In Western Sydney, Australia extreme temperature days pose serious health and socio-economic threats to its population.
Dr Milton Speer and colleagues from the School of Mathematical and Physical Sciences, University of Technology Sydney, Australia aim to quantify and explain what is driving the increasing disparity in extreme maximum summer temperatures between coastal and western inland Sydney.
Read the original research: doi.org/10.3390/cli11040076
Image Source: Adobe Stock Images / Tanapat Lek
Hello and welcome to ResearchPod. Thank you for listening and joining us today.
In this episode, we look at the work of Dr Milton Speer, Varsha Bubathi, a 2023 honours student, Professor Lance Leslie, Dr Josh Hartigan, Dr Joanna Wang and Dr Anjali Gupta from the School of Mathematical and Physical Sciences, University of Technology Sydney, Australia. The team’s motivation is to quantify and explain what is driving the increasing disparity in extreme maximum summer temperatures between coastal and western inland Sydney. The impact of heat waves in western Sydney will test the limits of human tolerance as the population explosion continues.
The Sydney Metropolitan area stretches about 60 kilometres westward away from the coast and extends approximately 50 kilometres from north to south. Summer officially starts on the 1st of December and ends on the 28th of February. However, the warm season extends into March, because of the lagging effect of slower-to-warm, adjacent sea-surface temperatures, compared with the land surface and atmosphere above. Sea-surface temperatures strongly influence summer maximum temperatures.
Typically, when extreme summer-maximum temperatures occur, the near-coastal suburbs can be up to 10-150C cooler than western Sydney. Because, since the 1990s, global warming has accelerated, the team decided to split the period from 1962-2021, into two halves – thereby creating the intervals 1962-1991 and 1992-2021. The team used temperature data from two Bureau of Meteorology observing stations, Observatory Hill representing coastal Sydney, and Richmond located 50kilometres inland, representing western Sydney. The team found that the median maximum temperature for December to March at both locations has increased significantly. Also, the variability of the December to March maximum temperature has increased at both locations, and also between the two periods.
Notably, for the two locations there are pronounced differences between the two periods. The number of days above 90% and 95% of values at coastal Sydney for both periods has not changed significantly and the number of days above 95% of values has even decreased by about 24%. Conversely, for western Sydney, the total number of days has increased by 120 for 90% of values and 64 days for 95% of values. Even so, 64 days representing a 35% increase does not represent a statistically significant increase.
A technique called Wavelet Analysis, which transforms the temperature data into both frequency and temporal signals reveals that both Sydney and Richmond summer maximum temperature data are dominated by sea-surface temperatures with a 2-5-year period in the central equatorial Pacific Ocean, an area known as Niño3point4; this signal has become stronger after the 1980s. This finding is confirmed as a dominant attribute related to summer maximum temperatures for Sydney and Richmond which will be explained later.
The mean monthly and summer period from December to March mean maximum temperatures were converted into box plots – which provide a visual summary of the variability in data set values, to understand better the distribution for each month and for the entire summer. The box plots show that the medians and the variances have increased in the second climate period at both locations. Pronounced differences in Sydney for December to March indicate a significant difference in medians across the two periods. The Richmond plots show a smaller increase except for February. For both regions, January is the warmest month. The Sydney plots also highlight larger upper quartile values in the second period compared to the first, suggesting an increase in upper extreme maximum temperatures over the last 30 years. A trend in the data is more negatively skewed in the second period compared to the first period, with more of the data shifting upwards, indicating a change in the median.
Permutation testing, which is a test on random samples – with replacement in our case – based on a number of permutations of a variable of interest, with the purpose of determining a probability of occurrence for that variable. 5,000 samples with replacement were performed to detect statistically significant differences between the two climate periods. The results for each month affirm the boxplots and percentile charts. Over the last 30 years, there has been an increase in median temperatures in Sydney, with statistically significant results for January, February and March. Despite the increasing trends previously noted for Richmond, most of the differences are not statistically significant, except for January. Notably, January has become significantly warmer in both Sydney and Richmond since 1992 and with the 90th percentile differences also being statistically significant, there is convincing evidence that extreme temperatures have increased in January.
There is confirmation that Greater Sydney summers have indeed extended beyond February. When the summer months, from December-March are aggregated to capture the entire summer period, there is stronger evidence for an increase in median temperatures for both Sydney and Richmond. As a measure of central tendency, the median is more appropriate than the mean because it is more robust to the presence of outliers such as upper extreme temperature values.
