Even here, where the moon is hidden

vivianimbriotis | June 13, 2022, 1:57 p.m.

We know the land

is disappearing beneath

the sea, islands swallowed

like prehistoric fish.

We know we are doomed,

done for, damned, and still

the light reaches us, falls

on our shoulders even now,

even here where the moon is

hidden from us, even though

the stars are so far away.

-   From Evening, by Dorianne Laux, 2019

In Tasmania, Australia, there is a mid-twenties man named Viv, who sporadically makes blog posts that no one reads. There is also a weather station on Ellerslie Road, which has been in operation since 1882. In Australia, there is a Bureau of Meteorology, that supplies public datasets for free to anyone who is interested.

Delightful that the study of weather,

From metéōros, “A thing from on high”,

Should now have overtaken the meteor

As the existential threat to our mind’s eye

As the angel spreads its feathered

Wings across the sky

And blows asteroids and CO2

And bleached coral, mosquitoes, march flies.

There is something sacred about this data. It predates computing. It is temperature measurement, predating the electric thermometer. Fahrenheit invented the mercury thermometer in 1714; 1882 is roughly halfway between then and now. Almost 150 years of data; about three quarters of the entire history of Australia’s colonization. It predates the University of Tasmania by seven years.

In 1880, the first issue of Science was published, backed by Thomas Edison. Ned Kelly was captured and hanged. Venn diagrams were invented. In 1881, Graham Bell and Edison formed the Oriental Telephone company, and Garfield is elected and then assassinated. In 1882, on April 1st, the maximum air temperature in Ellerslie Road, Tasmania was 20.6 degrees Celsius. This information is in a free CSV that you can download right now from the Bureau of Meteorology’s website.

What does the data look like?

The most obvious thing is the missing data. I wonder what happened in the 1910s? A whole decade of weather data gone, or maybe never collected. The metadata from the Bureau has this to say: Very few stations have a complete unbroken record of climate information...It is important to note that damaged instrumentation can result in gaps in the record…[and] the absence or illness of an observer may result in gaps in the record for all elements at a particular site.

This data is a sacrament and entangled in it is the lives (and illnesses, absences, and human stories) of the people who tirelessly collected, recorded, and preserved it. 

Looking at a scatterplot conceals the seasonal trends, which we can see as a sawtooth pattern on a line plot:

Hard to make out maybe, but zooming in shows a clear seasonal trend. You can also see that there's more temperature variability during the hotter months.

I built a very simple linear model – since February is the hottest month in Tasmania, I build the model with two coefficients, a linear time trend (showing how the average temperature has changed over the years) and a sine wave, fixed to have its antinodes (i.e. the highest/lowest part of the wave) in February and July. This sine wave will capture the variability in temperature due to the season. I then used the OLS estimators to get the best-fitting amplitude of the wave and of the linear trend. 

 Let's see it!

Zooming in shows that our model’s predictions pass the sniff test (i.e. they do seem to include about 19 out of every 20 observations).

So the temperature has risen, on average, by about 1 degrees Celsius since 1880; the difference between summer and winter temperatures is, on average, about 8 degrees. 

This model is pretty good. Its adjusted R² statistic (roughly speaking, the proportion of the variability in temperature that it can explain away) is 0.455. Admittedly, almost all of that is from the seasonal variation (the model with just the sine wave has an R² of 0.452), but this model is better (with high certainty).

Anyway, I guess there are two points to all this. 

The first is that you can find compelling evidence of global warming in your own back yard in about twenty minutes.

The second is that data lives and breathes and, sometimes, speaks.