By Steven Goddard and Anthony Watts
Fort Collins, Colorado is most famous for Balloon Boy, and Boulder, Colorado is most famous for Jon Benet and Ward Churchill.
Both are hotbeds of Climate Science, with familiar names like Roger Pielke (Jr. and Sr.) Walt Meier, William Gray, Kevin Trenberth and Mark Sereeze. Both are of similar size (Boulder 91,000 and Fort Collins 130,000) and located in very similar geographical environments along the Front Range – about 50 miles apart. The big difference is that Fort Collins has tripled in size over the last 40 years, and Boulder has grown much more slowly. Fort Collins population is shown in blue and Boulder in red below.
Sources:
http://en.wikipedia.org/wiki/Fort_Collins,_Colorado
http://en.wikipedia.org/wiki/Boulder,_Colorado
Until the mid-1960s, NCDC temperatures in the two cities tracked each other quite closely, as you can see below. Again, Fort Collins in blue, and Boulder in red – with Fort Collins temperatures shifted upwards by two degrees to normalize the left side of the graph. Since 1965, temperatures in Fort Collins have risen much more quickly than Boulder, paralleling the relative increase in population.
Source: NCDC Boulder Temperatures NCDC Fort Collins Temperatures
The graph below shows the absolute difference between Fort Collins temperatures and Boulder temperatures since 1930. There is some sort of discontinuity around 1940, but the UHI imprint is clearly visible in the Fort Collins record. The Colorado State Climatologist, Nolan Doesken manages the Fort Collins Weather station. He has told me that it has never moved or changed instrumentation. and that he believes the increase in temperature is due to UHI effects.
Roger Pielke Sr. further commented:
“the Fort Collins site did have the introduction of the CSU Transit Center a few years ago, although this is well after the upturn in temperature differences between Boulder and Fort Collins started to increase.”
From the promotional photo on the CSU website, the Fort Collins USHCN weather station (below) seems reasonably sited.
However when you look at the Google Earth street view, you realize that it is surrounded by concrete, asphalt, nearby parking, and a building just 7.5 meters away (By the GE ruler tool). It would rate a CRN4 by the surfacestations rating. It also appears to have been modified since the promo photo was taken as there is a new fence with shrubbery and wood chips surrounding it.
Besides the pressure of CSU expansion, Fort Collins has seen an increase of about two degrees since 1970, corresponding to a population increase of 90,000. This is probably a little higher than Dr. Spencer’s estimates for UHI.
The Boulder weather station is similarly sited since the concrete path is just under 10 meters away.
It is at the campus of NOAA’s and NIST’s headquarters in Boulder. Anthony Watts visited the station in 2007 and took photos for the surfacestations project. Like Fort Collins, it gets similar expansion pressure due to nearby construction as seen in this aerial photo.
Here are the temperature records fro these two USHCN stations:
NCDC Fort Collins Temperatures
There is some UHI effect visible in the Boulder record below, but much less than Fort Collins.
Conclusion:
We have two weather stations in similarly sited urban environments. Until 1965 they tracked each other very closely. Since then, Fort Collins has seen a relative increase in temperature which tracks the relative increase in population. UHI is clearly not dead.
Steve Goddard (12:48:15) : edit
Steve, I can only invite you to do the math. Assuming trends from eyeballing a graph is not how these things are determined. Here’s the procedure.
Go get the monthly GISS data for Fort Collins and Boulder. Subtract Boulder from Fort Collins.
Then take the trend of the difference (Ft. Collins – Boulder) for the following intervals:
1897-December 1941
February 1942 – present
February 1942 – 1970
1970 – present
I get ~ 0.2°C per decade for all four. Let us know your results and we can settle this once and for all.
Finally, I have no problem with removing a discontinuity which I can prove mathematically to exist. No, I don’t know why it is there … but the math proves that it is there, and allows us to calculate the size of the discontinuity.
However, the procedure I propose immediately above doesn’t depend on removing a discontinuity. It shows that the trend is not just 1970-present, it exists throughout the dataset.
I await your reply.
w.
