I use them because they directly measure temperature, pure and simple.
Proxies like tree-rings, while allowing for longer duration (since some trees have existed longer than the oldest thermometers) have confounding factors: the amount of rain, the availability of nutrients, other local ecological stresses, etc.
What I see is a varied rising or falling overall, depending on the location. If anything, it is fascinating to see some areas’ cooling trends amidst others trending warmer. I live in Redlands, not far from PS, and I think it’s getting warmer, but I know for a fact two things: it was always hot from time to time (childhood here in the 60s), a solid 5 month season of cough-inducing smog (not mostly gone), and I’m willing to recognize I am influenced by the constant barrage of news telling me the earth is burning up.
Personally, I am suspicious of human hubris that would somehow poopoo solar influence and think we are the ne plus ultra of influencers on planetary climate.
This is very good, seeing what we can make of trustworthy, thermometer temperature measurements. Using airports I would expect the temperature to go up because over those decades we have more flights and bigger engines as well as more people activity so more heat into the surroundings. I see no clear evidence for ‘global’ warming or cooling. I don’t think a global number has meaning unless it turned out to be really hotter everywhere.
I apologize but if you don’t mind can you can Into further depth on why you used thermostats over proxies(tree rings, etc) to determine long term climate temperature?
Weather is complex. Temperature is just one indicator. What about rainfall (same rain in longer period is better for vegetation)? Wind (dries vegetation)? Humidity?
The least squares fit to the medians is heavily influenced by very few data points. My suggestion is that you use a robust regression, such as a Theil-Sen estimator. You can also do a bootstrap on this estimator to obtain a confidence interval on the (robust) slope. Finally, you can do a histogram of the (robust) slopes across the different locations.
By the way, it would be better to subtract the local average temperatures before you compute these slopes, rather than the global average, as you have done. This way, the slopes are easier to compare.
And finally, a non-statistical remark: there is clearly a bias in only using airport data for your analysis. This is because the micro climate at airports is non-representative both because of the construction materials used, as well as because the locations were carefully selected to reduce the influence of certain environmental factors. This does not discredit the analysis but it is a caveat that should at least be stated.
Sorry I miss-wrote what I actually did and confused people. I just adjusted each detailed plot (the boxplots and median regressions) by the rigidly ofsetting the whole plot based on the median temp of the entire airport's individual time series (≥75 years) . All temperature changes are still shown in ºC relative to the separate constant (local) offset per plot.
And I recently talked to the guy who curates the dataset. He answered a lot of questions. These stations are pretty carefully sited to avoid bogus data. Other temperature records are mostly from city centers, which apparently suffer from the urban heat island as population increases over time. ("simply a matter of albedo").
I tried L1 regression and LOWESS regression and then (at your suggestion) Theil-Sen. Doesn't look very different.
You say “The global median temperature (of the entire dataset) is uniformly subtracted from every data point, to keep things relative to an arbitrary x-axis of 0.” and then seem to ask why most of the cities don’t vary much from 0. What does an equivalent plot of that global median temperature show? If it is rising over this 80 year period and you are subtracting that out, then why would you expect to see big changes in the city data?
The subtraction of the median for each data series (not the "global median" FWIW) does not effect the trends, the variance, or the shapes of any of the plots. It just shifts the origin (and not the scale) of each y-axis to be centered at an arbitrary zero (zero==the median for the entire series for the particular location).
I am not sure why this has confused people, but I will attempt to alleviate this confusion "next time".
"Global median" is the wrong word - sorry. What I mean is I subtract a constant scalar value (the overall median for each ~80 year timeseries) from all the values in that timeseries for the boxplots and median zooms. It just shifts each plot to be centered around zero. It does not change anything else: The variability of the medians, the size and shape of the boxes, the linear regressions all look the same as if the respective constant was not subtracted,. It is just a relabeling of the y-axis on each plot to make the vertical center of gravity of each plot show up near "zero".
