Observant readers will have noticed a renewed assault upon the meteorological station data that underpin some conclusions about recent warming trends. Curiously enough, it comes just as the IPCC AR4 report declared that the recent warming trends are “unequivocal”, and when even Richard Lindzen has accepted that globe has in fact warmed over the last century.
The new focus of attention is the placement of the temperature sensors and other potential ‘micro-site’ effects that might influence the readings. There is a possibility that these effects may change over time, putting in artifacts or jumps in the record. This is slightly different from the more often discussed ‘Urban Heat Island’ effect which is a function of the wider area (and so could be present even in a perfectly set up urban station). UHI effects will generally lead to long term trends in an affected station (relative to a rural counterpart), whereas micro-site changes could lead to jumps in the record (of any sign) – some of which can be very difficult to detect in the data after the fact.
There is nothing wrong with increasing the meta-data for observing stations (unless it leads to harassment of volunteers). However, in the new found enthusiasm for digital photography, many of the participants in this effort seem to have leaped to some very dubious conclusions that appear to be rooted in fundamental misunderstandings of the state of the science. Let’s examine some of those apparent assumptions:
Mistaken Assumption No. 1: Mainstream science doesn’t believe there are urban heat islands….
This is simply false. UHI effects have been documented in city environments worldwide and show that as cities become increasingly urbanised, increasing energy use, reductions in surface water (and evaporation) and increased concrete etc. tend to lead to warmer conditions than in nearby more rural areas. This is uncontroversial. However, the actual claim of IPCC is that the effects of urban heat islands effects are likely small in the gridded temperature products (such as produced by GISS and Climate Research Unit (CRU)) because of efforts to correct for those biases. For instance, GISTEMP uses satellite-derived night light observations to classify stations as rural and urban and corrects the urban stations so that they match the trends from the rural stations before gridding the data. Other techniques (such as correcting for population growth) have also been used.
How much UHI contamination remains in the global mean temperatures has been tested in papers such as Parker (2005, 2006) which found there was no effective difference in global trends if one segregates the data between windy and calm days. This makes sense because UHI effects are stronger on calm days (where there is less mixing with the wider environment), and so if an increasing UHI effect was changing the trend, one would expect stronger trends on calm days and that is not seen. Another convincing argument is that the regional trends seen simply do not resemble patterns of urbanisation, with the largest trends in the sparsely populated higher latitudes.
Mistaken Assumption No. 2: … and thinks that all station data are perfect.
This too is wrong. Since scientists started thinking about climate trends, concerns have been raised about the continuity of records – whether they are met. stations, satellites or ocean probes. The danger of mistakenly interpreting jumps due to measurement discontinuities as climate trends is well known. Some of the discontinuities (which can be of either sign) in weather records can be detected using jump point analyses (for instance in the new version of the NOAA product), others can be adjusted using known information (such as biases introduced because changes in the time of observations or moving a station). However, there are undoubtedly undetected jumps remaining in the records but without the meta-data or an overlap with a nearby unaffected station to compare to, these changes are unlikely to be fixable. To assess how much of a difference they make though, NCDC has set up a reference network which is much more closely monitored than the volunteer network, to see whether the large scale changes from this network and from the other stations match. Any mismatch will indicate the likely magnitude of differences due to undetected changes.
It’s worth noting that these kinds of comparisons work because of large distance over which the monthly temperature anomalies correlate. That is to say, that if a station in Tennessee has a particular warm or cool month, it is likely that temperatures in New Jersey say, also had a similar anomaly. You can see this clearly in the monthly anomaly plots or by looking at how well individual stations correlate. It is also worth reading “The Elusive Absolute Surface Temperature” to understand why we care about the anomalies rather than the absolute values.
Mistaken Assumption No. 3: CRU and GISS have something to do with the collection of data by the National Weather Services (NWSs)
Two of the global mean surface temperature products are produced outside of any National Weather Service. These are the products from CRU in the UK and NASA GISS in New York. Both CRU and GISS produce gridded products, using different methodologies, starting from raw data from NWSs around the world. CRU has direct links with many of them, while GISS gets the data from NOAA (who also produce their own gridded product). There are about three people involved in doing the GISTEMP analysis and they spend a couple of days a month on it. The idea that they are in any position to personally monitor the health of the observing network is laughable. That is, quite rightly, the responsibility of the National Weather Services who generally treat this duty very seriously. The purpose of the CRU and GISS efforts is to produce large scale data as best they can from the imperfect source material.
Mistaken Assumption No. 4: Global mean trends are simple averages of all weather stations
As discussed above, each of the groups making gridded products goes to a lot of trouble to eliminate problems (such as UHI) or jumps in the records, so the global means you see are not simple means of all data (this NCDC page explains some of the issues in their analysis). The methodology of the GISS effort is described in a number of papers – particularly Hansen et al 1999 and 2001.
Mistaken Assumption No. 5: Finding problems with individual station data somehow affects climate model projections.
The idea apparently persists that climate models are somehow built on the surface temperature records, and that any adjustment to those records will change the model projections for the future. This probably stems from a misunderstanding of the notion of a physical model as opposed to statistical model. A statistical model of temperature might for instance calculate a match between known forcings and the station data and then attempt to make a forecast based on the change in projected forcings. In such a case, the projection would be affected by any adjustment to the training data. However, the climate models used in the IPCC forecasts are not statistical, but are physical in nature. They are self-consistent descriptions of the whole system whose inputs are only the boundary conditions and the changes in external forces (such as the solar constant, the orbit, or greenhouse gases). They do not assimilate the surface data, nor are they initiallised from it. Instead, the model results for, say, the mean climate, or the change in recent decades or the seasonal cycle or response to El Niño events, are compared to the equivalent analyses in the gridded observations. Mismatches can help identify problems in the models, and are used to track improvements to the model physics. However, it is generally not possible to ‘tune’ the models to fit very specific bits of the surface data and the evidence for that is the remaining (significant) offsets in average surface temperatures in the observations and the models. There is also no attempt to tweak the models in order to get better matches to regional trends in temperature.
Mistaken Assumption No. 6: If only enough problems can be found, global warming will go away
This is really two mistaken assumptions in one. That there is so little redundancy that throwing out a few dodgy met. stations will seriously affect the mean, and that evidence for global warming is exclusively tied to the land station data. Neither of those things are true. It has been estimated that the mean anomaly in the Northern hemisphere at the monthly scale only has around 60 degrees of freedom – that is, 60 well-place stations would be sufficient to give a reasonable estimate of the large scale month to month changes. Currently, although they are not necessarily ideally placed, there are thousands of stations – many times more than would be theoretically necessary. The second error is obvious from the fact that the recent warming is seen in the oceans, the atmosphere, in Arctic sea ice retreat, in glacier recession, earlier springs, reduced snow cover etc., so even if all met stations were contaminated (which they aren’t), global warming would still be “unequivocal”. Since many of the participants in the latest effort appear to really want this assumption to be true, pointing out that it doesn’t really follow might be a disincentive, but hopefully they won’t let that detail damp their enthusiasm…
What then is the benefit then of this effort? As stated above, more information is always useful, but knowing what to do about potentially problematic sitings is tricky. One would really like to know when a problem first arose for instance – something that isn’t clear from a photograph from today. If the station is moved now, there will be another potential artifact in the record. An argument could certainly be made that continuity of a series is more important for long term monitoring. A more convincing comparison though will be of the existing network with the (since 2001) Climate Reference Network from NCDC. However, that probably isn’t as much fun as driving around the country taking snapshots.