If you have ever planned a wedding or a huge birthday party, chances are you've been tempted to look for a long-range forecast to see if you need to consider a wet-weather option.
While many sites and apps will happily give you forecasts extending months into the future, it's rare they will mention how accurate they are.
Spoiler: they're not accurate at all.
In fact, new research suggests that it may be impossible to accurately forecast beyond 15 days.
And with today's technology, we're not even close to that.
What's possible now?
Most people live in bands between about 23 and 66 degrees on either side of the equator called the mid-latitudes, and forecasts for those regions can be reasonably reliable out to as much as 10 days.
In reality, accuracy drops off pretty quickly.
The Bureau of Meteorology provides seven-day forecasts, and even they would admit it gets a bit error-prone towards the end of the week-long outlook.
The forecast for tomorrow and the day after is pretty much spot-on; up to day five it's not too bad, and by the time you reach day seven things get a little shaky.
That's because of how forecast models work.
You start by putting in as much detail as you can about the atmosphere as it is right now, then the model applies some very complicated mathematics to determine what's going to happen in the next time step.
A time step can be any length. The most accurate use one hour time steps.
The model then uses that information as the starting point for the next time step, and so on and so forth to the end of the forecast period.
One of these cycles is called a "model run", and they can take hours to complete — even for a supercomputer.
There are many forecast models, each using slightly different equations, and each model is run several times a day.
The initial conditions are incredibly important and incredibly hard to get right, especially in a place like Australia where observation stations are few and far between.
We often have to rely on inferring things like wind speed and direction from satellites rather than measuring them directly.
Any small errors at the start begin to grow, slowly at first, and more rapidly with each time step as they get carried forward during each model run.
This eventually means that the model stops producing reliable results.
Getting the forecast right
Since we began using numerical weather prediction models about 60 years ago, their accuracy and reach have steadily improved.
This is largely thanks to improvements in supercomputers. They can crunch much more data using more complex algorithms much faster than ever before.
Timing is key; the model needs to run quickly enough that the initial conditions are still roughly true, otherwise it might be putting out information that doesn't consider what's actually happening on the ground.
Even though computers will continue to get faster, researchers have found that there's an upper limit to how far in the future we will be able to make useful predictions.
Blame butterflies. Well, the butterfly effect.
That's where even small rounding on those initial conditions — say, measuring the temperature to only 1 decimal place — will always fail to produce the same results as you would get without any rounding.
And those small corrections add up to a lot of error.
So, a butterfly flapping its wings in China (a tiny effect that the models would miss) could change the course of a storm in Australia weeks later.
The future of forecasting
Researchers from Penn State University now report that they have found the limit of our forecasts.
They tested some well-established models with initial conditions that had only tiny differences. Called "initial twin" conditions, they look almost the same until you really get up close — just like real twins.
They found that even reducing the initial errors by an order of magnitude, the best we can achieve is a forecast about 15 days out.
And that's if the weather is fairly settled. Small scale but explosive events like thunderstorms and larger scale extreme weather predictions are harder to nail down.
Getting accurate predictions out to 15 days requires a few things that we don't have yet.
Faster computers will mean we can run better models, but the real key is getting the error as small as possible on the initial conditions.
Improvements in satellite coverage and the deployment of more automatic weather stations to remote parts of Australia are key to getting those observations to a high degree of accuracy.
Getting forecasts right is important. It helps farmers make effective plans, brings economic benefit across a wide range of industries, and ultimately can help to save lives.
And wedding plans.ABC