125: Queensland revisited - temperature trends STABLE to 1980

In Post 24 I interrogated the temperature data of Queensland. This Australian state had 28 long stations and a further 85 medium stations in its Berkeley Earth (BE) dataset as listed here. Averaging the anomalies from these 113 stations resulted in a mean temperature anomaly (MTA) that appeared to exhibit a warming of 0.7°C since 1900 as shown in Fig. 125.1 below. But is this the true picture? In this post I will show how the data can be reinterpreted, and thus deliver different results without altering the actual data.

Fig. 125.1: The mean temperature change for Queensland since 1887. The best fit is applied to the monthly mean data from 1901 to 2004 and has a positive gradient of +0.74 ± 0.08 °C per century.

The first problem with the data in Fig. 125.1 is that not all the 113 stations used are of equal length. That means that the MTA before 1905 is dependent on data from less than twenty stations rather than over one hundred as was the case in the 1980s, as the graph in Fig. 125.2 below indicates. This suggests that the data after 1920 will be more reliable than the data before.

Fig. 125.2: The number of station records included each month in the mean temperature anomaly (MTA) trend for Queensland in Fig. 125.1.

Then there is the impact of the fitting range. Applying linear regression to the entire range of data is often inappropriate because the data may have different behaviours or trends at different times. This appears to be the case for the data in Fig. 125.1 as data after 1975 is clearly behaving differently to data before that date.

So suppose we look just at data after 1920 and only fit to data before 1980. Then the picture changes from that presented in Fig. 125.1. The best fit trend line to the data now rises less steeply by only 0.2°C or so before 1980 (see Fig. 125.3 below), and while there is a larger jump after 1975 of about 0.3°C again, this appears to be temporary as the temperature returns to trend after 2010 (although that may just be a temporary reversal). So changing the interval of the linear regression fit can also change the result, or at least change our perceptions, interpretations and conclusions.

Fig. 125.3: The mean temperature change for Queensland since 1920. The best fit is applied to the monthly mean data from 1921 to 1980 and has a positive gradient of +0.29 ± 0.19 °C per century.

What is more, this temperature rise from 1921 to 1980 seen in Fig. 125.3 is more consistent with that seen for the longest temperature record for the state, Brisbane Regional Office (ID 152224), as shown in Fig. 125.4 below. Unusually, this record shows only modest warming despite coming from the middle of the largest urban area in the state. As I will show in future posts, the urban heat island (UHI) effect, where large urban areas lead to a greater warming of the local environment than is seen in more rural areas, or for the region as a whole, can be a serious issue. It usually results in greater warming for stations in large, dense, urban environments compared to the regional average, not less. But not here.

Fig. 125.4: The mean temperature change for Brisbane since 1887. The best fit is applied to the monthly mean data from 1901 to 2004 and has a positive gradient of +0.22 ± 0.08 °C per century.

All this indicates the difficulties in interpreting temperature data correctly. Not all times in history have equal quality of data, and even if they did, the natural variability in that data means that you need long time intervals to see the true trend. And even then your conclusions will be affected by your choices in how the data is analysed.

So which is the better interpretation of the data, Fig. 125.1 or Fig 125.3? My opinion is Fig. 125.3 because it focuses on the better data. The data analysis also fits to data that is less variable, and therefore more reliable. It is too early to know if the temperature rise after 1975 is part of a trend or whether it is just temporary, so the better approach is to treat it almost as a separate dataset and compare it with what went before. 

The temperature dips in Fig. 125.1 before 1910 are also of questionable veracity. Are they the latter part of an upward trend or are they just just natural variability? Without extra data before 1880 we don't know, and even if that upward trend exists, then why does it exist? Because it can't be caused by rises in CO₂ because those rises were negligible before 1910. In fact CO₂ levels in 1910 are estimated to be less than 300 ppm which is a rise of only 6% since 1800. That is nowhere near enough to produce temperature rises of 1°C or more. In fact it would barely result in rises of 0.1°C (see Fig. 87.3 in Post 87).

But of course not everyone sees things this way. One problem with climate science is the amount of data adjustments that are used to correct for perceived data flaws in the temperature data. But as I have shown repeatedly throughout this blog, those adjustments appear hard to justify from any statistical perspective. The raw data is far more reliable than is often assumed, and this can be evidenced by the repeated behaviours seen in temperature trends based on raw data from neighbouring regions that consistently correlate. Many (but not all) of these adjustments also appear to add warming more often than they reduce it, and so appear to exaggerate the amount of climate change that is occurring.

