Cliff Mass shows a graph, taken from the Seattle Times, of the hottest temperature each July from 1945 through 2022 at Seattle/Tacoma airport (SEATAC). He then says “… and there is very little upward trend! How could this be?”
Then he goes further:
Just to check on the Seattle Times… I did the same thing for July and August over the past 50 years, plotting the warmest observed temperature at both SeaTac and Pasco (see below).
Hardly any change in the extreme high temperatures each year at either site. No long-term trend….and you would expect a trend if global warming was important for the extreme heat waves!
Unfortunately for Cliff Mass, the data contradict him.
I retrieved the data for SEATAC, although the source from which I acquired it (Climate Explorer) doesn’t include this year, it only extends from 1948 through 2021. I then noted the hottest temperature during each year’s July/August, which looks like this:
The red line is a trend estimate from least-squares regression, and indicates that the hottest July/August temperature is rising at about 6 °F/century, and has gone up about 4.5 °F during the period of record. And — the trend is definitely “statistically significant.”
So … why does Cliff Mass say “No long-term trend“? He does have an extra year’s data — which is not really quite right, because this year (2022) isn’t over yet, the entire month of August is yet to come. More important, he doesn’t start his graph until 1971. If I restrict my trend analysis to the period from 1971 onward, I get the trend line shown here in blue:
It shows nearly the same trend as the full analysis, rising at a rate of 5 °F/century. Over that period of record, the total rise is 2.3 °F. Really, it’s the same trend (within the error ranges) but based on a shorter time span of data. And because of that, the trend estimate using only the shortened data set is no longer “statistically significant.”
Which is no surprise at all. When you shorten the time span, even if the trend doesn’t change at all, the statistical significance will go down. Shorten the time span enough, you’re sure to eliminate statistical significance. It seems to me, that is exactly what Cliff Mass has done.
One other thing strikes me as odd. In climate, “summer” is usually defined as June/July/August (JJA), yet Cliff Mass quite ignores the month of June. Thing is, the hottest day of the year often occurs during June, which makes his reference to “extreme high temperatures each year” incorrect, at least for the 20 years during the record when the “extreme high temperature” occured in the month of June.
If we look at the actual “extreme high temperature each year” at SEATAC, it looks like this:
Again, the red line is a trend estimate by least-squares regression. It indicates warming at a rate of 7.6 °F/century, and shows a total rise of 5.5 °F over the period of record. And it is most assuredly “statistically significant.”
Lest you think that’s only an accident because of last year’s super heat-wave, if I remove that year (2021) from the analysis, it’s still statistically significant.
Cliff Mass considers it a “dilemma” that there’s no trend in the hottest day of the year at SEATAC, and he promises to answer this dilemma in his podcast. I’d like him to answer a different question: How did you fail to see the rather obvious — and strongly statistically significant — trend in the hottest day of the year at SEATAC?
This blog is made possible by readers like you; join others by donating at My Wee Dragon.
I think the temperature scale on your graphs is in degrees F, not deg C. Otherwise things got awfully hot at SEATAC…
I see two interpretations of what Cliff Mass has done. Either he doesn’t know what he is doing, or he does. I think the second interpretation is even more damning than the first.
Bob – As I have previously stated regarding a certain UK “skeptic” of my acquaintance.
Either Mr. Mass doesn’t know what he is talking about or he is deliberately pulling the wool over the eyes of his faithful flock of followers.
The latter seems far more probable?
Has Cliff Mass been watching Tucker Carlson again ?
Now there’s a thought with no thought.
As someone who has lived in the area of the Seatac Airport for the last seventy+ years, I can say with authority that it looks nothing like it did when I was young. The airport itself has grown in size by a factor of ten, likewise the support facilities like freight terminals and parking lots, and that’s just the area around the airport.
Extend that outside of the airport boundaries, and what were once small agricultural communities covered in growing green things are now sprawling oceans of concrete, flat-roofed warehousing, strip malls and apartment blocks as far as the eye can see.
When you combine very substantial the heat island effect that now exists, the heat given off from internal combustion engines, air conditioning, and countless other conditions that raise temperature in the areas they occur, I am surprised that this area doesn’t show a more dramatic increase in temperature.
As my dad always used to say, “use your brain – that’s what God gave it to you for!”
Tamino…. I am encouraged that you are still reading my blog. In any case, there are serious problems with your analysis. When you have a time series with no real trend, you can falsely get a trend if you choose an optimal period. This is what you did. The Northwest was in the cold phase of the PDO from the late 40s to the early 70s. That is why you got a small upward trend in extreme high temperatures. If you go back further using longer record stations…there IS NOT TREND in extreme high temps. If you give me your email I can send you some graphics that provide it. I chose 50 years for my graphics because that is the period when the global warming signal is greatest and I avoid the mistake you made regarding the PDO. Why did I plot July and August. BECAUSE THAT IS THE WARMEST PERIOD in this region. Peak temps are last week in July and first week of August.
Anyway, I challenge you do this this yourself and do it correctly. Go back 100 years and plot the extreme highs. You will get no trend.
