Record World Cereal Outputs Forecast for 2023/24

“But max temps do *NOT* seem to be rising. Only minimum temps.”

Gorman keeps saying this and will ignore any evidence to the contrary. But for the record, USA maximum temperatures since 1970 are rising at a rate of 0.27°C / decade

https://www.ncei.noaa.gov/access/monitoring/climate-at-a-glance/national/time-series/110/tmax/ann/4/1970-2022?base_prd=true&begbaseyear=1901&endbaseyear=2000&trend=true&trend_base=100&begtrendyear=1970&endtrendyear=2023

Minimum temperatures are rising at an almost identical rate

https://www.ncei.noaa.gov/access/monitoring/climate-at-a-glance/national/time-series/110/tmin/ann/4/1970-2022?base_prd=true&begbaseyear=1901&endbaseyear=2000&trend=true&trend_base=100&begtrendyear=1970&endtrendyear=2023

There isn’t any USCRN data from the 70s, and the data we do have from USCRN shows similar levels of warming since 2005.

Of course it isn’t significantly different from “nada”, that’s just the nature of looking at trends in highly variable data over a short period. By the same argument it’s indistinguishable from a much faster warming rate.

If the trend is indistinguishable from zero, it’s also indistinguishable from a warming rate twice as fast.

So as the CRN data is warming at the rate of 0.5°F / decade, it could be less than zero or it could be more than 1.0°F / decade with the same level of confidence.

Care to expand on that assertion?

There may be ways of saying it’s more complicated than that, and it depends on what is meant by “indistiguishable from”. But I’m interested in what makes you think it is absolutely wrong.

That’s the point. Nobody knows what the true number is because it’s uncertain.

The calculated trend is the best estimate of what the true value is, the confidence interval is the best estimate of the likely range the true value could be and give the same result.

Heard it, understand it, have fun seeing examples of it.

But rather than using it as an insult, it would help if you explained exactly what part of my statement you disagreed with. I fail to see any contradiction between saying we don’t know what the true value is but can make best estimate of it.

Maybe that’s because I suffering from some cognitive blind spot, or maybe you don’t understand the point. Just hurling insults at each other won’t help us to find out which.

“Not an insult.” … “You have no clue what an uncertainty interval depicts do you?”

The uncertainty interval does not have to be equally “CERTAIN” throughout. In this case we are talking about a confidence interval, it tends to a normal distribution, and middle values have a greater likelihood than extreme values.

“Show some backup for your statement!”

What statement? All I’ve said is that the uncertainty is not the same throughout the “uncertainty interval” and that the calculated trend can be considered the best estimate of the actual trend.

I doubt any backup would satisfy you as you keep demonstrating you just don’t understand any of the sources you use.

“You have not even shown that the measurement distribution is normal.”

The temperature measurements do not need to be normal. What you may be thinking of is that the maximum likelihood estimate for a trend usually assumes that the errors are normal, and this is the justification for saying the best fit provided by an OLS regression is the best estimate, in the sense of the maximum likelihood. But that doesn’t justify any of your claims about the uncertainty interval being uniform.

Are the CRN monthly residuals normal? To my non-expert eye, I’d say not. It does, barely pass the Shapiro-Wilk tests (p-value 0.055), but that doesn’t prove they are normal.

It would be interesting to see what effect this has on the MLE for the trend, but I doubt it would make much difference. (I don’t have the time or the expertise to try it myself at this point.) But none of this changes the fact that the uncertainty is not constant throughout the interval.

Incidentally, the residuals for UAH monthly data over the new pause period fail the test Shapiro test, with a p-value of 0.025. Maybe you can mention it to Monckton next time he posts, or does you own cognitive bias only allow you to see a problem when you don’t like the result.

I would be a lot more useful if you answered the questions about your own CDD trends.

“Uncertainty is just that – uncertainty. It means unknown and unknowable. ”

Uncertainty does not just mean unknowable. It’s about figuring out how much uncertainty there is in a measurement – “parameter, associated with the result of a measurement, that characterizes the dispersion of the values that could reasonably be attributed to the measurand”

The measurement is the best estimate of the true value, and the uncertainty interval tells you how certain that estimate is. You don’t know what the measurand is, but you do know how likely it is that your interval contains the correct value.

