On average, around one ex-tropical cyclones - such as 2018's Gita - come within 550km of New Zealand each year. Image / JMA
Picking the track of a fast-rolling, constantly evolving tropical cyclone like the Category 3 Uesi is tricky enough to do for tomorrow, let alone next week.
So how do meteorologists do it?
First, it pays to understand what tropical cyclones are, and why they're so notoriously difficult to keep tabs on.
We can think of them, essentially, as low-pressure systems that form and build over warm waters in the tropics - but with extreme characteristics.
Gale-force winds - or those higher than 63km/h - are found at low levels near their swirling centres but can fan out for hundreds of kilometres.
Meteorologists consider a "severe tropical cyclone" – such as a system that might form to the northeast of Fiji later this week - one that blasts as hard as 118km/h.
Each year, over the November-to-April season, about 10 tropical cyclones form in the Southwest Pacific basin.
Only a few of those ever reach category 4 strength, where mean wind speeds are more than 159 km/h, or higher.
Vanuatu and New Caledonia typically experience the greatest activity, with an average of two or three named cyclones passing close to land each year.
At least one comes within 550km of New Zealand each year, usually around February and March.
To get down here, they have to make their way over much colder waters, while hitting strong upper-level winds as they move out of the tropics.
By the time they arrive, they are almost always re-classified as "ex-tropical" cyclones – as Uesi will be as it barrels past the South Island's West Coast on Sunday.
That doesn't mean they've weakened or been downgraded, but have morphed into a completely different type of beast.
And ex-tropical cyclones can still pack the potential for severe weather: under the right conditions, they can intensify and even muster lower pressures than they had before being reclassified.
Many of our most severe storms - such as 2017's Debbie, which brought the deluge that pushed a Rangitaiki River stopbank to breaking point, flooding Edgecumbe - have been ex-tropical cyclones.
In the tropics, the strongest winds and most intense rain associated with a tropical cyclone usually occur just outside the "eye", or cyclone centre.
But after it has been transformed in what's call an "extra-tropical transition", the systems lose their symmetrical cloud patterns.
The strongest winds and heaviest rain can then be found hundreds of kilometres from the cyclone's centre - usually in a large area south of the centre.
For meteorologists, that means the position of the cyclone centre is no longer a good indicator of where the most severe weather will hit.
In 1988's catastrophic Cyclone Bola, for example, the heaviest rain and strongest winds over New Zealand occurred well away from the centre.
That catastrophe illustrated why the stakes of cyclone forecasting are always high.
Across the world, there are six "regional specialised meteorological centres", or RSMCs, and six "tropical cyclone warning centres", or TCWCs, which are responsible for putting out advisories and bulletins in their regions.
MetService runs the Wellington TCWC, which monitors an area that stretches over the North Island and hundreds of kilometres east.
Tropical cyclone specialists track systems using several "ensemble" models that combine global and high-resolution regional models.
One pulls together more than 50, while another combines more than 100.
Model "runs" are made about twice a day, which forecasters use to produce new bulletins, and over busy periods, forecasters monitor the situation day and night.
The stronger co-relating patterns in the models become, the more confident forecasters are in predicting where tropical cyclones will move around the region.
Like many agencies around the world, MetService draws on the "big three" of the world's numerical weather prediction, or NWP, models.
Those are the Global Forecast System, from the National Centres for Environmental Prediction in the United States; the British Met Office's Unified model; and the renowned European Centre for Medium Range Weather Forecasting (ECMWF).
The latter two models run every 12 hours, while the US one runs every six. The models typically take between six and eight hours to run, and cover about a week into the future.
"These three systems capture the weather across the globe in reasonably high resolution, and give a best guess based on the most likely initial conditions," MetService meteorologist Andrew James explained.
"However, the exact state of the atmosphere at any given time is always slightly uncertain, because we only have so many observations. Initial conditions are of extremely high importance when forecasting tropical cyclones."
James also pointed out that tropical cyclones were complex systems embedded within the already complex system that was our weather.
"Uncertainty in initial conditions mean that it is important to avoid placing too much value on a single model run," he said.
"Because tropical cyclones are among the most damaging weather systems, we want to give as much warning as possible. However, balancing early warnings with communicating the uncertainty poses a challenge."
How did they strike that middle-ground? And how far out was too far out?
That all depended on what level of detail was being sought about a given system.
And in any case, James said, the track of a tropical cyclone was always published under what was called "a cone of uncertainty".
Once a tropical cyclone was named, tracks were produced every six hours out to 72 hours, or three days – yet the period for which MetService's meteorologists produced forecasts for New Zealand stretched to double that time.
That was why ensemble forecasting was crucial, as it acted to minimise changes between model runs.
"The details and timing often change slightly, especially with longer lead times, but the overall picture tends not to," James said.
"This is why meteorologists take model outputs at long lead time as indicative of a pattern, rather than as an exact answer."
Improving the forecasting of tropical cyclones, as with all of meteorology, was a global challenge.
As members of World Meteorological Organisation (WMO), MetService worked with other countries to improve observation networks across the Pacific, which were of great value to forecasters.
James said NWP models had improved significantly in the past 20 years, particularly in the long range, as have satellites.
"These advances are only possible because of the international and collaborative nature of the WMO."
Refining forecasts was also a big focus at Niwa, which recently began trialling its own model ensemble system.
Niwa meteorologist Ben Noll said that could eventually help fill the void that existed between Northern Hemisphere-based models and local ground observations, and which was having to be covered by satellite observations.
The grunt behind the new ensemble lay in Niwa's freshly boosted supercomputing capability. One of its new machines, the Cray XC50, was powerful enough to count all of the grains of sand in the world in only about 5000 seconds.
"Since we purchased the new supercomputer, that has basically enabled us more bandwidth to run the ensemble many times – and that is critical in terms of pin-pointing risks for extreme weather conditions," Noll said.
Going into the future, he expected artificial intelligence and machine learning would play increasingly bigger roles in forecasting.
"A lot of people bag us, but if those founding fathers of forecasting from 100 years ago could see what we're able to do today, they'd be shocked," he said.
"The modelling we have today is something to be proud of, and across the meteorological community globally, we've seen more and more collaboration."
The new free open-air saltwater pool is already proving popular.