Willis Eschenbach (14:11:21) :
Like I said in my first post: good but too vague. When you get down to the details of doing a comparison ( correlation, comparing trends) you find out just how deceiving the eye is.
Steve
Willis,
Without a physical explanation for the discontinuity there isn’t any way to make an estimate of the required mathematical fix. For all we know, it could be a homogeneity adjustment between sites. Maybe the Boulder site moved for a few decades? The only safe way to handle it without knowing the details is to throw the data out during that time period. The difference might have been increasing, decreasing, or staying the same. I don’t trust the data to any degree of precision between 1924 and (at least) 1950 from looking at that graph.
These resources might help you find some pristine data to work with
http://climate.colostate.edu/~coagmet/station_index.php
http://climate.colostate.edu/~coagmet/station_details.php?station=FTC01
http://ccc.atmos.colostate.edu/cgi-bin/monthlydata.pl
http://ccc.atmos.colostate.edu/cgi-bin/coagmet_map.pl
http://www.colostate.edu/Orgs/VegNet/coagmet/WSPICS/FTC01IMAGES.html
Sorry, meant to say I don’t trust the data between 1942 and 1970.
However, there is a clean break upwards around 1970.
http://docs.google.com/View?id=ddw82wws_465qwp7dgc7
You said “Assuming trends from eyeballing a graph is not how these things are determined. ”
Quite often that procedure is completely adequate, and is much safer than blindly applying a mathematical correction without any physical basis to justify it. The tone of your remark is quite authoritarian, like I would expect from Gavin or Tamino.
Another close by
http://www.colostate.edu/Orgs/VegNet/coagmet/WSPICS/FTC03IMAGES.html
http://www.colostate.edu/Orgs/VegNet/coagmet/WSPICS/ALT01IMAGES.html
The data is there to do some comparisons with data that doesnt make it into USHCN. rural data less than 1200km away ( hehe)
http://www.colostate.edu/Orgs/VegNet/coagmet/WSPICS/LCN01IMAGES.html
http://www.colostate.edu/Orgs/VegNet/coagmet/WSPICS/FTL01IMAGES.html
U want to look at RURAL trends versus URBAN trends.
there are plenty of of these in the same area. Yes the weather is different for them. But take the rural stations. Look at the correlations month to month. Look at all the trends. you could even average the rural.
That gives you a “climate” trend for the time period. If the rural trend is
ZERO and the trend at boulder and fort collins is positive, then you have
some sense of the UHI. looks like you have 10+ years of data.
I’ve done the same thing in california using the ag system.
One thing is sure: the ag system is held to close examination. Crops depend on it.
have a look
Steve Goddard (14:18:59)
Do the math. Calculate the trends. You complain about me removing a mathematically demonstrable single month step change, yet you “don’t trust” and want to throw out the data from 1924 to 1950 based on your intuition? Are you a scientist, or not?
There is a recognized mathematical procedure to determine discontinuities in difference datasets (e.g Ft. Collins minus Boulder). I described it above. There is a more detailed description here. If you use it you will see that there is only one discontinuity in the dataset, in January 1942.
DO THE MATH. Calculate the trends. Identify the discontinuity. Report the results.
I await your response.
Wllis, to demonstrate the fallacy in your mathematics, I took a simple physical model and replotted below, which made the trend disappear prior to 1965
https://spreadsheets.google.com/oimg?key=0AnKz9p_7fMvBdElxNDA4Vlh2OGhvOUdEX1N0bm1CeWc&oid=5&v=1268434079371
I assumed that there was a movement of the Boulder station to a colder location from 1942 to 1953, and added 0.7 degrees onto the delta for those years. Presto, your trend disappears. You can’t do math tricks like you did without a physical basis.
@ur momisugly Willis
Sorry, at this point I have to side with SG here. Given the procedure you’ve described, even if there was no discontinuity in the time series one of those few hundred combined RSS values is going to be the smallest. Apologies if you’ve already provided this info, but one of the key bits of information is how much smaller that combined RSS is compared to all of the others and whether it’s unusally small compared to what you’d expect if there was no discontinuity in the time series. What would also be interesting is looking at what dates rank 2nd, 3rd, 4th, etc smallest. If the top15 or 20 are all some time around 1942 that would strengthen an argument that perhaps something unusual happended around about then. But if they’re spread across the date range that would weaken the argument.
