dkidi

Satellite Imagery and AI Agent-Based Crop Condition Monitoring and Investment Decision Support Program

위성영상과 인공지능 에이전트 기반 농작물 작황 모니터링 및 투자 의사결정 지원 프로그램

What it does

dkidi turns satellite imagery into actionable crop intelligence. We collect Google Earth Engine MODIS daily 250 m NDVI for tens of thousands of sample points across top producer countries, compute a Standardized Vegetation Index (SVI) against a rolling 10-year baseline, and surface crop condition as a 6-tier grade (Excellent → Critical) on a 3D Cesium globe.

Real-time satellite signals are paired with USDA WASDE, NASS, NOAA, EIA, and CBOT futures (corn / wheat / soybean) so investors and analysts can move from satellite signal to price impact in a single workflow. The entire data stack is queryable in natural language through an Anthropic Claude–powered AI agent.

Live product screenshots

3D Cesium globe showing satellite NDVI status dots colored by SVI 6-tier grade (Excellent to Critical) across wheat, maize, rice, and soybean producing countries
3D globe — global crop condition by SVI grade
Multi-year NDVI time-series chart for a single sample point with 10-year baseline median band, season shading and SOS/EOS/Peak markers, current-year overlay
Cell chart — multi-year NDVI vs 10-year baseline
Admin jobs page showing CSV sample upload, validation status, NDVI collection job progress percentage, and bulk delete actions
Admin jobs — CSV upload & NDVI sweep monitoring
AI agent chat interface integrating satellite NDVI, news, WASDE supply-demand, and CBOT futures into natural-language answers
AI agent — natural-language queries across data layers

Key capabilities

  • Satellite NDVI pipeline — Google Earth Engine MODIS MOD09GQ daily 250 m, QA-masked, 3×3 neighborhood, batch FeatureCollection or per-point modes
  • Standardized Vegetation Index (SVI) — z-score against rolling 10-year baseline, mapped to 0–100 score and 6-tier grade with percentile clipping
  • Hybrid season detection — GEOGLAM CM4EW lookup → NDVI auto-detection v2 → none fallback; SOS / EOS / Peak DOY extraction
  • Stall detection — in-season green-up stall escalates country to Watch tier (one-way upgrade)
  • 3D globe visualization — Cesium WebGL Earth with multi-crop layer toggle, sample point dots, country-level color dots, off-season override
  • Time-series chart — Recharts multi-year NDVI comparison, baseline median / mean / max / min lines, smoothing, season shading, today + last-observation markers
  • AI agent (Claude) — Anthropic claude-sonnet-4-6 model, natural-language queries that synthesize satellite + news + WASDE + futures with cited reasoning
  • Financial integration — USDA NASS, WASDE, NOAA, EIA, yfinance CBOT corn / wheat / soybean futures (15-minute delayed)
  • Weekly automated reports — Korean + English PDF / DOCX with matplotlib crop-condition map, futures chart, and WASDE summary parsing
  • Admin console — Jobs (CSV upload + sweep monitor), Quality (outlier detection + sample CRUD), Seasons, VCI Stats, Users, Permissions, Access Stats
  • Role-based access — admin / pro / free tiers, JWT Bearer auth, AI agent gated to pro
  • Cropland validation — ESA WorldCover v200 (10 m) automatic filtering of mis-located sample points

Tech stack

Python (FastAPI) backend with Google Earth Engine SDK and Anthropic Claude SDK; React 18 + Cesium 3D frontend; PostgreSQL 14 with JSONB sparse storage; alembic migrations; deployed on Oracle Cloud Infrastructure (Always Free A1 VM, Ubuntu 22.04). Source code registered for Korean program copyright (2026-06).

Try the live product

The live application is fully functional. JavaScript must be enabled — the interactive React + Cesium app mounts on top of this summary page. If you are seeing this static summary, your browser or crawler did not execute the JavaScript bundle.

Launch dkidi