Weather Channel Founder John Coleman: There is no significant man-made global warming at this time

c grier
October 29, 2014 at 5:05 pm

Let me start by saying that I’m not an expert. But no, you don’t want to use differentiation to try and reduce the noise. When looking for a small effect you need to integrate the data over time. Brian H was on the right track. By doing a “diff” analysis you are filtering the data and coming out with noise – all the signal is gone. If you want me to find a citation for this I’m sure there’s on on a statistics website somewhere.

Then think of it as an anomaly. It’s basically daily sampling with the daily 24hr solar cycle removed. What’s left is the seasonal signal + the weathers signal + Co2 warming + any other underlying trends.
Then it’s all integrated together.

If you look at the first graph you posted on Oct. 23 at 9:43 you’ll see what I mean. The Max temp anomaly and the Average Temp Anomaly are almost *exactly the same*. This means that the high averages are being “cleaned up” in a different way than the low averages. (Or that the Max and Ave calculations are not doing what they are supposed to do) The Min. Temp Anomaly (whatever that really means) is showing big spikes because of bad data points in the data set.

Except they are not, and that Min temp data is the basis of the average value as it’s calculated from the min and max values.
Again what I was digging for is any sign of a reduction in night time cooling. But calculating a Diurnal swing based on a single calendar day (which is what they’re doing) totally misses the point, what matters is how much temps went up yesterday and how much it falls that night, which requires data from 2 calendar days. Day over day min temp difference is this value.

To put it a different way, truly random noise is hard to find, but eventually it should return to zero – if it is summed over a sufficiently long period. If you start looking for signals inside of noise you’ll be surprised what you find, or what you think you’ve found:

Long period of time or a lot of samples, I’m working with over 120 million samples, for most of the last 30 or so years there’s 2 to 3 million samples per year.

I guess what I’m saying is that your specific findings should be reproducible by other teams, or using other data sets. So if there is a significance to the odd behavior in the early 70s, based on your analysis, it should show up when comparing to other work. Have you found any other studies by skeptics showing similar flat results over the same periods?

The early 70’s is an artifact of a very large reduction in samples.
Other teams could probably reproduce this if they tried. The only real caveat is I don’t infill for missing samples and un-sampled areas. For instance there is no arctic amplification in the actual surface measurements, so the only place they are coming from is the temp model being used, it’s made up. Now there might be arctic amplification, but it’s not in the measurements we have.
So, I don’t think your observation is correct, But I’m not an expert either, so I am looking for an expert who can either help make this better, or who can articulate some fatal flaw that I haven’t thought of over the last 5 or so years I’m been working on this, please speak up. I’ve been writing on this topic for a while, did you look through much of that, or gotten any of the data from sourceforge?

c grier commented

If you’d like to take this off line, I’ll be happy to do that.

Here’s fine, until someone says otherwise I guess.

You’ve obviously spent a lot of time running these numbers, so I’m sure you want to know if there’s something that’s gone missing.
I think we’re talking past each other a bit, as you don’t seem to be checking my points against your data

I pretty much have the code memorized, and I’ve tested it a lot. I did find a mistake a year ago, thought I was looking at min diff, but was actually looking at max diff. I’ve been even more careful reviewing it since then. But I checked it again this morning for you 🙂
There’s 2 areas that do the work. During read for each station I read all of that stations data files in order and construct the differences by saving the first temp record to use when I insert the data row, I can calculate the difference value.

ymxtemp – ymntemp,
ymxtemp – to_number(trim(SubStr(file_line,111,6))),
ydate,
to_number(trim(SubStr(file_line,103,6))) – ymxtemp,
ymxtemp,
trim(SubStr(file_line,111,6)) – ymntemp,
(ymxtemp – ymntemp) – (ymxtemp – to_number(trim(SubStr(file_line,111,6)))),
ymntemp,
trim(SubStr(file_line,25,6)) – ytemp,
ytemp);
ydate := to_timestamp(SubStr(file_line,15,8),’YYYYMMDD’);
ymxtemp := trim(SubStr(file_line,103,6));
ymntemp := trim(SubStr(file_line,111,6));
ytemp := trim(SubStr(file_line,25,6));
else
ydate := to_timestamp(SubStr(file_line,15,8),’YYYYMMDD’) + 1 ;
ymxtemp := trim(SubStr(file_line,103,6));
ymntemp := trim(SubStr(file_line,111,6));
ytemp := trim(SubStr(file_line,25,6));
end if;
else
if ymxtemp -199
and trim(SubStr(file_line,111,6)) -199
and trim(SubStr(file_line,103,6)) -199
then
ydate := to_timestamp(SubStr(file_line,15,8),’YYYYMMDD’);
ymxtemp := trim(SubStr(file_line,103,6));
ymntemp := trim(SubStr(file_line,111,6));
ytemp := trim(SubStr(file_line,25,6));
else
ydate := to_timestamp(SubStr(file_line,15,8),’YYYYMMDD’) + 1 ;
ymxtemp := trim(SubStr(file_line,103,6));
ymntemp := trim(SubStr(file_line,111,6));
ytemp := trim(SubStr(file_line,25,6));
end if;
end if;