If we look for something hard enough, we’re going to find it even if it’s not there. It’s kind of similar to the multiple testing problem in statistics. If we analyse 20 data sets that truely have no systematic trend in them, and testing for a trend using a statistical hypothesis test with a type I error rate of 5%, then we’re going to find a ‘significant’ trend in 1 of those 20 data sets on average, even though it’s not really there. This is why we have things like Bonferroni adjustments.
Something to consider: both Boulder and Fort Collins are college towns and the student populations are included in the cities’ population counts. (At least I’m sure that’s true of Boulder, am assuming it’s true of Ft. Collins too.) Boulder has about 26K undergraduates and Fort Collins about 20K. (I didn’t see info on graduate populations.) Most students leave town during the summer months, as well as Christmas break, so the UHI effect may be diminished during then. Boulder has a higher percentage of its population being students (over 25%) than Fort Collins (about 15%) which could impact the differences between them in terms of UHI.
The point is Willis, that you made an assumption of a physical model (a one time shift) which has no basis to back it up. I chose a different arbitrary physical model (no less valid than yours) and the trend from 1930-1965 completely disappears. Without more information we can’t know how to correct the data. That is why I choose to throw it away.
I do however believe that Fort Collins has seen 2+ degrees of UHI warming, because I can and do measure it riding in and out of the city center and campus.
BTW – I didn’t describe my model correctly before. I meant to say that the model was based on the idea that Boulder station moved to a 0.7 degree warmer spot from 1942-1953.
Steve, I ask again that you calculate the following trends:
1897-December 1941
February 1942 – present
February 1942 – 1970
1970 – present
Those require no “assumption of a physical model”. They do not require that we change any data. They are just simple measurements of the existing data. DO THE MATH and report back on what you find.
You keep insisting that we should make judgements based on eyeballing a graph, and assume arbitrary models. Science doesn’t work that way.
For example, I have given a mathematical method that clearly identifies a discontinuity. It is a single-step discontinuity in January 1942. Therefore, my model is definitely NOT a “model … which has no basis to back it up” as you claim. It has a statistically significant mathematical calculation which identifies the discontinuity.
You, on the other hand, pick a totally arbitrary model. If you can show mathematically that there is a discontinuit in 1942 and again in 1953, then fine, your model deserves consideration. Do you have such mathematical evidence???
The reality is, if you increase the values from 1942 to 1953, mathematical analysis no longer finds a discontinuity in 1942 … but now it finds one in 1953. So your arbitrary model is mathematically demonstrably false. DO THE MATH.
So far, all you’ve given us is handwaving. But hey, let’s assume that your model is correct. Calculate the after-your-changes trend from 1953-1970, and 1970 to the present and report your findings … your claim that the trend started in 1970 doesn’t pass examination.
Willis,
You are assuming a particular physical model. You are assuming that there was a one time shift in 1942 were no important changes after 1942. I chose a different model (two shifts) and clearly demonstrated that the trends you claim disappear.
https://spreadsheets.google.com/oimg?key=0AnKz9p_7fMvBdElxNDA4Vlh2OGhvOUdEX1N0bm1CeWc&oid=5&v=1268434079371
Your interpretation, whether you recognize it or not, is based on a model you have chosen to believe – i.e. a one time shift in the data in 1942.
MinB (15:59:30) :,
The disappearance of students in the summer doesn’t change the amount of streets and buildings, and is probably more than compensated by increased driving and number of tourists.
Smokey (14:11:05) :
You seem to be following Dr. Jones quite closely.
Let me know if they ever fire him.
Or reinstate him as Director.
Willis Eschenbach (01:27:43) :
Hey Willis, I like your graphs! What are you using to produce them? Also curious on the package you are using for series analysis, it must have gaussian smoothing builtin, do you mind listing that also?
I am going to do some deep analysis and really don’t want to write it in C, Cpp, Cs from scratch though I can if necessary. Have most of the code from a equitites charting program I wrote years ago.