Finally, even doing it the way you did, you really prove my point. Even your problematic analysis also shows a small upward trend of the record highs…certainly doesn’t explain last June! Truth is important here …cliff mass
Response: Cliff Mass,
I got all the data for SEATAC I could find through Climate Explorer, and used all of it (except 2022 which isn’t over yet — August is yet to come). You say I only found a trend because I decided to “choose an optimal period.” That’s a lie; I didn’t choose any period, Cliff. I used all of what I had available.
You certainly seem to have the SEATAC data yourself, at least from the GHCN — but you *did* choose not to use it all. And you did so in order to say there was no trend. That’s not a lie, Cliff, it’s the truth. Readers need only look at your own post.
Your feeble attempt to blame the upward trend on the PDO is — feeble. It’s also simply nonsense for you to say that (let me paraphrase) “There’s no trend because the trend is caused by the PDO.” Do you really not get the logic fail here?
Now you say “Go back 100 years and plot the extreme highs. You will get no trend.” Why didn’t you use that data? You claim to have this “proof” going back 100 years, but you choose to use only 50 in a data set which covers 74 years? (even excluding 2022) Smells fishy to me.
As for Ignoring the month of June, it doesn’t matter that summer temperature peaks on average in July/August, what matters is that in 20 of the 74 years of the data the year’s hottest day was not in July or August. Ignoring that is just a simple mistake, and admitting that would have made people think highly of you; doubling down on this idiocy makes you look *really* bad.
And I have to wonder, why do you keep saying “no trend” when you seem to have avoided the rather important step of first doing trend analysis? Do you just “eyeball” a graph and decide you’ve seen what you wanted to see?
I doubt you are able to be honest with us, because I don’t think you’re able to be honest with yourself.
You would expect a grown-up who insists “Truth is important here” to have some small grasp on the subject he is pronouncing on and not be just resorting to bullshit. But then there are those who are anatomically grown-up but still psychologically child-like. And this Cliff Mass appears to be such a one.
I’m no expert on sourcing US temperature records but this webpage sets out a record of Highest Temperature for Each Year for Seattle back to 1894 (excepting 2009 & 1945) so allowing us to “go back 100 years and plot the extreme highs” to see if, as this Cliff Mass says, “you will get no trend.”
The first thing that springs off the listing is the timing of these “peak temps” and how you would have to be mind-numbingly stupid to insist they occur in the “last week in July and first week of August.” While 25% occur in the last 10 days of July and first 10 days of August, 15% of them occur in June, with 6% in May and 9% in September. There’s even one listed in April.
And that “you will get no trend” claim? That also appears to be but puerile nonsense. The data shows a +0.45ºF/decade trend.
Running a Mann-Kendall test for a trend on the June-July-August data from 1948, I get that yes, there is a trend (p = 0.011). Chopping off that last high value for 2021, there is still a trend, with p changing to 0.025.
Using only data from 1971 onwards, p changes to 0.122.
These results agree with the post, not a surprise since it’s just another way to look at the same data.
Running a LOWESS procedure with a 30-year wide window, it appears that the rate of increase of the max summer temperature slowly decreased from 1948 to about 2000, by when it had roughly leveled off. Then it started to increase rather faster than the linear least squared line. For just the data from 2000 – 2021, the Mann-Kendall trend test is True with p = 0.041, once again statistically significant. This supports the LOWESS- fitted curve.
[Response: I’ve seen evidence in a lot of records of yearly high temperature, that the trend is not simply linear. In particular, there are signs that things significantly picked up speed in some places in Europe in the 1990s, and in the 2000s over much of the continent.]
Very glad you handed Mass’s derriere to him on a platter.
Good morning Cliff (UTC),
Would you mind digging my comment from yesterday evening out of your Blogspot moderation queue?
It’s hiding under the comment where you state that “Tamino is not [a] meteorologist or climatologist”.
That’s true, Tamino is not a meteorologist or climatologist, he’s a statistician, something Mass clearly is not.
I’ve noticed a real “trend” in deniers starting to throw out all the data except aggregating single extreme values from various time periods and then stating that since single extreme values aren’t increasing dramatically, warming is not occurring. Coupla’ points:
1. All temp observations are a combination of weather and climate. Throwing away all the data except single extreme values in the aggregation process essentially maximizes the effect of individual weather events and minimizes the effect of climate in the sample under study*. Cliff–and others using this meme–blithely appear to believe that one record day of temperature X provides more information than a study of all temp values in the aggregation period.
2. Cliff–and others using this meme–also apparently would have us believe that one really hot day with a weather event producing temperature X means that that month was “hotter” than another month from a different aggregation with 15 days of temperature X-.1 degrees. That is patently ridiculous, yet that is what Cliff apparently would have us believe. I cannot believe he’s really that stunned, so I go with Bob L’s option #2 above.
*One would think if one were actually interested in studying climate, one would want to use underlying observations that minimized individual weather events, not maximized them. But then maybe Cliff’s interest here is not actually studying climate.