It might have been smarter to not do this, but keep them centered some other way. Some would be centered around zero, some around 20ºC, etc. The shapes of the boxes and curves, and the scales of the y-axes and the distance between the tickmarks in ºC would all be unchanged but the numbers labeling each y-axis tickmark would be different.
One big problem with "time series from around the globe" is that long-duration direct temperature records are *very* sparsely sampled. Most of them are from large, old city centers which tend to have suffered from the urban heat-island effect over time. This latter problem is less severe for airports, which tend to be at least a little outside of city centers.
> The urban heat island has had only a minor impact on estimates of global trends of LSAT. Its impact is much smaller than the 0.74◦ C global warming between 1906 and 2005. The impact is small because assiduous efforts have been made by the compilers of global surface air temperature records to avoid or compensate for urban warming. This is confirmed by analyses using only rural stations or using only days and nights with windy weather. Further support comes from comparisons with marine surface temperature trends in the light of the expected augmentation of trends over land.
Figure 2 in that paper is particularly worth a look, because it tries to summarize the uncertainty introduced by both the sparsity of historical data and urbanization.
Let me clarify: airport runway temperatures have to be actual temperatures. Not adjusted in any way. A pilot needs to know the air temperature to know if he has, literally, enough runway to take off. Hot air is less dense and provides less lift.
"The impact [of UHI] is small because assiduous efforts have been made by the compilers of global surface air temperature records to avoid or compensate for urban warming."
That may be so (it sounds a little self-referential to me) but I was looking at thermometer readings made into timeseries with - I have been assured by the curator - no adjustment.
I may try next to look at city centers vs. airports for the same cities. There are some places with records for both that go back ~80 years.
In my experience it is always good to check the data records themselves, if only to understand how the "consensus" has been achieved.
I was curious if there is a known effect specific to airport. One commenter already speculated it could be changing runway materials. Or perhaps air conditioner upgrades that increased efficiency. Or who knows. I’ll search some more but in the meantime, found this. Want to run your process on some airports in Greece? https://link.springer.com/article/10.1007/s10584-019-02634-z
You maybe missing the trees for the forest. Global or local mean or median temperature is a nebulous (no pun intended) indicator. It's the increases in daily weather variance, incomprehensibly-fast melting of major glacial masses, rising sea levels, and frequency and intensity of storms, fires, famines, and species' extinctions that are the "forest". These are consequences of anthropogenic GHGs that reduced Arctic sea ice (nearing an inevitable blue ocean event) and increased global sea thermal energy absorption which destabilized the jet stream from more predictable cycles. GHGs lead to a -11 F snowpocalypse in Austin, megadrought in US West and SW, "Hurricane Alley" shifting eastward, 100 F in the high Arctic town of Verkhoyansk, water trains of India, and slumping and melting of Siberian and other permafrost areas releasing absurd amounts of methane.
The variance (locally) is shown in the plots. What do you see re: variance?
"Incomprehensible" glacier movement? As I understand it nine time in the life of the human species the glaciers have advanced to cover much of N. America and then retreated to where there was little or no polar ice at all.
Frequency and intensity of storms? I suggest downloading the data on all hurricanes for the last ~100 years (as long as they've been recorded) from the NOAA national hurricane data center. I did that. I plotted it. Guess what I found?
Fires... I lived in Napa California during the big Napa fires. How much was due to climate change, and how much was doe to terrible forest management by the forest service and the state? (don't thin the forests, don't clear dead trees, etc).
Species diversity is very complicated. A whole different matter. So much to say at another time.
Famines? Have you looked at famines vs. time for the last 1000 years? Other than the global covid mismanagement disasters of 2020-2022 (lockdowns, etc) famines are becoming a thing of the past - though if we continue killing our soils with mono-crop soy, etc they will recur.
Arctic (and Antarctic) sea ice is another deep topic.
Austin was a fluctuation. They happen. Don't kill your local electric utilities - you may want them when things get dicey (or icy :).
Look into the history of draughts (and floods) over the last 200 years in the W and SW US.. Write a post and send me the link.
