Top Debate – Have your say — With the surge in tropical lows lately we’ve been talking more and more about the computer models we use.  Here in New Zealand you don’t often hear forecasters talk about them – but for those who have ever watched US television as a tropical storm approaches it’s all you hear about.

All weather forecasters, including MetService, rely heavily on global computer models.   The two main ones are GFS (American) and ECMWF (European).  There are many others out there too.  Forecasters heavily rely on these models to give us long range predictions – but for more shorter term we use more of our own forecasting skills, tied in with the models/data.

Someone asked me the other day, “Why bother with forecasters at all?  Especially when all of your stories about this tropical trouble at the moment have mentioned computer models”.  A fair question – so let me explain it.

For those who have relied on long range weather forecasts that other websites offer (10 days) you’ll notice something – the forecasts chop and change wildly.  That’s because they 100% using computer models and not human input.

For example, if we had simply reported what the models were predicting seven days ago for the North Island we’d be saying “sunny” one day “torrential rain” the next, then “sunny” again, then “windy but dry” etc.  This is because the models are working through the various possibilities – sometimes they can be way off the mark and so it takes a human forecaster to look at it and work out the most logical forecast.

No one is perfect – but a blend of computer guidance, a forecaster and an understanding of local topography is the best formula for accuracy.  Remove just one of those things and accuracy starts to drop significantly.

The other issue is which model to trust.  The two main models both have strengths – but they both have significant downsides too.  One model may pick 9 out of 10 storms right, then spectacularly get it wrong the 10th time.  We need to be aware of that every time we make a prediction – we never make a prediction based on one model or data update.

So yes, 10 days ago, mostly based on computer models, I first tweeted about the risk of a tropical storm cyclone north of NZ for this week.  WeatherWatch.co.nz tried to keep the message as consistent as possible all last week, despite the models wildly chopping and changing each day – and both disagreeing with each other up until the weekend (making it especially tricky to work it out).

In the end all models were actually off the mark – but you can see why.  A very large low has now formed that stretches from the North Island to Fiji.  None of the models picked the size – but when you combine the three we use, they actually painted an area of low pressure similar to what we have now.  In other words – they all had the right ingredients but the cake only baked when you combined them.  And you know what?  Next time it won’t work that way – which is why every single time we need to look at more than just what the models are saying.

They are simply there to help guide us, but not tell us.

The current low is so big It has now sucked cyclone Daphne and a couple of other lows (or soon to be lows) in to the equation.  Now we’re seeing a low so large it is filled with blue sky and light winds in the middle – and the wind and rain stretches around the outside for a few thousand kilometres.  The central part of the low is roughly 1000kms wide with no rain and limited wind and cloud within it.  Not the usual sort of low we get.

But either way – we’d be lost without the models in this day and age, but without human forecasters you’d find a far more frustrating forecast to deal with, that changes wildly from update to update.

And even in the future, as the models become more accurate with improved technology, you still need human eyes to check it and add local knowledge.  After all, a plane can land on autopilot, but most of us would prefer the pilot doing it and looking out his cockpit window at the same time – the same goes with weather forecasting and the use of automated data.

– By head weather analyst Philip Duncan