The land area of South America is more than twice that of Australia while its population is more than sixteen times greater. Yet it has fewer high quality temperature records than New South Wales.
Overall South America has about 1000 temperature records, but only 21 have more than 100 years, or 1200 months, of data. Of these long station records, the longest is that of Rio de Janeiro (Berkeley Earth ID: 152852). Its earliest temperature data dates from 1832 and clearly shows that temperatures in the city at that time were higher than today (see Fig. 35.1 below). The average temperature then declined throughout the 19th century before recovering over the last 100 years. This behaviour is not unique to South America. We have seen it in both Australia and New Zealand previously, and if other countries in the Southern Hemisphere had longer records, we might have seen it in many other places as well.
Fig. 35.1: The temperature trend for Rio de Janeiro since 1832. The best fit is applied to the interval 1921-1990 and has a gradient of +2.06 ± 0.15 °C per century. The monthly temperature changes are defined relative to the 1951-2000 monthly averages.
However, the extent to which the Rio data reflects the overall temperature trend in South America is harder to determine. Of the 1000 or so sets of station data from South America, only about 318 have more than 40 years, or 480 months, of data, and as I have shown in previous posts, even this length of data is insufficient to ascertain the recent overall temperature trend, let alone its longer term context.
The approximate locations of the long and medium stations in South America are shown on the map below. It can be seen that they are fairly evenly spread throughout the continent, but with the Amazon region, not surprisingly, being less well represented.
Fig. 35.2: The locations of long stations (large squares) and medium stations (small diamonds) in South America. Those stations with a high warming trend are marked in red.
In order to determine the overall temperature trend for South America we need to average the temperature anomalies from all the different stations in the region, as I explained in Post 5. These temperature anomalies are not the actual average monthly temperatures for each location, but the amount by which those average monthly temperatures change relative to a defined reference value for that month. That monthly reference temperature (or MRT), is usually an average of the values for that month over a given period, say 1961-1990. However, as I have pointed out in many previous posts, the way anomaly data used by climate groups like Berkeley Earth is calculated is not quite that straightforward.
The anomaly data used by climate groups is also subject to adjustments via processes such as homogenization and breakpoint alignment. In homogenization, neighbouring sets of station data are compared and averaged to determine the monthly reference temperature values (MRT) for the target station. Once this homogenized MRT is subtracted from the monthly temperature to yield the anomaly for that month, a number of breakpoint adjustments are then made to different segments of the anomaly data in that dataset, supposedly to correct for bad data and other behaviours seen in the data that appear to be incongruous. The problem with both these interventions is that they are not neutral. Analysis from almost all the previous data I have examined so far in this blog has shown that both these interventions tend to add significant warming to the temperature trends after 1900. The data presented here for South America is no exception.
It is relatively easy to calculate the magnitude of the adjustments made by Berkeley Earth, because conveniently Berkeley Earth list both the adjusted and unadjusted anomaly data in each data file, together with the original raw temperature data. If we average the Berkeley Earth adjusted anomaly data from all the long and medium stations identified in Fig. 35.2 above, the result is the curves shown below in Fig. 35.3.
Fig. 35.3: Temperature trends for all long and medium stations in South America since 1832 derived by aggregating and averaging the Berkeley Earth adjusted data. The best fit linear trend line (in red) is for the period 1900-1999 and has a gradient of +0.85 ± 0.02 °C/century.
The temperature trends shown in Fig. 35.3 above clearly exhibit a strong warming of more than 1.0 °C after 1890. The other notable point is how well the data in Fig. 35.3 agrees with the official Berkeley Earth version shown in Fig. 35.4 below. This can be verified by comparing the pattern of peaks in the 12-month moving average in each case. This shows that using different weightings for each set of station data when calculating the average is not needed.
Fig. 35.4: The temperature trend for South America since 1850 according to Berkeley Earth.
So, what happens if we perform the same averaging of station data, but use the raw temperature data without homogenization and breakpoint adjustments? This means first constructing the anomaly data by calculating the MRT using only the actual station data.
Ideally when calculating the MRT values it is best to choose as long a time interval as possible for the reference average so that those values have less uncertainty associated with them. This means that ideally you want to include as many years of data as possible in the calculation of each of the twelve MRT values. But this poses a number of problems.
(i) Different station records generally have different amounts of data.
(ii) Different station records often encompass different epochs.
(iii) Most station records have periods of missing data or gaps in their record. These may be a single month of data, or they could be several years in length, and there are often multiple gaps.
