You’d think the answer would be obvious, but here we have a NOAA operated USHCN climate station of record providing a live experiment. It always helps to illustrate with photos. Today I surveyed a sewage treatment plant, one of 4 stations surveyed today (though I tried for 5) and found that for convenience, they had made a nice concrete walkway to allow servicing the Fisher-Porter rain gauge, which needs a paper punch tape replaced one a month.
Here is what you see in visible light:
Here is what the infrared camera sees:
Note that the concrete surface is around 22-24°C, while the grassy areas are between 12-19°C
This station will be rated a CRN5 by this definition from the NOAA Climate Reference Network handbook, section 2.2.1:
Class 5 (error >~= 5C) – Temperature sensor located next to/above an artificial heating source, such a building, roof top, parking lot, or concrete surface.”
Now a caveat: There had just been a light rain, and skies had been overcast, it had just started to clear and you can see some light shadows in the visible image. Had this rainfall and overcast not occurred, the differences between grass and concrete temperatures would likely be greater. Unfortunately I was unable to wait around for full sun conditions. The air temperature was 58°F (14.4°C) according to my thermometer at the time.
Here is another view which shows the NOAA sensor array, the sky, and the evidence of recent rainfall as evidenced by the wet parking lot:
Why NOAA allows installations like this I’ll never understand. And this station is a USHCN climate station of record, used in who knows how many climate studies.
I’ll tell you more on this station and others I surveyed tomorrow.
The web community appreciate and congrats You for the good job!!
Thank You Very Much!
Unspecified standard?? Apparently you know nothing about what’s going on here. There are very specific standards for placing these stations, and they’ve been violated left and right, that’s what this is about. Violations cause problems. I suggest you find out before you open your piehole.
So concrete appears and disappears from day to day does it? Interesting property. I wasn’t aware of it.
It appears and does not disappear. The jump in temperature is not adjusted for and is recorded an an increase in trend. After that, the trend itself is exaggerated by the Heat Sink Effect.
The problem is that exurban creep is overtaking the stations at a MUCH faster rate than it is taking over the surface of the globe. This results in a false reading of both temperature and trend.
Only on an individual station basis, and the change points are very infrequent. Concrete doesn’t come and go like rain.
Unfortunately the switchover from Stevenson Screens to MMTS equipment has directly resulted in a massive number of violations, all occurring since the 1980s. The switchover to MMTS is why there are so many CRN4 violations.
More to the point, GISS and HAD correct for this stuff.
One would have though and expected so. But, according to SHAP records, unfortunately not. Yes, it is both inexplicable and inexcusable.
Don’t be too harsh on the man, Jeff.
His questions and concerns are easily answered. Let us merely correct hat we see as his errors, direct him to what we deem to be the accurate answers and allow him to assuage his concerns. He considers himself to be in “enemy territory”. It is therefore incumbent upon us to put the case to him in as reasonable and open a manner as possible.
I didn’t mean “doomed” in that sense, sorry if you took it that way, and I’m glad you got back safely.
I also think it’s excellent that you’re providing this information to NCDC. What I would focus on is the somewhat poor state of the enclosures you often turn up. Those are built to a standard design, and if they’re not properly maintained the measurements become unreliable.
What I am trying to say is that you’re drawing too much out from your observations.
1. You’re trying to extend the standard design of the box to the siting of the box. They’re supposed to be measuring what’s actually there, not some measurement of what might be there if you completely rebuilt the site.
[REPLY: No I’m not, I don’t know where you get this from.]
2. What you’re looking at is noise. All data series have noise. How often does that concrete path get replaced? Not too often and it will only cause a single step jump in the series, not invalidate the whole lot
[REPLY: Noise is omnipresent, but a step change is not noise, it is a constant bias up or down, and if not removed/adjusted for, will create a false trend. Station moves or siting changes create those steps, that’s the issue.]
3. Changes in the site will not always cause that step jump to be one directional. As I say noise. [REPLY: As I point out, it could be up or down. Shading/tree growth sometimes makes Tmax go down, but could make Tmin go up due to IR reflectance at night. Point is, we shouldnt have to disentangle this stuff from the data in the first place. Gavin Schmidt at GISS says we could get by with 60 stations, I’m betting we’ll be able to find that many in the USHCN data set that are relatively bias free.]
