Introduction

Since 2010, there have been over 60,000 laser incidents involving planes in the United States and territories according to the Federal Aviation Administration (FAA). Laser strikes, or “lasing” has become increasingly prevalent in the United States given the ease of access to lasers and the power that exists in a sub-$20 device. These devices can cause vision loss when pointed directly at the human eye and can have destructive consequences when aimed haphazardly. Lasing is a threat to aviation safety and is a federal crime. The Federal Bureau of Investigation (FBI) reports, “When aimed at an aircraft from the ground, the powerful beam of light from a handheld laser can travel more than a mile and illuminate a cockpit, disorienting and temporarily blinding pilots. Those who have been subject to such attacks have described them as the equivalent of a camera flash going off in a pitch black car at night.” (For more information, watch this short FBI video on lasing.

Given the potential for serious danger from “lasing” attacks, the FBI and FAA work collaboratively to mitigate these attacks. As part of their mitigation effort, the FBI and FAA encourage the public to report “lasing” incidents and make the reported data available on the FAA website. I decided to explore the ‘Reported Laser Incidents for 2022’ dataset in Tableau Public to see what trends might exist within the sphere of “lasing.”

My Questions of Interest:

1. Under which conditions is a laser incident likely to occur in the United States?
   
2. Under which conditions is injury from a laser incident is more likely to be observed?

Methodology

All of my visualizations were created in Tableau Public as I wanted to use a visualization tool where I could easily collaborate and share my work with others. From a data perspective, I used the Reported Laser Incidents for 2022 Dataset available on the FAA website, which currently contains data from Jan 1, 2022 – May 31, 2022 in .xlsx format. (For those interested, I became aware of this dataset through the weekly Data is Plural newsletter.) The FAA has some existing Tableau Public visualizations that were useful in designing my own dashboard.

The following fields were included in the FAA laser incident dataset:

Field	Data Example
Date of the Incident (MM/DD/YY)	9-May-22
Time of the Incident (HHMM)	0355
Flight ID	09Q
Aircraft	C172
Altitude (in feet)	2500
Airport	VPS
Laser Color	Green
Injury (Y/N/Unknown)	No
State	Honolulu
City	Honolulu

I chose to limit my data cleaning to Tableau Public in the spirit of reproducibility and data integrity. While I can clean data more confidently and quickly in Excel, the lack of an audit history would make it difficult to understand exactly which permutations were made to the data in the event I wanted to recreate this visualization in the future. Furthermore, I want my visualizations to continue working as more data is introduced for the remaining months of 2022. Adding in a manual Excel cleaning step would make this more difficult.   

Importing the data was largely straightforward. Prior to pulling it into Tableau, I did a quick Excel scan to make sure the data was in a tidy format. There were some blank values in the FlightID and AircraftID columns, but I chose to leave them as-is given there were 10 blanks out of 3,000+ total entries.

Only one category required grouping – the laser color. Some of the values were straightforward (e.g., blue, green) while others were more challenging to deal with (e.g., continuous green, white and red and blue). I ultimately made six groups for the laser colors based on the frequency of the values as well as the type of color being captured. The final groups were: Green, Blue, Unknown, Purple or Red, Multi-Colored, and Yellow or White. While I considered grouping Purple or Red and Yellow or White together as ‘Alternate Single Color’, I recognize that when data for the rest of the year is added there might be meaningful variation that should be reflected.

I also chose to incorporate some binning for the quantitative altitude variable as there were a large variety of values that made the spread of the data more challenging to understand. For the variable of altitude, there wasn’t a huge loss of an information when similar values were grouped together, so I moved forward with this approach to improve the interpretability of the visualization.

I chose to make a simplistic dashboard with a blue-green color scheme to reflect the prevalence of these laser colors. After looking through all of the fields available, I identified five which were important to make available interactively as visualizations on the dashboard.

