powered by  bravenet.com

Essential Reading 

Discuss climate change online

posted 07-04-2002 09:17 AM

To answer whether the world is warming because of an enhanced greenhouse effect, we should first know whether it is warming at all, and whether warming is accelerating as the gases take effect. Given the timing of rises in greenhouse gases, and taking note of the variety of records available for different periods, it is convenient to look at trends over the past 20 and the past 40 years. The latter takes us back to about 1960. About two thirds of the increase in greenhouse gases has occurred since that time. (NASA GISS link) Moreover the warming effect of the extra gases is usually considered to be subject to substantial delay. Depending on the estimate used for that delay, temperatures since 1960 should reflect somewhat more or less than 80 per cent of the anthropogenic greenhouse effect to date.

Here are the different sources of information about global temperature and what they show since c.1980 and c.1960:

• aggregates from near-surface thermometers: roughly 0.15° C warming per decade over both periods
• satellites: 0.06° C warming per decade since c.1980 in lower troposphere; no earlier data
• balloon thermistors: agree with satellites since 1980; similar rise 1960-80 [result varies depending on whether 1958 and 1959 are also included; these first three measures are compared by the IPCC
• balloon pressure-derived temperatures: agree with thermistors and satellites [Pielke, Sr., R.A., J. Eastman, T.N. Chase, J. Knaff and T.G.F. Kittel, 1998: 1973-1996 Trends in depth-averaged tropospheric temperature. J. Geophys. Res., 103(D22), 28909-28912; and same authors: Errata to 1973-1996 Trends in depth-averaged tropospheric temperature. J. Geophys. Res., 103(D14), 16927-16933]
• compilations of proxy records: show troughs in decade c.1970 and little overall change from 1960 to the end of records about 1990, broadly similar to the balloon record (Tom, your chart)
• “carbon-dioxide thermometer”: not a test of long-term trends, but of year-to-year changes. Correlates better with the satellite/balloon record than with the surface [Calder, N., 1999: The Carbon Dioxide Thermometer and the Cause of Global Warming, Energy & Environment, Vol. 10, No. 1, pp. 1-18, updated by John Daly. Whether Calder’s unorthodox cosmic-ray theory is correct does not, of course, affect the short-term CO2-temperature correlation.]

All measures show temperatures have risen since 1960. But the average rise is small and no measure shows that the rise is accelerating.

All the measures have problems. The satellites are new technology, constantly being tweaked. The balloon sites are widely scattered. The proxy records are rough, of varying quality, and subject to influences other than temperature. The carbon-dioxide thermometer only measures short-term relative temperature, and is also subject to other influences.

Nevertheless, these five measures correlate well, and suggest trend warming of about two-tenths of a degree between 1960 and 1980, and about another one-tenth from 1980 to now.

The sixth measure, the near-surface data, show warming of about three tenths of a degree in each 20-year period.

So, is the warming three-tenths of a degree since 1960, or six tenths? To answer this question, attention must focus on the near-surface thermometer data, since although there are recognised faults in the other measures, they are independent peer-reviewed sources of empirical data about temperature, and have remarkably high correlation.

Debate in your forum has concentrated on whether the near-surface data show an artificial warming trend from the urban heat island (UHI) effect. The scientific literature on the UHI is extensive. Nevertheless it is of limited usefulness in this context, for although the effect has been closely studied and quantified, there has been no systematic attempt to apply the results to temperature aggregates.

For example, the Jones data set used by the IPCC mixes urban and rural records, and only makes one-off “step” adjustments where the compilers detect a clear discontinuity in the record from some local effect. These local effects are many, and include urban heat, site and instrument changes and other factors. But the Jones methodology does not capture the effects of “creeping” warming from the gradual increase in city population and infrastructure. [Wood, F.B., 1988: Comment: On the need for Validation of the Jones et al. Temperature Trends with respect to Urban Warming. Climatic Change 12, 297-312].

