The California Urban Heat Island Effect

Last week Anthony Watts had a post at WUWT in which he talked about a new effort to find out just how much the urban environment was affecting the temperatures at Californian weather stations. The study is being carried out in conjunction with the EPA, and the announcement came by e-mail rather than a more conventional press release.

I was interested since, as part of series that I carried out looking at the US Historic Climate Network (USHCN) data, I plotted temperatures for the stations in each state as a function of latitude, longitude, elevation and local population. The first three values were identified with the information at each station. The local population for a town can be found on the web in several different places, and very largely I relied on the city-data web sites for information (see, for e.g. this for Sacremento).

The question arose as to which particular temperature should be used for that of the station, since the USHCN provides annual average temperatures, as raw data, Time of Observation (TOBS) corrected and “adjusted.” When the original post for California was written, only the last of these was available, and thus it formed the basis of the analysis. Shortly thereafter, in 2010, the USHCN site also provided the raw data, and the TOBS temperatures for each station, each year. The data was therefore re-analyzed using the TOBS values. But the plot that was originally generated was plotting the current population against the average temperature since 1895.

As the study grew to include more states, that plot seemed to be an error, since populations can change very rapidly, and go up as well as down. So, towards the end of the series the average temperature was taken only for the past five years, since this was likely to reflect the impact of current populations. At the same time, since there is little difference between the two sets of values in this period, the “adjusted” values were used to derive the plot. It looks like this:


Figure 1. The comparison of average California station temperature plotted relative to adjacent population, with a log-normal plot.

Now the “discovery” of a log-normal relationship is not new. Oke has been studying the topic for decades, and has proposed such a relationship. But it does have a side effect. Consider what happens when the trend line is shown on a normal plot:

Figure 2. The comparison of average California station temperature plotted relative to adjacent population, with a normal scale on both axes.

There is a “kick-over” in the rate of temperature rise at around a population of 10,000. (In fact this is a curve and the sharp transition is an artifact of the software, but it illustrates the trend). Temperature gains for smaller gains in population are higher below that level, while those above that population require a larger population growth to get the same increase. (Failure to recognize this is one of the underlying faults of the Berkeley Earth Project work on the topic.) Since the GISS data on temperatures also does not recognize any difference in population size below 10,000 it is also a fault of that data set.

I am curious to see how the California study pans out, I did drop a note with this finding to William Dean, as the e-mail suggested, and he was courteous enough to reply noting that this was “an interesting approach.”

As I pointed out to him, the strength of that relationship is, perhaps, borne out not only by the R^2 value, but by the consistency of the coefficient over the plots for a number of states. The tabulation is as follows:



I have had to cut the list in two to allow screen capture.


Figure 3. Correlation Coefficients for the relationship of temperature to local conditions with temperatures in degrees C.



And similarly for the table where I have converted the temperatures to def F.


Figure 4. Correlation Coefficients for the relationship of temperature to local conditions with temperatures in degrees F.

California temperatures, the TOBS data

Back in April I first commented on the historic California temperature changes, using the homogenized data then available from the USHCN. Since then they have posted both the raw data and that corrected for Time of Observation (TOBS), as well as the data that has been adjusted to apply other perceived corrections. At the moment I am going back and seeing how the TOBS data is changed by those corrections. So this will examine that data, using the same form as originally. (Note that following the comment by Kinuachdrach the evaluation of Death Valley at the bottom of the piece has been UPDATED).

So first I go to the US Historical Climatology Network and download the 54 station set of data (there will be a short pause while I do this). And then there are the four GISS stations whose data we got the first time around (we aren’t going into those adjustments at this time). Looking at the difference between the few GISS stations and the average for the USHCN set it is interesting to note that while the GISS stations got successively warmer than the state average, in the raw data the reverse was the case, although the decline was very small.



The average difference between the two sets is 1.63 deg, which is the same as for the homogenized data. For the state as a whole, over the past 115 years there has been a steady warming trend, though the raw data suggests an average increase of 2 deg F per century, while the homogenized data suggests only 1.3 deg.


Looking at the effects of geography, there is the correlation with latitude:


There is virtually no change in the relationship of temperature with latitude, though there is sensibly no correlation with longitude, similar to the result with the homogenized data.


Given that California goes all the way to the sea, the effects of elevation should be significant, and in both cases they are.


The regression is a little higher than with the homogenized data (r^2 0.4) but otherwise they are much the same. Which speaks for having a significant number of stations in the state, since there were a significant number of years where there were dropped data from several of the stations. This becomes obvious as the standard deviation shows that scatter getting worse as the collective number of stations reporting peaked. Interesting with the raw data it is leveling off, whereas it was getting worse when the data was homogenized.


And even having to make some assumptions about the number of inhabitants of some of the more remote stations there is still a logarithmic correlation with population.


The correlation is slightly worse however (it was 0.10).

And just for grins, this is what has been happening in Death Valley, and it looks as though that has been getting hotter too.


California was the first state that I have looked at that was on the coast, and so it will be interesting to see if there is a coastal effect and define what it is, and how far inland it stretches. But for that we need more data.

Kinuachdrach made a comment on the Death Valley data (see below) and so I went to see what the condition of the weather station was, and discovered something that I thought readers might find of as much interest as I did. The full story is at the reference by John Daly.

There are, actually, two weather stations now in Death Valley, where the all-time highest temperature in the United States was recorded on July 10th, 1913, at 57 degC. Using the GISS data for California, the plot that is given is this:

And the site has a plaque that says:

"During the summer of 1998 - the warmest year on record - we recorded the hottest air temperature anywhere in the world of 53.06°C ±0.1°C (128°F) on 17 July 1998 at 3:15 pm local standard time."

