Google Research has announced Google’s New Generative AI Model to Reduce Weather Forecasting Uncertainty. The new generative artificial intelligence (AI) model was announced on Friday.
Google claims that the new AI Model can lessen the unpredictability and errors in weather predictions. The artificial intelligence approach, known as Scalable Ensemble Envelope Diffusion Sampler (SEEDS), is based on denoising diffusion probabilistic models rather than the conventional probabilistic model of weather forecasting. The software giant has already released two weather forecasting models: MetNet-3, a high-resolution model for a 24-hour period, and GraphCast, a model that can predict weather up to 10 days ahead of time. This is not the first weather forecasting model that the company is working on.
Rob Carver, a research scientist at Google Research, and senior software developer Lizao Li announced the announcement in a blog post. The group has an article in the Science Advances journal on the generative AI model called SEEDS. According to the statement, the AI model will improve weather forecasting in two different ways: by increasing its accuracy and lowering its cost.
Also Read: Google Pixel 6a is no longer available paving the way for the Pixel 8a
The report highlighted the two main problems with contemporary weather forecasting, stating that models now produce what are known as “probabilistic forecasts.” In essence, they use the starting conditions to provide a primary forecast, and then the model adjusts itself to produce a more accurate forecast when the conditions change and additional data are received by the weather models. Longer-duration projections can accommodate greater uncertainty, according to Google.
The study team noted that large supercomputers operating extremely intricate numerical weather models, where forecasts must be produced continuously to achieve an accurate result, can be expensive.
According to the research article, SEEDS uses diffusion probabilistic models that have been denoised and built by Google Research. It was trained using skill-based metrics, including the continuous ranked probability score (CRPS), the root-mean-squared error (RMSE), and the rank histogram. According to the paper, even though the model has a little computational cost, it also increases the first prediction’s accuracy, reducing the number of forecasts that must be generated within a certain time frame.
The study team discovered that the AI model provided more reliability than the Gaussian model when it was used to predict the weather. It noted, citing a geopotential trough to the west of Portugal as an example, “Although the Gaussian model predicts the marginal univariate distributions adequately, it fails to capture cross-field or spatial correlations. This hinders the assessment of the effects that these anomalies may have on hot air intrusions from North Africa, which can exacerbate heat waves over Europe.”.
According to Google Research, SEEDS can consider these things to make better predictions. Peer review has not yet been completed, and the model may eventually be transformed into a commercial model based on its viability.