Armchair air crash investigator

Given that the most common cause of aircraft accidents (by quite some margin) is human error, I'll take my chances with the computer.

The data points are in the neighborhood of 15 seconds apart. There are 4 points labeled 2:25PM, 3 points labeled 2:26PM, and 5 points labeled 2:27PM.

The altitude data is all rounded to the nearest 100ft. Obviously this is coming from transponder data. The mode C transponder carried by an airliner reports pressure altitude back to the interrogating radar at 100ft intervals.

Vertical velocity is all multiples of 60fpm. I have no idea where that would be coming from. As far as I know, vertical velocity is not transmitted by either radar transponders nor the newer ADS-B position broadcasting equipment. Simply taking the derivative of reported altitude wouldn't produce these results, because with 100ft altitude intervals and 15s reporting intervals, that gives you a vertical velocity resolution of only 400fpm.

Let's skip that question for the moment. The raw altitude data does show an increase from 100ft to 200ft at the end. But what conclusion can we draw from this? Essentially nothing. Because the altitude is so quantized, this could in theory indicate anything between a 0ft and 200ft increase in altitude, non-inclusive. Of course, the instruments are not that precise, so in practice the range is even larger. This is especially true given that the instrument was in the middle of a fairly violent plane crash when it gave that last reading. From the video of the crash, it's clear the plane never got very high above the runway. Since the runway is only 13ft above sea level, and the plane didn't crash into the ocean, the amount of climb possible is very much limited. The transponder may well have increased in altitude a bit for this last reading, but if so, the alignment with the reported altitude increase is pretty much just coincidental.

Now back to the vertical velocity. Where did it come from? I can only speculate that they're applying some kind of curve-fitting or smoothing to transform the coarse altitude data into smoother vertical velocity figures. These numbers are therefore even less reliable than the altitude figures, and can probably just be ignored.

Just to briefly support that last bit, there are some obvious numerical anomalies in the vertical velocity figures earlier in the flight. At 9:38AM, the data source briefly switches from "FlightAware Transoceanic" to "Oakland Oceanic" and then back again. The two Oakland data points show an altitude of 37,000ft, while the FlightAware data shows 34,000ft. The vertical velocity for these four points is shown as 2280fpm, 9960fpm, -4320fpm, and -1800fpm. These changes obviously didn't actually happen (otherwise we'd have heard about people being flung about the cabin in mid-flight!) and are just some sort of reporting anomaly. The ridiculous vertical velocity figures (nearly 10,000fpm climb accompanied by a minor increase in horizontal speed!) are obviously just some sort of post-processing going crazy trying to fit the bad altitude data.

The handling of the aircraft is a small part of the job. Autoflight systems have been capable of flying and landing since the 1960s. It is a mature technology although it does need to be monitored by humans as equipment failures do occur. I am confident a system could be developed with sufficient fail safes to remove the requirement for monitoring, and a fully automatically flown aircraft could be built tomorrow although at a higher cost than the current monitored setup.

Even if an unmonitored autoflown aircraft was developed, judgement decisions about the operation of the aircraft still need to be taken, and these are best done by a human. Examples of decisions I make:

1. The best fuel consumption maybe achieved at a higher flight level, but I elect to stay low and burn into contingency because I've heard reports of turbulence at the higher level. I judge that I probably won't need the contingency fuel later based on previous experience of the route and destination. So I give my passengers a smooth ride, which helps repeat custom for the airline.

2. I experience radio crackle whilst in cloud and elect to change my routing as I suspect we are near a developing thunderstorm that may soon become active. I have to balance the extra fuel burn of a non-optimal route versus the risks of being caught in a thunderstorm which could destroy the aircraft.

I'm not an AI expert, but it seems to me these are value judgements that are best made by a human. So the fully autoflown aircraft will still need a human in the loop at some point.

It is possible to remotely site this human, and maybe make efficiency savings by a single human controlling many aircraft but this would require a bulletproof, near realtime datalink. Again, I'm not an expert in the field, but I would suggest this is, if not technically impossible, then a lot more expensive than the cost of employing human pilots. I imagine this datalink would also be of interest to people with ill intent- no fun when you get the airborne equivalent of a rickroll.

TL;DR: A human is required in the loop, not to handle the aircraft but to make decisions. It's cheaper to have them on-board than the cost of a datalink. It's cheaper to have them trained to take manual control than the cost of an autopilot sufficiently fail safe that it can be left unmonitored.

My dream (well, the realistic one) is to get a CH-801 STOL kit plane with a diesel engine so I can live in Central Washington and get to I-5 in a reasonable amount of time, although weather probably prevents that often. That's only $150k or so, which would be saved in taxes and property cost differential several times over.