You raise a lot of points. (or less politely, you are shifting the goalposts 12 times in one paragraph). I am just plotting local temperatures for 75-100 years. Nothing more.
The people who say the planet must reduce the billions of humans on the planet, also fund "climate science".
The energy sector is used because it allows them to control all facets of our lives.
Tony Heller has been pointing out the climate data fraud for years. He has an enormous collection of unaltered data sets, and newspaper articles going back more than a century that point out the lies about "hottest year ever".
Why does the display and trendlines of the data use the median of temperature data instead of the mean? The mean would be the natural choice as a higher average temperature necessitates a higher overall energy density, while the median can disguise the temperature changes in extremes. (For example, there seems to be a trend in newfoundland of the coldest temperatures growing warmer over time, but the median doesn't change much at all.)
The extremes are shown in my graphs. All the boxplots show minima and maxima as well as variance (interquartile range, outliers). I discussed the provenance of the data with the curator. Bogus outliers (which effect the mean but not the median) are present in the dataset for various reasons. However, there are so many datapoints (approximately 29,000 each) in each 3-year bin that the means will look almost the same as the medians.
Given this similarity, why do you think that almost all climate science seems to be focused around average temperatures rather than median temperatures? (As a cursory example, multiple examples of average temperature changes are shown in wikipedia versus no references to median temperature.)
An increase in the mean temperature of a location tells us that there is a lot of new energy in the system.
Recall that the big understood mechanism of 'global warming' is that carbon dioxide in the atmosphere traps heat coming from the sun.
Think about a big cubic kilometer out in the middle of the ocean somewhere. If the average temperature of that water goes up by a full degree this year compared to last year, that means that the system has absorbed enough energy to raise *all* of the water in it by one degree. (You can sum up the 3 hour temperature changes yourself to do the math if you would like.)
Now, we also get something from the temperature at the end of those two years. If the average temperature increased this year, but the temperatures were the same at the end of the year, that means all of that energy went somewhere else. Some of it was radiated out of earth's atmosphere, and maybe if our KM of water is in a shallow part of the ocean some of it is conducted or convected to the seafloor. But most of it is exchanged with other water or sunk into the enormous energy requirements of water evaporating into water vapor.
When you multiply this amount of energy by the number of cubic KM in the ocean, you can see that even a sub degree increase in temperature of the earth's oceans is a *lot* of new energy. If you look at the average temperature of every cubic kilometer of the earth's upper surface and conclude that some have increased and some have decreased, then adding together the amount of energy gains and losses needed to produce this temperature change in each kilometer can tell you the amount of net new energy that has been added into the part of the earth that we live in.
I think they are focused not so much on temporal averages in particular locations, but on "global average temperature" vs. time. I am not sure how one can define "global average temperature" in an obvious way? How do you factor in different altitudes? The ocean? etc. But it is beyond my level of expertise.
Interesting approach! You might find my own "from scratch" analysis worth reading, it used NASA's dataset and compared to a random walk of yearly deltas:
Interesting work. However, I would expect the climate to have some amount of homeostatic (self-regulating) forces, and*not* be a random walk. A random walk can take arbitrarily large excursions, which would not be good for us.
This is true at the macro scale, but on short timelines (vs eons) there's enough volatility to justify analyzing the YoY changes that way. If you looked at the period from the ice age to now it would reject the null of random deltas precisely because of the clear secular trend.
To be clear, I'm saying that a significant trend would overwhelm the homeostatic forces. I found no (statistically significant) trend in the NASA data.
I had a similar question a few years ago and grabbed the weather data for the Sydney observatory itself (not Sydney airport). The observatory had a longer time series than the airport FWIW. Data sources here: http://www.bom.gov.au/climate/data/stations/
There are github links in there for a Jupyter notebook to reproduce it.
I was doing a different analysis (basically, what was happening to minimum and maximum daily temperatures), and it sure looks like there's a clear effect there.
I'm struggling to come up with why there's a difference in our analyses, though.