D MacKenzie (15:59:08) : edit
If you are still asking for information, and you haven’t done the math yourself, why on earth would you be siding with one side or the other?
In any case, here’s the results of my analysis. FIrst, here’s the residual analysis of the original data:
<img
Figure W4. Residual analysis of Ft. Collins minus Boulder temperature data, using the procedure described here.
As you can see, there is a single significant discontinuity in January of 1942. Next, here is the result of Steve’s proposed adjustment, adding 0.7°C from 1942 to 1953:
Figure W5. Residual analysis of Steve Goddard’s proposed adjustment to the Ft. Collins minus Boulder temperature data, using the same procedure.
As I mentioned above, all that this does is displace the discontinuity from 1942 to 1953. Finally, here is the result of my adjustment, adding 0.6°C from 1942 to the end of the record:
Figure W6. Residual analysis of my proposed adjustment to the Ft. Collins minus Boulder temperature data, using the same procedure.
This reduces the total sum of the residuals, and does not leave any obvious discontinuities. (A true discontinuity shows up as a deep “V” shaped drop in the values, as shown in the above graphs)
Like I said, it is necessary to do the math. You can’t just propose a random adjustment as Steve has done. You need to use math to find the discontinuity, and to estimate the size, and to check the results of your adjustment.
Finally, the existence of the discontinuity in January 1942 is confirmed by looking at the trends for all of the data except for that date. Take a look at the trends:
1897-December 1941
February 1942 – present
February 1942 – 1970
1970 – present
I get ~ 0.2 for all of those.
w.
Willis,
Try the same analysis where you add 0.7 on to Fort Collins from 1942-1953, instead of subtracting from Boulder.
Please post your code.
Willis,
You are attempting to fit a straight line, because that is the model you want to see. Looking at the graph, it was obvious that shifting Fort Collins upwards by 0.7 degrees in 1942 would create a straight line.
https://spreadsheets.google.com/oimg?key=0AnKz9p_7fMvBdElxNDA4Vlh2OGhvOUdEX1N0bm1CeWc&oid=2&v=1268444596121
In my model, it is not a straight line and it obviously won’t fit your linear regression.
Willis Eschenbach (17:23:09) :
Nice graphics.
Could you explain how you included them in a comment ? I’ve seen the occasional image or video embedded on WUWT, but I would like to know how the few people like yourself are doing so.
Steve Goddard (17:38:14)
Steve, I am analyzing the difference between Fort Collins and Boulder, which is:
Diff = FC – B
For the change, I used
Diff = FC – (B-0.7°)
You propose using
Diff = (FC + 0.7°) – B
Expand both terms, and tell me what you get …
Steve Goddard (17:48:05)
So we’re back to “just look at the graph” again? Could you determine that the step discontinuity was in January 1942 by just looking at the graph?
Steve, forget about discontinuities for a moment. Using the original Ft. Collins minus Boulder data, I get the following trends:
February 1942 – present: ~ 0.2
February 1942 – 1970: ~ 0.2
1970 – present: ~ 0.2
You have claimed that the temperature increase began in 1970 … I’m sure you see the problem. How do you explain that?
Willis Eschenbach (17:23:09)
“If you are still asking for information, and you haven’t done the math yourself, why on earth would you be siding with one side or the other?”
Well, sitting in a hotel room in Malaysia doing some other work I didn’t have the time to recreate your analysis. Apologies if I unintendly got offside with you, I was just noting that given the evidence that had been presented so far, I wasn’t convinced that you hadn’t just found a random result. Thanks for posting the additional information, that provides a lot of additional important detail.
Willis,
Please post your code.
Willis,
Look at the graph of the entire time series from 1895.
https://spreadsheets.google.com/oimg?key=0AnKz9p_7fMvBdElxNDA4Vlh2OGhvOUdEX1N0bm1CeWc&oid=2&v=1268448520528
The trend from 1895-1941 is clearly not linear. There is a step function up in 1920 and a step function down in 1942. The first few years all over the place.
How did you pick your various start dates of 1897 and 1910? They look cherry picked to me.