Since the beginning of the COVID-19 outbreak, governments worldwide have implemented strong measures to counteract its spread. In Dubai, these measures have included social distancing (#StayHome), lockdowns and curfews. Meanwhile, many businesses observing the impact of COVID-19 in countries such as Italy, Spain, Iran and the UK implemented work-from-home opportunities even before these government initiatives began.
COVID-19 has therefore had a substantial impact on transport behavior across Dubai. It is generally understood that the outbreak has decreased public transport trips, lowered traffic accidents, and improved air quality – this paper seeks to explore these assumptions using and analyzing open data related to Dubai. This paper details, at a high-level, both the technical elements used to obtain, process and output data results, but more importantly it focuses on the precise insights this data uncovers and how this can help cities like Dubai plan for the future.
Methodology
To test the above hypotheses, our data science team at Surface Mobility Consultants downloaded disparate open datasets available from Dubai Pulse and Dubai Municipality to visualize and analyze what (if any) affects COVID-19 has had on transportation, accidents and air quality within Dubai.
Datasets used in the analysis were as follows:
Whilst Public Transport data was available at HH:MM:SS level, this was deemed too granular for the final output – especially if the end visualization were to display 3+ months of data. It was decided that a daily level would provide adequate comparative trend analysis from which users could see overall changes in public transport behavior in light of COVID-19 events.
The above datasets totaled 300 csv files with a cumulative data size of around 40GB – presenting a challenge since our data science team are working from home on laptops with limited processing resources. It was clear from the outset that using Python Pandas was not an appropriate option due to the above limitations and so Spark was used instead.
Spark is built on top of the Scala language and has been successfully used for Big Data processing across many interfaces including Python, Scala, Java, R and SQL. It enables partitioning of data, lazy evaluation and distributed computing, and allows for very large datasets to be processed column-wise as well as row-wise within a single node or cluster. Through Spark, 40GB of data was processed in just 1.5 hours with normal CPU resources.
Once data was prepared, it was loaded into MicroStrategy Desktop – a freely available dashboarding tool, which allows users to join, wrangle, and visualize their data onto a user-friendly interface. MicroStrategy has collaborated with ESRI in order to provide map-based visualizations to users, thus presenting a fantastic, free means of visualizing our data.
A further visualization was then created using Kepler – a web interface created by Uber which enables easy data visualization and is particularly appropriate for displaying geographic data.
Results
The end MicroStrategy visualization consisted of a single dashboard with four user-configurable views: (1) Incidents Analysis, (2) Public Transport Analysis, (3) Check-In and Check-Out Analysis, (4) 2019 vs 2020 Trips Analysis.
Each of these views allow users to select dates for analysis and day of week (i.e. Sunday – Saturday). Users can select locations for analysis which includes Dubai communities, public transport stops and can even select individual transport modes (i.e. Bus, Metro, Tram, Marine) for analysis.
The dashboards are fully dynamic – selecting a date or location from one of the visualization widgets would prompt the other widgets to change accordingly, allowing easy configuration and analysis. Outputs and insights from these dashboards are explored below.
Traffic Incidents and Air Quality
The ‘Incidents Analysis’ view maps traffic incidents and provides a trend analysis of traffic incidents and air quality, enabling users to view whether there has been an overall increase or decrease in accidents and whether there has been an overall improvement in air quality (measured by O3 and PM2.5).
It can be seen that there was a slight drop in road traffic accidents starting from around the 23rd March – this is the date that the UAE government began its #StayHome campaign encouraging people to socially distance and only leave their homes for essential purposes. However, the rate of incidents did not consistently decrease – rather, levels fluctuated between 21 and 121 on any given day between 23rd March and 8th April.
It is interesting to note that air quality did not demonstrably improve – PM2.5 levels in January 2020 were at an ‘Unhealthy’ level at 65.41 micrograms and increased in April 2020 to 91.77 micrograms, meanwhile Ozone (O3) levels remained largely consistent at a ‘Good’ level around 20 ppb. It is likely that a substantial decrease in these levels was not observed since private vehicles only produce a proportion of air pollutants – other sources being power stations, oil/gas refineries, cement manufacturing, constructing/demolition activities, quarries, and diesel generators. Furthermore, HGVs and other commercial vehicles have remained in operation. Lastly, it should be noted that Dubai is an emirate built within the desert – as such, dust storms which naturally occur between February and March have also contributed towards the elevated PM2.5 levels.
Public Transport
When analyzing the 2020 public transport data, it became apparent that traveler check-in behaviors did not noticeably change until around the 7th March 2020, with significant changes only starting from 21st March 2020.
It is worth noting that 8th March 2020 was the date on which school closures were announced in Dubai and Italy was placed under lockdown. Between the 22nd and 25th March 2020, Dubai had begun its sterilization and #StayHome campaigns, Malls had been shut across the UAE, and Emirates airlines had announced that flights would be cancelled. By 26th March 2020, there was a night curfew imposed, meaning that only those with a permit would be allowed to travel between the hours of 8pm and 6am, causing a further drop in customer check-ins. The Metro services did not officially close-down, however, until the 29th March 2020 by which point customer check-ins were at an all-time low.
Nonetheless, following the 29th March, public transport check-ins again started to increase – this could be due to the increasing number of COVID-19 recoveries (52 recoveries) at this point in proportion to the relatively low death count (2 deaths).
Whilst the closure of Metro could have initiated an uptake of bus ridership, this did not appear to be the case – with numbers remaining low for bus usage throughout the end of March 2020.
When comparing the check-ins for 2020 against the same month and day for 2019, it was apparent that despite the COVID-19 outbreak, public transport trips until 16th March 2020 were still generally higher in 2020 than during 2019 (accounting for the offset weekends between the two years). Indeed, the only significant drop before the government shutdown of the Metro and Tram was on the 28th February 2020.
A significant decrease in ridership was not observed until around the 17th March 2020 – the date on which the first two deaths from COVID-19 were announced in the UAE. Further declines in check-ins can be attributed to government #StayHome initiative, nighttime curfews, and the eventual closure of the metro and tram.
Using Kepler’s data-in-motion capabilities, we observed traveler patterns over the three-month period. As expected, the most substantial changes could be seen in Dubai’s central locations – along Sheikh Zayed Road, Dubai Marina, Downtown Dubai and Business Bay.
Conclusion
Prior to government intervention, little change was observed with regards to public transportation behavior and the number of traffic incidents. Indeed, the most significant changes in public transportation behavior happened after the shut-down of the Metro and Tram and traffic incidents only marginally decreased towards the end of March, possibly due to increasing legal restrictions on public travel.
Concerning the minimal decline in accident rates, this could possibly be attributed to more irrational driving from motorists on roads which were previously congested and typically low speed. Many of these roads are now uncongested (particularly during AM and PM peaks) allowing much higher speeds. Further analysis of the accident data would be required to validate such a theory. Both of these insights highlight the importance of government authorities in slowing the spread of COVID-19.
Whilst much public discourse has been made concerning the benefits of lockdown on air quality, PM2.5 and O3 levels stayed largely consistent with PM2.5 levels remaining within the ‘Unhealthy’ range. This may be attributed to a combination of factors, including Dubai’s desert climate and the continuation of industrial practices throughout the lockdown.
A PDF copy of this article also available at: Download PDF
Report written by myself and Hafiz Shehbaz Ali
Stephanie Robinson
A classically trained musician, who found similarities between playing the piano and coding and turned her attention to data science.