Kubernetes Meets Climate Science: Building Large-scale Feature Detection From Climate Data Records
Talkat KubeCon + CloudNativeCon Europe 2025 · on Friday, 2025-04-04· at 11:45· Level 0 | ICC Capital Hall | Room J·
summary
European Weather Cloud hosts a Kubernetes platform on OpenStack with Rancher, JupyterHub, Label Studio, Nucleo and a cloud‑native Postgres/PostGIS feature DB, using two A6000 GPUs presented as vGPUs and time‑slicing for sharing. Models evaluate faster R‑CNN as best for storm detection, scanned 2,000 images containing ~5,000 tropical storms and use IBTrACS as ground truth, but manual expert review remains required. SAM segments satellite imagery well and is integrated for assisted labeling, the MLOps team is tiny (~4) and the pathfinder setup is working with scarce GPU resources.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 3:35European Weather Cloud concept
- 5:42Earth system feature detection goals
- 6:47Tropical storm detection
- 9:17Segment Anything Model for satellite imagery
- 10:22Collaborative labeling platform blueprint
- 12:31Kubernetes setup
- 16:25Platform challenges
- 17:51Design principles and toolchain
- 19:56User experience and Jupyter Hub
- 21:18Nuclio for Earth observation
- 22:44Label Studio integration
- 24:53Labeling workflow demo
- 26:39Future plans
- 27:24Q&A
related talks
31mKubernetes For GPU Powered Machine Learning Workloads In Academia
22mKubernetes Meets Climate Data: Designing Analysis-Ready Dat...
09mGen AI Workflows for Climate Risk Modeling in Food Systems
46mMachine Learning with Kubernetes
30mDesigning a Hybrid AI Platform on Kubernetes
06mKubernetes Love Machine Learning, Even on Private Cloud
22mBuild and Monitor Machine Learning Services in Kubernetes
44mEmpowering Data Science with Kubernetes | Samuele Chiocca and Andrea Rota