Bermuda Triangle : a case of anomaly or sensationalism?
Bermuda Triangle is infamous for its high-profile disappearance events. But disputes arise as to whether the frequency of these events actually constitute a statistical anomaly.
How to answer this phenomenon? We examined the aviation and marine casualty data over the last few decades to make comparisons between this region and elsewhere in the globe.
Marine Casualties: The data available online are mostly fragmented by regions or have limited timespan.
The most comprehensive one we found relevant to this analysis was the Marine Casualty & Pollution Data for Researchers dataset provided by the United States Coast Guard (150,000+ records, spans 1982-2015).
This dataset reports only investigations pursued by the Coast Guard pertaining to US-related activities. Which is a source of bias. But it remains suitable for our usage. Since it covers the Bermuda region as well as other oceanic regions around the globe.
We pre processed the dataset. We are omitting some records due to the presence of typos. We are formatting mismatch and missing data. The final result is shown below.
Figure 1 Marine Casualty & Pollution Data for Researchers, all reported incidents within the Bermuda Triangle (red) and elsewhere (green). Disappearance incidents are marked with their last known locations.
It is apparent from the figure above that the reported incidents are clustered around coastal areas. (further illustrated in Fig. 2) As they account for the majority of collision events as well as accidents that occur while disembarked.
Since we are primarily interested in studying the conditions of ocean regions, only incidents that happen at least 80 miles (80th percentile and above) from the nearest coast are considered.
Figure 2 The distance to land frequency graph shows a quickly decaying pattern. Generated using land geometry data provided by the geopandas package.
All in all, 9096 incidents are considered, 361 of which fall within the Triangle (here we define it as the area spanned by Miami, Bermuda and San Juan, Puerto Rico).
Qualitatively, we can detect the presence of unusual activity. We compare the causes of casualties that occur within the Triangle and elsewhere.
As shown in Fig 3. The types of incidents in this region mostly resembles the worldwide pattern. With an expected lower share of capsizing events. This is possibly due to its proximity to major trading ports.
Interestingly, there was no incidence classified under ‘Disappearance’ reported within the Bermuda Triangle from 1982-2015.
Figure 3 Cause of casualties in Bermuda Triangle (left) and elsewhere (right)
Furthermore, we can determine whether the frequency of incidents within this region falls within an expected range.
To this end, kernel density estimation is utilized to model the underlying distribution of the observed data (shown in Fig. 4). This allows us to visualize the density of data points around the globe and identify regions with particularly high rates of marine incidents.
Figure 4 Kernel Density Estimation plot, with parameters chosen to minimize cross-validation error
The KDE plot indicates that naval incidents are particularly concentrated around three regions. The Gulf of Mexico, the Gulf of Alaska and the north-western part of the Pacific Ocean.
The Bermuda Triangle is situated on the eastern edge of the Gulf of Mexico. Which could imply that the Triangle itself is not an isolated region of unusual activity. But rather characteristic of the marine region around southeastern US.
What remains unanswered however, is whether this concentration of incidents is due to high marine traffic? Or is it due to the severity of the weather conditions.
To discern between these factors, the next step is to integrate data on global shipping traffic into the analysis.
Unfortunately, both the governmental and private data sources (Offshore Waters Vessel Traffic Data, MarineTraffic.com) are hidden behind paywall. And cannot be presently obtained.
We can however still utilize meteorological data to infer the influence of weather on incidents within the Triangle.
The chart below illustrates the seasonal effect of tropical storms on capsizing and collision events – both of which are commonly cited as explanations for missing ships. With a Pearson correlation score of 0.71 (p<0.01), the marine incidents do indeed exhibit a clear correlation with adverse weather conditions, giving some credibility to the explanation.
Figure 5 Number of tropical storms affecting the Bermuda island (bar) compared with a subset of marine incidents (line)
Aviation Casualties: With notorious incidents such as the disappearance of the Flight 19 bombers and multiple passenger aircrafts.
Aviation incidents in the Bermuda Triangle deserve a closer inspection.
We found the Aviation Accident Database provided by the National Transportation Safety Board to be suitable for this task. As it covers all investigated accidents. Both by US authorities from 1962 up to present. And it is inclusive of both commercial and military flights.
The low rates of incidents within the Triangle during this period presents difficulty in drawing any claims based on quantitative comparisons.
As listed in the table below. Of the 53 incidents recorded within its boundaries, the vast majority occurs during landing/take-off (particularly in airports in the Bahamas). Thus are not directly linked to our investigation.
Of the 11 incidents that occur during the journey, only 5 lead to fatalities. 2 are missing. Both of which are personal flights by a single pilot.
The only documented commercial flight accident during this period is attributed to the pilot’s poor judgement of the weather conditions.
Bermuda Triangle Elsewhere Notes
Total Recorded 53 84,216
Take-off/Landing 42 64,471
Cruising/Others 11 19,745
Personal Flights 5 11,625 Cruising/Others Fatal Incidents 5 6,548 Cruising/Others
Table 1 Statistics of incidents occurring within the Bermuda Triangle and elsewhere (landing and takeoff excluded),
Conclusion: Combining the insights from analyzing marine and aviation casualties from the past few decades, we can reasonably conclude that the Bermuda Triangle is not a particularly anomalous region.
The notable incidents were all reported during the pre-radar and satellite weather tracking era, hence it is very likely that capsizing events and flight accidents were simply not communicated, and their wreckage swept away by the currents.
Of course, the authenticity of publicly available data is itself an uncertainty, and further steps can be taken to cross-validate with alternative data such as marine/aviation traffic density, accident reports and trends in the media coverage to detect any incoherence.
Additionally, this analysis reveals that cluster regions such as the Gulf of Alaska and the north western Pacific Ocean deserve closer examination due to their evident high number of incidents despite being situated in open waters.
It could be fruitful to apply data science to analyze marine/aviation casualty causes in conjunction with meteorological data to find ways to reduce these fatality events.
Written by Jin Hong Kuan
Edited by Jeremy Knopp & Alexander Fleiss