But perhaps one of the most concerning aspects of temperature adjustments is that they are not permanent. The same data often continues to be readjusted over time, and more often each adjustment makes the claims for the warming trend even greater. As exhibit #1 I give you the Australian Bureau of Meteorology (BoM). According to the BoM the climate of Queensland has warmed by about 1.65°C since 1910 as shown in Fig. 125.5 below. Yet the raw data in Fig. 125.1 suggests that the warming is less than half this value and Fig. 125.3 suggests it may be less than 0.3°C. The problem is that the data shown in Fig. 125.5, which is the official BoM version for July 2022, is rather different from the version published in 2010.

Fig. 125.5: The mean temperature change for Queensland since 1910 according to BoM in 2022. The best fit line has a positive gradient of 1.5 °C per century.

In 2010 the temperature trend for Queensland according to the BoM was as shown in Fig. 125.6 below (h/t Ken's Kingdom). Yes it has twelve years less data, but that is not the only difference. Many of the temperature anomalies before 2010 have rather different values compared to now, and so too does the linear trend which was only 1.0°C per century; this despite there being no change in the 30-year reference period of 1961-1990. Now some of the change in linear trend may be due to the extra data after 2010, but not all. It is quite clear that most of the annual anomalies before 1980 in Fig. 125.6 are larger or less negative than was the case for anomalies for the same year in Fig. 125.5 above, while anomalies for most years after 1980 in Fig. 125.6 have smaller values when compared to the corresponding anomaly in Fig. 125.5.

Fig. 125.6: The mean temperature change for Queensland since 1910 according to BoM in 2010. The best fit line has a positive gradient of 1.0 °C per century.

It may be difficult for some readers to spot the difference because we are talking of changes of less than 0.2°C in the height of the bars, but if we overlay the data from Fig. 125.6 on top of that from Fig. 125.5 the differences become more apparent. This is done in Fig. 125.7 below with the 2010 data from Fig. 125.6 coloured green (for positive vales) or sea-green (for negative values) and being slightly translucent so that the red and blue coloured bars from 2022 can be seen underneath.

Fig. 125.7: A comparison of BoM annual temperature anomalies for Queensland from 2022 (red and blue) and 2020 (green).

What Fig. 125.7 shows quite clearly is that the temperature anomalies before 1980 were up to 0.2°C greater back in 2010, while those after 1980 were up to 0.2°C smaller in value. In other words, the extra adjustments made to the data since 2010 have added up to 0.4°C of warming. And yet neither set of data is comparable to the unadjusted data in Fig. 125.3 where the warming is estimated at less than 0.3°C.

Summary

Re-analysis of the unadjusted Queensland temperature data from Post 24 shows that the state may have warmed by as little as 0.3°C since 1920 (see Fig. 125.3).

The most extreme analysis of the unadjusted data indicates that the warming since 1900 is less than 0.8°C (see Fig. 125.1).

According to the Australian Bureau of Meteorology (BoM) in 2010, there had been 1.0°C of warming from 1910 to 2010 (see Fig. 125.6).

In 2022 the BoM now claims that warming since 1910 has increased to 1.65°C (see 125.5).

Adjustments made to the 1910-2010 data by the BoM since 2010 appear to have added up to 0.4°C of warming (see Fig. 125.7). So up to 60% of the 0.65°C temperature rise claimed by the BoM since 2010 could be due to data readjustments for data before 2010.

124: Arctic temperature trends - a comparison

In the previous five posts I examined the temperature changes for five different territories in the Arctic region (Greenland, Iceland, the Faroe Islands, Jan Mayen and Svalbard). All appeared to exhibit similar trends with a peak in temperatures in the 1930s followed by a dip, and then another rise after 1980. In this post I will compare and examine these five trends in more detail.

In Post 11 I demonstrated how the correlation between temperature trends from different stations depends on their separation: the further apart they are, the less well correlated they are. In fact if the distance between them exceeds 1500 km their correlation becomes very weak. On that basis we would not expect any great correlation between the the Faroes and Svalbard as they are over 2,000 km apart. In contrast, Greenland, the Faroes and Jan Mayen are all less than 600 km from Iceland, so the correlations of their temperature data with that from Iceland should be must stronger. The trends in Fig. 124.1 below attempts to do just that, compare the trends of Greenland, the Faroes and Jan Mayen with that of Iceland. For clarity the trends of Greenland and the Faroes are offset vertically by -3°C and +3°C respectively, as are their Icelandic comparator curves.

Fig. 124.1: The 5-year moving average temperature trends for Greenland, Jan Mayen and the Faroe Islands all compared against the equivalent trend for Iceland.

The data in Fig. 124.1 can be summarized as follows.

