more injury accidents in the past 1 year (Transport Department, 2018).
There were 40 Junction Blacksites in Quarter 4 of 2018. They are mostly located at the urban areas of Kowloon, with several situated at the new towns such as Tsuen Wan, Yuen Long and Shatin.

Relationship between Junction Blacksites and neighbourhood attributes

The base map shows the population density in different large tertiary planning units (LTPU). Darker colour represents higher population density.
As expected, the junction blacksites are usually found in areas of higher population density.

Relationship between Junction Blacksites and neighbourhood attributes

The base map shows the employment density in different large tertiary planning units (LTPU). Darker colour represents higher employment density.
The Junction Blacksites do not exist in areas of the highest density of employment opportunities (i.e. Central, Admiralty and Wan Chai).
Some Junction Blacksites appear in the typical downtown areas of Kowloon with relatively high employment density (i.e. Mong Kok ad Tsim Sha Tsui).
There are significant implications on public health as well. The residents in the most densely populated areas ((both residential and activity nodes) are more exposed to the risks of traffic hotspots.

Relationship between Junction Blacksites and traffic flow

The base map shows the annual average daily traffic flow (AADT) in 2017. Darker colour represents higher AADT.
The relationship of AADT and Junction Blacksites seems to suggest different patterns. Some of the traffic crashes are located at road sections of higher AADT, whereas some are located at a moderate level of AADT.
This suggests that traffic crashes are not only related to traffic volume. Other factors such as human and road environment are relevant to traffic crash occurrence.

Based on the hotspot analysis (Junction Blacksites), we can identify specific road intersections of a relatively high number of traffic crashes. However, dangerous road locations do not only include road junctions (i.e. points). There are also some road sections (i.e. lines) that are quite dangerous and require corrective measures. An illustrative example is that almost no hotspots are identified in highways.
In the past several years, there are several major incidents of traffic crashes along road sections outside the traffic hotspots. The pictures below demonstrate several serious traffic crash locations that are not identified before. Therefore, it is important to identify the dangerous road sections in order to devise preventive measures in advance, and to further reduce the number of traffic crashes and promote road safety.

A fatal traffic crash in Tai Po Road that caused 19 fatalities and 60 injuries in 2018 (Source: SCMP)

A fatal traffic crash happened near Tai Lam Tunnel in Pat Heung that caused 1 person killed and 14 injured in 2019 (Source: SCMP)

Research questions

Where are the most dangerous road sections in Hong Kong?
How can we use GIS techniques to develop a systematic and consistent methodology of identifying dangerous locations?
What are the environmental features of the identified road sections?
What feasible countermeasures can be implemented to reduce the traffic crashes of the identified road locations?

Methodology

01 / 08

To identify the hazardous road locations falling outside the intersections, we need to examine the dangerous road sections (i.e. lines). Loo (2009) develops a hotzone methodology to identify continuous road segments of a certain threshold of traffic crashes.

The difference between hotspot and hotzones analysis is shown in the figure. Essentially, a hot spot refers to a dangerous road location where the actual crash rate is higher than the critical crash rate, whereas a hot zone refers to dangerous road locations made of two or more continuous segments that contain an actual crash rate higher than a threshold value for each spatial unit (Satria et al., 2020).

In the diagram, the roads are first segmented into the basic spatial units (BSUs). Suppose we classify road sections as hotzones by using the threshold of at least 3 crashes at each BSU. Then, any 2 or more continuous road segments that are recorded with at least 3 road crashes each will be identified as a hotzone.

Overall, the hot zone methodology takes network contiguity into account instead of identifying single hot spots of road crashes. It is very useful in road safety analysis because a hot spot might be a result of randomness rather than a systematic deficiency (Loo, 2009).

Hotzone Generation Add-in

The hotzone methodology includes mainly three essential steps: (i) network segmentation, (ii) finding intersected points on the road network and (iii) finding the consecutive road sections above the minimum threshold (Loo, 2009). In order to facilitate and speed up the analysis, a GIS-based add-in integrated with Python Scripting is developed. The button below can link to the webpage of this add-in. Free download is available!

Free download (version 0.9)

The add-in is a GIS-based add-in integrated with Python Scripting (ArcGIS 10.3 or above) that identifies hazardous road locations and detects hot zones (also called black zones) of road traffic crashes in a systematic and scientific manner.

Major values of the add-in

High practicability: The add-in can be universally used by road safety administrations, consultancies, as well as researchers all over the world to identify hot zones of road crashes given precise spatial locations of crash data.