Reading Cliff’s reply I note 2 additional things:
1. He reiterates his analysis is of extreme or record high monthly values FOUR times in a short reply. So he is QUITE aware he is saying that there is no warming going on based solely upon single extreme values. And then he throws away a large number of single extreme annual values that actually occurred to boot as pointed out above as they occur in other months. And then he wraps himself up in “Truth is important here”. Bull shit.
2. Cliff goes on to state that he studies only single extreme values in a restricted part of the year because that’s where the warming signal apparently is shouting: “BECAUSE THAT IS THE WARMEST PERIOD”. I guess, warmer winters, springs, falls and early summers in the other 10 months of the year don’t mean a warmer climate to Cliff.
I was going to download the Seatac data from Climate Explorer, but the user interface there does not make it easy. Anyway, Al Rodger and others have dug out some of the numbers I was going to try to look at.
The oddity of saying the trend isn’t there because the trend is caused by X has already been noted. The aspect of late July/early August being claimed to be the warmest period – except when it isn’t – has also been noted.
And Cliff says he used the last 50 years because that is where warming is the greatest, and then says “If you go back further using longer record stations…there IS NOT TREND in extreme high temps.” seems rather inconsistent at the very least.
Which period he chooses seems entirely dependent on what result he wants to see. It’s hard to find a better example of Maier’s Law in action.
Whether you want to note it or not, there is a problem using trends starting in the the 1950s in the NW…it is in the middle of the cold phase of the PDO. My central point, which I clearly have not communicated well to you, is that we are looking for trends in extremes….like record temps. There is very little trend in these. Using Tmax trends like you did is problematic, because some years (particularly ones in the early part of the record you used) are quite low. So the Tmax does not necessarily indicate an extreme situation. Look at how the top envelope is changing…or even better see how daily records in the NW change over time. You will see that it is quite flat.
There is no doubt there is global warming going on. You will see it in mean temperatures (1-2F over the past 50 years). But the uber extremes are hardly changing. Yes, you can get a black swan year like 2021…but that is mainly natural variability. Anyway, please stop calling me names…and terms lie liar…not a good look and inappropriate for scientific discourse…cliff mass
[Response: This is truly bizarre. You now say “Using Tmax trends like you did is problematic because some years (particularly ones in the early part of the record you used) are quite low.” That’s an admission that you left out that data because it contradicts your assertion.
And do I really have to point out that you are the one who decided to use Tmax trends? It’s what your post was about. You are the one who cherry-picked the data you would include and exclude, to get the result you wanted, and it is looking more and more like you never did any trend analysis at all — you really did just “look at a graph” and decide you could see what you wanted to see (as long as you omit the data you don’t like).
Also bizarre is that you keep telling me to “look at” this or that quantity, obviously unaware that I already have. Then you declare that it definitively doesn’t do what the math tells me it does.
Readers probably already figured out that this isn’t a “scientific discourse,” it’s a pathetic attempt to defend your post.]
Tamino…Re. ” you are the one who decided to use Tmax trends?” Plus, as you point out, Cliff decided not to include 27% of the ACTUAL Tmax’s for some odd reason. Plus, I’d be curious if Cliff could tell us why an annual Tmax in July is an “extreme situation” while an annual Tmax in June is not?!
Seriously, Cliff? You say “Using Tmax trends like you did is problematic, because some years (particularly ones in the early part of the record you used) are quite low.”, and argue that “we are looking for trends in extremes….like record temps. “?
You are aware that the first value in a series represents both a record high and a record low in that series? And that the next value will be either a new record high or a new record low – or will be a tie for both? And that the third value has a good chance of being a record high or low? And that as the series accumulates, the probability of seeing record highs or lows decreases? So, as time goes on, your collection of “record values” has fewer and fewer points in it (per unit time)?
The collection of record values has exactly the characteristic that you are claiming to want to avoid in Tmax. You really have not thought this out, have you?
In the previous thread, on the recent UK temperature extremes, bindidon provided two graphs for US and global temperature extremes. You would do well to look at those and think about what they mean before you go off contradicting yourself again.
Mass is generally following a standard trick: the smaller the time period, the smaller the land area, and generally the less data included, the more likely the trend is to not be significant. I actually agree with him that 50 years is a reasonable period to look at because 1970 is about when the current warming trend is most visible in both the global and continental US datasets (e.g. https://www.epa.gov/climate-indicators/climate-change-indicators-us-and-global-temperature), so he didn’t mess up there… but then he takes a _single_ airport thermometer which severely truncates the data, and then looks only at a single data point for each year which truncates the data even more. So it is no surprise that the data is so noisy that it is hard to get significant trends!
This is the same line of examples as contrarians who claim that Arctic sea ice has stabilized since 2012 (time period is way to short), or who look only a tropical tropospheric temperatures (too small an area), or continental temperature records (only 7 data points, and the ones many decades ago are not going to be as precise as more recent ones), etc. etc.
Trends in extremes are always going to be hard to measure, because by definition extremes are infrequent, which makes the statistics wonky. This includes single-day heat records, hurricanes, rainfall extremes, etc. But the thing is that climate has been changing fast enough that scientists have already detected trends in a number of these extremes, despite the challenges involved, which should say something!