“The assumption that the distribution across an uncertainty interval approaches normal is the old canard you always fall back on”

In this case the measurement is the slope of a linear regression and the uncertainty interval is the confidence interval of the best fit. That does tend towards a normal distribution, by virtue of the CLT.

If you don’t think it’s approaching normal, what do you think the distribution is?

“all uncertainty is random, Gaussian, and cancels”

Nonsense.

“That way you don’t have to worry about actually addressing uncertainty ”

I’m the one trying to address uncertainty here. You kept insisting there was no uncertainty in a linear regression, you never worry about the uncertainty of the pause, or your claim that GDD’s are declining in the US, or your own CDD trends. I specifically saying there is a lot of uncertainty in the trend using USCRN given the large variability in the monthly values, and the short period of time it’s been in operation.

“Uncertainty does *NOT* have to tend toward normal.”

I’ve never said it does. I keep addressing the possibility that individual measurements might have non-normal error distributions. What I’ve said is that when, what you are measuring is a mean, or a slope, the CLT says that the uncertainty should tend towards a normal distribution, regardless of the distribution of the population, or of individual measurement errors.

“ and how it adds as you combine measurements of different things using different measuring devices.”

I’ve addressed this numerous times, and we’ve had endless pointless arguments over it. The fact you don’t understand how uncertainty propagates when taking an average of different measurements is not my problem. I’ve also tried to explain, with equal futility, that when looking at samples, whether to find the mean or a slope, any measurement errors are largely insignificant compared to the random nature of the sampling. The large uncertainty in the USCRN trend has nothing to do with any uncertainty in the individual stations, it’s entirely based on the fact that month to month variability in the US is large, and we only have about 18 years of data.

“…Something like +0.3C/-0.2C. Or vice versa for rising temperatures. All due to the thermal inertia of the measuring device. ”

If you have a problem with the reliability of the CRN stations, take it up with those who insist it is the only data we should be using. I’m merely pointing out what it actually shows to those who claim they show no warming.

Following that you might want to consider how reliable all your CDD and GDD estimates are, and why you are happy to talk about trends for them with no uncertainty.

“If the distribution of uncertainty is normal, guess what? All uncertainty would cancel.”

I have no idea why you would think that, or what you even think it means. In this case we are talking about the uncertainty in the slope of a linear regression, so I’ve no idea how you think that would cancel.

“The mean would not only be the true value, but it would also be 100% accurate!”

A tautology, and I’ve still no idea what you are talking about here. Do you?

The very fact that we are talking about a confidence interval is because we know the slope is not 100% accurate.

“One data point does not make a distribution.”

Again, it can be an assumed distribution. It isn’t something that only exists once you have generated an infinite number of data points.

Pretty much everything else you rant on about are things you know are not things I’ve said. They are just the strawmen who live in your own head.

An uncertainty distribution may or may not be Gaussian. Being Gaussian does not mean all errors cancel. Random errors will tend to cancel regardless of the distribution.

“Uncertainty is uncertainty. It is an unknown. When you assume a distribution then you are saying that the uncertainty is KNOWN.”

Again, if uncertainty can never be known there is no point studying it, and there is no point in describing it. You keep confusing uncertainty with knowing the true value, when it means the opposite. Saying something is uncertain means you do not know the true value. Uncertainty in sense it’s used here is a known (or at least estimated) interval which describes what you do know about the measurement.

Uncertainty: parameter, associated with the result of a measurement, that characterizes the dispersion of the values that could reasonably be attributed to the measurand

That is not saying “you know nothing”, it’s describing what we do know about the uncertainty of a measurement.

“When you assume a distribution then you are saying that the uncertainty is KNOWN. If you can’t recognize the cognitive dissonance in that then you are simply too dense to ever understand uncertainty.”

Take it up with the GUM.