Idea 1: Could there be an airport-specific effect happening there? Change of runway materials or airport design? (The Australian bureau of meteorology data is just a zip of a CSV file, so I suspect it wouldn't require any changes in your R code. That would you let confirm whether the phenomenon is airport-specific.
Idea 2: Maybe if I re-do my analysis on Sydney airport and I'm able to confirm a growing loss-of-winter effect there too (it's not that far away from Sydney observatory, so hard to see how there couldn't be)... maybe your results and the loss-of-winter effect can both co-exist? I'm not sure what that means though.
Just noticed that you did Tasmania and Broome rather than Sydney. (Typical Sydney-sider, if it's not east of the great dividing range it doesn't exist).
Anyway I dug around and I actually did do a very similar analysis to yours using Sydney observatory (and a few other cities) about a year later:
Where you did Broome, I did Alice Springs. (Which has an airport too.)
Do you want to try Alice Springs, Melbourne or Sydney? Any of them should show a strong effect.
If it doesn't, then there's something very strange going on because there is a trend on minimum temperatures, so if the median temperatures aren't rising, then that's one very strange statistical distribution.
I use thermometers, not thermostats :-)
I use them because they directly measure temperature, pure and simple.
Proxies like tree-rings, while allowing for longer duration (since some trees have existed longer than the oldest thermometers) have confounding factors: the amount of rain, the availability of nutrients, other local ecological stresses, etc.
Nothing to add - I just can't get over how refreshing it is to come across people who think clearly! Thanks SO much.
What do I see from your data? I'm no expert but I see very little going on beyond "normal" oscillations.
What I see is a varied rising or falling overall, depending on the location. If anything, it is fascinating to see some areas’ cooling trends amidst others trending warmer. I live in Redlands, not far from PS, and I think it’s getting warmer, but I know for a fact two things: it was always hot from time to time (childhood here in the 60s), a solid 5 month season of cough-inducing smog (not mostly gone), and I’m willing to recognize I am influenced by the constant barrage of news telling me the earth is burning up.
Personally, I am suspicious of human hubris that would somehow poopoo solar influence and think we are the ne plus ultra of influencers on planetary climate.
This is very good, seeing what we can make of trustworthy, thermometer temperature measurements. Using airports I would expect the temperature to go up because over those decades we have more flights and bigger engines as well as more people activity so more heat into the surroundings. I see no clear evidence for ‘global’ warming or cooling. I don’t think a global number has meaning unless it turned out to be really hotter everywhere.
I apologize but if you don’t mind can you can Into further depth on why you used thermostats over proxies(tree rings, etc) to determine long term climate temperature?
Weather is complex. Temperature is just one indicator. What about rainfall (same rain in longer period is better for vegetation)? Wind (dries vegetation)? Humidity?
Thanks for this very interesting analysis.
The least squares fit to the medians is heavily influenced by very few data points. My suggestion is that you use a robust regression, such as a Theil-Sen estimator. You can also do a bootstrap on this estimator to obtain a confidence interval on the (robust) slope. Finally, you can do a histogram of the (robust) slopes across the different locations.
By the way, it would be better to subtract the local average temperatures before you compute these slopes, rather than the global average, as you have done. This way, the slopes are easier to compare.
And finally, a non-statistical remark: there is clearly a bias in only using airport data for your analysis. This is because the micro climate at airports is non-representative both because of the construction materials used, as well as because the locations were carefully selected to reduce the influence of certain environmental factors. This does not discredit the analysis but it is a caveat that should at least be stated.
Sorry I miss-wrote what I actually did and confused people. I just adjusted each detailed plot (the boxplots and median regressions) by the rigidly ofsetting the whole plot based on the median temp of the entire airport's individual time series (≥75 years) . All temperature changes are still shown in ºC relative to the separate constant (local) offset per plot.
And I recently talked to the guy who curates the dataset. He answered a lot of questions. These stations are pretty carefully sited to avoid bogus data. Other temperature records are mostly from city centers, which apparently suffer from the urban heat island as population increases over time. ("simply a matter of albedo").