(a related question: how strong would the trend from 1971 have to be to be statistically significant? and/or how high a trend could you have before the existing data since 1971 rules it out?)
Look at the third figure (the second one of Tamino’s). Look at the blue line compared to the red line. Do they look like the slopes are significantly different? If you took any other 50-year period out of the full record, how much do you think the slopes would vary? Is there any justification in the data to say that the period 1948 to 1970 is different from 1971-2021?
The PDO has been negative since Jan 2020 and neutral since early 2016. Why did we set the heat record in 2021 during the negative PDO cycle? Why isn’t a timeseries starting PDO negative and ending PDO negative legitimate? How long does the ground temperature a Seatac take to respond to changes in the PDO cycle, shouldn’t it be on the order of 6-12 months?
Good morning once again Cliff (UTC),
My comment over on your Blogspot is still invisible
It’s the one in which I idly enquire what your qualifications as a “statsologist” are.
Don’t feel bad. I posted a comment in his blog where he said that tamino was a (mere) climate activist by pointing out the peer reviewed journals he has published in. It has not appeared!
Cliff certainly loves presenting all the actual data even there, doesn’t he?!!!!
It sounds as though you were commenting on the same thread as me.
There’s still no sign of either comment. I cannot help but wonder how many others are being trampled underfoot on Cliff’s cutting room floor?
It is easily possible to lie by telling (a portion of) the truth. In Propaganda 101 one technique the student learns in the first week is to search for tiny subset of the data–a factoid–which “disconfirms” the conclusions garnered from a competent analysis of the dataset as a whole. The propagandist then focuses wholly on that factoid so that the naive reader will (utterly wrongly) generalize said factoid to the whole field. We saw this in tobacco “science” where the propagandists focused on one Grandpa somewhere in Kentucky who smoked every day and never got cancer or heart disease where the “inference” was “therefore” tobacco doesn’t cause cancer. In small, we see the same technique used by teens talking to their parents and in marketing every day. We see it many other places. Cliff is just being another such practitioner here.
Let’s even posit that there is a temp ceiling effect of some kind–I mean when sampling only single extreme max values from any distribution, one mathematically expects a skewed result with higher values being rarer than lower values in the resulting distribution, after all, which is at least a partial ceiling effect. (Sampling only maxes from a uniform distribution with a defined ceiling skews far more.) That absolutely does NOT imply that dangerous warming is not occurring in the PNW or anywhere else. Living more and more days at or near the same record highs is NOT benign to the PNW as Cliffs wants the reader to infer. Cliff’s suggestion in the comments below his article that “vulnerable folks should get AC” is especially ridiculous when considering the environment as a whole.
Cliff…my Gramps chewed tobacco all his life spitting the juice into 1 lb coffee cans all over the farm house, lived to his 90s and never got throat or lip cancer. This simply is NOT “proof” that chewing baccy doesn’t cause certain cancers and other significant problems in the whole population so matter what the Tobacco Institute “scientists” said.
Bob Loblaw: “Is there any justification in the data to say that the period 1948 to 1970 is different from 1971-2021?” There may not be justification in the data, but the data is not everything we have. If there isn’t a global or national trend from 1948 to 1970, then doesn’t that imply that maybe the local trend from 1948 to 1970 might not be anthropogenic GHG related? (there can be subtleties that I’m not addressing here, but this is the first order logic)
To me “maybe…” gets trumped by “but the data shows”.
The question at hand is not the cause of the trend, but whether or not a trend exists. And being selective about the period before looking at the data is not a Good Idea ™. And when that selection of a subset of the data does not display a different trend – but only affects the significance of the trend – and then that difference is used to justify “the data shoes no trend”, then my spidey-sense gets triggered.
Post-hoc claims of “but I picked that period because of X” start to look like sorry excuses hiding an agenda of “I wanted to make sure I got a non-significant result”. And doubling down on the cherry-pick after it is pointed out…
Another look at Seattle Tacoma Intl AP
Out of the data provided by GHCN daily
USW00024233 47.4444 -122.3139 112.8 WA SEATTLE TACOMA INTL AP
I generated a time series with absolute temperatures, and split these into the usual winter/spring/summer/fall seasons (DJF, MAM, JJA, SON):
The first winter season is 1948/1949; the last seasons are all in 2021.
Cliff Mass’ arguing is completely untenable because the trends since 1971 are, with the exception of the spring season, all higher than the trends since 1949 (numbers for DJF / MAM / JJA / SON):
– 1949-2021: 0.12 / 0.25 / 0.25 / 0.28
– 1971-2021: 0.16 / 0.20 / 0.32 / 0.32
(All estimates in °C / decade, ± 0.05 CI on average.)
A very interesting, unusual point was for me that for this Seattle station (and possibly way around it), the fall season shows absolute temperatures perfectly similar to those measured during the summers.
For those who like to look at the full data in anomaly form:
It’s always nice to see how trustworthy a linear trend across the data of a time series looks when it is the same as the trend of the data generated by the low-pass smoothing of the original data :-)
FWIW, uBlock blacklists postimage.cc as containing potential malware:
“… sites documented to put users at risk of installing adware/crapware etc.”