The proper use of the pool of available information for a Type B evaluation of standard uncertainty calls for insight based on experience and general knowledge, and is a skill that can be learned with practice. It should be recognized that a Type B evaluation of standard uncertainty can be as reliable as a Type A evaluation, especially in a measurement situation where a Type A evaluation is based on a comparatively small number of statistically independent observations

So much nonsense, just to demonstrate how little you understand, but you you still won’t explain your own trend lines. No mention of the uncertainty there, it’s only CRN data you find suspect. I can’t think why.

“”A single data point” does not define a distribution.”

Who ever says it does. The point, you keep avoiding seeing, is that if you can estimate an uncertainty distribution – a type B uncertainty to use the GUM’s jargon – you do not need multiple data points to define it. It’s already defined.

By the same token, two data points do not define a distribution, nor 3 or 4. In fact the only way to completely define a distribution from data is to have an infinity of data points. Anything less is an approximations, and then only if you accept the statistical arguments you keep trashing.

As the GUM says, and I’ve already quoted to you, a type B uncertainty distribution based on expert assessment can be more reliable than one based on a small number of data points.

“If a Gaussian distribution is not symmetric then it is not a Gaussian distribution.”

Correct – but I’ve no idea why you needed to state it. It’s got nothing to do with what I said.

“If it is not Gaussian then none of the common statistical descriptors apply, not the standard deviation, not mean, and not the variance, none of them describe the distribution properly.”

Absolutely incorrect. Please do yourself a favour and actually read up on the subject rather than relying on your own misconceptions.

Half the time I’m being attacked for not stating a variance every time I mention an average, and now you insist that the variance is meaningless.

The mean, standard deviation and variance are statistical parameters that apply to any distribution.

This is all new to me, so take the following under advisement, but an alternative way of calculating trends which should not depend on normality, is quantile regression, which fits to the median (or any other quantile) of the data, rather than the mean.

https://data.library.virginia.edu/getting-started-with-quantile-regression/

Using the quantreg for R, and default settings, I find that for the US CRN monthly data, the trend for the median value is slightly larger than using standard liner regression. With a trend of

0.74°F / decade, that is 0.41°C / decade.

This graph shows the OLS regression in blue and the median regression in red.

Overall, the difference between the two isn’t significant, and there doesn’t seem to be any reason to reject the OLS trend. It’s just that this is a short period and the trend may be influenced by the few outliers.

Yes, I’ve used the data from that very page. The problem is that it’s very difficult to see any actual trend just by looking at the highly variable monthly data. It’s difficult enough with he annual data. You have to actually do the calculations.

Based on the monthly data I make the trend 0.52°F / decade, with a 95% confidence interval of ± 0.49. This is just about significant, but I’m not correct for any auto-correlation. If I did, it would certainly not be significant.

The total warming is about 1°F over the last 19 years – or 0.56°C. But look at the scale of the graph. Each line is 4°F. The total warming only goes up a quarter of one of those gaps. It’s impossible to see just be looking at the monthly data without actually drawing the line.

“I keep providing you the proof but you can’t remember it for more than 24hours!”

I may well be getting senile as I don’t remember you presenting this particular “proof” before – could you provide a link to the last time you made the argument?

“I just pulled Cooling Degree-Day data for four randomly chosen locations in the US going back to 2000.”

You were claiming maximum temperatures were not rising. But no you pivot to CDD. And my data was from 1970, you are only starting in 2000. And I was showing the data for the whole USA, you are just using 4 “random” locations – with no link tot he actual data. How much certainty do you have in a random sample of 4 stations?

“Here are the trend line equations for the four.”

Could you actually say what those trends actually mean? You’ve made a habit of this for years, quoting a figure with no reference to the units.

“+.0000515 + 1.14 (F)”

Is +0.0000515 the trend or the intercept? What is your starting date, and what time units are you using? And is this for an annual total of degree days, or what?

“If maximum temps were going up globally you would expect to see at least one of these four showing cooling degree-day values going up as well.”

You are showing 2 going up.

