Weather forecaster vs AI algorithm - who would you choose?

Weather forecaster vs AI algorithm - who would you choose?

When building a wind farm and the forecasted weather for wind, waves or swell are close to your operational thresholds the smart move is to talk to your weather forecaster. The forecaster will explain to you the evolution of the weather and the associated uncertainty or possible bias in their forecast. The advantages of regularly having these conversations can help keep people safe, and save your project tens of millions of pounds in unnecessary weather downtime.

Semisubmersible installing blades pic ref 2.

Yesterday I did an internet binge on GoogleMind`s new AI generated weather forecast (thanks K2 Management ). I hoped that when I woke up this morning I would have a clear chain of thought - or be more positive about the future.

GraphCast's 10 day synoptic chart forecast for forecast from ECMWF`s website pic ref 3.

I am in favour of the democratisation of science and meteorology (although many good quality forecasts are available free or at a low cost).

One argument in favour of AI forecast is that "AI models run 1,000 to 10,000 times faster than conventional NWP models, leaving more time for interpreting and communicating predictions" [1]

However ?"it will be another two to five years before people can use forecasting from machine learning approaches to make decisions in the real-world" [1]

?[1] https://www.nature.com/articles/d41586-023-03552-y#:~:text=AI%20models%20run%201%2C000%20to,in%20the%20Atmosphere%20in%20Colorado.

But how will forecasters interpret the forecast if it is produced by a black box system? If you procure your forecast from an AI supercomputer - who do you phone if you are uncertain in marginal weather conditions?

These are important as there are 2 parts to making a weather forecast. The first is forecasting the weather, and the second is appreciating the uncertainty.

When I have analysed forecast uncertainty in the past, I have found that 1 in 7 weather windows were wasted due to forecast uncertainty. When scaled up through the cost of installing a project this can easily equate to tens of millions of Euros, and hundreds of millions though the operational lifetime of a project.

Students learning from NOAA Lead Meteorologist about whether to make a decision on whether to issue a Severe Thunderstorm Warning. pic ref 4.

So for now I expect that installation and marine weather experts like the ones at K2 Management will pay attention to the developments in AI forecast but I will still advise our clients to use proper human (oxygen and experience in -> carbon dioxide and advice out) , sat in a chair meteorologist to add confidence when they are unsure if the weather will be suitable to safely work or not.

What do you think?

Reference:

[1] Carissa Wong`s article in Nature:, doi: https://doi.org/10.1038/d41586-023-03552-y


Picture credits:

1. 234H Global Forecast System ensembles from: https://www.netweather.tv/charts-and-data/ensembles.

2. Heerema`s semi sub installing blades from: https://www.maritimeeconomy.com/admin/uploads/news-img/79914278.jpeg

3. GraphCast's 10 day synoptic chart forecast for forecast from ECMWF`s website

https://charts.ecmwf.int/products/graphcast_medium-mslp-wind850?base_time=202311160000&projection=opencharts_europe&valid_time=202311251800

4. Students learning from NOAA Lead Meteorologist

https://www.weather.gov/lmk/Students2023




Laurence Fahrni

Senior Naval Architect [GM]

11 个月

No doubt you've seen the reporting around this University of Copenhagen/Victoria study? AI assisted insights on rogue waves, and their regularity. https://www.pnas.org/doi/10.1073/pnas.2306275120

回复
Rob Hutchinson

Meteorological Consultant / Team Lead at Meteomatics Ltd: Helping organisations address complex weather-related challenges

11 个月

Great post Jake! We've come to much the same conclusions - whilst AI will certainly have a role in the future of weather forecasting, the consensus is that the recent buzz around AI approaches is overstated. Chat GPT just gives a blurry image of the internet, and AI just gives a blurry image of ERA5 with no real ability to resolve important fine-scale details (especially in complex near-shore environments). Much higher resolution is required here and physics always wins...

David McMillan

Reader in Wind Energy at University of Strathclyde

11 个月

Fascinating post Jacob. Im just fresh from a workshop where we discussed the current usage of forecasts & data. One of the things i've observed is how different communities use the uncertainty info. The traders are all over this bc they understand the link between uncertainty and risk in their financial decisions. Operations community have been a bit slower and I think thats partially bc its more safety critical and decisions tend to be made on a more conservative basis (Ian Bonnar & Elena González know a lot about this on the T&I/ O&M side). You could argue that a black box is very undesirable in that case. Its also a visualisation thing - I know this is something you have been in the forefront of as well. We did a bit of thinking on this within the ORACLES project with Jethro Browell & Ciaran Gilbert see paper below https://iopscience.iop.org/article/10.1088/1742-6596/1222/1/012040 now thinking about it within the context of Satellite measurements within the OSCAR project.

While I can see its practical uses, I am against its debilitating effects. I don't want to give up science! BTW, I would call this ML, nothing intelligent about it!

Why choosing if one can have both and combine the forecasts?

要查看或添加评论,请登录

社区洞察

其他会员也浏览了