I tried L1 regression and LOWESS regression and then (at your suggestion) Theil-Sen. Doesn't look very different.
You say “The global median temperature (of the entire dataset) is uniformly subtracted from every data point, to keep things relative to an arbitrary x-axis of 0.” and then seem to ask why most of the cities don’t vary much from 0. What does an equivalent plot of that global median temperature show? If it is rising over this 80 year period and you are subtracting that out, then why would you expect to see big changes in the city data?
The subtraction of the median for each data series (not the "global median" FWIW) does not effect the trends, the variance, or the shapes of any of the plots. It just shifts the origin (and not the scale) of each y-axis to be centered at an arbitrary zero (zero==the median for the entire series for the particular location).
I am not sure why this has confused people, but I will attempt to alleviate this confusion "next time".
"Global median" is the wrong word - sorry. What I mean is I subtract a constant scalar value (the overall median for each ~80 year timeseries) from all the values in that timeseries for the boxplots and median zooms. It just shifts each plot to be centered around zero. It does not change anything else: The variability of the medians, the size and shape of the boxes, the linear regressions all look the same as if the respective constant was not subtracted,. It is just a relabeling of the y-axis on each plot to make the vertical center of gravity of each plot show up near "zero".
It might have been smarter to not do this, but keep them centered some other way. Some would be centered around zero, some around 20ºC, etc. The shapes of the boxes and curves, and the scales of the y-axes and the distance between the tickmarks in ºC would all be unchanged but the numbers labeling each y-axis tickmark would be different.
> How do you even define "global temperature"?
This is an interesting question. Doing some Googling, it looks like the answer is, very roughly:
1. Collect all the time series from around the world that you can.
2. Define a grid that covers the globe.
3. Average the temperature time series located within each grid cell, to get a single time series for that cell.
4. Average all the time series across all the cells to get a single "global average temperature."
Here's an example of this process, including all the data cleaning steps I didn't include above: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2005JD006548
One big problem with "time series from around the globe" is that long-duration direct temperature records are *very* sparsely sampled. Most of them are from large, old city centers which tend to have suffered from the urban heat-island effect over time. This latter problem is less severe for airports, which tend to be at least a little outside of city centers.
Here's a review paper of sorts that looks at the impact of the UHI effect on these averages:
http://danida.vnu.edu.vn/cpis/files/Papers_on_CC/CC/Urban%20heat%20island%20effects%20on%20estimates%20of%20observed%20climate%20change.pdf
> The urban heat island has had only a minor impact on estimates of global trends of LSAT. Its impact is much smaller than the 0.74◦ C global warming between 1906 and 2005. The impact is small because assiduous efforts have been made by the compilers of global surface air temperature records to avoid or compensate for urban warming. This is confirmed by analyses using only rural stations or using only days and nights with windy weather. Further support comes from comparisons with marine surface temperature trends in the light of the expected augmentation of trends over land.
Figure 2 in that paper is particularly worth a look, because it tries to summarize the uncertainty introduced by both the sparsity of historical data and urbanization.
Let me clarify: airport runway temperatures have to be actual temperatures. Not adjusted in any way. A pilot needs to know the air temperature to know if he has, literally, enough runway to take off. Hot air is less dense and provides less lift.
"The impact [of UHI] is small because assiduous efforts have been made by the compilers of global surface air temperature records to avoid or compensate for urban warming."
That may be so (it sounds a little self-referential to me) but I was looking at thermometer readings made into timeseries with - I have been assured by the curator - no adjustment.
I may try next to look at city centers vs. airports for the same cities. There are some places with records for both that go back ~80 years.
In my experience it is always good to check the data records themselves, if only to understand how the "consensus" has been achieved.
