New research conducted using machine learning has highlighted the importance of focusing on specific seasons and regions when using major modes of climate variability to predict rainfall. 

Large-scale modes of climate variability, such as the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD), are often identified in advance by major weather organisations around the world. They play an important role in forming seasonal forecasts, and can give an indication of whether we could receive more or less rain in the months ahead.

In work published by the journal npj Climate and Atmospheric Science, a team of researchers in Australia has used machine learning to show the relationship between the major climate modes and precipitation is more complicated than previously assumed.

Highlights

Chief among this complexity is the fact that the relationship between climate modes and precipitation is nonlinear. This means that one phase of the mode can significantly affect rainfall in one region, but the other phase may not. For example, El Niño is not a good indicator of the intensity of rainfall in Australia, but La Niña can have a more robust impact on spring rainfall. 

The study has also highlighted that the relationship is often limited to one or two seasons, and not the entire year. The characteristics and impact of the relationship greatly depends on the region studied. For instance, La Niña can influence spring rainfall over large areas of east Australia, but it does not have an effect on the eastern seaboard itself, so we should not expect Sydney’s seasonal rainfall to be highly correlated with ENSO.

According to Dr Sanaa Hobeichi of UNSW, a Senior Research Fellow at the ARC Centre of Excellence for 21st Century Weather and the lead author of the paper, the work shows that we should “limit expectations of predictability from modes of variability in all but a few select regions and seasons.”

Conducted by members of the ARC Centre of Excellence for Climate Extremes, the ARC Centre of Excellence for 21st Century Weather, and the UNSW Climate Change Research Centre, the study shows that predictability is enhanced in certain regions for particular seasons. 

The results challenge the common idea that whether we are in El Niño or La Niña dictates future precipitation. Broad statements suggesting it will be wet or dry during certain phases of climate modes hide important details and could risk misinforming decision makers, while suggesting greater certainty than is supported by evidence.

For instance, while the 2023-24 El Niño event generally matched expectations around the United States in December, January and February, it surprised many on the east coast of Australia, which experienced a wet summer despite anticipated dry conditions.

This research highlights that communication about the influence of modes of climate variability to the public and decision makers requires nuance, and the setting of realistic expectations.

Go Deeper

When it comes to large-scale modes of climate variability, the range and scale of their complex interactions should not be underestimated. For example, Atlantic sea surface temperatures influence the relationship between ENSO and rainfall in South America, Europe, and the United States. The Indian Ocean Dipole has a say in how ENSO affects the Asian monsoon. And precipitation in Australia is impacted by the interaction of IOD, ENSO, and the Southern Annular Mode (SAM), as well as the Madden-Julian Oscillation (MJO).

Although knowledge of these modes has advanced considerably, the complexity of their interactions and their collective impact on rain has traditionally been explored using linear frameworks. Research in this field is also often limited to just one region or just one characteristic of a climate variability mode. 

By contrast, in this work, the researchers employed machine learning techniques and artificial intelligence methods to understand these more complex impacts. They used both linear and non-linear methods to capture the relationships between a range of modes of climate variability and unusual precipitation across land globally. 

Critically, these interactions were tested with data that was not used to establish the relationships in the first place, enabling a fair assessment of the contribution by modes of variability to precipitation around the world.

The team identified which modes of variability allow for the greatest precipitation predictability in different regions. They then worked out the amount of variability explained in unusual precipitation across different regions and seasons that the modes can provide. By doing so, they are able to offer a concrete basis for determining the degree of precipitation predictability that we might expect modes of variability to offer.

Spotlight On Results

The median rainfall explained by these modes of variability across all land areas is only 8.7%, with 75% of global land not exceeding 15%. Not surprisingly, most of the land areas with a relatively larger proportion of rainfall variability associated with those modes are situated in the tropics, where the oceans have a stronger influence on precipitation.

It was also found that regional predictability can be considerably higher in particular seasons. For example, Australian precipitation predictability is high during September, October and November, and reaches 60% in some areas, but is significantly lower during other seasons. 

These findings highlight the need to consider the timing of these relationships with the variability modes in specific regions to plan for potential seasonal.
Associate Professor Andrea Taschetto of UNSW, a Chief Investigator at 21st Century Weather and a contributor to the paper, said: “The best way to stay informed is to consider seasonal forecasts that already take into account the non-linearities and interactions between climate models in dynamical weather prediction models.”