“The three that show an increase are out in the hundred thousandths digit – far beyond the ability to measure accurately using today’s field measurement devices.”

So now you are claiming three are rising.

Don’t blame me if your data doesn’t show anything useful. Look for better data is you need to prove that maximum temperatures are not rising.

“HDD and CDD values are *FAR* better measures of climate than trying to trend averages of daily mid-range temperatures”

I was showing you maximum and minimum trends. That’s what you were claiming – no increase in maximum temperature.

Here’s NOAA’s chart for CDD since 1970

https://www.ncei.noaa.gov/access/monitoring/climate-at-a-glance/national/time-series/110/cdd/ann/10/1970-2023?base_prd=true&begbaseyear=1901&endbaseyear=2000&trend=true&trend_base=100&begtrendyear=1970&endtrendyear=2023

An increase of 65°Df / decade.

You’ll reject this of course, as you donl;t like the way they calculate CDD (i.e. the same way they calculate GDD in the paper you keep going on about.)

“No! I am not going to waste my time trying to find posts well in the past that showed EXACTLY THE SAME THING.”

As I thought. Maybe your own memory is playing tricks. I do remember that time when you were making some point about maximum temperatures in one site not warming, and refused to accept that your data was obviously wrong. And then when you finally, after I’d had to download your own spreadsheet to show you were wrong, you tried to correct it and resulted in even worse errors, which left the warming rate at around 1°C / decade.

You were also at that point claiming that you could turn a positive trend into a negative one, simply by converting from Fahrenheit to Kelvin – again based upon a simple failure to check your own data.

I bring this up to explain why I don’t immediately assume your claimed trends this time are beyond reproach.

“RISING TEMPS WOULD CAUSE CDD TO GO UP!”

Not necessarily. And you were accepting that temperatures are rising, just not maximum temperatures. And, again, your argument isn’t any more persuasive by being written in all-caps.

“We’ve been over and over and over this in the past. And you never figured it out!”

And I’ve never suggested it was anything else than how much and how long it is over a base value. For some reason you have to believe you have a superior understanding of this.

“All you are doing now is whining!”

At least I’m not shouting.

“You are getting so senile you can’t even remember what I said in the post you are replying to! ”

Rather than resorting to more ad hominems. You could have tried answering my questions. In particular what units are your trends meant to represent?

“And now we are back to you being the troll!”

Try answering the question.

““So now you are claiming three are rising.”
NO! THAT IS *NOT* WHAT I SAID! Can’t you read at all?”

Your exact words were

“The three that show an increase …”

If that doesn’t mean that three are showing an increase, what does it mean?

“There is no useful trend that can be established because the slope lies within the uncertainty interval for over twenty years!”

Which just shows there isn’t much use in looking at just 20 years of individual stations. You’re the one claiming maximum temperatures are not rising, and now that CDD’s are not rising globally. If your method of showing this is to use data that couldn’t possibly show any rise, then your method is not very useful.

“Do you see an uncertainty associated with that graph ANYWHERE? How do you judge if this data is useful at all?”

There was no uncertainty given for your data, but you claimed it was meaningful.

There’s no uncertainty quoted in the agricultural paper you think is meaningful.

“This appears to be evidence against the theory of GHGs being responsible for increasing global average temperatures”

Could be. It could also be evidence against the warming being caused by UHI. I just point out what the data shows.

I should say that whilst daytime maximum temps seem to be warming at around the same rate as minimum temps, there is more of a difference between cold and hot seasons.

E.g. for the same NOAA data the US winter months (Dec – Feb) since 1970, show minimum temps warming at 0.34°C / decade, whereas summer maximums only increased at the rate of 0.21°C / decade.

I doubt it’s quite as simple as that. CO2 does affect day time temperatures, a) because it still works as a green house gas during the day, and b) because warmer nights will also tend to lead to warmer days.

I think you would be correct to say if there was only warming at night and none during the day it would indicate that this was unlikely to be caused by the sun, but that doesn’t mean warming during the day can only be caused by the sun.

You know full well what I meant. Warming in the context of days and nights getting warmer over time.