4. (key point) You’re arguing that noise invalidates the signal itself. How? (This is the sense I meant with “doomed”, your venture to use noise to invalidate a signal is doomed)
[REPLY: noise, true noise, is bideirectional in amplitude over time, again step changes are what we are looking for. Put a slab on concrete under a sensor, you get a step change that remains. Put a sensor closer to a building, you’ll get a step change that remains, build up an asphalt parkign lot around the sensor where there was none before, you’ll get a step change that remains. If these are not detected and appropriately adjusted, you’ll end up with a time series trend that is not truly representative of the air temperature measurment for “climate” but one of the changes in the local measurement environment.
THOUGHT EXPERIMENT: Think of it this way, you have a slow flashing on/off lightbulb and a photocell in a room, and you have a meter connected by wire to the photocell to measure peak output when on, but you don’t get to look into the room. If you are trying to measure the decay in luminance peak output on the incandescent light bulb over time as the filament ages, and somebody started adding other syncronized slow flashing light bulbs to the test room, at random, your graph of light output would change. If you don’t know when those were added you might miss them depending on the magnitude of the bias. Many small additions might not be caught, yet they would add to the overall peak output. ]
Last quibble. Please don’t dig into your blog data, extract my name and use it publicly – I have good reasons for not wanting you to do that, including contracts I’ve signed in my employment.
Do you have a privacy policy here? Cause I think you may have just violated it.
REPLY: Sorry I just addressed you as I would anyone, you name in email appears in my form. Removed.
I doubt any reasoned explanation will help. He obviously came here to poke fun and call people names, not to debate.
A step change – aka ‘glitch’ is not noise? If I render these data series as an audio signal, that step change will sound like a click. Noise in other words.
And so long as I look at deltas rather than absolutes, it will look even more like noise.
Sorry, don’t buy the argument.
(Thanks about the name thing)
REPLY: I’m familar with audio circuitry and signal processing, and while a step change will indeed sound like a click, it essentially becomes a DC bias to the signal that shifts the AC component of the signal upwards or downwards, depending on what the step change is.
Here’s a good example of a step bias from about 1998 to present:
http://data.giss.nasa.gov/cgi-bin/gistemp/gistemp_station.py?id=425724650000&data_set=1&num_neighbors=1
Ok, noise, but a DC bias. I can agree with that. Good metaphor, but I can remove the bias by using deltas. I still think it’s noise, but nevermind.
And sorry I can’t see any bias in your Goodland/Renn example. When did the environmental/siting change occur and what was it?
Can I approach this from another direction? How many stations are there in the US and the world? Lots. We’ve agreed that these step changes occur infrequently, and can go in both directions.
How often does a positive (meaning increased temperature) change occur in the population of stations? 5% a year you reckon (sounds about right, once every 20 years)
How long’s the modern warming period? 30 years. So approx 7% of stations have seen a one-off step change.
Now in the combined series that constitutes noise, a fair bit of noise no argument, but not enough to obscure the signal.
What I think you’re trying to do is to argue that the noise is driving the signal. That there is more or less no signal and that what we are finding is not a signal, but a positively skewed DC bias.
Can I be a bit rude and characterize that as disingenuous? GISS (and also Hadley I believe) control for this by comparing nearby stations. So if there are 10 (say) stations in a cluster and 1 undergoes an environmental change every 3 years or so, that will constitute only a small error which will be washed out in the processing. Errors like this occur all the time in all sorts of different situations and we’ve got pretty good at compensating for them, particularly in the physical sciences where ropy data is often used to validate a model. Arhennius’s heat balance model is not particularly complex, so ropy data can go a long way to supporting it. Which is what we find.
Now I believe that GISS didn’t start to apply that correction until after the “heat island” criticism was leveled at them, and that they reran their datasets, and found the differences to be marginal. Isn’t that correct?
REPLY:
“…but I can remove the bias by using deltas. I still think it’s noise, but nevermind. And sorry I can’t see any bias in your Goodland/Renn example. ” My point exactly. If you don’t know the bias was introduced by some other means, how would you know without a site survey? Right now, using USHCN1 methods, they simply don’t know. They have no metadata to correlate site quality and potential bias. Which is why stations like Miami, AZ and Lamapsas, TX get missed, and homogenization makes them worse. USHCN2 will do better at catching changepoints, but it won’t catch all.