The trends of Greenland, Jan Mayen and the Faroe Islands all appear to follow the same broad pattern. Temperatures peak in the 1930s, then decline by about 2°C by the 1980s before peaking again after 2000. Only in Jan Mayen is the peak after 2000 higher (by about 0.5°C) than the one in the 1930s.

The trend from the Faroe Islands is most closely correlated with that of Iceland. Not only is the broad trend the same, but the smaller peaks and troughs also align well.

The smaller peaks for Greenland are not closely correlated with those of Iceland or the other two regions. This may be because the mean temperature anomaly (MTA) for Greenland is the result of averaging anomalies from stations over a much larger area than is the case for Iceland, Jan Mayen and the Faroe Islands. So some of the fine detail may be lost by the averaging of stations that are themselves not well correlated.

Jan Mayen shows better correlation with Iceland but its peaks and troughs are larger in size. This may be the result of it having a more extreme climate (due to being inside the Arctic Circle) where the temperature anomalies are naturally larger.

Fig. 124.2: The 5-year moving average temperature trends for Greenland, Iceland and Svalbard all compared against the equivalent trend for Jan Mayen.

If we repeat the process used for Fig. 124.1 but instead use the temperature trend of Jan Mayen as the reference comparator, then we get the trends shown in Fig. 124.2 above.

What we see from Fig. 124.2 is similar to what we saw in Fig. 124.1 with temperatures peaking in the 1930s, then declining by about 2°C by the 1980s before peaking again after 2000.

Once again the smaller peaks for the trend of Greenland are not closely correlated with those of the comparator (in this case Jan Mayen), and the reason is probably the same.

This time, though, the greatest correlation of the smaller peaks and troughs in each trend line is between those for Jan Mayen and Svalbard. This is perhaps not a surprise given that they are near(-ish) neighbours and both are well inside the Arctic Circle.

Summary and conclusions

The general long-term temperature trends of Greenland, Iceland, Jan Mayen and the Faroe Islands are well correlated over timescales of more than 20 years. This suggests that there is no need to adjust the temperature data because the data is correct.

Correlations on shorter timescales (5-10 years) are generally weaker. The two notable exceptions are firstly Jan Mayen and Svalbard, and secondly Iceland and the Faroe Islands.

98. What happened to Louisiana temperatures in 1957?

Fig. 98.1: Global average land temperatures since 1850 according to Berkeley Earth.

In my previous post looking at the temperature trend for Louisiana (Post 97) I showed that the mean temperature in the region had declined by almost 0.2°C in the last century or so. This is in sharp contrast to the claim from most climate scientists that average temperatures have increased by almost 1.2°C in that time, and that this increase is even greater on land. In fact Berkeley Earth claims the increase in land temperatures since 1850 to be in excess of 2°C (see Fig. 98.1 above). But while analysing the Louisiana data one feature stood out that makes me query both the results of my last post and the analysis processes of Berkeley Earth (BE). 

In 1957 the temperature appears to drop suddenly and permanently by about 0.615°C (see black arrow on Fig. 98.2 below). What makes this feature significant is that similar temperature falls at identical times can be seen in the most of the individual temperature records for Louisiana. But they can also be seen in the temperature trends of neighbouring states like Texas. 

So is this temperature drop due to a sudden and large, natural change in the local climate? Or is it due to a change in the data measurement and analysis? If it is the latter then it needs to be corrected for and that will change drastically the true temperature trend. If it is the former then it raises serious questions about how the climate changes over time. In this post I will look at this feature in more detail and try to answer those questions.

 

Fig. 98.2: The mean temperature change for Louisiana relative to the 1951-1980 monthly averages. The best fit (white line) is applied to the monthly mean data from 1911 to 2010 and has a negative gradient of -0.38 ± 0.15 °C per century. The arrow and red line indicate the position and size of the data discontinuity.

The data in Fig. 98.2 above is the part of the same data that was presented previously in Fig. 97.1 of Post 97. In this case I am concentrating only on data after 1910 which, as I pointed out in Post 97, is the most reliable as it all results from an averaging of over forty distinct temperature records (see Fig. 97.2). The white line in Fig. 98.2 is the best fit to the data from 1911 to 2010 and has a strong negative gradient of -0.38°C per century. This is somewhat more negative than the trend in Fig. 97.1 because the fitting range is different. This shows how the value of the best fit gradient can be strongly influenced by the data range, particularly when the data exhibits large fluctuations.

