A talk hosted by the Wanaka Branch of The Royal Society of New Zealand will provide an overview of some of the latest developments in climate science.
The speaker is Dr Greg Bodeker, of Bodeker Scientific in Alexandra, who will share some of the work of his team.
Identifying extreme weather events often began by using data from MetService.
‘‘MetService are constantly running numerical weather prediction models, to do standard weather forecasts.’’
Identifying extreme weather events was looking for events that typically were at the 99th percentile.
‘‘So something that maybe only happens one in every hundred days, or might only happen once or twice every decade.’’
One of the projects they were involved with was called the emergence of the climate signal, funded by the Ministry of Business, Innovation and Employment.
‘‘It is all about how will we, over the coming decades, experience that kind of emerging signal over and above day-to-day weather variability.’’
The work programme Dr Bodeker was leading was about making sure models used incorporated the right level of understanding of the physics of the atmosphere to get extreme weather events right, he said.
When thinking of an extreme weather event, ‘‘especially massive rainfall or massive snowfall’’, they tended to be ‘‘very small, very energetic’’ events that happened in a contained area.
‘‘So a very, very localised event that just dumps a lot of rain in a very focused spot.’’
That often was what caused a lot of damage, for example, roads washed away.
The challenge for weather prediction models was being able to be very specific.
Climate models might run at a larger ‘‘resolution’’ for example 50km by 50km, but they would not help show the effects of these very localised storms.
Machine learning could help with this by running larger climate models and then teaching the computer how to predict what more localised weather events could look like.
‘‘The learning process can be slow and expensive, but once it is trained it is extremely fast at being able to simulate high resolution weather, and that is what you need to get these model estimates of what these future extremes might look like.’’
‘‘One way to think of these models is wearing glasses where everything is a bit fuzzy, everything is blurry.’’
A model projection of rainfall over New Zealand might be ‘‘a bit blurry, you are just seeing general shapes’’.
‘‘This machine learning approach now gives you 20/20 vision — you get really, really focused vision.’’
Being able to have more accurate prediction models was always about hazard, exposure and venerability.
A hazard, like rainfall, would have an effect on a vulnerability, like a road.
‘‘The risk of something going wrong is the hazard, times the exposure, times the vulnerability.’’
An example might be if rainfall was the hazard, exposure was the severity or volume of rain and vulnerability might be the strength of a road in being able to withstand that level of rainfall.
‘‘So everything we are doing is about the hazard — the rain falling out of the sky.
‘‘Whether or not it washes the road away, well that is now an issue about vulnerability.’’
★ Dr Bodeker will speak at the Presbyterian Community Centre in Tenby St, tomorrow from 6pm.