Previous work, using high-quality temperature and rainfall data sets averaged across Australia, has demonstrated that the relationship between annual mean maximum temperature and rainfall changed in the early 1970s. The relationship was still strong and negative (i.e., high temperatures tended to accompany droughts) but the temperature in the period 1973–1992 tended to be higher, for any value of rainfall, than had been the case in the earlier years. This study again uses high-quality data averaged across Australia and the year, and demonstrates that this “anomalous warming” has intensified. Since 1993, temperatures have tended to be higher, for a given rainfall, than was the case in the 1973–1992 period (which in turn was relatively warmer than previous years). Thus the warming observed over Australia over the last few decades does not reflect changes in rainfall.
Nicholls et al.  used high-quality temperature and rainfall data sets across Australia to document apparent changes in the relationship between rainfall and mean maximum temperature in the second half of the 20th century. They showed that, for any value of annual rainfall averaged across the country, mean annual maximum temperature was higher after the early 1970s than would have been the case in earlier years. Artificial changes, such as changes in instrumentation or urban warming, were considered unlikely to account for this change, because the data had been carefully selected to avoid such problems (e.g., mainly rural stations were used) and adjusted for any changes in instrumentation or exposure [Torok, 1996]. Mean maximum temperatures during 2002 set a new record across Australia [Watkins, 2002]. High daytime temperatures are normally found during droughts, so the question arises whether the record temperatures occurred simply because 2002 was also a severe drought (the 2002 mean rainfall was also very low, at about 340mm)? Or has the “anomalous warming” (i.e., warming not associated with a rainfall decline) noted in Nicholls et al.  continued through the 1990s and thereby contributed to the record mean maximum temperature? Nicholls  and Nicholls and Powers  concluded that this recent warming had exacerbated the 2002 drought. Here high-quality rainfall and temperature data for Australia from 1910–2002 have been used to determine whether the anomalous warming has continued, or intensified, in the past decade. The focus here is on temperatures in all years, not just the 2002 drought.
Figure 1 shows time series of annual rainfall and mean daily maximum temperature, both averaged over Australia, for 1910–2002. A strong negative correlation is evident between rainfall and temperature – when rainfall is low, maximum temperature tends to be high (and vice versa). The correlation between the variables is −0.47 (n = 93). Despite this negative correlation, both rainfall and temperature have increased since the middle of the 20th century. That is, the long term change in the two variables is of the same sign – whereas the short term variations are negatively correlated. If the long-term warming evident in Figure 1 was associated with long-term changes in rainfall we would have expected a decrease in rainfall across Australia to have accompanied this warming, rather than the observed increase in rainfall. This suggests that Australia may have exhibited continued “anomalous warming”, i.e. increases in temperatures for any value of rainfall, on independent data not available to Nicholls et al. .
Figure 2 shows a scatter diagram of annual mean Australian maximum temperature versus rainfall (i.e., the data of Figure 1 shown as a scatter plot). Different markers are used for three different periods: 1910–1972, 1973–1992, and 1993–2002. Linear regression lines between temperature and rainfall are also shown for each period. In each period, there is a strong negative relationship between rainfall and temperatures, i.e., droughts tend to be associated with high temperatures. However, the line of best fit shifts upwards from the earliest to the latest period, i.e., for each possible rainfall, temperatures in the later periods are higher than in the earlier periods. Thus the tendency for warming unrelated to rainfall changes, noted in Nicholls et al. , has continued through the last decade, and intensified.
 The relationships between rainfall and mean maximum temperature averaged across Australia are shown for various periods in Table 1 and Figure 3. The Table lists the correlation between rainfall and mean maximum temperature, for various periods. Also listed are predictions, from the regression for each specific period, of the expected mean maximum temperature for a year with Australian rainfall equal to the 1961–1990 mean, i.e., 472 mm. The predictions of the mean maximum temperatures listed in Table 1 for years with mean annual rainfall of 472 mm illustrate the anomalous warming of recent decades. Thus the prediction of the expected mean maximum temperature in a year with average rainfall during the most recent decade is 0.78°C warmer than the prediction from the earliest (1910–1972) period, and 0.25°C warmer than would have been the case in the 1973–1992 period. The warming started around the early 1970s but appears to have intensified over the past decade. Table 1 also shows that, prior to this warming of the last few decades, there were some decadal-multidecadal variations, presumably reflecting natural variations of the climate system that have been documented elsewhere [eg., Allan et al., 1996].