Fast computing time: The add-in can generate hot zones over a vast amount of road crash data and dense road network in a short period of time. In our trial, the add-in can process the road crash data of a year in Hong Kong within 30 seconds, with over 15,000 road crashes distributed on more than 4,000 km of roads.

How the add-in works?

1) Network segmentation

Intersections and road ends in the road network are considered to be nodes. Links are checked to see whether they exceed the specified BSU length. If so, they are further cut up at a standard length from the starting nodes. The process repeats until none of the links exceeds the standard length.

2) Finding the intersected points

To avoid double-counting (the edge effect), a crash that occurs at the end nodes of more than one BSU is assigned to the BSU with a smaller x coordinate.

3) Finding the consecutive roads of a minimum crash threshold

Only when contiguous BSUs (more than 1 BSUs) are found to be exceeding the minimum threshold of crashes are considered as a hot zone. In other words, a hot zone must be formed by at least two BSUs.

How to use the add-in?

Step 1. Prepare and import two different datasets in Shapefile format: (i) a road network (in line) and (ii) geo-validated road crashes (in point).

In this study, we use the datasets of "Road Crashes in 2017" and "Road network of Hong Kong.

Step 2. Select the road network shapefile in the drop-down menu of “Roads” and the geo-validated road crashes shapefile in the drop-down menu of “Crashes”.

Step 3. Specify the length of basic special units (BSUs). The drop-down menu of “BSU length” provides the option of 50m, 100m and 200m but users can define specific values (manual input) based on specific local context. In this study, we set the BSU as 100 meter, which is a common guideline in hot zone analysis.

Step 4. In the box of “Crash threshold”, specify the minimum threshold of road crashes in a BSU to be considered as a dangerous road section. Only when contiguous BSUs (more than 1 BSUs) are found with exceeding the minimum threshold of crashes are considered as a hot zone.

In this study, we set two crash thresholds. The first is "3 crashes per BSU", which means there will be at least 6 crashes in each hotzone (at least 2 continuous BSUs with each higher than 3 crashes).

The second is "6 crashes per BSU" as a tighter threshold to identify the most dangerous locations, which means there are at least 12 crashes in each hotzone.

Step 5. Click the “Fire” icon to generate the hot zones.

Step 6. Look at the results! A new shapefile of Hotzones is generated. The demonstration here also shows the number of traffic crashes in each identified hotzone.

Results

In this project, we utilise the Hong Kong traffic road crash data in 2017. The locations of road crashes are geovalidated before the analysis. All results are generated by the Hotzone Generation Add-in.

Traffic crash hotzones in 2017

When the threshold is set to be 3 traffic crashes or above in each basic spatial unit (BSU), and with each BSU of 100 meter, there was a total of 228 hotzones in 2017.

The number of traffic crashes in the hotzones ranges from 6 to 106, with a mean of 13 crashes in each hotzone.
These hotzones accounted for 61.8 km of the roads in Hong Kong.
The highest crash density is 17.5 crashes per km. This is located at Tai Mei Tuk Road.

Relationship of traffic crash hotzones and neighbourhood characteristics

The overall picture of the hotzones (3 crashes per BSU) is displayed on the map in association with population density. Darker colour represents higher population density.

As shown on the map, some traffic hotzones are located in areas of very low population density. These are the key findings and illustrate the benefits of using the hotzone methodology. Indeed, some hotzones are on highways and road sections out of the intersections. This is where more road safety strategies can be implemented to tackle the problem.

A simple correlation analysis also suggests some relationships between hotzone density (length per km2) and the neighbourhood attributes. Some variables are observed with significant association (p<0.01).

Road density = 0.621 **
Junction density = 0.492 **
Population density = 0.652**
Employment density = 0.508**
Percentage of elderly = 0.421**

The hotzone density is positively correlated to road density, junction density, population density and employment density. It is noted that the percentage of elderly in the neighbourhood also illustrates a positive correlation with the hotzone density.

The most dangerous traffic crash hotzones in 2017

To highlight road sections of the most serious traffic crashes, we use a threshold of 6 traffic crashes or above in each basic spatial unit (BSU), and with each BSU of 100 meter. Overall, the results show a total of 14 hotzones.

In order to understand the road-specific features of these hotzones, the next section will outline the environmental features of these hotzones and suggest several feasible countermeasures to improve traffic safety.

What the major groups of hotzones based on the envrionmental characteristics?

1) Highways and major trunk roads

Several traffic crash hotzones are identified along highways and major trunk roads.

There are several characteristics of these road sections:

Relatively high traffic volume
Some trunk roads are of high travel speed

The identified hotzones are shown below.