(iv) Different sets of station data have different temperature trends. Moreover, the longer the temperature records, the more the trend can vary over time within that station record, or the greater the difference in the mean temperatures will be at either end of the record.
The result is that you need to compromise by choosing a time interval for the MRT average that maximizes the total number of stations that can be included in the average, reduces the uncertainty in that average, but also addresses the four issues listed above. The solutions to those four problems are as follows:
(i) Use the same length of time interval for the MRT average for all stations. This is normally 20 or 30 years.
(ii) Choose the MRT time interval so that it overlaps with the epochs of as many station records as possible. In the case of South America this was achieved by using the time interval 1971-2000.
(iii) Set a minimum threshold for the number of years of data required for the MRT for that month to be of sufficient quality. For the data analysed here that was set at 40% (i.e. 12 out of a possible 30 years).
(iv) Set the length of time interval for the MRT average to be shorter than the timescale over which most significant temperature trends are seen. This is normally 20 or 30 years.
Fig. 35.5: The temperature trend for South America since 1832. The best fit is applied to the interval 1900-1999 and has a gradient of +0.54 ± 0.05 °C per century. The monthly temperature changes are defined relative to the 1971-2000 monthly averages.
After the MRT has been calculated for each of the twelve months in each station record, these MRT values are then subtracted from the raw temperature data for that station to yield the monthly anomalies. The anomalies from all the stations are then averaged to provide the regional temperature trend. This trend for South America is shown in Fig. 35.5 above.
The data in Fig. 35.5 clearly shows an upward temperature trend since 1920 of about 0.5 °C. This is similar to trends we have seen in other regions such as Australia and New Zealand, but it is much less than the 1.0 °C or so that is expected based on the HadCRUT4 data. However the picture before 1900 is less clear. There is some evidence of higher temperatures in the late 19th century, but the data is not extensive enough to provide definite proof. There are only about a dozen stations in the whole continent with data that dates from before 1880. Nevertheless, there are definite similarities between the data shown here for South America and other Southern Hemisphere data from Australia and New Zealand. The fact that all three regions exhibit similar trends before 1900 and also after 1900 suggests that the higher temperatures seen in Fig. 35.5 before 1900 are probably real.
What is evident, though, is that the temperature rise in the trend based on the raw data shown in Fig. 35.5 is not as severe as that published by Hadley-CRU, or that published by Berkeley Earth and shown in Fig. 35.4 above. The differences between the two sets of data (from Fig. 35.5 and Fig. 35.3) are highlighted in Fig. 35.6 below.
Fig. 35.6: The contribution of Berkeley Earth (BE) adjustments to the anomaly data after smoothing with a 12-month moving average. The linear best fit to the data is for the period 1901-2010 (red line) and the gradient is +0.33 ± 0.02 °C per century. The orange curve represents the contribution made to the BE adjustment curve by breakpoint adjustments only.
What Fig. 35.6 demonstrates is that the adjustments made to the data by Berkeley Earth are (again) not neutral. In this case they add at least 0.4 °C to the warming since about 1900. But, this is not new. Similar impacts have been seen previously on the temperature trends from many other regions in the Southern Hemisphere (see my other country-based and regional blog posts 8, 18-26, 30-34).
Conclusions
1) There has been about 0.5 °C of warming in South America since 1920. This is much less than is claimed by climate groups and the IPCC.
2) There is some evidence of higher temperatures before 1900, similar to those seen in Australia and New Zealand, but it is heavily dependent on only one or two temperature records, principally Rio de Janeiro as shown in Fig. 35.1.
3) Adjustments made to the data by Berkeley Earth have added significant warming to the temperature trend since 1900.
Addendum
The temperature trend shown in Fig. 35.5 above is the result of averaging almost 300 temperature records. While all these records have over 480 months of data, very few extend back beyond 1900. In fact between 1860 and 1900 the number of temperature records involved in the average for the trend varies from 4 up to 18. This implies an uncertainty in the trend in Fig. 35.5 of between 0.25 °C and 0.5 °C, whereas after 1960 this falls to about 0.05 °C. Thus the uncertainty in the trend in 1860 is about ten times greater than in 1980. This is probably reflected in the differing amounts of natural variation in the 5-year moving average temperature trend for those different epochs.
It should also be noted that the temperature data shown in Fig. 35.5 before 1860 is dominated by the temperature data from Rio de Janeiro shown in Fig. 35.1. Its reliability is therefore questionable.
Fig. 35.7: The number of sets of station data included each month in the temperature trend for South America shown in Fig. 35.5.