I was curious if there is a known effect specific to airport. One commenter already speculated it could be changing runway materials. Or perhaps air conditioner upgrades that increased efficiency. Or who knows. I’ll search some more but in the meantime, found this. Want to run your process on some airports in Greece? https://link.springer.com/article/10.1007/s10584-019-02634-z
You maybe missing the trees for the forest. Global or local mean or median temperature is a nebulous (no pun intended) indicator. It's the increases in daily weather variance, incomprehensibly-fast melting of major glacial masses, rising sea levels, and frequency and intensity of storms, fires, famines, and species' extinctions that are the "forest". These are consequences of anthropogenic GHGs that reduced Arctic sea ice (nearing an inevitable blue ocean event) and increased global sea thermal energy absorption which destabilized the jet stream from more predictable cycles. GHGs lead to a -11 F snowpocalypse in Austin, megadrought in US West and SW, "Hurricane Alley" shifting eastward, 100 F in the high Arctic town of Verkhoyansk, water trains of India, and slumping and melting of Siberian and other permafrost areas releasing absurd amounts of methane.
The variance (locally) is shown in the plots. What do you see re: variance?
"Incomprehensible" glacier movement? As I understand it nine time in the life of the human species the glaciers have advanced to cover much of N. America and then retreated to where there was little or no polar ice at all.
Frequency and intensity of storms? I suggest downloading the data on all hurricanes for the last ~100 years (as long as they've been recorded) from the NOAA national hurricane data center. I did that. I plotted it. Guess what I found?
Fires... I lived in Napa California during the big Napa fires. How much was due to climate change, and how much was doe to terrible forest management by the forest service and the state? (don't thin the forests, don't clear dead trees, etc).
Species diversity is very complicated. A whole different matter. So much to say at another time.
Famines? Have you looked at famines vs. time for the last 1000 years? Other than the global covid mismanagement disasters of 2020-2022 (lockdowns, etc) famines are becoming a thing of the past - though if we continue killing our soils with mono-crop soy, etc they will recur.
Arctic (and Antarctic) sea ice is another deep topic.
Austin was a fluctuation. They happen. Don't kill your local electric utilities - you may want them when things get dicey (or icy :).
Look into the history of draughts (and floods) over the last 200 years in the W and SW US.. Write a post and send me the link.
You raise a lot of points. (or less politely, you are shifting the goalposts 12 times in one paragraph). I am just plotting local temperatures for 75-100 years. Nothing more.
The people who say the planet must reduce the billions of humans on the planet, also fund "climate science".
The energy sector is used because it allows them to control all facets of our lives.
Tony Heller has been pointing out the climate data fraud for years. He has an enormous collection of unaltered data sets, and newspaper articles going back more than a century that point out the lies about "hottest year ever".
Thanks for your work here.
Why does the display and trendlines of the data use the median of temperature data instead of the mean? The mean would be the natural choice as a higher average temperature necessitates a higher overall energy density, while the median can disguise the temperature changes in extremes. (For example, there seems to be a trend in newfoundland of the coldest temperatures growing warmer over time, but the median doesn't change much at all.)
The extremes are shown in my graphs. All the boxplots show minima and maxima as well as variance (interquartile range, outliers). I discussed the provenance of the data with the curator. Bogus outliers (which effect the mean but not the median) are present in the dataset for various reasons. However, there are so many datapoints (approximately 29,000 each) in each 3-year bin that the means will look almost the same as the medians.
Given this similarity, why do you think that almost all climate science seems to be focused around average temperatures rather than median temperatures? (As a cursory example, multiple examples of average temperature changes are shown in wikipedia versus no references to median temperature.)
Median means (roughly) 1/2 of the time the temperature is higher, and 1/2 of the time it is lower. Mean temperature means what, exactly?
An increase in the mean temperature of a location tells us that there is a lot of new energy in the system.
Recall that the big understood mechanism of 'global warming' is that carbon dioxide in the atmosphere traps heat coming from the sun.
Think about a big cubic kilometer out in the middle of the ocean somewhere. If the average temperature of that water goes up by a full degree this year compared to last year, that means that the system has absorbed enough energy to raise *all* of the water in it by one degree. (You can sum up the 3 hour temperature changes yourself to do the math if you would like.)
