AIRBENDER is a real-time, personalized air-quality intelligence platform that transforms NASA TEMPO satellite data into individualized forecasts and actionable guidance.
Developed by Team NAVI for the NASA Space Apps Challenge 2025, it bridges space-based science and everyday life - helping people breathe safer, live smarter, and stay informed in real time.
AIRBENDER turns NASA TEMPO Level-2 and Level-3 Version 3 data - including NO₂, O₃, HCHO, and Cloud Fraction (CLDO4) - into personalized, human-centric air-quality intelligence.
Unlike conventional dashboards showing city-level indices, AIRBENDER adapts forecasts to each user's location, routine, and health context, combining cloud computing and large language models (LLMs) to deliver empathetic, actionable insights.
AIRBENDER continuously retrieves and processes NASA TEMPO data through a cloud pipeline built on the Google Cloud Platform (GCP).
It generates hourly nowcasts and short-term forecasts that are:
- Location-specific
- Context-aware
- Personalized to user profiles
Based on these predictions, a large language model (LLM) converts scientific outputs into natural, human-like messages - adjusting tone, timing, and phrasing to fit each user’s lifestyle and preferences.
- Connects to the NASA Earthdata Cloud using the earthaccess API
- Retrieves TEMPO NO₂, O₃, HCHO, and Cloud Fraction (CLDO4) datasets
- Filters by geographic coordinates (e.g., NYC: −74.3 to −73.6°, 40.4 to 41.0°)
- Converts NetCDF → structured time-series using xarray and pandas
- Stores cleaned data in Google Cloud Storage with version control
- Uses a lightweight AI model on Google Cloud Run
- Combines historical TEMPO data with meteorological covariates
- Cross-references forecasts with each user’s encrypted metadata:
- Occupation
- Outdoor exposure
- Commute schedule
- Health sensitivity
- Produces individualized forecast layers for each user
- Integrated LLM (Qwen 2.5-VL) generates human-like recommendations
- Adapts message tone and timing dynamically
- Sends alerts via push/email notifications (e.g., “07:30 morning brief”, “18:00 preview”)
- Modular design supports easy integration of:
- AOD (Aerosol Optical Depth) data
- Ground-station PM₂.₅ / PM₁₀
- Future TEMPO V04 products
- Scalable architecture enables global deployment
- Provides real-time, personalized forecasts per user
- Issues actionable, empathetic recommendations like
“Delay your commute by one hour for cleaner air.” - Protects sensitive groups (children, elderly, respiratory patients)
- Encourages preventive behavior (mask use, ventilation timing, exercise scheduling)
- Makes NASA TEMPO data accessible and interpretable to the public
- Supports urban planners and local authorities in identifying pollution patterns
- Promotes data-driven awareness and community-level behavioral change
- Demonstrates how TEMPO L3 V03 can power citizen-facing AI services
- Provides a reusable framework for merging satellite data + AI personalization
- Fosters public engagement with NASA open data
- Envisions a world where everyone receives personalized environmental intelligence
- Expands beyond TEMPO to other satellite missions, connecting people globally to NASA’s Earth science ecosystem
AIRBENDER aims to evolve into a cloud-based environmental intelligence agent -
a personal assistant that interprets NASA Earth observation data in real time.
It embodies the Space Apps theme “From EarthData to Action” by turning complex atmospheric measurements into personalized, meaningful actions that empower individuals and communities to live sustainably.
- Python, JavaScript, HTML/CSS
- Python: Data ingestion, preprocessing, forecasting
- JS/CSS: Interactive visualization and control
- NASA Earthdata (earthaccess API) for TEMPO datasets
- xarray, pandas, NumPy for NetCDF handling and preprocessing
- Google Cloud Storage for data management
- Google Cloud Platform (GCP) components:
- Cloud Run – AI forecasting API
- Cloud Storage – dataset repository
- Cloud Functions – automation and orchestration
- Qwen 2.5-VL (Alibaba Cloud LLM) for adaptive, context-aware notifications
- Dynamically adjusts tone, timing, and message style
- Google Gemini Veo 3 for animated, data-driven storytelling
- GitHub for source code and version tracking
- Notion for documentation, dataset logs, and progress tracking
Members
-
Jeong Ah Yoon – Project Lead & AI Personalization Designer
github.com/jjyoon012-git -
Jihoon Jeong – Data Pipeline Architect & Visualization Engineer
github.com/jeehun3020 -
Jiho Ryu – Data Scientist & Environmental Data Engineer
github.com/ryujihos0105
Challenge
From EarthData to Action: Cloud Computing with Earth Observation Data for Predicting Cleaner, Safer Skies
(NASA Space Apps Challenge 2025)
For inquiries or collaboration:
Team NAVI – [email protected]
GitHub: https://github.com/nasa-navi