Just by coincidence, I’ve just worked up a similar graph as bindidon. It shows the SEATAC summer max temps extracted from the Climate Explorer. The data is for 1948 through 2021, and I added the hottest temp for 2022 even though August isn’t over. I fit the data with a LOWESS routine, and overlay it onto the HADCRUT global mean temperature anomalies.
The graph shows:
1. The SEATAC data points, without connecting lines. The data has had its mean subtracted so it represents anomalies, not absolute temps;
2. The LOWESS fitted line with 2-sigma bands;
3. The HADCRUT global anomaly data from 1850 through 2021.
Hopefully, inserting an anchor element here will display the graph:
SEATAC, Global Temps
If the graph doesn’t show up, here is the url for it:
It’s interesting that if you plot the data points as symbols without connecting them with lines, the increasing nature of the extremes (i.e., the hottest highs and the lowest highs both increase dramatically over time) is instantly obvious.
The increase from 1948 to 2022 of the LOWESS fit is about 5 deg C, or 9 deg F. With the worst case statistical fluctuations for both the first and last points, the span could be as low as about 2 deg C. But that is extremely unlikely because most of the other points in between will fall much closer to the fitted curve, thus supporting a larger trend. Also, fluctuations in the end points could just as well go in the other direction, increasing the span.
It looks like the graph didn’t display. Does the blog software want to see a Markdown statement instead?
Bob Loblaw: data plus physical understanding almost always trumps “just the data”.
And you can see on Tamino’s own website many examples of starting in 1970, such as:
It all depends on what question you are asking. If the question is, “is there a trend in the data?”, then, yeah, use all the data. If the question is, “does the data show that human GHG emissions have contributed to increasing summer maxima”, then maybe only use the portion of the data where we expect to see a strong human signal. As I noted, by restricting the analysis to a single station, and to a single day (in a 2 month period), that will guarantee highly noisy data, which maybe makes it hard to find significance in a trend analysis… but we also have physical understanding that suggests there _should_ be a trend, so using bayesian reasoning, the positive trend (5 F/century) that Tamino found even if non-significant combined with our physical reasoning suggests that humans are contributing. In fact, 5 F/century is pretty similar to the Berkeley Earth annual rate of warming for Seattle (a bit over 4 F/century) or the July rate of warming (a bit over 6 F/century)… so that all confirms nicely. But we have less grounds to find that the hottest day is warming faster or slower than the average annual or summer temperature… and it is an interesting question whether we would expect it to, but there’s complex interactions with soil dryness and other factors that are hard to model and think all the way through.
I think we are all in agreement that Mass is incorrect, I’m just saying that of his dubious choices, the choice to start in 1970 is the one that I wouldn’t necessarily criticize.
(to be precise, the Berkeley Earth trends i calculated were for starting in 1970 through the end of 2020 from http://berkeleyearth.lbl.gov/auto/Local/TAVG/Text/47.42N-121.97W-TAVG-Trend.txt)
Re. ” If the question is, “does the data show that human GHG emissions have contributed to increasing summer maxima”, then maybe only use the portion of the data where we expect to see a strong human signal. ” you are correct in an objective sense. However, beyond data + physical understanding is purpose. Mass’ purpose is clearly not to communicate any objective fact or understanding, It is to produce a false inference in his readership that no practical change is occurring through various statistical manipulations he figures his audience won’t see.
That is the problem here with his statements.
Well that and the fact he clearly hides relevant statements while promoting out and out false statements like my own comment on his blog where I replied to his specific assertion that tamino is not a scientist but an activist by simply listing the top-level peer-reviewed journals his research has appeared in and he refused to publish on his site. “Truth is important here” is what he says, but then he tells falsehoods and writes to mislead his readers. Absolutely hypocritical.
Folks….since you all seem so interested in SeaTac, let me note that one should not use it for any climatological analysis or claims of global warming. Why? Hopelessly undermined by the construction of a third runway and massive development at and near the airport. I am going to do another blog considering better stations. And by the way, the length of record changes the trends. For example, there was warming in the early 40s and 30s in the region. And folks…please stop the name calling…please…cliff mass
[Response: SEATAC: First you “use it for “climatological analysis” and make “claims of global warming.” When your “analysis” turns out to be an embarrassment you tell us that one “should not use it for any climatological analysis or claims of global warming.”
I think you should stop digging.]
Translation: “Do what I say, not what I do”.
Please publish folks’ comments on your blog.
Cliff: What a truly embarrassing reply.
If you know R, run the following script on your computer:
We can notice that if the max temperature record at SEATAC had exactly tracked the global means temperature anomalies, the large amount of noise we see in the data would have prevented us from drawing any useful conclusions about its trend.
Temperatures over land would normally vary more than at sea (heat island effect or no), but even so, a positive trend would not support (or contradict) any conclusions about global warming.
So this topic could be about any number of issues, such as attempting to see how much information can be wrung out of noisy data, but it cannot – on its own – be useful for drawing conclusions about global warming.
Re. “And folks…please stop the name calling…please…cliff mass”
To what names are you objecting? Be specific with quotes. Or are you just playing the victim card without actually being a victim which seems to be popular in certain circles these days?