1. Laser Color
   
   This stacked bar visualization uses normalized percentages to understand which laser colors appear most frequently across incidences. Given the majority of incidents involved a green laser, colored percentages in a stacked bar felt more appropriate compared to a pie chart or side-by-side bar graph. The color of the bar matches the color of the laser for clarity.
   
2. Incident Frequency by Weekday
   
   This line graph uses the date time series field in order to understand if there are certain days of the week that are more (or less) prone to laser incidents. Given this was a partial 2022 dataset, a weekday analysis felt more appropriate compared to a day-by-day or month-by-month visualization as in-progress trends could be misleading.
   
3. Incident Location (State View)
   
   This choropleth map visualization allows users to explore the prevalence of laser incidents in specific states and also filter the rest of the dashboard by selecting a specific state or region.
   
4. Flight Altitude during Incident
   
   This histogram visualization enables one to understand whether or not there is a relationship between flight altitude and the occurrence of a laser incident.
   
5. Injury status
   
   This filter field notes whether or not an injury occurred as a result of the incident and is key to answering the second research question re: conditions leading to an injury.
   

Results & Interpretation

Some learnings from the dashboard were as follows:

• Most laser attacks occur at lower altitudes, with ~77% occurring under 12,000 ft
• Green lasers are responsible for over 80% of the incidents, with blue being responsible for 11% of incidents.
• There is not dramatic variation across days of the week for total # of incidents. Largest gap was 131 incidents between Saturday (593) and Monday (462). Unclear if this is because there are a lot of flights on a Saturday and relatively fewer on a Monday.
• California had the most incidents thus far in 2022 with 617. Texas and Washington followed with 383 and 265 incidents respectively.’
• Only 10 incidents resulted in injury. Of the 10 incidents, 8 of them occurred at an altitude below 4,000 feet and 9 of them involved a green laser. These were not concentrated to one specific state six states had 1-2 laser incidents resulting in injury.

It appears that most laser attacks occur at relatively low altitudes with green lasers, with California, Texas and Washington accounting for ~35% of all cases. This trend held true for the incidents resulting in injury as well.

Reflection

While I developed a stronger understanding of the conditions in which laser attacks are most likely to occur, I realize that I am limited to observing trends, rather than suggesting causation. These trends are valuable to understand and were made clear in my dashboard design, but they fail to provide anything conclusive. Perhaps green laser attacks are reported more because green is an easier color to perceive as a laser vs. red or blue. This data is unable to tell me whether or not green is more dangerous, just that when injury occurred, a green laser was involved 9 / 10 times.

More specifically, I don’t think that I had a sufficient volume of data to answer my second question of “when an injury is recorded, what are the conditions?” We can partially answer the question using cognitive science – a green laser can emit significant levels of green light and cause significant eye damage to humans and animals. Furthermore, when reflected off of an optical surface (e.g. car window, mirror) a laser can do a similar amount of damage as pointing it in one’s eyes. But with only 10 responses in which injury was reported, we fail to meet an n = 30 threshold for statistically significant findings. When 2022 concludes there may be enough data for a more meaningful exploration into laser injury trends to occur, but in the current state we cannot be confident in the results.

In a future version of this dashboard, I would like to expand my dataset both longitudinally, as well as through the incorporation of additional related fields. It would be fascinating if there was a dataset focused on the perpetrators of laser attacks. I’m very curious as to whether or not these attacks are intentionally harmful, or if many of them are children playing innocuously with a laser and creating danger unintentionally. State population would be another interesting overlay to see if the frequency of incidents is really a result of population size rather than something unique about different states. Flight frequency might also be valuable here – do California / Texas / Washington have more flights than the rest of the country and thereby more incidents? This is all unknown given the limitations of my existing dataset.

Finally, from a design perspective it would be great to identify a way to make the names of my filters applied to appear dynamically on the dashboard for clarity. Specifically for the state view, I think it’s challenging to know exactly what population one is analyzing when selecting multiple states and/or altitudes to analyze. This enhancement would greatly reduce the prevalence of reporting errors and make the dashboard more valuable to end users.