Joanna Donnelly: Focus for forecasters is on how to communicate uncertainty

Limitations in how far we can predict the weather, and how accurately we can describe the future changed climate, has been limited by chaos at the starting point or description of the boundary conditions

Sat, 14 Feb, 2026 - 01:00

Joanna Donnelly

The only certainties in life are death and taxes.

So why do people struggle to accept that there is uncertainty in so many things? Perhaps it’s due to the fact that, because so much information can be obtained in a quick online search, the concept of not being able to find something out for certain seems alien.

This is why when discussing how weather prediction is communicated, for both climate forecast and day-to-day forecasts, the focus of many professional forecasters is now on how best to communicate uncertainty.

Here in Ireland, forecasters have traditionally shied away from using percentages to communicate things like the chance of rain — partly due to the subjective nature of understanding chance.

A person that has something very important that must be done, that would have to be cancelled if it rained, will read a 60% chance of rain as implying rain is less likely, than someone who would be happy to cancel a planned event. This is a gambler’s attitude and a perfectly human approach.

In an economic world grown on mathematical tables that govern risk management, this doesn’t suffice. Industry is calling for certainty in an area of science, weather, and climate forecasting, where it is still only available for a very limited window. At government level, trying to plan for a future changed by climate — we know it is changing — is made extremely difficult when we can’t pin down precisely how those changes are going to manifest or when.

With the introduction of AI, the science of weather prediction is taking a turn. To date, all the weather models around the world have been based on the laws of physics — the mathematics described above. Now we’re looking at pattern matching.

AI is absolutely made for pattern matching. That’s its prime skillset. Searching the input it has, looking for patterns, and predicting the outcome.

Of all the uses that AI can be put to, pattern matching weather maps is about as perfect as it gets

Google’s AI programme, Deepmind, has weather forecast technology called WeatherNext that has been able to fairly accurately forecast the future formation of a tropical cyclone in the Sea of Japan based on pattern matching alone.

Warmer air holds more water, therefore it will rain more — that is certain. Where that rain will fall is uncertain. Warmer oceans will provide more energy for storms — that is certain. Where those storms will land is uncertain.

The Earth is heated unevenly due to a variety of reasons. The tilt of the Earth on its axis means the sun’s radiation lands on the equator directly through the thinnest layer of atmosphere, so it gets most of the heat. The poles are in the shade for half the year for the same reason, so they get significantly less direct radiation through a thicker layer of atmosphere. Before we even look at the make up of the planet, the heat is uneven.

When we look at surface of the Earth, we can add in more inequity. Sand gets hot quickly, water gets hot slowly. You know this if you’ve ever been to a beach in the summer. The converse is also true: The sand gets cold quickly, the water gets cold slowly. The surface of the planet is mostly water, the major deserts of the world are distributed across the continents. Then there’s the major forests of the world, the rainforest of the Amazon, and the snow forests of Siberia and Canada. The snow-capped poles also play a very significant part in the distribution of the heat.

Laws of physics

Between the tilt and the composition, we’re already dizzy with the amount of variables we need to deal with and, if you’re not dizzy enough, remember we have to spin. The Earth is spinning on its axis at a speed of around 1,600km/h at the equator, and around the sun at a speed of about 107,000km/h.

Despite all of that, scientists have still managed to work out the mathematical equations that represent the laws of physics describing the movements of air in our atmosphere.

This has been the foundation of what is known as numerical weather prediction. These are the weather models. Those equations, similarly to the simplest equations of a line, only describe outcomes if they have data.

When weather forecasting as we know it today began in the mid-19th century, observation points around the coasts of Britain and Ireland took measurements of the weather, the air pressure being the most important.

Those measurements were used to feed data into the mathematical equations that then produced a weather forecast. In fact, many of those same stations are still open today. From Malin Head to Mizen Head, the coastal reports as they were then are today.

In the early days, there were so few data points, the forecasts were extremely rudimentary. Over the 150 years since then, we have added thousands of more data points from the surface of the Earth to the top of the atmosphere. Satellites and radar imagery feed in more and more information into the weather models, and weather prediction has improved hugely. That is still not enough.

As soon as a measurement, say at Valentia in Ireland, is taken, it changes

Air pressure, temperature, or wind speed is changing all the time. The smallest variability at the starting point of the model — time zero — causes changes in outcomes as we move forward in time.

To date, forecasters have tried to overcome this variability at the starting point by running the model multiple times and looking at the spread of outcomes. This spread represents the uncertainty.

Imagine standing at the top of the Hill of Tara on a snowy day. You build a snowball, drop it from the top of the hill, and watch it roll to the bottom.

The ball may get larger and land in a particular spot. You do the exact same thing again, build a snowball and drop it down the hill. The ball is neither the same size nor in the same spot as the previous one.

Trying to reproduce the exact conditions to have the snowball fall to the exact place at the same time and in the same shape is a good analogy of trying to start the atmosphere off so that we can describe what it’s going to do exactly. This is called chaos theory, and this is what we’ve been working with until very recently.

Mathematicians, building on the shoulders of giants from thousands of years of study, have taken us so far.

Now, we’ve built machines that can take those equations and leap learning forward in giant steps previously inconceivable.

Changing technology will soon be able to give us the certainty we’ve been missing. In fact, it may take a while for the public to accept that the future is predictable — especially if they don’t like what they hear.