The GISS compilation uses a different approach, which at first sight looks more promising. Data from “urban” sites, i.e. those with a population of over 50,000 in the GHCN database, are adjusted to match trends from neighbouring non-urban sites. But there are two problems here:

• First, as one of your correspondents has pointed out, the effect of urban heat on a temperature trend depends not on the population at a certain moment in time, but on the rate of population growth over the period in question. The UHI effect is greater for the earlier increments of population, and gradually levels off according to a logarithmic function. [As mentioned in your forum, one recent article on this is by Torok, S.J., Morris, C.J.G., Skinner, C. and Plummer, N. 2001: Urban heat island features of southeast Australian towns. Australian Meteorological Magazine 50: 1-13. The classic book on the subject is Landsberg, H. E.: The Urban Climate, 1981.]

• Second, the GHCN population data for countries other than the USA are generally at least 20 years out of date [Hansen, J.E., R. Ruedy, Mki. Sato, M. Imhoff, W. Lawrence, D. Easterling, T. Peterson, and T. Karl 2001: A closer look at United States and global surface temperature change. J. Geophys. Res. 106, 23947-23963: .]http://www.giss.nasa.gov/gpol/papers/2001/2001_HansenRuedyS.pdf]. The largest error I have found is for Hambantota, Sri Lanka, to which GHCN/GISS assigns a population of 11,000 but where the actual population (shown by world-gazetteer.com) is 532,000. This is an extreme but not an isolated case and in general, the GHCN population figures are too small by a factor of between 2 to 20. The fact that the GHCN figures are actual data from the 1970s shows the vertiginous growth of urban population over the past two decades, and hints at the magnitude of the UHI signal in the data for that period.

Several attempts have been made to quantify this UHI effect. They generally follow the procedure and come to the conclusion that you outline, Tom:

When these calculations are done, the increase in temperatures over the past century is lower, but it is only lower by 0.05ºC (as reported on Page 39 of the report from the US National Academy of Sciences). There are two reasons why the effect is so small. Firstly, only 27% of weather stations are located in urban centres, according the Global Climate Observing System. The rest are either in small towns (19%) or rural (54%). Secondly, although large urban heat islands can sometimes be several degrees centigrade warmer than their surroundings, most of them have been warmer than their surroundings since records began. In addition, in some islands the population density has actually decreased, and the energy efficiency has often increased (i.e. less heat is vented to the atmosphere). The overall effect is that the rate at which the temperature of heat islands has increased is not much greater than the background rate.

The method to which you allude aims to compare temperature change in rural areas and small towns with that in urban areas. The three categories are distinguished according to whether their populations below 10,000, between 10,000 and 50,000, or above 50,000 in the GHCN database. This was the method used by Easterling et al. in 1997 and by Petersen et al. in 1999 [Easterling, D.R. et al., 1997: Maximum and minimum temperature trends for the globe, Science, 277, 364-367: Peterson, T.C., K.P. Gallo, J. Lawrimore, T.W. Owen, A. Huang, and D.A. McKittrick, 1999: Global rural temperature trends. Geophysical research letters , 26 (3), 329-332. Both studies concluded that the UHI effect, as measured by the difference in trends between urban stations on the one hand, and small town and rural stations on the other, was negligible.

Nevertheless, the approach of these papers is not satisfactory, because of the two problems noted above. No account is taken of the rate of population growth in different places, and the population data used are in any case typically understated by up to, or even more than, an order of magnitude [further examples of population data problems] As the Torok et al. paper points out, a further problem is that UHI effects are detectable at populations of as little as 1,000, and decrease logarithmically as population increases. This means that substantial UHI warming could have occurred at places with current true populations of less than 50,000, or even less than 10,000.

The conclusion is that the elements exist in the peer reviewed literature to enable a calculation of the effects of urban heat on aggregates of near-surface temperature:

• temperature data exist over time for numerous locations;
• population data exist over time for the same locations;
• formulas exist for estimating the UHI effect of a given population.

But the work necessary to estimate the UHI impact on the station records used to compile near-surface aggregates has not yet been done. In the meantime, one suspects that those aggregates do contain a spurious warming trend from urban heat that may be greater than the estimates made to date. Of course, there may also be offsetting spurious cooling effects. However, given that five other methods of measuring atmospheric temperature indicate less warming than the surface aggregates, the balance of probabilities suggests that the warming trend in the near-surface aggregates is overstated.