“We’ve agreed that these step changes occur infrequently, and can go in both directions.” Yes they can go both directions, but the majority are positive. As for “infrequently”, no we have not agreed on that, that is what you think. As it stands now, only 13% of stations on the USA surveyed so far meet the siting criteria of either the older 100 foot rule or the newer CRN siting system. That means 87% have some sort of bias. Most observed biases are positive. ROW is even less quality controlled.
See the chart on the front page of http://www.surfacestations.org
Any experiment where 87% of the data has bias issues would not be considered acceptable by any engineering or scientific standard, yet here we are, arguing the issue. Go ahead, defend it. The real point is, NCDC has recognized that the old USHCN network is systemically polluted with biases of varying magnitude, as we’ve demonstrated, and NOAA is replacing it. See the press release on April 24th. Your argument about replacing sites is past tense. The decision has been made.
As for adjusting with nearby stations, not knowing (and they don’t, GISS has no site quality factor, neither does Hadley) which are contaminated by microsite biases and which are not then homogenization makes assumptions without evidence to back it up.
“What I think you’re trying to do is to argue that the noise is driving the signal.” No. I’m not. Noise and offset are different. What I’m arguing is that there has been a systemic series of offsets, mostly positive, and that has created an overall systemic bias.
“Can I be a bit rude and characterize that as disingenuous?” Sure why not, that’s been your MO so far. Look, if it was disingenuous, NCDC would not have invited me to speak there, and offered full access and interaction with the staff on April 23rd/24th. If the work wasn’t relevant, they’d never have offered. They’d just ignore it.
The point is, they agree that siting matters. They are replacing the USHCN network with a new one that will have quality control, and they want me and the volunteers to continue documenting the condition of the old network. Why? Because they have no metadata to show them the site quality, unlike the new CRN which has that from the beginning.
If siting quality is not important, the new CRN would not exist. The new USHCN-M network which will replace the old USHCN would also not be happening.
This is just a comment, but I can’t let this go:
“[REPLY: No I’m not, I don’t know where you get this from.]”
Your posting here is basically a recommendation that concrete paths not be built under weather stations. So what I get from that is that you *are* saying that the site should be rebuilt.
I got my comment directly from the implication of this posting. The only alternative is that you’re saying something along the lines of “measurements from this site are rubbish and should be ignored, but let’s not do anything to correct the problem”
So should this site be rebuilt or not?
REPLY: Well its a moot point because NOAA is abandoning all these older USHCN sites for the reasons I’ve specified, undocumented biases, and setting up new ones. See the press release from my day 2 visit to NCDC. My goal was to find the best sites of the USHCN network so that a bias free signal could be examined. That is still the case. The issue is finding the one that are uncontaminated so that the best record can be extracted from them. NOAA has no adjustments to properly account for microsite bias in USHCN1 algorithms. The new USHCN2 algorithms may fix some of these issues but may not catch them all.
JM: It’s not two-way bias that’s being introduced. They are almost exclusively heat sink and waste heat biases. These biases are not only not being adjusted for, but they are corrupting the UHI adjustment because urban sites are being compared with corrupted rural sites.
Consider that a sort of “positive reinforcement” of warming bias.
Anthony, are you going to respond to my (18:40:56) comment that remains unmoderated?
Once that’s done, I’ll respond to your reply to my (18:58:49) comment.
REPLY: Pushy aren’t you? You don’t even give a guy time to type it out. Done
JM (22:22:52) :
“So concrete appears and disappears from day to day does it? Interesting property. I wasn’t aware of it.”
Sure it does. On sunny days it has a huge effect. On cloudy days it has a small effect. On sunny days in June (especially northern US) it has a bigger effect than sunny days in December. When it gets cold and snowy enough to be under a couple feet of snow it has no effect.
Hey, how about if we specify some standard for placement of weather stations such that the data may not quite measure what people experience in their daily activities but that the data is the least corrupted by variables that are being mostly ignored. Better yet, let’s use the standard you want to discard.
Hey, I got another idea – how about if climate researchers looking at ground stations pay more attention to wind, clouds, snow cover and use it all to come up with an order of magnitude more questionable studies. That should keep the dialog frothing.
Seriously – there ought to be some interesting statistics lurking in some of that data. Not much point in digging into it given how much controversy we have with just temperature data. Still, there are a lot of good reasons to keep raw unadulterated data around for future use.
