Are AI models the future of weather forecasting?
An Opinion piece by Nikko
As stated this is the opinion of myself only, based on my personal experience with AI modelling and what I’ve read from the developers of the technology. Some people may agree or disagree and thats fine. In fact debate is welcome on this subject either here or on the forums. Links will be provided below.
So let’s get in to this. With the increase in AI modelling, one has to ask, is it the answer for the future of weather forecasting.
Lets first examine the basics of these models. Currently they are in there experimental stage which means they are not fully trained to assess all aspects of forecasting. As of writing this piece they cannot fully predict tropical cyclone strengths nor have they been fully trained in the intensity of low and high pressure systems. They do show these systems but the full picture is not available. What they have been trained for is general weather outlooks. When it comes to general weather, these models have proven to have a high accuracy in the short term and hold a fairly decent accuracy out to 7 days for certain weather features, however as these models are new, they will still produce missing data. AI models definitely shine in the short term, providing the data fed to them is correct.
Key differences between AI and Numerical modeling.
• Numerical modeling is a proven established method of forecasting, whereas AI is in its infancy and relies on data fed to it from those established numerical models, this is the training process.
• Both struggle with complex systems. Although AI struggles a little less due to it learning from past weather data similar to the complexity seen.
• AI models lack the human touch, it may be able to make intelligent like decisions, it cannot make on the fly decisions that a human Meteorologist would identify and correct quickly. This may change in the future but as of writing this, AI fails in this aspect.
• Numerical modelling offers higher resolution modelling whereas AI is only available in lower resolution currently.
• AI has proven superior compared to traditional numerical models in the short term with general weather however if sudden changes happen, it fails to identify those.
• It is less likely to have major fluctuations in outlooks unlike numerical which can show huge variations between one run to another. AI has lessited this although still suffers from it occasionally. We see less fluctuations in the longer term modelling as well on AI.
• Numerical modeling is based on equations and mathematical models, AI relys on data to learn from.
So overall AI does have its advantages over traditional models even in these early days, however do I feel it should be relied upon solely. No way. Not yet, and maybe never. It all depends if it manages to get past its flaws. It should never replace humans, no amount of learning can fully replicate a humans touch as yet plus I believe real people are essential for accurate forecasting. As much as these new models are Impressive, they are far from being the only data we should rely on.
Conclusion:
AI modelling has been fairly impressive to date, showing great promise but it definitely has a long way to come before becoming a viable option instead of traditional numerical models. It has many aspects to learn before it can replace numerical, if it ever can. Its a completely different method of modelling. AI will struggle with extreme and unprecedented weather conditions as there is not enough data for it to learn from. There are some aspects of weather we are simply unsure if how it works, so data for that is missing. Example being Tornadoes, yes we know the atmospheric conditions needed to produce a potential tornado spawning supercell however we still do not know what invisible things that come together to actually create a tornado from one Supercell but the one 10 kms away doesn’t produce one even if atmospheric conditions are the same. These are the type of situations that until we have the data for, AI modelling will struggle with. As it cannot be taught what we do not know.
Like the image below, AI is far from perfect but if we like it or not, its a part of our lives now.
To discuss, please comment below or follow the link to the forums on my website here:
You will need to sign up for a free account to access.
-Nikko-
