An AI-native geospatial application that combines Google Earth Engine (GEE) satellite analytics with multi-decadal climate modeling to forecast urban heat risks for the top 25 most populated US cities.

This agent doesn't just show a map; it reasons over 22 years of climate data to identify specific neighborhood-level vulnerabilities and thermal trajectories.

• Satellite Engine: Processes MODIS (2003–2026) to calculate a robust warming trend using Sen's Slope and Landsat 8/9 for high-resolution 30m surface temperature mapping.
• Predictive Analytics: Projects 2026 thermal baselines by merging historical multi-decadal trends with current median surface temperatures.
• Interactive UI: A high-performance Streamlit dashboard featuring dynamic opacity toggles, interactive color ramps, and server-side GEE tile injection for seamless performance.

• Language: Python 3.11+
• Frontend: Streamlit
• Geospatial: Google Earth Engine (Service Account Auth)
• Mapping: Leafmap / Folium
• Data Sources: NASA/USGS Landsat Collection 2 Level 2 & MODIS MYD11A2.061

├── .streamlit/          # Streamlit Secrets (GEE Credentials)
├── src/
│   ├── __init__.py      # Package identifier (Critical for Cloud Import)
│   └── engine.py        # GEE Logic: Sen's Slope & Tile Generation
├── app.py               # Main Entry Point (UI, Sliders, & Mapping)
├── requirements.txt     # Dependencies (earthengine-api, leafmap)
└── README.md            # Documentation

Explore the interactive heat maps and 2026 forecasts here: 👉 https://uhichat.streamlit.app/

1. Historical Trend: We extract 22 years of MODIS daytime Land Surface Temperature (LST) to calculate the "Warming Trend" ($^\circ\text{F/year}$) using a robust linear regression.

2. Current Detail: We utilize Landsat 8/9 thermal bands (TIRS) to create a 30m resolution "Current Baseline," allowing for street-level heat island detection.

3. The 2026 Forecast: The projection is calculated by applying the localized historical slope to the high-resolution baseline: $$\text{Forecast}{2026} = \text{Landsat}{\text{Median}} + (\text{Sen's Slope} \times 2)$$

4. Visualization: Data is pre-visualized on the GEE server-side using a fixed color ramp (85°F to 115°F) to ensure consistent visual analysis across different geographic regions.

• Cloud Masking: While the engine filters for <40% cloud cover, some artifacts may appear in high-humidity coastal regions (e.g., NYC or Miami).
• GEE Quotas: If the map fails to load, the service account may have hit its concurrent request limit. Refreshing usually resolves this.

Developed for the uhichat project.