Korea Typhoon Research Institute (KTRI)
AI weather model evaluation for tropical cyclone forecasting.
Research portfolio for real-time AI model forecasts, ensemble experiments, tropical cyclone analysis tools, and conference poster archives.
Live system
AI Real-time Prediction
Deep Learning Model Forecast, latest run: 2026-08-22 06Z
Typhoon archive
2025–2026 Northwest Pacific forecast pages
Operations
Server Monitoring
Access the dashboard to check server status and performance metrics.
Managed Services
Three production web services
Operating the Typhoon Research Center service (React · Spring), the KTRI institute site with independent SSL and HTTPS API proxy, and this realtime research portal.
Stack & Automation
Nginx · SSL · scheduled pipelines
Per-domain Nginx and SSL configuration with cron · PM2 · systemd schedulers. A hybrid filesystem + metadata DB (10 tables) backs 16 automated collection and post-processing jobs for KMA, JTWC, RSMC, GK2A, OSTIA, and HYCOM data.
10 DB tables
16 scheduled jobs
1982–2025 archive
Security Practices
Audit-driven hardening
Periodic security audits with risk-based triage: parameterized SQL, input · file · path validation, XSS sanitization, hardened cookies and tokens, centralized CORS, and secrets kept in environment variables.
Domain isolation
No wildcard vhosts
HTTPS everywhere
Archive
Posters
01
Analysis of AI-based Global Models at WP
Performance Analysis of AI-based Global Models in Tropical Cyclone Forecasting
AOGS 2025
02
Assessing AI-based Models at TC track and intensity
Assessing AI-based Global Weather Models for North Pacific Tropical Cyclone Track and Intensity Forecasts
ECMWF workshop 2025
03
Intensity & Track Correction Comparison for WP, EP
Intensity calibration of WP, EP using various methods. In particular, focus on XGBoost correction.
KSO 2025Directories
Research Fields
01
AI-MODELS
AI-models for TC
NVIDIA FourCastNet
FourCastNet-v2
Huawei Cloud Pangu-Weather
Fudan Univ. Fuxi
Shanghai Univ. Fengwu
Google DeepMind GraphCast
Gencast
Microsoft Aurora
Update AI model Microsoft's Aurora to 'Aurora' folder
02
Ensemble
Ensemble Result of Gencast and AI-MODELS
03
Tools
All of management and analysis tools
Update AI model's start-end time verification to 'Analysis' folder
Highlights
Research Highlights
Ensemble Research
BV-perturbed AI ensemble for TC tracks
Applied Bred Vector initial perturbations to AI weather models to evaluate ensemble predictability of tropical cyclone tracks over the Western North Pacific.
🏆 2026 수로학회 우수 논문 발표상Model Pipeline
7 AI models, one automated pipeline
GraphCast, Pangu-Weather, FourCastNet-v2, FuXi, FengWu, Aurora, and GenCast run through a unified inference → TC tracking → verification → visualization → realtime publishing workflow.
Post-processing 80+ min → 20 min
Aurora inference −65%
Data Quality
Typhoon dataset cross-validation
Cross-checked initial times and IBTrACS storm IDs against 2021–2025 Northwest Pacific typhoon records, cataloguing missing model×time cases for automated re-collection.
4,891 initial times
417 storm IDs
1,364 gaps identified
Intensity Correction
Statistical & XGBoost calibration
Compared statistical and XGBoost-based corrections to reduce tropical cyclone intensity bias of AI weather models over the WP and EP basins.
KSO 2025 poster