PM

Tom Rees
posted 07-07-2002 02:58 PM

Here’s my response to PM. There are two parts to the post. First, an analysis of temperature trends (sonde, paleoclimate, MSU, CO2 thermometer) shows that the GISS/CRU record is out of step with other records for the period 1980-2000 (although not before then). Second, an explanation for this deviation is proposed: that the current methods for estimating UHI systematically underestimate the effect in the period 1980-2000. I’ll call this ‘hypothesis A’. The alternative explanation (B) that I’ll propose is that the surface records are good, but that, in the short term, tropospheric and surface temperatures respond differently to the various forcing factors (most importantly, stratospheric cooling caused by ozone depletion). I’ll first talk about some specifics of the temperature records, then take a look at how they all correlate.

To illustrate my points, I downloaded a whole bunch of data, and spent some time calculating correlation coefficients and drawing graphs (this is what took the time to respond). The datasets I’m using are as follows: the satellite-derived MSU 2lt, Goddard Institute for Space Studies surface temperature (GISS) <excluding sea surface temperatures>, Climate Research Unit surface temperatures (CRU CRUTEM1v) <including sea surface temperatures, radiosonde data of Angell and Sterin, the global paleoclimatic reconstructions of Jones 1998 and Man 1998, and CO2 data from Mauna Loa.

NB: All correlation coefficients are presented as the more useful r-squared, rather than r. Also, Jones provides only hemispheric data: I took the average of N & S hemisphere. The dataset of Mann et al only runs to 1980, that of Jones et al to 1991.

Radiosonde data for the lower troposphere show good correlation with MSU2lt, and this is rightly regarded as providing validation of MSU2lt (Figure 1a). In turn, the correspondance between MSU2lt and radiosonde-derived temperatures supports the valifity of these ballon-derived data.

Over a 20-year period, both satellite and radiosonde data show a low or zero trend.

Figure 1a. Comparison of lower tropospheric temperature records from ballon (Angell, Serin) and satellite (MSU 2lt) over the past 20 years.

Overall, the underlying trend for the 40-year radiosonde LT data  is more similar to the surface data than is the 20-year radiosonde data. This supports the idea that the separation between MSU2lt and surface data is a short-term fluctuation (supports hypothesis B).

Figure 1a. Comparison of lower tropospheric temperature records from ballon (Angell, Serin) and satellite (MSU 2lt) over the past 40 years.

Figure 2. Surface temperature comparisons. Warming of the 70s and 80s continues into 80s and 90s. Records from sonde stations agree with global temp network of GISS/CRU.

Since these involve proxies that are not located in urban areas, they should not be influenced by UHI. The figure I provide in my FAQ shows only 30-year Gaussian weighted means, and it splits out southern and northern hemisphere – so it isn’t very useful for our needs (where we’re looking at decadal global trends). I’ve therefore drawn a new figure, Figure 3. On the whole, the trends seem to be recreated reasonably well, given the limitations of the technique (for more, see below). This indicates that, over the period of the paleoclimatic record, there is little evidence that surface temperatures are driven by UHI.

Figure 3. Comparison of available surface temperature reconstructions: GISS & CRU (thermometer based), Jones & Mann (proxy reconstructions), and Angell (radiosonde). Heavy lines indicate 10-year moving average.

The theory behind the CO2 thermometer is that yearly fluctuations in CO2 (but not the underlying trend) are driven by fluctuations in temperature. However, a high correlation between two datasets requires two factors: first, that the yearly fluctuations are similar, second, that the underlying trend is similar. The MSU2lt and dCO2 correlate better than surface and dCO2, but this is simply because the underlying trend of MSU 2lt and dCO2 are similar: approximately zero. However, the fact that CO2 is increasing by about the same amount each year is not caused by the (alleged) near-constant surface temperature. Rather, it’s driven by fossil fuel emissions and carbon cycle feedback. The higher correlation between MSU 2lt and dCO2 is spurious. These correlations, therefore, simply tell us that the underlying trend of MSU 2lt and CRU are different: something we already knew!

Interestingly, there is an even better correlation on the table provided (although the authors choose not to highlight with in bold type and asterisk…): the correlation between surface and CO2 (r-squared = 0.64 compared with 0.49 for MSU2lt vs dCO2). The reason for this high correlation is simple: the surface temperatures correlate with both the yearly fluctuations in CO2 (which are caused in part by fluctuating surface temperatures) and with the underlying trend (since the underlying trend in surface temperatures is driven by the underlying trend in CO2. This is a crude version of more sophisticated analyses (Thomson 1995, Kaufmann & Stern, 2002), which find similarly.

