By their nature, disasters are extremely tricky to respond to. The call to ‘do things better’ across the board in how we face natural hazards is understandable and genuine.

We already know what a good warning looks like. Research that tested the wording and structure of warning messages under actual hazard conditions have provided Australian emergency agencies with a better understanding of how messages are understood and translated into direct action. These include using clear, direct language, structuring information in easily understood formats, and links official warnings to other credible information sources. Warning messages today differ greatly from what was produced even a few years ago.

Much of the current discussion is around warnings of an urgent nature; when the river has broken its banks or the thunderstorm is above the town. We are asking a lot from warnings delivered at the most urgent point of time, when the hazard is at its most chaotic, and when both emergency workers and those threatened, are under extreme stress. Research is looking at how we can better warn people under such circumstances – how the warning can be more personal and precise to an individual’s actual situation, how the warning can better communicate the uncertainties around how a hazard will occur, and how to warn people whose circumstances make them difficult to reach.

We know that even when warnings are received in time, research shows they do not always motivate a safe and timely response.

Our research after the 2022 floods in New South Wales and Queensland showed that over 60% of residents surveyed did not evacuate, despite receiving a warning. This decision was motivated by other factors. For some, the decision to stay was part of their flood plan; many had stayed in previous floods and had been safe. Others stayed to lift up their belongings, to protect against looting or to be around to quickly start the clean-up. Some stayed to look after less mobile dependents, care for pets and livestock, or because they had nowhere else to go. Even when clear, accurate and locally specific warnings are communicated in a timely manner, people’s pre-existing social context greatly shapes their ability to act on information.

The term ‘unpredictable’ is frequently confused with ‘variable’. The concept of weather predictability must also be considered in the light of the ‘natural variability’ in a given location.

The dynamics of the atmosphere dictate that the atmosphere becomes more variable as one moves away from the Equator and towards the poles. For example, the maximum temperature on 80% of days in Darwin in January lies in the 5oC range of 29-34oC (with an average of 32oC). By contrast, Melbourne is exposed to the wild storms which regularly cross the Southern from west to east. A consequence of this is that on 80% of days the range of Melbourne maximum temperatures is 16oC (range of 20-36oC, average of 27oC). Hence if one uses the average temperature in Darwin to ‘predict’ tomorrow’s max most of the time there would be an error of less than the 2oC. Following a similar method for Melbourne most of the time there would be an error of less than a much greater 8oC.

It follows that the concepts of ‘accuracy’, ‘skill’ and ‘predictability’ have different meanings. The skill of a forecast must be seen in terms of its accuracy relative to the background variability.

A significant proportion of atmospheric variability is tied up with the presence of moisture. With global warming the atmosphere is able to hold more water vapour, which in turn can mean cyclones and atmospheric fronts are more energetic. This leads to increased variability and more extremes. Hence the ‘prediction’ task becomes more difficult.

It should be pointed out that the BOM (along with all other weather centres in the world) base their forecasts on the fundamental physics governing atmospheric and climatic phenomena. These fundamental laws are encoded into computer models of the atmosphere. With increasing computer speed and memory these models, representing our best understanding of the atmosphere, are advancing. So even though the atmosphere is becoming more variable there is, overall, a progressive and demonstrable improvement in forecast accuracy.

It is important to remember that recent events tend to assume a greater importance in our minds than those of the past. As such the recollection of the impact of past events tend to diminish with time.

These flash-flood-producing thunderstorms are difficult to forecast at the best of times. It’s always a forecasting challenge when severe weather warnings are involved. There are unusual features of this El Nino that can be attributed to global warming.

From November, atmospheric moisture has significantly increased right through the upper levels to about 10,000 metres and beyond, where weather processes occur, in addition to near the earth’s surface.

With heat at low levels to destabilize the atmosphere, we have seen a constant barrage of lines of heavy showers or thunderstorms.

The southern annular mode (SAM) is one of the main factors responsible for the depth of atmospheric moisture, but it is more complicated than that because when it all started in November the SAM was neutral. It was very strong south of Australia but counteracted by a negative SAM near South American longitudes.

Currently, the SAM is strong and moisture in the upper levels is apparent right across southern Australia. This was partly the reason for damaging thunderstorms yesterday in Perth. These atmospheric circulation changes that produce such deep atmospheric moisture so far south in summer are a sign of global warming.

Emergency services both in regional areas and cities are impacted by these difficult meteorological forecasting conditions and, of course, cropland farmers are particularly aware of the damage that these conditions pose around harvest time.

Well, it appears that more ‘severe El Niños’ (very warm water in the central and eastern Pacific) tend to ‘overshoot’ Australia.

The current pattern in the upper atmospheres over the Pacific and Indian Oceans is remarkably similar to that in 1997/98 – the so called ‘El Niño of the Century’ (also an intensely strong El Niño).

This pattern tends to produce major subsidence - drought impacts - over the eastern Indian Ocean (and some parts of WA) and so the lack of rainfall tends to affect the Indian Ocean rather than eastern Australia.

We recognised this was happening in 1997, but it seems to have been forgotten this time.

Some seasonal climate forecast systems picked this up (eg ECMWF) but also the Queensland Government system that is strongly statistical.

BoM’s seasonal climate forecast system is a hybrid between its own models and that of the UK Met Office. I'm not sure it did so well.

This is a rare event since we don’t have many of these extremely severe El Niños (two in the past 70 years).

However, our analysis (which includes climatologists and geographers from Europe) shows this current (extremely intense) Niño and that of 1997/98 (which produced remarkably similar rainfall patterns in Australia to the current pattern) are the only events  to display this current upper-level dynamics, globally

Forecasts beyond a few days, including those for El Niño, are by their nature probabilistic. So we predict the odds of something happening. But what does it mean to have a 60% chance of rain (or an El Niño occurring) and how do we evaluate such forecasts?

Well, we look at many (say 100) forecasts that said 60% chance of rain over the past years. If in 60 of the cases it rained, and in 40 it did not, we know our forecasts work. In short, evaluating the quality of a probabilistic forecast on a single event cannot be done - we can not declare success or failure that way.

We may not like it, but it is the nature of forecasting the weather and climate beyond a few days!