Google has also used AI to improve many of the services and technologies it offers to its users. One of its recent advances is SEEDS. Thanks to this new model, Google will use Artificial Intelligence to forecast the weather. Hopefully this will translate into more accurate and easier to make weather predictions.
SEEDS is the acronym for Scalable Ensemble Envelope Diffusion Sampler, a generative AI model capable of efficiently creating ensembles of weather forecasts at scale, at a much lower cost than traditional forecasting models. Without a doubt is a technology that opens new horizons within the fields of meteorology and climatology.
Probabilistic forecasting: the new paradigm
Although scientific progress in the study of climate and atmospheric phenomena has advanced spectacularly in recent decades, There is still a large margin of error when it comes to predicting what the weather will be like. in a region and at a certain time. Not in vain are there many humorous comments that we usually hear about it: "The weatherman is always wrong." They are not fair judgments, but they hide a bit of truth.
These predictive errors are due to the high cost (we refer to the computational cost) of generating the probabilistic forecasts. Large, powerful computers are needed that meteorological agencies cannot afford. For this reason, more traditional methods of observation and prediction are used, whose degree of precision is far from perfection.
Now, thanks to AI, generating probabilistic forecasts is no longer a pipe dream and becomes a real possibility. The old human dream of knowing what the weather will be like tomorrow with total reliability will be possible. A new paradigm that changes everything. Or, at least, that's what they say from Google.
How SEEDS works
Let's see how SEEDS works, the great tool with which Google claims it can accurately forecast the weather. a high level of accuracy and reliability.
The new technology is based on probabilistic denoising diffusion models (a generative AI method pioneered by Google Research). These models generate are based on the calculation and assignment of probabilities about the climate, generating more accurate predictions in less time and with much less computational cost.
One of the highlights of SEEDS is its ability to generate highly detailed images and videos. This is very useful when generating forecasts and applying them to classic weather patterns. In other words, Google's weather forecasting technology does not replace previous methods, but rather improves them. A hybrid forecasting system in which some climate aspects are calculated with a physics-based model and others through AI to achieve, together, a much more efficient predictive model.
Since SEEDS directly models the joint distribution of the atmospheric state, realistically capturing a lot of data and technical magnitudes that influence each other (the mean sea level pressure or the generation of winds in the troposphere, for example). For a layman, like the person writing this article, all this sounds like Chinese, but meteorologists know how to appreciate what this technology provides to the fullest extent.
Hats wing date, the results are promising. The models projected by SEEDS have been compared with real meteorological data a posteriori yielding a really high degree of coincidence. There is still considerable room for improvement, but everything indicates that the development of this technology based on artificial intelligence is on the right track.
Conclusions
SEEDS proposes an alternative weather prediction model that could revolutionize this field. The notable savings in computational resources that it represents can be used in two directions: increase the degree of precision of weather forecasts or increase the frequency with which these forecasts are issued. In any case, both paths lead to the same destination: more precision and reliability.
This is also a very good example of how AI can accelerate the progress and development of climate-related scientific fields. A hot topic that currently generates great concern. If these advances prosper, it would be possible in the not too distant future project the arrival of certain meteorological disasters (storms, hurricanes, floods...) and, if not avoid them, at least alleviate their consequences.
It must also be said that it is a technology in the making. For now, the SEEDS model and other technologies in development will be included within other Google projects such as MetNet-3 and GraphCast. Until its use is perfected and widespread, we will have to continue resorting to other tools such as apps to predict the weather, which don't work badly either.