In fact, the correlation coefficients vs Mauna Loa are as follows: CRU= 0.68, Angell Surface=0.60, whereas Angell LT = 0.20, MSU=0.07. This provides good evidence that the surface temperature is being driven largely by CO2 (i.e. not by UHI), whereas the tropospheric temperatures are also being driven by something else (i.e. ozone depletion).

Correlation coefficients

These are shown in Table 1. As discussed above, the highest correlations will occur when both the underlying trend and the yearly fluctuations are similar. The relative importance of these two factors depends on the length of the series (the underlying trend is more important in long time series).

Table 1. Correlations between surface and lower troposphere (LT) datasets. GISS & CRU correlations are provided for 20th century (1900) and for 1958+. Data of Mann et al finish at 1980, those of Jones et al at 1991. Sterin & Angell data are from 1958+, those of MSU2lt 1979+.

Some observations based on the various correlation coefficients:

1. The trend between surface measurements and the data of mann et al is good over the course of the 20th century. This suggests that the underlying trends in CRU and GISS are not an artefact of UHI (i.e. supports B)
2. The correlation between surface measurements and mann et al since 1958 is lower (although still reasonable). This suggests that the paleoclimate and surface measurements agree on the overall trend, but disagree on yearly fluctuations.
3. The trend between Jones and other datasets (surface, LT, and Mann!) is not so good.
4. The correlation between paleoclimatic data and LT data since 1958 is similar to the correlation between the paleoclimatic data and the surface data. Therefore, the paleoclimatic data since 1958 don’t help us to choose between A and B.
5. The correlation between CRU/GISS and Angell surface is better than with Angell LT. MSU2lt correlates better with Angell LT than with Angell surface. This suggests that LT temperatures really are responding differently to surface temperatures (supports B).
6. Of the correlations between surface and LT, that with MSU2lt is weakest. This is likely because differences between surface and LT are short term and tend to ‘even out’ in the long term (supports B).

Urban Heat Islands

Looking at the data, I can’t see any evidence for hypothesis A. However, PM, you make a series of reasonable points about urban heat islands and the potential for an influence on the record. As you say, until someone actually factors these in, any speculation about the impact they are likely to have is just that: speculation. However, a little speculation never hurt anyone, and my speculation is that part of the reason there is no evidence for the effects you mention in the surface temperature record is that they are non-systematic (Yes, they are likely to introduce error, but it is error without bias. On average, non-systematic, random error will tend to cancel out).

In addition, there is a problem with the two points you make: they appear to me to be contradictory. First, the Turok paper makes the case that the UHI effect is relatively independent of population size - as you say, “substantial UHI warming could have occurred at places with current true populations of less than 50,000, or even less than 10,000”. This would tend to reduce the impact of the other point you make, namely “the vertiginous growth of urban population over the past two decades”.

Finally, I can see no reason to assume that the nature of these changes is substantially different now than it was in the period 1958-1980, or in the period 1900-1980. Therefore, I speculate that, although populations have increased and moved around over the past century, the impact on the change in UHI is likely to be small.

Conclusion
So, I suspect that the UHI effect is not as you propose, although neither of us will know until the effect has been properly quantified. However, given all the above, given that there is a correlation between GHG and the surface record that is difficult to explain if the surface trend is driven by UHI, and given that there are known physical mechanisms the have been properly quantified that can explain a substantial part of the discrepancy between surface and MSU2lt, I doubt that the surface record can be explained away on the basis of UHI effects.

Jeff Norman
posted 07-08-2002 08:17 AM

Tom,

Some thoughts...

quote:

---

The alternative explanation (B) that I’ll propose is that the surface records are good, but that, in the short term, tropospheric and surface temperatures respond differently to the various forcing factors (most importantly, stratospheric cooling caused by ozone depletion).

---

Possibly but then doesn't this suggest that the GCMs are inadequate if they cannot model these inconsistencies?

quote:

---

However, Angell also provides near-surface data for the same network. These data show good correlation with GISS and CRU (fig 2, more on this later). This strongly supports hypothesis B.

