Accessing WeatherNext on Vertex

This page guides you through using the WeatherNext model. You can access its Model Card on Vertex AI Model Garden in the Google Cloud Console.

This page covers the first steps: Getting access to the model and deploying it in your Google Cloud project.

Only WeatherNext 2 is available on Vertex.

Build custom forecasts with WeatherNext 2 on Vertex

WeatherNext 2 on Vertex AI gives you access to Google's state-of-the-art weather model. Instead of consuming pre-computed, static forecast datasets, you can now build applications that generate custom forecasts on-demand, tuned to your specific technical requirements.

This approach provides a powerful set of capabilities:

  • Batch Inference API: Deploy your own private instance of WeatherNext 2 using Vertex Batch Inference API . You can use a Colab Enterprise notebook to manage and test the model and manage inference runs.
  • Granular Forecast Configuration: Control the model's output using parameters to tailor each forecast to your use case:
    • Probabilistic Forecasting: Specify the number of ensemble members to quantify forecast uncertainty and build risk-aware applications.
    • Custom Time Horizons: Set the forecast lead time in hours for short-term operational needs or long-range strategic planning.
  • Cloud-Native Integration: The model is designed for seamless integration with the Google Cloud ecosystem. Use the Zarr output in Google Cloud Storage to trigger event-driven workflows with Cloud Functions , run large-scale data transformations with Dataflow , or load results directly into BigQuery for analysis.

Early access program

WeatherNext 2 is available through an Early Access Program (EAP), which means access is granted to a limited number of customers. The goal of this program is to gather feedback and ensure the model meets the needs of a diverse range of applications. View WeatherNext 2 page on Vertex AI

To express interest in joining the EAP, fill out the WeatherNext 2 EAP Interest Form.

To utilize these models for generating real-time forecasts with this service, you must apply for access. Your application will be reviewed, and if approved, you will be added to an allowlist.

Prerequisites

Once you have been accepted into the EAP and your project is allowlisted, you will need the following:

  1. A Google Cloud Project with billing enabled.
  2. The Vertex AI API enabled in your project.
  3. IAM Permissions: Ensure your user account has the Vertex AI User role or equivalent permissions.
  4. Request appropriate quota for Vertex Custom Training accelerators.

Finding and deploying the model

Once your project is ready, you can deploy WeatherNext 2 from the Model Garden.

  1. Find WeatherNext 2 in Model Garden: Navigate to WeatherNext 2 on Vertex AI.
  2. Deploy the Model: Select Open Notebook. Follow the instructions in the Colab Enterprise Notebook to deploy WeatherNext 2.
  3. To return to the notebook after you've deployed the model, you can navigate directly to the Colab Enterprise page in the console.

Dedicated compute allocation (single-tenant)

When you configure your deployment, you are provisioning a single-tenant environment. This means the model is deployed on dedicated hardware (GPUs) that is allocated exclusively to your endpoint. It is not a shared, multi-tenant "model-as-a-service" API.

During the deployment configuration, you will be able to select the specific type and number of accelerators (e.g., NVIDIA H100 80GB GPUs or A100 80GB GPU) for your endpoint based on your performance and cost requirements.

Things to be aware of

Your use of WeatherNext 2 on Vertex is subject to the Google Cloud terms of use, which you should familiarise yourself with. In particular, you (and your end users) must not:

  • Use WeatherNext 2 or the forecasts it generates to develop a similar or competing product or service;
  • Use generated forecasts to: (a) substitute, replace, or circumvent the use of WeatherNext 2, directly or indirectly, or (ii) create or improve models similar to WeatherNext 2 unless as part of any authorised fine-tuning; and
  • Reverse engineer or extract any components of WeatherNext 2.