Do you mean like: “tamino is not meteorologist or climatologist, but a climate activist”?
And when are you going to allow my NON-name calling comment through on your blog which merely lists a number of rather high-level peer-reviewed journals which have published his research?
Folks….here is a more details analysis… let me know if you think anything is wrong with it!…cliff mass
Is this Cliff Mass a professional comedian?
His first attempt at analysis of Olympia Airport highest annual temperature is done using only July & August data. Yet the record 1942-22 has 18 of the 81 years posting a highest annual temperature outside July & August. three falling in September, fourteen in June and one even as early as May. And it does appear that the timing of annual highest temperature is falling earlier with the passing years with half the last decade falling in June (and all three Sept highs occurring in the fist couple of decades).
So is this Cliff the Comedian having a laugh? Or is it Mass the moron?
Not only does the maximum fall outside of Jul-Aug in eighteen of those years, but in some of those years there is more than one daily maximum temperature higher than the Jul-Aug maximum. Those 18 years include a total of 31 days warmer that the Jul-Aug maximum of the same year. In 1963, there were four. In 1957, 1989, and 2021 there were three. In 1969, 1992, 2000, and 2013 there were two. Of the remaining ten years with only one day exceeding the Jul-Aug maximum, three are in 2016, 2017, and 2019. From 2013 to 2021, five years have the maximum temperature in a month outside of Jul-Aug.
Talk about an effective method to “hide the incline”.
Note the following quickie results of this R script which takes 50 years of maxes from a 61 day span versus using the mean of those spans:
q = replicate(50,mean(rnorm(61,0,1)))
r = replicate(50,mean(rnorm(61,.1,1)))
# t = -3.6477, df = 97.69, p-value = 0.0004267
q = replicate(50,max(rnorm(61,0,1)))
r = replicate(50,max(rnorm(61,.1,1)))
# t = -1.2134, df = 97.999, p-value = 0.2279
So, even though we built in a true difference in the underlying distributions, which statistical procedure has the statistical power to show an increase and which does not?
More importantly: Would it be correct to say we’ve “shown” there is no increase in max temps?
As has been pointed out repeatedly: When you throw away the contributions of the vast majority of the data, power gets shot to hell. You prove nothing with a test that has no power to show a true difference even if it’s there.
Good morning once again Cliff (UTC),
Thanks for the heads up, but I do have a minor quibble. Please let me know in what way this currently invisible comment of mine violates your blog’s terms and conditions:
At a glance, a few things that seem wrong to a non-specialist:
1 – Why are you using only a few stations to analyse extreme heat events in the Northwest, rather than all stations?
2 – If your analysis intends to look at extreme heat events generally, why are you restricting your extreme heat events to July/August when extreme heat can happen any time of the year?
3 – If your analysis intends to look at extreme heat event in Summer only, why are you restricting the analysis to July/August when international definition of Northern Hemisphere meteorological Summer is from 1st of June and ending 31st of August?
4 – You state at the start of the post “My basic point is that the peak temperature, duration, and frequency of extreme heat events are not rapidly rising in the Northwest.”.
4.1 – How is a graph plotting peak temperatures per calendar year and a few tables showing 6 day peak temperature averages per calendar year supposed to analyse duration and frequency of extreme heat events?
5 – More broadly, what is your definition of “extreme heat event”? How can you claim to analyse something you do not define?
Lowlander…check my blog..I answered your questions there. cm
1. You never address the issue of lack of vastly reduced statistical power in your statistical procedures to actually have the claim you make that there is no practical difference. Essentially you are saying “well I don’t see any of these ‘cell’ thingies those scientists are talking about through my 3x loupe, therefore they don’t exist.” That is not good science.
2. You never address the Jul/Aug issue except to shout “BECAUSE THAT IS THE WARMEST PERIOD” as if shouting makes it true. It doesn’t.
Oh, and tamino had another pub just a couple of weeks ago in Journal of Climate as I think I mentioned elsewhere–including in an unpublished comment on your blog that broke zero of your conditions. Pretty good for a mere “activist”, wouldn’t you say?
I see your answer in your blog and posted a reply which should be on your moderation mailbox.
Don’t hold your breath if you don’t agree with him lockstep!
(1) Don’t understand your statement “lack of vastly reduced statistical power in your statistical procedures to actually have the claim you make that there is no practical difference.” What statistical procedure?
(2) Check out my second blog on the topic. I tried it for all summer and all year…SAME result. Much more of a trend in mean and minimum than T max
and even less trend for records or extreme warmth. Plenty of folks in the refereed literature have found this. In any case, nothing wrong looking at only the warmest months (July, August).
There are a lot of activists and hobby types ending up as contributors to papers…including folks from WUWT. Tamino is obviously not objective about the topic (look at the title of this blog!). And I suspect that is true about a number of contributors to the comments. ..cliff mass
If you don’t believe me, about the best I can tell you is talk to an actuary: They make a profession out of quantifying extreme events. Or to tamino who makes my quite doctoral level stats knowledge look like a Chevette.