**Fanling Highway** (Google Street View)

**Lung Cheung Road** (Google Street View)

**Kwun Tong Road (near Choi Hung)** (Google Street View)

There are several feasible countermeasures to reduce the number of traffic crashes:

For urban trunk roads

Enhance road user awareness by installing more warning signs ahead of lane changing
Install centreline rumble strips
Reduce speed limit on sections of the road with difficult roadway geometry over a substantial length
Increase public awareness of road accidents by undertaking campaigns of good practices on roads (signs on highways such as the dangers of using cell phones while driving, etc.)

For suburban highways

Create more clear zones and improve existing ones
Ensure sufficient sight distance of dangerous sections of the road
Enhance road user awareness by installing more warning signs ahead of sharp curves
Install additional lighting before, in, and after intersections
Install more barrier reflectors, at smaller intervals, over dangerous road sections

2) Around road junctions

Several traffic crash hotzones are identified around the road junctions. There are several features of these hotzones.

Some of the hotzones around road junctions are observed with high volume of pedestrian flow and a moderate level of traffic flow.
Pedestrian-vehicular conflicts are pervasive in these areas.
Some hotzones are without pedestrian crossing facilities but with higher traffic flow.

**Nathan Road - Mong Kok Road** (Google Street View)

**Yen Chow Street - Un Chau Street** (Google Street View)

**Chatham South Road** (Google Street View)

**Pui Ching Road near Ho Man Tin** (Google Street View)

These road sections around junctions require further investigation. There are several feasible countermeasures to reduce the number of traffic crashes:

Enhance road user awareness by installing more intersection warning signs
Enhance road user awareness by installing more warning signs ahead of stops
Construct clear and enlarged road signs on and above the passageways
Improve junction design based to reduce speed (e.g. rear control, rumble strips and markings)
Improve the overall visibility (i.e. blind spots, vegetation cover and lighting conditions)
Construct tunnels and bridges to reduce pedestrian-vehicular conflicts
Enforce speed reduction or low-speed driving areas
Conduct targeted speed enforcement

3) Urban and suburban roads mainly with vehicular traffic

Several traffic crash hotzones are identified along urban roads and suburban roads with heavy vehicular traffic flow. There are several features of these hotzones.

Multiple intersections (including the unsignalised) along these roads are observed.
There is a mixture of urban and suburban roads.
Lane-changing instruction might not be clear in these locations.

**Kwun Tong Road near Kwun Tong MTR** (Google Street View) (Google Street View)

There are several feasible countermeasures to reduce the number of traffic crashes:

Urban roads

Improve junction design based to reduce speed (e.g. rear control, rumble strips and markings)
Improve the overall visibility (i.e. blind spots, vegetation cover and lighting conditions)
Develop roadblocks to check on risky driving behaviour (e.g. speeding, drink-driving, drug driving and text-driving)
Conduct road safety campaigns to high-risk groups such as motorcyclists and young drivers

Suburban roads

Construct warning signs before the dangerous road sections ahead
Develop clear signage of speed limit
Introduce visual aids to warn the dangerous locations (e.g. curve marking and transversal white lines)
Construct a lamppost near the intersection between two rural roads
Invest in protection infrastructure (e.g. crash barriers)
Improve the overall visibility (i.e. vegetation cover and lighting)
Reduce the number of unsignalised intersections

Conclusion

To summarise, this project helps answer the four research questions:

Where are the most dangerous road sections in Hong Kong? There are different types of dangerous traffic hot zones in Hong Kong, including (i) highways and trunk roads, (ii) around road junctions and (iii) urban and suburban roads maily with vehicular traffic. Some of these locations are not detected by the hotspot methodology before.
How can we use GIS techniques to develop a systematic and consistent methodology of identifying dangerous locations? The Hotzone Generation Add-in integrates the three-step approach and provides an efficient and consistent analysis of the traffic crashes in Hong Kong. Specific road sections with a high number of traffic crashes can be identified.
What are the environmental features of the identified road sections? What feasible countermeasures can be implemented to reduce the traffic crashes of the identified road locations? A wide range of environmental features of the identified road sections are observed, including roads of heavy traffic flow and fast speed with a few intersections, roads with a high level of pedestrian-vehicular conflicts and some roads of high junction density. This suggests that place-specific / road-based analysis helps policy makers to devise more effective countermeasures to improve road safety. A combination of feasible engineering and enforcement measures are suggested. Future research can further improve the accuracy by incorporating traffic crashes of different years.