Introducing the Geospatial Studio, a toolkit for researchers and developers to quickly and easily build geospatial AI models for our evolving planet.
We are in the early stages of development and are looking for interested users. Sign up for updates and early access to help shape the product and its features.
How it works
- Choose a pre-trained model - Begin by selecting one of our foundation AI models that have been pre-trained on massive amounts of geospatial data to recognize patterns and features that can inform a variety of tasks.
- Prepare your data to create a custom model - Gather and organize your specific geospatial dataset for the model to learn new patterns specific to your analysis.
- Train the model - Run your model tuning without having to worry about complex AI model requirements like setting hyperparameters and accessing GPUs (all handled automatically).
- Use your new model to make predictions - Provide the model with the time and location of interest and receive back answers to make informed decisions based on more accurate predictions.
Benefits of Geospatial Foundation Models
Less labeled data required
Geospatial foundation models reduce the volume of labeled data required to build AI models that target key applications. Obtaining a sufficient amount of labeled data is a recognized obstacle when building AI models for geospatial applications.
Generalizability
Geospatial foundation models understand the highly complex patterns in Earth system data, including from Earth observing satellites and climate models. This general representation allows the fine-tuned models to generalize across applications, space, and time.
Less time and compute
Using the foundation model as a baseline can save fine-tuning time and compute. Because the model has already learned the general representations of the geospatial data, there is less reliance on HPC at the user side to create a fine-tuned model for your specific application. This leads to faster fine-tuning, inferencing, and deployment of your own models.
How the Geospatial Studio helps you to build and use your models
Data access and creation for fine-tuning
Fine-tuning a model requires a custom training dataset, which can be complicated to assemble for non-expert users. The Studio simplifies data access across various sources and offers tools for creating & managing custom training data & labels.
Support for fine-tuning
The Studio includes base TerraTorch configurations for common tasks which can be used or customized by users for fine-tuning. The flexibility remains for an expert user to provide a new or bespoke configuration.
Provenance of models and data
The history of all model training is tracked and recorded within the Studio, allowing monitoring and assessment of model accuracy and traceability of models and tuning datasets.
One-click deployment of models to inference
Once a model has been fine-tuned within the Studio (or onboarded from TerraTorch), a simple API call or a button click in the UI triggers model deployment to the inference service. This enables rapid availability of the models for evaluation or production.
Powerful and intuitive visualization
Geospatial analysis is a visual process and visualization is especially important for validating results and rapidly gaining insights from models. The Studio has a UI to visualize the data used for model training and the inputs and outputs from inferencing workflows.
Deployable anywhere, anytime
The Geospatial Studio is readily deployable on OpenShift, SaaS, multi-cloud, and on-prem environments and integrated with watsonx.ai where fine-tuned geospatial foundation models are now available for use.
From the blog
Disclaimer
IBM's statements regarding its plans, directions and intent are subject to change or withdrawal without notice at IBM's sole discretion. Information regarding potential future products is intended to outline our general product direction and it should not be relied on in making a purchasing decision. The information mentioned regarding potential future products is not a commitment, promise, or legal obligation to deliver any material, code or functionality. Information about potential future products may not be incorporated into any contract. The development, release, and timing of any future features or functionality described for our products remains at our sole discretion.