The Causes Behind the SQ321 Flight Turbulence Event
The turbulence event from Singapore Airlines SQ321 flight resulted in the death of one passenger, making it the first time in over 20 years* that a death in a commercial flight is solely attributed to turbulence. It's a stark reminder on both the dangers of nature, and the extreme level of safety that the aviation industry has managed to achieve.
* Based on data from the Aviation Safety Network. To our knowledge, the United Airlines accident in 1997 in Japan caused the last fatality solely attributed to turbulence.
What happened?
At 7:49 UTC, 21st May 2024 the SQ321 flight from London to Singapore encountered severe turbulence when cruising at 37,000 ft over Myanmar. The pilot was able to activate the seat belt sign 8 seconds before the major section of turbulence happened, likely saving many people from major injuries.
The information retrieved from the flight data recorder shows the following events:
What type of turbulence did SQ321 hit?
The SQ321 flight was likely hit by a near-cloud turbulence event coming from the thunderstorm clouds present in the area. Other sources of turbulence such as jet streams or mountain waves are unlikely since the the plane was well below the 30 degree latitude of the subtropical jet stream and the topography of the region is a flat river delta, with the Arakan mountains more than 200 km away.
Near-cloud turbulence refers to turbulence released from the body of a cloud, typically in the form of an outburst or a gravity wave. Unlike clouds, which can be spotted with the naked eye, these near-cloud events develop in clear air and thus can't be seen by the pilots.
Satellite images show thunderstorm clouds beside the area where the turbulence event occurred. Between 7:30-8:00, the clods appear to grow in size and height, indicating that they are in the initial growth stage characterized by strong upwards current. An outflow from these upwards currents might have been the cause for the turbulence event.
Despite the evidence of thunderstorm activity in the area, the size of the storms is not very big, specially when compared to the large thunderstorm cells located south-east. Why exactly a seemingly isolated, medium sized thunderstorm cloud was able to deliver such a powerful blow highlights the chaotic and unpredictable nature of this type of turbulence.
Clear-air turbulence?
The term clear-air turbulence (turbulence that you cannot see) has been used in many news outlets to characterize the type of turbulence encountered by SQ321. Despite its widespread in the aviation industry, this term is just too general, and it often results in more confusion than clarity.
For example, clear-air turbulence could be the one generated by jet streams, mountain waves (in the absence of lenticular or rotor clouds), near-cloud outbursts, morning thermals, wakes etc.. If we stick to the definition, even the turbulence behind the wake of a runner can be labeled as clear air turbulence.
The SQ321 likely encountered near-cloud turbulence, which is a type of clear-air turbulence.
Was this turbulence predicted by turbulence forecasts?
The resolution of global turbulence forecasts is currently 9-13 km, and this is not enough to resolve thunderstorm clouds, let alone the turbulence within them or the near-cloud events that they can generate.
Current turbulence forecasts predict well turbulence induced by large scale phenomena such as jet streams and mountain waves. This is known by the pilots, so they were certainly not relying on a turbulence forecast alone when navigating though the clouds of Myanmar.
What can pilots use to predict near-cloud turbulence?
Some countries have developed quick refresh forecasts that only cover up to 2 h in the future but that are very useful for tactical decision making. These combine a traditional coarse resolution turbulence forecast with radar data and live turbulence measurements from other aircraft that feed into a machine learning model that has been trained to predict near cloud turbulence events.
As of 2024, the US is the only country having an operational model serving this, the GTG-N. Europe has also developed one and it is planned to become operational in 2025.
Keeping a 20 miles distance
For large thunderstorm clouds, the Federal Aviation Administration (FAA) recommends maintaining a 20 mile (32 km) distance to it to avoid hitting any of its near-cloud turbulence. This guideline has been working exceptionally well so far, but there is also plenty of evidence that moderate or greater turbulence can extend well outside this value, even up to 30 miles. The SQ321 flight happened to hit one of them.
Radar measurements are very useful to determine the clouds movements, sizes, and the path needed to keep the 20 miles separation. However, not all countries are covered by radar since it's an expensive infrastructure to maintain. The area where the SQ321 incident happen was in fact not covered by ground radar.
If ground radar is unavailable, pilots can always look at their own onboard radar, located at the nose of the plane. This radar is constantly scanning the area in front of them and providing high resolution maps, but it cannot give the full picture of the thunderstorm circulation pattern surrounding the area.
Satellite data
Satellite data is also very useful to track storm development in real time. Using sensors at different infrared frequencies, they can track the cloud water vapor content at low, mid and upper altitudes. However, this information is still far from what radar can deliver, since these can give a complete 3D picture of the cloud.
Machine learning has also been applied to satellite images with the aim of predicting near-cloud turbulence, both in the research domain and operationally by weather agencies. Still, the prediction skill is not very good (particularly at high altitudes) due to the low vertical resolution of the satellite data.
High resolution models
Regional models run the weather agencies of a country tend to use a much finer resolution of 2-3 km that allows them to capture thunderstorm development better. They are known as “convective-permitting” models, and this a good name since it conveys exactly what the models do: they allow predicting clouds formations, but they are still far from the ~100 m resolution needed to actually resolve them.
Another important feature of these models is that they are non-hydrostatic. Global weather forecasts use a simplified equation of motion for the vertical direction to save computational time at the expense of a negligible loss of accuracy: they do not bother trying to resolve what they could not resolve from the beginning due to their course resolution. On the other hand, high resolution forecast can capture more detailed physics, so they go with the full equations of motion.
In the US, the high resolution HRRR and RRFS models do much better than the coarser GFS in predicting thunderstorm activity. Nevertheless, they still struggle to predict the correct initiation time and final size of the storm.
Unfortunately, the power of regional forecasts has not yet been transferred to turbulence prediction for cloud turbulence. We might see developments of this in the near future.
Bottom line
The SQ321 turbulence event is a very unfortunate, but also extremely rare. Given the strict safety standards of the aviation industry, it will likely accelerate the already ongoing development of models to better predict convective turbulence.