Since the GISS plot stops before that, I am, being curious, just going to go to the GISS Site and download the latest version of the graph:


Note that since the high temperatures were single days, and the data plotted is the average for the year the records aren't evident. However the station was apparently changed in the 1990's to give the record at Badwater, rather than the earlier data that was recorded at Furnace Creek, according to John Daly. The differences are, among others, that Badwater (as documented) is more of a sun trap than the old station, 20 miles away out in the open, at Furnace Creek.

But it is worth commenting that over at Watts Up With That, Steve Goddard has noted that there seems to be some work on re-adjusting the data back to 1998 that removes the overall high temperature that they showed then (and which agreed with other records) in favor of lowering that temperature so that the overall graph now shows a steady increase in temperature over the last three decades (which does not agree with other records). So when I show you the GISS record (which as with the USHCN data is modified from the original raw data) until I have worked out (or more likely someone else has) what was done with that data before it was plotted, all I can give you is what is on the record. And in this case, I suspect it is only a part of a continuing story.

California temperatures, GISS USHCN, E.M. Smith and Anthony Watts

This is the day that I am going to look at the temperature records for California. It was the truncation of the number of stations in the GISS analysis that led E.M. Smith to his post, that started me off into taking a significant look into the temperature records. His concern was initiated by the discovery that the number of stations being used by GISS to monitor CA temperatures had been cut to four, all located near the coast. So does this have any meaning? The task begins, as I outlined at the beginning, by getting the data from the 50 USHCN stations and also inputting the data from the GISS stations.

There will be a slight pause while I do this. And after loading in the data from the 54 USHCN stations there are a few observations. Firstly the data from Death Valley is missing three data points (1896, 1897 and 1899). So noting that 1896 was 0.29 degrees above the state average, and that Death Valley is on average 16.62 degrees above the state average, suggests that in 1896 the temp there would have been 75.92 deg. And so we enter that and do the same for 1897, and 1899. And in passing I note that there are a couple of stations (Death Valley and Indio) that are below sea level. Wonder how that will work out. There don’t seem to be that many in the heights, but we’ll see how the graphs plot out. This is the correction that I explained in more detail when I was looking at the data from Colorado and found some values missing.

Now I get the four station data from GISS that Chiefio lists, which are San Francisco, Santa Maria, Los Angeles, and San Diego, which I download from the GISS site and at first none of these are on the GISS list. So I go back to the station locator and type in San Francisco and I get four stations and checking with Chiefio’s list by grid reference the top one is the one he cites. (and it is the one that has data from 1880 to 2010). So the next one on his list is Santa Maria, try that through the station locator and there are two locations, but neither has a full set of data!! In this case it is the lower of the two, which gets me information from 1948 on. Los Angeles data is all there, as is San Diego’s, though again one has to choose the longer history site from the four available. Phew!

OK so what have we got? With the varying conditions in the state (and particularly since we are coming down from the mountains to the sea) I will expect that there will be some influence of longitude, but given the concentration of data along the coast and at low elevations, I am not sure how it will end up. And then there are a lot more towns with larger populations that we have seen in the states we have looked at until now. Checking populations Cedarville was too rural for the usual city-data site, so I got the population from neighborhoodlink . Cuyamaca is a State Park with zero inhabitants (I put down 1). Electra also appears to be on none of the lists – (So looking at the one site with info, I put down 10). Lake Spaulding is a fishing camp (no data – suggest 5).

And having put in all the data (using the elevations of the airports for the GISS stations) one finds some interesting results. Firstly how do the GISS stations compare with the USHCN data?

While the GISS stations are on average 1.6 degrees warmer than the USHCN stations, the difference between the two is increasing:


(Note however that the small number of GISS stations relative to the number of USHCN stations means that when one does a total average for the state the differences induced are quite small.) Even without the GISS station contribution, the temperature in the state has been increasing, though the rate seems relatively constant since about 1900.


The state is a relatively long one, and there remains a strong influence of latitude:


I had expected, since the mountains are on the East and the sea is on the West, that there would also be an influence of longitude.


And that, at any significant level is apparently a wrong assumption.

Hmm! Well how about height, that has been fairly consistent.


And so it is again, though note that those below sea level seem to be even hotter than predicted.

Now one thing that we also can check on, given that there are significantly more stations in this state, is as to whether the scatter in the data is getting worse. This is something that would be suggested by Anthony Watts survey of stations. The premise has been that as the maintenance on the stations declines so the scatter in the results would get worse. This would be reflected in an increasing standard deviation across the stations in the state.


And while there was no such trend in other states, it is very clear and significant in California.

And with respect to the influence of population, I am still seeing that logarithmic fit, with the knee of the curve being at around 10,000 folk. Thus if GISS is cutting off all the influence of towns below that size, and just calling them rural, the evidence continues to suggest that this is a significant error.


Well I also suppose that this is one of the states where there are enough larger cities in the data bank that we can plot this on a log scale:


Correcting the individual temperatures as though the stations were at the center of the state (i.e. adjusting for latitude, as I did last time), one gets:


And then if one looks at the effect of elevation with latitude taken out of the data, one gets:


Again a clear correlation, except that there is a considerable scatter as one gets down to the range below about 100 m. Since this brings in the possible effects of the nearby ocean, and I am not sure how to isolate that at this time, we’ll leave that part of the analysis until we have more information.

The average elevation of California, by the way, is 884 m, and if you look at the plot above you can see that there are only 9 stations (out of 58) that are above that height. So if you do just an average of the data (which is what I have been doing) it will be weighted by the stations below the average elevation, and this will bias the results. By how much? Well we’re going to have to find more data before we can answer than question, and I suspect that it has to do with making adjustments for being close to the ocean.