Now, we look at Machine Learning, a subset of Artificial Intelligence that uses algorithms trained on data to produce mathematical models that perform complex tasks to support the analyzing, reasoning and learning involved with human decision making.
The research team used 8 Machine Learning techniques and a possible 45 known climate-related attributes or predictors singularly and in combinations to find the most dominant predictors related to Sydney and Richmond December to March maximum temperatures. The 8 techniques included 2 linear methods and 6 non-linear methods. The 2 linear methods were forward and backward selection using linear regression. Forward selection begins with one climate driver and sequentially adds others. The process continues until the testing error is minimized. On the other hand, backward selection starts with all climate drivers, or predictors, and continues to eliminate them until the testing errors are minimized.
However, the complex interactions between climate drivers lend themselves more to non-linear than linear methods. Linear methods are useful only when there is a direct relationship between the variable to be predicted such as temperature, and the change in temperature over time. Briefly, the team used 3 non-linear methods comprising Random Forests, Support Vector Regression with a radial based kernel function and a polynomial based kernel function, all with both forward and backward selection, making a total of 6 non-linear methods.
In summary, machine learning attribution revealed that Sydney is mostly influenced by Tasman Sea sea-surface temperature anomalies, due to its coastal location. Tasman Sea sea-surface temperatures, global sea-surface temperature anomalies and global temperature, all indicators of global warming, are prominent as attributes selected. Appearing independently and through interaction terms, we can infer that the impacts of global warming are amplifying the influence of other atmospheric and oceanic drivers.
The influence of ENSO is also represented through the appearance of Niño3point4 in a high percentage of sliding windows for both regions. Moreover, the Niño3point4 area, interacting with global temperature, ranks higher for Richmond than for Sydney. This result confirms the analysis of the Australian Bureau of Meteorology that the strongest impacts of ENSO events are usually felt inland, with more varied effects on coastal eastern Australia. The assessment of model performance revealed that Support Vector Regression and Random Forests performed best.
Maximum temperature medians and ranges have increased significantly for both coastal and inland Sydney between the climate periods 1962-1991 and 1992-2021. In general, the largest shifts are noted for January, implying that January, the month with the warmest summer temperatures, also is warming fastest. In terms of climate drivers, some of the highest recorded summer temperatures correspond with moderate to extreme El Niño phases. ENSO influences are also noted with 2-7-year signals influencing temperature in both areas, supporting the reliance on ENSO. In Sydney, mean maximum temperatures have increased strongly over the last 60 years, for all months except December. Richmond monthly maximum temperatures have not changed significantly, except for January. However, for entire summer periods December-March, maximum temperatures for both locations have increased.
Daily temperature analysis provides stronger evidence for the disparate temperatures experienced in both areas, with a sharp rise in the number of days over the 90th and 95th percentile for Richmond, compared to Sydney. In contrast, the team found that extreme temperature days in the summer months have been decreasing in Sydney’s coastal suburbs over the period 1992-2021 relative to the period 1962-1991. The team do note that for the unexpectedly slight decrease in the number of extremely hot, that is greater than the 90th percentile, and extreme heat – greater than the 95th percentile – days, a possible explanation is due to increased differential heating between the hot interior and the near-ocean water which thereby enhances the sea-breezes at the coast.
This effect is also known from more recent studies. Having noted January as being the hottest month, both Sydney and Richmond exhibit the strongest evidence for January having experienced an increased mean and 95th percentile of maximum temperature. This reflects the fact that higher temperature values are increasing for both areas, and clearly more so for Richmond.
As the research team remark:
“Attribution to the known climate drivers of observed maximum summer temperatures at western and coastal Sydney using machine learning techniques revealed a marked disparity in the percentage of summer days above the 95th percentile during the accelerated climate change period – 1992-2021 – between western Sydney, with 35% more days, and coastal Sydney, with 24% less days, relative to 1962-1991. The climate drivers detected as attributes were similar for both coastal and western Sydney but, as expected, coastal Sydney was more affected than western Sydney by oceanic climate drivers.”
That’s all for this episode, thanks for listening. Links to the original research can be found in the show notes for this episode. And be sure to stay subscribed to ResearchPod for more of the latest science.
See you again soon.