But I’ll try:
Aggregating means melds the information from each and every datapoint into the values studied. Let’s say we have 30 years of monthly data. We end up collecting 360 means which contain the information from a bit less than 11K values. This results in low confidence intervals and high sensitivity–i.e., with both alpha and beta errors (relatively speaking compared to below, of course).
Aggregating bimonthly maxes melds pretty much nothing. It simply takes the information from ONLY the 30 maxes from the 61 day spans for those 30 years (or other spans as you state). You end up with FAR less significance (in a beta error sense) and FAR, FAR less power in an alpha error sense (actual sensitivity to a true difference).
Interestingly, many, many years ago (15+???) I disagreed with tamino on whether to even work with monthly aggregations for temps rather than annual aggregations for some of these same sensitivity reasons. Even said so here. Of course after these years there is certainly enough data now where no one can deny using the finer aggregation. That said, let’s return to the point: Aggregating is not a simplistic procedure in the least, and there is still a goodly measure of professional art and judgement needed. In particular there is a huge difference between aggregating individual values from spans versus aggregating statistics from those same spans.
Whoops…alpha and beta are switched in error in the 2nd to last paragraph.
Ahhh, Cliff. So you are not familiar with the standard statistical Power Test.
That explains a lot. You certainly should not be passing judgement on other people’s statistical abilities. You need to read AND UNDERSTAND the example jgnfld gave on August 4.
Never knew anyone ever actually read my posts!
Well, even though I only have a passing familiarity with R, I not only read it, but I understood exactly what it was demonstrating.
For Cliff, I suspect the only thing he got was a nice whooshing sound as it went over his head.
When your analyses are published with top scientist co-authors in peer-reviewed journals, please do let us know.
Cliff Mass actually replied to my question:
“5 – More broadly,what is your definition of “extreme heat event”? How can you claim to analyse something you do not define?”
“(5) I am not claiming anything. Showing plots of some key parameters related to heat.”
There you go!!! I guess its all right then…
“Nothing to see here! Please disperse!” – says Lt Frank Drebin from Police Squad
How’s that comment moderation going?
Surprising to see you around here discussing your blog post, telling people to read it through and writing and (attempting I guess?) to defend your extraordinary assertions in said blog post saying… err…. stuff… when back in your own blog, direct comment(s) specifically addressing issues with your post and your follow on explanations remain trapped in moderation limbo?
Strikes one as odd behaviour… one must say…
“nothing wrong looking at only the warmest months (July, August)”
Except, of course, when they are NOT the “warmest months”.
Don’t bother to publish this if I am getting tedious, tamino. But here is a better example R script for cliff to consider since he actually showed an interest:
set.seed(1234L) # to replicate these results
# required library for colstats functions…load if not present:
# Generate 10K random normal series w/ mean=50
# and sd=10 (series are cols here) to simulate temp readings
q = replicate(10000,round(rnorm(61,50,10),1))
q.mean = colStats(q, FUN= “mean”)
q.max = colStats(q, FUN= “max”)
# required library for describe()…load if not present:
describe(q.mean) # se of estimate is .01
describe(q.max) # se of estimate is .05
# se of max is 5X se of mean
hist(q.mean, breaks= 30)
hist(q.max, breaks= 30)
# Note that the mean in the vast majority of these series
# is between 46 and 54 whereas the various maxes
# range mostly from 61 to 91 and show a definite skew.
Which parameter (as much as max even is a ‘parameter’ like mean/var, etc.) has the power to see small changes and which does not? Which parameter makes it easy to hide any changes over these intervals even if they are occurring?
The point is here as I’ve tried to say repeatedly before: Aggregating actual stats which employ all values (or at least a high number of them as in tamino’s 2 different high values aggregations) in their calculation leads to a quite different sampling universe than that found when aggregating single selected extreme values.
Bob…. facts and truth do matter. Here are the climatological daily average max temps for Seattle during the warm season:
Please check your numbers….they are not correct.cliff
Oh, my, Cliff. you are really digging yourself a hole here.
You justify looking at extreme temperatures (highest single daily maximum each year in July-August) by arguing that trends in extremes behave differently from averages.
…and then you justify ignoring extreme highs that occur in June or September – because the average highs for July and August are the two warmest months?????
Make up your mind. Do you want to look at averages, or do you want to look at extremes? “Facts” and “truth” need to be accompanied by consistency.
You check your numbers. Al Rodger already pointed out that the maximum temperature in the year falls outside of July-Aug for 18 years of the record:
I’ve already pointed out that in some of those years, the temperature exceeds the July-Aug maximum more than once (up to four times!) outside the July-Aug period.
The fact is that you have selected a collection of data points that you call “the extreme high temperatures each year” that excludes values that exceed the values in your data set.
Just a followup to myself. Note that the original post here covers SEATAC, and both the post and the early comments note that this “July-Aug” period misses the warmest temperatures in many of the years. The two comments I linked to in my August 9 comment are both looking at the Olympia data that came up later.
In other words, both locations exhibit the same behaviour: extremes outside Cliff Mass’s preferred data range.