Now, we also get something from the temperature at the end of those two years. If the average temperature increased this year, but the temperatures were the same at the end of the year, that means all of that energy went somewhere else. Some of it was radiated out of earth's atmosphere, and maybe if our KM of water is in a shallow part of the ocean some of it is conducted or convected to the seafloor. But most of it is exchanged with other water or sunk into the enormous energy requirements of water evaporating into water vapor.
When you multiply this amount of energy by the number of cubic KM in the ocean, you can see that even a sub degree increase in temperature of the earth's oceans is a *lot* of new energy. If you look at the average temperature of every cubic kilometer of the earth's upper surface and conclude that some have increased and some have decreased, then adding together the amount of energy gains and losses needed to produce this temperature change in each kilometer can tell you the amount of net new energy that has been added into the part of the earth that we live in.
I think they are focused not so much on temporal averages in particular locations, but on "global average temperature" vs. time. I am not sure how one can define "global average temperature" in an obvious way? How do you factor in different altitudes? The ocean? etc. But it is beyond my level of expertise.
Interesting approach! You might find my own "from scratch" analysis worth reading, it used NASA's dataset and compared to a random walk of yearly deltas:
https://statisticsblog.com/2012/12/03/the-surprisingly-weak-case-for-global-warming/
Interesting work. However, I would expect the climate to have some amount of homeostatic (self-regulating) forces, and*not* be a random walk. A random walk can take arbitrarily large excursions, which would not be good for us.
This is true at the macro scale, but on short timelines (vs eons) there's enough volatility to justify analyzing the YoY changes that way. If you looked at the period from the ice age to now it would reject the null of random deltas precisely because of the clear secular trend.
To be clear, I'm saying that a significant trend would overwhelm the homeostatic forces. I found no (statistically significant) trend in the NASA data.
I had a similar question a few years ago and grabbed the weather data for the Sydney observatory itself (not Sydney airport). The observatory had a longer time series than the airport FWIW. Data sources here: http://www.bom.gov.au/climate/data/stations/
My analysis is here:
http://blog.ifost.org.au/2017/01/farewell-to-cold-winters-and-hello.html
There are github links in there for a Jupyter notebook to reproduce it.
I was doing a different analysis (basically, what was happening to minimum and maximum daily temperatures), and it sure looks like there's a clear effect there.
I'm struggling to come up with why there's a difference in our analyses, though.
Idea 1: Could there be an airport-specific effect happening there? Change of runway materials or airport design? (The Australian bureau of meteorology data is just a zip of a CSV file, so I suspect it wouldn't require any changes in your R code. That would you let confirm whether the phenomenon is airport-specific.
Idea 2: Maybe if I re-do my analysis on Sydney airport and I'm able to confirm a growing loss-of-winter effect there too (it's not that far away from Sydney observatory, so hard to see how there couldn't be)... maybe your results and the loss-of-winter effect can both co-exist? I'm not sure what that means though.
Just noticed that you did Tasmania and Broome rather than Sydney. (Typical Sydney-sider, if it's not east of the great dividing range it doesn't exist).
Anyway I dug around and I actually did do a very similar analysis to yours using Sydney observatory (and a few other cities) about a year later:
http://blog.ifost.org.au/2018/12/the-cities-of-endless-summer-how-is-our.html
Where you did Broome, I did Alice Springs. (Which has an airport too.)
Do you want to try Alice Springs, Melbourne or Sydney? Any of them should show a strong effect.
If it doesn't, then there's something very strange going on because there is a trend on minimum temperatures, so if the median temperatures aren't rising, then that's one very strange statistical distribution.
Very cool. Your charting gave me a few ideas for a project I'm working on that's similar but only for the US (https://wxrecords.com). The data source is the RCC-ACIS API (https://www.rcc-acis.org/docs_webservices.html).
Another fun tool is the query tool attached to the API - https://xmacis.rcc-acis.org/