USGS Partners with AI/ML Experts to Improve Landsat Flight Operations
January 15, 2026
The U.S. Geological Survey Technology Transfer Office has entered into three Cooperative Research & Development Agreements (CRADAs) with a.i. solutions, Inc., KBR, and Parsons. The goal: apply AI and ML to streamline Landsat flight operations and strengthen day-to-day decision-making.
Initial focus areas include optimizing satellite anomaly triage and improving telemetry trending and analysis. The team will build on existing processes to monitor and manage Earth-observing satellites moving at nearly 17,000 mph in an orbit roughly 438 miles above Earth.
What this means for operations teams
- Faster, clearer anomaly handling: ML models can prioritize issues by impact and likelihood, reducing noise and shortening time-to-triage.
- Telemetry with fewer surprises: Trend detection and predictive thresholds help flag drift early, cut false alarms, and reduce alert fatigue.
- Repeatable playbooks: Pattern recognition from historical data supports standard responses, handoffs, and escalation paths.
- Knowledge captured in tooling: Expert judgment gets translated into features, labels, and rules that new operators can use from day one.
Early research scope
Research is underway using telemetry data libraries from Landsat 8 and Landsat 9. The work traces back to a December 2024 Sources Sought Notice requesting expertise to apply AI/ML to future Landsat mission operations.
Implementation details Ops leads should plan for
- Data readiness: Clear definitions for events, labels, and severity; consistent time sync; quality checks on gaps and outliers.
- Model governance: Approval gates, versioning, drift monitoring, retraining schedules, and rollback procedures.
- Human-in-the-loop: Operators remain final authority; UI should expose why a model made a call and how to override it.
- Integration: Connect models to existing telemetry pipelines, paging, ticketing, and reporting-start in shadow mode before enabling actions.
- Verification and validation: Backtesting with historical incidents, red-teaming edge cases, and simulating anomalies before production use.
- Security and reliability: Access control for data/models, audit trails, and fail-open behavior if ML services are degraded.
How to prepare your team now
- Map your top 10 incident types and define triage classes, success metrics, and acceptable false positive/negative rates.
- Stand up a clean telemetry baseline: standardized schemas, unit consistency, and a feature catalog that operators can understand.
- Pilot a small anomaly triage model on a narrow subsystem, measure MTTR/MTTD changes, and iterate.
- Document decision rules in plain language; convert them into labels and tests that can be automated later.
USGS plans to share updates on this research when appropriate for all parties involved in the agreements.
For additional context on Landsat missions, see NASA's overview here. For background on CRADAs, see the Federal Laboratory Consortium resource here.
If your operations team is building AI skills for telemetry, triage, and automation, you can explore role-aligned options at Complete AI Training.
Your membership also unlocks:
