OceanScope AI

AI-powered 3D subsurface ocean temperature reconstruction for the North Indian Ocean.

Overview

OceanScope AI reconstructs daily subsurface ocean temperature fields across 15 depth levels (surface to ~1,000 m) on a 0.25° grid using satellite surface dynamics (DUACS SLA, OSCAR currents, CCMP winds, SST, SSS) trained against GLORYS12 reanalysis and evaluated against independent ARGO float observations.

Repository Layout

• ml/: Python package oceanscope_ml, configuration, training pipelines, and evaluation.
• backend/: FastAPI application serving precomputed predictions, metrics, and chat.
• frontend/: Next.js web application with MapLibre canvas overlays and interactive ocean exploration.
• fixtures/sample_outputs/: Contract-exact synthetic sample outputs for frontend and backend development.
• docs/: Product Requirements Documents (PRDs), decision logs, phase execution roadmaps, and progress tracking.

Development Checks

• ML: cd ml && ruff check . && ruff format --check . && mypy src && pytest -q
• Backend: cd backend && ruff check . && ruff format --check . && mypy app && pytest -q && python -m app.export_openapi --check
• Frontend: cd frontend && pnpm typecheck && pnpm lint && pnpm test && pnpm build