Using Cliff’s methodology, you could have a major climate shift in seasonality, with extremes coming earlier and earlier into June, and Cliff would still prattle on about “there is no trend” because he ignores that data. The Morton’s Demon is strong in that one.
Still responding to my own comment.
I don’t know if Cliff Mass has posted any comments that are still in moderation. Or if Cliff Mass will return here to defend his Jul-Aug selection process (with anything more than just asserting it is fine). It has already been pointed out that many Tmax values occur outside Cliff’s selected period. I’ve taken the Olympia data and done a little more to demonstrate how unsuitable that selection is.
To start, during the period 1941-2021, Tmax can occur as early as May 22 (in 2001, tied with the value on Aug 10) or as late as Sept 24 (in 1943). The average date is day 220, or July 19 (and the median date is July 21). Thus, the central tendency is at least 10 days earlier than Cliff’s mid-point of Jul 31/Aug 1. Cliff’s early cut-off is only 18 days before the average date, while his late cut-off is 43 days after the average date.
The distribution of annual Tmax dates is also not normal, or symmetrical. Most of the “later than average” Tmax values are found between day 200 and day 230 (Aug 18) – 41 of 47. This 30-day period is the peak of the distribution, so we have skew. The “earlier than average” Tmax dates are more spread out – 23 of 34 are in the 30 days preceding the average date. Only three of the “late” Tmax dates are after Cliff’s Aug 31 cut-off, whereas 21 of the “early” Tmax dates are before his July 1 cut-off.
Cliff may try to argue that including the extra data will not change his result. If his result is “the trend is not significant”, then he would be correct. As has been pointed out, “the trend is not significant” is not the same as “there is no trend”. (As has also been pointed out, the use of Tmax suffers from having low statistical power.) If you use Cliff’s Jul/Aug restriction, the trend in the Olympia Tmax values is 0.79 C/century. If you include all the Tmax data, it rises to 1.38C/century. R-squared rises from 0.0057 to 0.019 – both low, as the data are noisy, but a noticeable difference.
We can also do a regression on the annual difference between Cliff’s Jul/Aug Tmax and the full year Tmax. That has a slope of 0.59C/century – but we could have figured that out with the previous results, because the three regressions are simple additions, just as the data are related by simple addition/subtraction. Tamino noted that 2021 was an unusually hot year at SEATAC. Olympia is no different: 2021 stands out as a sore thumb. It also stands out as a sore thumb when we compare Cliff’s Jul/Aug Tmax to the full Tmax: Cliff’s approach gets 35.6C for Tmax (Aug 12), while using the full year produces a scorching 43.3C. When does the 43.3C happen? On June 28 – a full three days before Cliff’s cut-off. For a matter of three days, Cliff eliminates a day with a Tmax value 7.7C higher than his preferred value.
We can also look at how the date of Tmax relates to year. Linear regression gives a slope of -0.158 days per year – about 12 days difference over the 83-year record. Still not significant, but if the trend were to continue for another 83 years then the average date of Tmax would probably fall before Cliff’s cut-off. (There is no physical reason to expect that trend to be linear, or continue for another century, but the trend value of -0.158 days per year is just as much of a fact as most of Cliff’s facts.) Note that the way I have determined these dates involves taking the earliest day when that Tmax is found – so if there are ties, only the first date is used.
The fact is, Cliff’s restriction of only considering July and August Tmax values eliminates a lot of relevant data – if your purpose is to look at overall extreme temperatures. So far, his excuse for reducing the data he uses is just hand-waving.
In the interest of trying to make as much mileage as possible out of limited data, I have estimated the derivative – the trend – of some of the temperature records in play here. After all, if we want to think about whether temperatures at a certain place are “really” rising or not, we should look at the trend if we can. The data are very noisy, and the derivative of noisy data is much noisier, so heavy smoothing was necessary. The smoothing method was a series of LOWESS passes with various window widths. With such smoothing, one should not pay close attention to details near the extreme ends of the data, either early or late years. But the general patterns are clear.
I looked at Lind, WA since that is one of the places in Cliff Mass’s post under discussion here. As a comparison, I also got data for Pullman, WA, a town 65 miles away (104 km) from Lind. Its station is about half a mile northwest of town. BTW, some houses on the south side of Lind were burned down a few days ago as a wildfire came through. The Lind station is northwest of town, and let’s hope it’s is still operational.
The graph below show the trends for the annual maximum temperatures for both Pullman and Lind, and also for the annual mean temperature for Lind:
One interesting feature is that both mean and max temperatures at Lind tended to *decrease* until the late 20th century, but the *trends* were actually rising most of that time. If the pattern of greater increases in the trends continues to apply, there should be some max temperature records broken in the not very distant future.
I got the underlying data from the Climate Explorer site (there are several with that name; this one is at https://climexp.knmi.nl.
Pullman max temps: https://climexp.knmi.nl/data/tlist_temperature_-117:_:47__1_-1_:___max.dat
Lind max temperatures: https://climexp.knmi.nl/data/xgdcnUSC00454679.dat
Lind mean temperatures: https://climexp.knmi.nl/data/vgdcnUSC00454679.dat