The y variables are “yesterday”, ymntemp is yesterday’s min temp. I check each min and max temp to make sure it’s in the range of -199 F, if both aren’t valid, I skip them, go to the next record. Mean Temp, is calculated from min and max, they are identical for all records (ie mean temp = min + max/2). In the data record temp (mean temp) is the 6 characters starting at position 25, max is 103, and min is 111. This line

trim(SubStr(file_line,25,6)) – ytemp

creates the Average temp diff, today’s temp – yesterday’s temp. This is the only code that creates this value.
Then, I average them together here.

when Type_ = ‘Y’
then
Selectfield := ‘avg(max_temp) as MaxTemp,
avg(min_temp) as MinTemp,
avg(rising_temp_diff) as Rising,
avg(falling_temp_diff) as Falling,
avg(Diff) * 365.25 as YrDiff,
avg(Diff) as Diff,
var_pop(Diff) as V_Diff,
0.141 * count(Diff) * 2 / power(count(Diff) * 2,2) as Diff_Error,
avg(case when temp > -199 and temp < 199 then temp else null end) as Temp,
0.316 * count(temp) / power(count(temp),2) as Temp_Error,
avg(MNDiff) as MNDiff,
avg(MXDiff) as MXDiff,
avg(AVDiff) as AVDiff,
sum(MNDiff) as MNSum,
sum(MXDiff) as MXSum,
sum(AVDiff) as AVSum,
avg(case when dewpoint < 9999 then dewpoint else null end) as DewPoint,
avg(DewPt_to_RelHumidity(temp,dewpoint)) as RelH,
avg(case when SEA_LEVEL_PRESSURE < 9999 then SEA_LEVEL_PRESSURE else null end) as SeaLevelPressure,
avg(case when STATION_PRESSURE < 9999 then STATION_PRESSURE else null end) as StationPressure,
avg(case when PRECIP < 99 then PRECIP else null end) * 365.25 as Rain,
Count(temp) as Sample';
Selectlist := 'avg(maxtemp) as MaxTemp,
avg(mintemp) as MinTemp,
avg(rising) as Rising,
avg(falling) as Falling,
avg(YrDiff) * 100 as YrDiff,
avg(Diff) as Diff,
var_pop(V_Diff) as V_Diff,
avg(Diff_Error) as Diff_Error,
avg(temp) as Temp,
avg(Temp_Error) as Temp_Error,
avg(MNDiff) as MNDiff,
avg(MXDiff) as MXDiff,
avg(AVDiff) as AVDiff,
sum(MNDiff) as MNSum,
sum(MXDiff) as MXSum,
sum(AVDiff) as AVSum,
avg(dewpoint) as DewPoint,
avg(relh) as relh,
avg(SEALEVELPRESSURE) as SeaLevelPressure,
avg(STATIONPRESSURE) as StationPressure,
avg(Rain) * 365.25 as Rain,
sum(sample) as Sample';
Grouplist := 'year';
Insertlist := 'MaxTemp,
MinTemp,
Rising,
Falling,
YrDiff,
Diff,
V_Diff,
Diff_Error,
Temp,
Temp_Error,
MnDiff,
MxDiff,
AvDiff,
MnSum,
MxSum,
AvSum,
DewPoint,
relh,
SeaLevelPressure,
StationPressure,
Rain,
Sample';
InGrouplist := '''9999'' as year';
GroupBy := '' ;

If it’s a Y yearly report, I create a table of these values, and group them year, I do this on a set of stations I select by lat, lon and date.
A single line makes the data in that chart.

avg(AVDiff) as AVDiff

This BTW is the same as how min and max temps are processed, I also do some different things trying to get an idea of error bars, but I don’t know what I’m doing to I don’t “publish” those. But this is the same source data that the core of all of the temp series made, there just aren’t that many stations. But if any of them are correct, which I think they are, they are all correct, or at least a faithful representation of the measured value. You’re welcome to get the code yourself, just follow the url in my name.
So, I am comfortable that graph represents the data, I was surprised when I saw it, I don’t particularly understand it, same with the regional swings in min, but that is what the average of those values is, that I’m sure about. This is why I give Mosh such a hard time, these are the fracking measurements, what everyone else does with it looks completely different, I have to presume it’s because they decompose the temp by location, and then use what they have to fill in everywhere they don’t have a value. What they do is so complicated they have to make up fake data to try and test it. What I’m doing a patient 4th grader could do on paper. What they have yet to figure out is the more complicated something is like this, the more creative they have to be in coming up with ways that it might go wrong, and have to make specific test for them, but they can only test what they think of. All I got to do is check some simple math. I also presume I know less about how the value might be wrong than the people who recorded it and don’t try to fix it.

But what you are really missing is that the very very small changes to the maximum rate of cooling during the night are not present in the data after you do your number crunching.

Yes, had I found hourly data instead of min/max data I would have. But I accepted the longer integration time of min and max, and fundamentally they are doing the same. But, hourly data would have it own difficulties, how to you know if the swing in temps is the storm, or the clear skies after the passage of a cold front? Early on I spent a lot of time pondering just these differences, and decided what I was doing the superior analysis. Plus, I can turn the Sun off and see it’s impact, no one else does that. I do the same processing, just aggregate it daily. You can see how much a handful of seconds changes the surface temp, with all of the other things that clutter up this measurement averaged out.
Oh, perfect analogy, from mph through the traps, and weight of the car, you can tell how much hp it has, and how quick you should be able to go through the qtr mile, with the only real variable being tire spin, which artificially raises mph some. I love how a 8,000+ HP car can break 5-10 feet from the starting line, and still go quicker than really fast cars.
So you tell if you have a performance problem.