Terradue has integrated a comprehensive MLOps framework into the Geohazards Exploitation Platform (GEP), extending the platform’s ability to support AI-ready Earth Observation workflows.
These new capabilities enable users to develop, train, deploy, monitor, and improve machine learning models for geospatial and geohazard applications directly within a scalable cloud-native environment.
This strengthens GEP as a platform for advanced EO analysis, operational geohazard monitoring, terrain motion analysis, disaster response, critical infrastructure assessment, and environmental monitoring.
What is now supported?
GEP’s MLOps framework supports the full machine learning lifecycle, from data preparation to operational deployment.
From EO data to operational AI services
High-level view of the GEP MLOps workflow: data, model development, training, deployment, inference, monitoring, and feedback.
The new framework allows AI models to be managed as operational assets rather than one-off experiments.
A typical workflow can include:
- Ingesting EO, geospatial, and in-situ datasets
- Preparing training datasets and features
- Developing models in cloud-hosted environments
- Training and validating models with reproducible configurations
- Deploying models as containerised services
- Exposing results through APIs, GEP services, or thematic applications
- Monitoring model performance over time
- Updating or retraining models when new data becomes available
This approach helps users move from research prototypes to repeatable and scalable AI-driven services.
Cloud-native and interoperable by design
GEP’s MLOps capabilities are designed for cloud-native deployment and scalable execution.
They support:
- Containerised model execution
- Automated deployment pipelines
- Microservice-based architectures
- OGC-compliant service integration
- Scalable cloud resources
- Operational monitoring and logging
- Feedback loops for continuous improvement
This makes the framework suitable for both research experimentation and operational EO services.
Continuous monitoring and feedback
Operational AI services need to remain reliable as new data and real-world conditions evolve. The GEP MLOps framework supports monitoring and feedback mechanisms that help identify:
- Model performance degradation
- Data drift
- Changes in input data characteristics
- Execution issues
- Validation gaps
- Opportunities for retraining or model improvement
These capabilities are especially relevant for dynamic geohazard applications, where conditions may change rapidly after events such as earthquakes, landslides, volcanic activity, floods, fires, or subsidence episodes.
Why this matters for GEP users
The integration of MLOps into GEP enables new possibilities for AI-assisted geospatial analysis.
Potential application areas include:
- Geohazard monitoring
- Terrain motion analysis
- Event response
- Critical infrastructure assessment
- Environmental monitoring
- Change detection
- Risk screening
- Large-scale EO data analysis
- Thematic application enhancement
For researchers, the framework supports reproducible experimentation and model validation.
For service developers, it supports packaging, deployment, and execution of AI-enabled workflows.
For operational users, it provides a path towards scalable and monitored AI services that can adapt as new data becomes available.
Next steps
More information will be shared through GEP as these capabilities are made available to users, projects, and stakeholder communities. If you are interested in developing, testing, or deploying AI-enabled EO workflows on GEP, contact:
Please include a short description of your use case, the type of EO data involved, and whether you are interested in model development, training, deployment, or operational monitoring.