Flood Mapping System

Problem

• → Manual processing steps sat on the critical path — flood products lagged during active global disaster events when hours mattered.
• → Sensor acquisition → product generation → dissemination crossed teams and environments with fragile handoffs.
• → Emergency-management users needed repeatable, trustworthy maps — not one-off runs that only the on-call engineer could reproduce.

Decisions

• → Automate end-to-end from raw sensor ingestion through geospatial product generation — remove humans from the latency path under urgency.
• → Containerize processing stages so the same pipeline is deployable and recoverable across environments, not a snowflake workstation.
• → Integrate outputs with emergency-management and research distribution networks so “done” means disseminated, not merely generated.

Outcome

• → Near real-time flood inundation maps during active disaster events worldwide — latency measured in hours, not overnight queues.
• → Fewer manual handoffs under urgency — consistency came from the pipeline, not who was awake.
• → Supported peer-reviewed research on global water and flood mapping (GeoHorizons).

Team & leadership

Leadership mode

Technical lead for a mission-critical geospatial pipeline

Team & partners

~4 engineers and science partners, plus emergency-management users from sensor acquisition through global dissemination

What I unblocked

Removed manual handoffs that forced late-night heroics during active disaster events.

Hard decision

Prioritized automation and fault tolerance over ad-hoc speed — latency dropped because the team system was reliable under urgency.