Forecasting the Future: Huawei's Billion Dollar Weather AI Revelation

0:00Today on the podcast, we're going to be jumping into some science mixed with AI. and we are talking about predicting the weather with AI. There's actually two groups of researchers that have recently come out with some very interesting research and some very interesting techniques. So today on the podcast, we're going to dive into what is going on in the world of weather prediction and AI. So the headline story here is essentially that there is a team, a couple teams, that one of the teams was from Huawei Cloud, and essentially they developed a system dubbed the Pengu Weather, which essentially uses historical data of the weather to predict weather conditions a week in advance.

0:38And the system was trained using 39 years of weather data, using current weather patterns to make predictions. Really interestingly, I think it performs its predictions a lot faster than existing systems, but it doesn't actually provide predictions about precipitation amounts. So instead, it actually estimates temperature, wind speed, air pressure, and other weather-related data. So that really leads humans to estimate predictions based on the provided information. So simultaneously, a collective team from Tsingha University, which is essentially the China Meteorological Administration, and also one of their associates at the University of California, Berkeley, they designed what's called NaocastNet.

1:19So unlike Pengu Weather, NaocastNet focuses on predicting precipitation levels for the next six hours using both historical data and also physical rules, and it demonstrated accuracy compared to traditional systems and also delivered results more rapidly. So currently the most accurate weather forecasting is done using numerical models which apply mathematical and physical formulas to current weather data. And these systems, while generally reliable for major metropolitan areas, are CPU intensive and can take hours to generate results. So both the Pengu Weather and now CastNet systems both actually promise to significantly speed up this process.

2:01Eam Ebert Uhoff and Kyle Hilburn of the Cooperative Institute of Research in the Atmosphere at Colorado State University have recently kind of jumped into this whole conversation. And they did this by actually publishing a weather predictive systems report recently in which they called News and Views article in Nature. and so essentially they highlighted the challenges of building AI weather predictive systems and also the accomplishments of the team behind these two new platforms that have been developed so I think you know overall this is really impressive while the accuracy of weather prediction is really crucial to a variety of sectors like if we're talking about agriculture disaster management you know being able to accurately predict the weather is a really really important for those sectors particularly agriculture predicting when rain or snow or frost might affect your crops um with you know either kill them or help them so this is really really important um billions of dollars trickles down below that um particularly through agriculture and so this is something that people really want to get right and so i think because of that the potential of these ai based weather predictions is really just the beginning um we're really just beginning to kind of tap them.

3:16So I think though these systems are still in the kind of test of principle stage, the promising results point to the possibility of AI based weather forecasting becoming essentially the standard approach in the not too distant future. And of course, you know, we haven't been doing this in the past, we haven't been doing a lot of this AI weather predictive systems. But now that we have this AI, and we have these new AI power capabilities that are super powerful, I think we're going to be seeing a lot more of this in the future. I think that these advancements mark a really significant step forward in the use of AI in meteorology, which in my opinion suggests a future where weather predictions are a lot faster, they're a lot less resource intensive, and maybe even more accurate, right?

4:00So I think it's interesting with both of these new models that have come out, they both lacked something, but together they were able to predict the precipitation and the weather patterns based off of data and historical information and what we're seeing today. So I think that's really impressive. Combining the two, they do get a really massive chunk of this prediction and they get it right to a high level. And I think the results of these two kind of teams that are working on this are really impressive. They both published reports in the journal Nature, so you can read more about them there. But overall, I think this is going to have some massive implications for a lot of different industries, for weather, for news, for agriculture, a lot of different spaces.

4:40Of course, everyone wants to know what the weather's like and if you could more accurately predict that i think people would appreciate that i've definitely have lived places before where um weather predictions were never very accurate and uh some people just never checked what the weather was because the accuracy was not super high i think today you know by and large we have fairly accurate weather prediction results in a majority of places but being able to take this up to a higher level um and i think it's going to be really impressive especially when we're looking at historical data Something I view as being very impressive, very interesting is, you know, being able to look at a further forecasted weather into the future.

5:19Now, I know that that is that is sort of difficult due to a lot of different variables and a lot of the, you know, the weather predictions is essentially looking at historical data than looking at the data that we have today about, you know, how the weather and atmosphere and everything around us is going. So I understand there definitely are challenges there, but I think overall this is going to be a really interesting space to follow in the future. there definitely are some big implications into how this affects some, you know, multi-billion dollar industries.