What I think you’re trying to do is to argue that the noise is driving the signal. That there is more or less no signal and that what we are finding is not a signal, but a positively skewed DC bias.
I (generously) estimate that the one-directional noise exaggerates the trend by a factor of two since 1980.
“I (generously) estimate that the one-directional noise exaggerates the trend by a factor of two since 1980.”
How do you know that when you don’t even know the magnitude of the “bias”?
Take this site. Your argument is that the thermometer is being heated by concrete right? Concrete that can be as much as 4-5 degrees hotter than the air? Is that right?
How does that radiative heat reach the thermometer (which is in vented box)? It has to heat the air above it which by your own account here was about 5C less.
And that’s a large mass of open air for that concrete to heat, any hotter air would mix pretty quickly. A radiator 5 degrees hotter than the ambient temperature is not much of a radiator.
Now since you disagree with on this, I’ll give you a challenge. Take the record from this site, and find the bias. See if you can identify when the concrete went in. You won’t be able to find it. I betcha.
And if you can’t do that, find out what this station recorded as the temperature for the time you were there. I’d bet on pretty close to the 14.5C you measured yourself (perhaps less as I presume your thermometer was both hand held and exposed to direct sunlight)
“Any experiment where 87% of the data has bias issues would not be considered acceptable”
Firstly, I don’t accept your estimate of 87%. Let’s assume that over 30 years 87% of the stations had an environmental change that positively affected them. On average that change would occur at the 15 year point. Therefore only 43% of the data has “bias issues”
Secondly, in nearly every scientific experiment there are what you call “bias issues”. They can be filtered and controlled for, and they are.
Thirdly as I said to you before, I can’t see radiative energy in this case having much of an effect.
And I’m also perfectly ready to accept that there are problems with this data. It’s called noise, and analysis can still recover the signal. I do not believe that the noise is the signal itself.
But I can be persuaded otherwise if there is sufficient evidence. Do the exercise with this site and tell me the magnitude of the “bias” and the date on which that concrete was placed. If your argument is correct, you should be able to find it in the daily record (it can’t have taken more than a day to lay that concrete).
REPLY: Well I guess for now, until I can get the daily data in ASCII form we’ll have to agree to disagree. At least you aren’t calling the project “stupid or doomed” anymore. NCDC does not publish daily in ASCII form, only monthly. I have asked them for several data sets, when and if they provide it I’ll be able to d an analysis.
“It’s called noise, and analysis can still recover the signal. I do not believe that the noise is the signal itself.” Ok fair enough for now, but we still differ in opinion. But here’s the thing. Prior to now, (and still pending for network wide use) NCDC was not doing an analysis to remove this type of noise aka “step functions”. I have test case sites already where we know when the change was made and we’ll be able to run that and see if the new USHCN2 catches them.
In the meantime, see the Yilmaz paper which shows 7.5C for over concrete exposure compared to grass if you don’t believe “radiative” effects. Both Lampasas,TX and Miami, AZ show effects of placement change also.
Finally, ‘Firstly, I don’t accept your estimate of 87%.” Its not an estimate, it’s a measurement from survey data of station siting now. Right now 87% of the USHCN network has siting issues by either the older NOAA 100 foot rule or the CRN rating system. NCDC accepts it, and one of the USHCN2 authors used it in a study he presented to me there, which is good enough for me.
Oh, and please try to find the temperature recorded at this site for the time you were there. You say you have good relations with NCDC, you shouldn’t have too much trouble getting their help.
“NCDC would not have invited me to speak there, and offered full access and interaction with the staff on April 23rd/24th. If the work wasn’t relevant, they’d never have offered. They’d just ignore it.”
This doesn’t surprise me at all. You’re driving round the country doing free site surveys for them (that they probably haven’t had the money for) and basically making a business case to get that money. Money they need to improve their service and fulfill their mission.
You should be congratulated, it’s a public service.
It still doesn’t mean that noise is a signal.
Yilmaz: You can’t expect me to take that seriously. Refer figure 3 of his paper for the locations of the stations (the study was conducted at an airport, it also has nothing to do with climate but is talking about the comfort of cities and whether grass is a better surface for open space than concrete)
Concrete: Right next to the runway. Jet engines are hot you know
Grass: Between 2 runways, but not that close to either
Soil: Quite some distance away in what looks like it might be a service area
Do the exercise for this site please. If the effect is there, you’ll be able to find it.
