Eleven Years Of Watching The Sky. It Was The Humidity.

ITHACA, N.Y. — Ray Pell has stood on the roof of this building at 06:00 and 12:00 for eleven years and written down how long the trails last.

44,102 observations. Time, bearing, estimated flight level, and persistence in minutes, in pencil, in eleven notebooks, transcribed each Friday.

We gave the whole thing to somebody who had never heard of us.

The question

Persistence varies enormously. Some days a trail is gone in twenty seconds. Some days the sky is a lattice at noon that is still there at four.

That variation is the observable fact at the centre of every claim about atmospheric dispersal, and this desk has been recording it for longer than anybody has been asking us about it.

The question is what predicts it.

The analysis

Dr. Yelena Marchetti-Obuya was paid $6,200, at her standard rate, to answer that. She was given the observations, the matched radiosonde soundings, and no indication whatsoever of what the desk thought the answer was.

She tested nineteen candidate variables: relative humidity at flight level, temperature, pressure, wind shear, time of day, month, bearing, estimated altitude, day of week, aircraft count, and ten interactions between them.

One variable does the work.

Relative humidity with respect to ice explains 71 per cent of the variance on its own. Adding temperature and shear takes the full model to 0.78 and nothing else moves it.

Day of week — the variable that would have to matter if persistence were driven by a schedule rather than by weather — explains 0.2 per cent, which is indistinguishable from nothing across 44,000 observations.

What she said about that

Her report contains one paragraph that is not statistics, and this newspaper asked her permission to lead the story with it. She declined, and asked that it appear where she put it, so it is here:

I want to be careful about what this shows. It shows that in this data set, from this location, persistence is nearly entirely explained by whether the air at flight level is supersaturated with respect to ice. That is the textbook mechanism and I would have been extremely surprised by any other result. It does not prove anything about anyone's intentions. It shows that you do not need to invoke intentions: the sky already accounts for the variation. If somebody were flying a programme, they would have to be flying it on exactly the days the atmosphere was going to hold ice anyway, which is a strange programme.

The dispersal index

Pell's own nine-point dispersal index, which this desk has published daily since 2011, correlates with the humidity figure at 0.66.

That is a good correlation. It means the index is, in effect, a crude and manually-observed humidity proxy — which is a real thing, and is not the thing the index has spent fifteen years implying it is.

The index moved out of the forecast box last Monday, into its own panel, after a reader in Binghamton made an argument this desk could not answer. This analysis is the second half of that answer.

"I built a humidity gauge and called it something else for fifteen years," Pell said. "I would like that sentence in the story, please, and I would like it in the same size as everything else."

The data

The complete observation set has been downloadable from this site since 2018. It stays up. Dr. Marchetti-Obuya's code and report are published alongside it, unaltered, including the two places where she is critical of how the observations were recorded.