Table 1. Correlations Between Australian Mean Annual Rainfall and Annual Mean Maximum Temperature, for Various Periods
Predicted temperature anomaly (°C)
Also shown are the predictions, from the linear regression appropriate to the period, of annual mean maximum temperature in a year with rainfall equal to the mean 1961–1990 rainfall of 472 mm.
Figure 3 shows the linear regression lines between rainfall and mean maximum temperature anomaly, for decades starting with 1913–1922 and ending with 1993–2002. These decades were chosen so that the last decade ended with the final year of record (2002). This meant that the first three years of record (1910–1912) were not used in this comparison. The three most recent decades are identified and shown as full lines, whereas the earlier decades (up to 1963–1972) are shown as broken lines. Although the earlier decades exhibit some variability in their linear regression lines, they remain clustered near each other. The three lines derived with data from the most recent decades are clearly separated from the earlier decades. The most recent decade is shifted upwards relative to the two previous decades (1973–1982 and 1983–1992). The Figure demonstrates that the shift in the regression lines in recent decades is substantially different from the inter-decadal variability in the rainfall-temperature regression exhibited in earlier years. There is still some interdecadal variability in recent decades that is not simply a relative warming. Thus the 1983–1992 decade regression line falls slightly below that of the previous decade. However, the interdecadal variability of the last 30 years is substantially shifted from that of earlier decades.
 The tendency noted in Nicholls et al.  for Australian mean maximum temperatures to be warmer in recent years, irrespective of the rainfall, has continued and apparently intensified. Thus there has been a continued anomalous warming across Australia that is not associated with a decline in rainfall. Given the strong negative relationship between temperature and rainfall, the warming should have been associated with a decline in rainfall, if the changes simply reflected random variations in the occurrence of droughts or wet years. In fact, the sense of the rainfall change has been opposite to that expected from the strong negative correlation, i.e., rainfall has increased rather than decreased as temperatures have warmed over the past few decades. Across the entire range of observed rainfalls, temperatures have been warmer in the past decade than was the case in the previous two decades which, in turn, were also relatively warmer (for any specific rainfall) than the previous two decades.
 The Intergovernmental Panel on Climate Change [IPCC, 2001] concluded that “…most of the observed [global] warming over the last 50 years is likely to have been due to the increase in greenhouse gas concentrations”. The possibility that this “anomalous warming” observed over Australia is also likely due to the enhanced greenhouse effect warrants consideration. If the warming trend had been associated with a decline in rainfall, as might have been expected from the strong negative relationship between temperature and rainfall, the trend could well have been attributed to natural climate variability such as a random increase in the frequency or intensity of droughts. Such a trend could, of course, still be caused by factors other than natural climate variability. But the likelihood of natural climate variability producing rainfall and temperature trends so different from those expected from the close temperature-rainfall relationship would appear to be lower than if the trends had reflected this relationship. This in turn suggests that the likelihood that the “anomalous” warming is due to anthropogenic influences is higher than would be the case if the trends had been more like those expected from the negative rainfall-temperature relationship.
 An alternative explanation of the anomalous warming is that the apparent changes in the rainfall-temperature relationship reflect changing biases in observations. Nicholls et al.  noted that the likelihood that the observed changes in rainfall-temperature relationship was the result of changes in instrumentation or other instrumental biases was low, since the data had been carefully selected and adjusted for artificial changes caused by changes in instrumentation or siting. As well, the most recent change in the rainfall-temperature relationship documented here since the early 1990s appears unlikely to be the result of instrumental changes - changes in siting or exposure are now well-documented, and seem unlikely to have introduced new biases. The network is largely non-urban, so urban warming is unlikely to have contributed substantially, if at all, to the effect documented here.
 The results presented here suggest an approach to the regional detection and attribution of climate change. As noted above, the demonstration that the recent warming is not simply reflecting the well-established rainfall-temperature relationship coupled with an appropriate trend in rainfall (i.e., a trend that reflects the sense of the rainfall-temperature relationship) reduces the likelihood that the warming trend is due to natural climate variability. If climate models forced with scenarios of enhanced atmospheric greenhouse gas content produced warming across Australia similarly unaccompanied by rainfall trends of the sense expected from the rainfall-temperature relationship, this would increase the likelihood that the warming is the regional expression of an enhanced greenhouse effect [Power et al., 1998; Walsh et al., 2003].