---

Not necessarily. How many radio sondes are released from city centres (UHI)? There are vast areas of the world where there is a good correlation between the local surface, the radio sonde and MSU records. The U.S. is a good example. The areas where there isn't a good correlation are those where the surface records show significant warming suggesting flaws in those local surface records as opposed to a general atmospheric phenomena.

General Question

quote:

---

In fact, the correlation coefficients vs Mauna Loa are as follows: CRU= 0.68, Angell Surface=0.60, whereas Angell LT = 0.20, MSU=0.07.

---

The 0.20 and 0.07 are rubbish, but what can we say about 0.68 and 0.60, other than the 0.68 is slightly better than the 0.60?

Tom Rees
posted 07-08-2002 02:21 PM

Hi Jeff,

quote:

---

Possibly but then doesn't this suggest that the GCMs are inadequate if they cannot model these inconsistencies?

---

Whether or not the GCMs can odel it depends on exactly what you're talking about. They do a reasonable job of modelling changes over the past 40 years (e.g. Tett 1996, as depicted in the IPCC [6 years out of date but still pretty good]).

Quoting Gaffen 2000, who looked specifically at tropical temperatures (which is where the greatest discrepancy between MSU & surface occurs):

The tropical lower and mid troposphere experienced greater warming (0.11 to 0.26 K decade1 at 700 hPa and 0.12 to 0.26 K decade1 at 500 hPa), a pattern consistent with model projections of the vertical structure of tropospheric warming associated with increasing concentrations of well-mixed atmospheric greenhouse gases (13, 22). However, although the surface warmed (0.05 to 0.28 K decade1) during 1979-97, lower tropospheric temperatures experienced a small, and at many locations not statistically significant, decrease (0.22 to +0.08 K decade1 at 700 hPa and 0.26 to +0.08 K decade1 at 500 hPa)

So what we are talking about is not so much whether the models can simulate the 40-year trend, but whether they can simulate the shorter fluctuations. Santer 2000 goes into this in some detail. They find that models predict that surface and tropospheric temperatures can diverge quite a lot over a 20 year peroid - the maximum divergence they found from a suite of models & 300-year runs was 0.08 degC/decade - compare the current discrepancy of ~0.07 deg/decade. So a lot of the variation might be just a matter of chance.

Talking of chance, it's worth bearing in mind at all times that the trends are not statistically significantly different.

Then Santer et al go on to explore whether climate models can explain the 20-year difference (see fig 5 & sequential text). They show that ozone & volcanic effects, when added to the models, add a differential of 0.052° ± 0.031°C and 0.067° ± 0.030°C, respectively, for the period 1979-1997.

So the models can do provide a reasonable stab at simulating the effects - although, as I point out in my FAQ, this area is remains one of the ones in which the models fall short.

quote:

---

How many radio sondes are released from city centres (UHI)?

---

AFAIK, radiosondes aren't launched from urban areas for simple reasons of practicality, although the website doesn't give the location of the sonde sites for Angell, and I don't have access to the original publication. The angell dataset is a small set of 63 sites selected on the basis of the data quality, so I expect they would have excluded anything with any obvious potential for bias. Gaffer 2000 (above) do a similar thing. Eyeballing their fig 1, there doesn't seem to be any obvious urban association of their locations (although I know that this doesn't prove much!).

quote:

---

The 0.20 and 0.07 are rubbish, but what can we say about 0.68 and 0.60, other than the 0.68 is slightly better than the 0.60?

---

I think that the difference for the correlations with CRU (0.68) and Angell(surface) (0.60) is insignificant - essentially, they have the same relationship with Mauna Loa. However, the difference between angel (LT) (0.2) and MSU (0.07) is more interesting. The angell (LT) trend goes back to '58, wherease MSU only goes back to '79. So my interpretation of what this is saying is that, the longer the tropospheric trend, the greater the correlation with CO2 (because of the recent disturbances due to ozone etc).

One thing that I haven't been able to find is a regional analysis of the differences between MSU & surface - except ones that say that the main difference lies in the tropics. To me, this also militates against the cause being UHI - if it were UHI, then I don't see why you would get this trend (certainly, I have seen not a single publication that explores this concept in any systematic way, and I suspect that this is because the experts believe that the explanation lies elsewhere, e.g. the article I cited before by MSU meister Christy on SST temperatures: http://zeus.nascom.nasa.gov/~pbrekke/klima/christy_GRLtxt07.doc )