Sorry a reference in case you can’t find it: http://www.ejournal.unam.mx/atm/Vol21-2/ATM002100202.pdf
REPLY: No worries, Evan Jones posted it here oroginally.
“Right now 87% of the USHCN network has siting issues by either the older NOAA 100 foot rule or the CRN rating system”
I think you missed my point here. The data extends over 30 years (or even longer), and I don’t think very many stations would have been installed initially in violation of those standards. So the siting issues must come from changes that occurred in the surrounding environment *after* installation. On average those would have occurred at the half way point.
That means that only about half the data exhibits the “bias” you’re referring to.
You’ve also pointed out that stations are frequently resited, presumably because the old site is no longer compliant but the new site is.
So while I can accept that there is probably some positive skew, I can’t accept that it is all positive. Resiting most probably introduces negative “bias”
it also has nothing to do with climate but is talking about the comfort of cities and whether grass is a better surface for open space than concrete)
Yes, which I noted earlier. it makes me more likely to believe it, not less, as it is not tied up in AGW politics.
Also, if it was conducted under typical airport UHI, the differences would be masked and made to appear smaller than they actually were.
But Yilmaz is merely in support of the LeRoy estimates from 1999, which are included in the NOAA/CRN handbook.
(Rev: You could have included the “HOT-L Baltimore” observations.)
You’ve also pointed out that stations are frequently resited, presumably because the old site is no longer compliant but the new site is.
. . .
Resiting most probably introduces negative “bias”
Your presumption, while not unreasonable, as such, turns out, on closer examination, to be not correct (or even close). Don’t blame yourself. Blame (grossly) inadequate SHAP on the part of the NOAA.
The MMTS switchover was the main reason for resiting. And that is the main reason there are so many CRN4 violations. Cable issues drew them right next to their housing. Result: massive numbers of violations. (Not adjusted for.)
For increase in trend (as opposed to offset), see:
LaDochy, Medina, Patzert. 2007. Recent California climate variability: spatial and temporal patterns in temperature trends. Climate Research, 33
http://www.int-res.com/abstracts/cr/v33/n2/p159-169/
“Also, if it was conducted under typical airport UHI”
Aren’t we talking about microclimate? An airport runway is a pretty hot microclimate. I agree differences would be masked – by engine exhaust – and this study doesn’t compensate for it. Rather than smaller, the differences would be larger. (In fact, given the heat and volume of air from turbine exhausts the differences shown are more likely to be somewhere around 100% due to engine exhaust IMHO)
I can’t see the energy being radiated by a few square meters of concrete being anything like the same as required to get a plane in the air, it’s just silly.
What you have over the little slab is a small volume of air being heated and mixing with the much larger volume in the surrounding environment. No doubt you’ve been in front of a bar radiator? You’ll know then that you get very little convective heating, it’s all radiation, which the thermometer is shielded from.
I just don’t buy it. But if you want to prove me wrong, by inferring the installation date of the path and the magnitude of the “bias” from the data record – be my guest. How long before Anthony gets the data? A few days or will it be in June?
ps – LaDochy is looking at different regions and locations, not surfaces – at least as far as I can make out from the abstract, so I don’t think it’s that relevant to this particular site.
No need to buy it. There will be more studies. But considering specific examples (e.g., lampassas and Baltimore) I would be willing to hypothesize what those future results would be.
Jet exhaust effect would cover both concrete and nearby grass. This would not exaggerte the effects. If anything it would minimize them. Assuming he chose active areas of the airport (which I doubt).
I am happy to wait until further results are in. (I also guess the effects would be even greater at sea level.)
And, of course, there is LeRoy (1999) which is where NOAA got those estimates.
The gold speck in what LaDochy shows is that a constant sink exaggerates not only the offset but the actual trend. I repeat, the trend.
“Assuming he chose active areas of the airport (which I doubt).”
It’s at the end of the runway! That thermometer is going to get regular blasts of hot air.
The grass one is directly between the two runways (which are not close together) ie. at the side of both away from direct exhaust blasts.
Secondly, your measurement site here is dealing with radiated heat, not convected heat. Thermometers are protected from radiation but open to convection.
And he’s an architect. Since when do architects get training in experimental design?