The point of interest in the data above is in 1957 (as indicated by the large black arrow) where the mean temperature appears to drop suddenly and permanently by about 0.615°C. This can be seen clearly in the yellow line which is the 5-year moving average of the monthly anomaly data. It is also illustrated by the red line which is effectively two separate lines: the average temperature for 1921-1960 and the average for 1961-1990. In both cases the discontinuity is clear. The magnitude of the vertical discontinuity can be estimated from the discontinuity in the red line and is 0.615°C. 

Fig. 98.3: The mean temperature change for Louisiana after breakpoint adjustment. The best fit is applied to the monthly mean data from 1911 to 2010 and has a positive gradient of +0.54 ± 0.15 °C per century.

The next step is to remove the discontinuity by shifting upwards all the data after the start of 1958 in Fig. 98.2 by the size of the discontinuity, 0.615°C. The result is shown in Fig. 98.3 above. Two things are striking about the result. First, the gradient of the best fit is now strongly positive (+0.54°C per century) suggesting that the climate is warming. And secondly, the data just looks better with a more consistent trend. Of course just because data looks nicer does not prove that it is more reliable or more accurate.

 

Fig. 98.4: The total contribution of Berkeley Earth (BE) adjustments to the Louisiana temperature data. The orange curve shows the contribution just from breakpoint adjustments. The blue curve represents the total BE adjustments including those from homogenization. The linear best fit (red line) to the total BE adjustments for the period 1911-2010 has a positive gradient of +0.731 ± 0.004 °C per century.

The process I have employed here is virtually identical in concept to the breakpoint adjustments used by Berkeley Earth (BE). The main difference is that I have only applied one adjustment to the final mean temperature data whereas Berkeley Earth apply multiple adjustments of differing magnitudes and times to almost every station dataset. The sum total of those BE adjustments for the Louisiana data is shown in Fig. 98.4 above and the result is a huge warming trend of +0.73°C per century. This is warming that is added to the original data as I showed in Post 97. Yet the 0.6°C discontinuity in the middle of 1957 still remains in the adjusted BE data even after their adjustments have been made as the arrow in Fig. 98.5 below indicates. So the BE adjustments have not corrected the most glaring issue with the original data, which does rather raise a lot of questions regarding the accuracy and validity of the BE adjustments that are made.

Fig. 98.5: Temperature trends for Louisiana based on Berkeley Earth adjusted data from the 90 longest station data records. The best fit linear trend line (in red) is for the period 1911-2010 and has a gradient of +0.37 ± 0.05°C/century.

This is not the first time I have encountered these sudden jumps in temperature data. A similar upward jump in temperature of over 1°C can be seen in the temperature trend for Europe in 1988 (see Fig. 44.1 in Post 44). So what is the cause? At the moment I can only think of two explanations: a natural phenomenon that suddenly changes the local climate, or a sudden change in measurement equipment or methodology that is applied across all stations in a region simultaneously. But so far I can find no evidence for either. Of course the natural phenomenon may not have occurred in 1957 or at any other recent time before that. The complex dynamics of the Earth's climate could mean we are seeing the ripples now of forcing events many centuries ago. In Post 9 and Post 17 I have investigated chaotic effects in the temperature record and found evidence of fractal behaviour that can persist for centuries.

Fig. 98.6: The mean temperature change for Texas relative to the 1961-1990 monthly averages. The best fit (white line) is applied to the monthly mean data from 1911 to 2010 and has a negative gradient of -0.15 ± 0.15 °C per century. The arrow and red line indicate the position and size of the data discontinuity.

What is clear is that this temperature discontinuity is not restricted to Louisiana. The same data anomaly can be seen in the temperature trend for Texas that I analysed in Post 52. This is shown in Fig. 98.6 above with the breakpoint adjusted temperatures shown in Fig. 98.7 below.

 

 Fig. 98.7: The mean temperature change for Texas after breakpoint adjustment. The best fit is applied to the monthly mean data from 1911 to 2010 and has a positive gradient of +0.56 ± 0.15 °C per century.

After the breakpoint adjustment the temperature trend for Texas is now positive and virtually identical to that of Louisiana in Fig. 98.3. There also appears to be a strong correlation between the 5-year moving average (yellow curves) of each. This suggests that the region could have warmed by about 0.5°C over the last one hundred years. However, as I pointed out in Post 52, direct anthropogenic surface heating (DASH) or waste heat equating to about 0.7 W/m² is probably currently warming Texas by up to 0.3 °C compared to 1850. That only leaves about 0.2°C for carbon dioxide induced climate change. This in line with the temperature rise I estimated in Post 87 and a long way short of the 2°C claimed by Berkeley Earth and others. So even with this adjustment there is little evidence to support severe carbon dioxide induced climate change in Louisiana or Texas.