ClimateLens

See how Earth is changing through AI-powered satellite analysis.

Overview

ClimateLens is an AI-powered satellite monitoring dashboard for exploring environmental change across time. It detects patterns such as vegetation change, urban growth, water change, and ice/glacier variation using Sentinel-2 imagery when available and a location-aware fallback mode when live imagery is unavailable.

The platform combines:

• interactive map overlays
• time-series trend charts
• timeline exploration
• Simple / Expert modes
• AI-generated explanations and recommendations

Features

• Multi-year environmental change analysis
• NDVI, NDWI, and NDBI computation
• Land cover classification and change detection
• Interactive Leaflet map with overlays
• Plotly trend charts and forecasts
• Timeline explorer
• Simple Mode and Expert Mode
• AI assistant for context-aware explanations
• Sentinel-2 live mode with graceful fallback support
• Caching for analysis and imagery

Quick Start

1. Create and activate a virtual environment

python -m venv .venv
.\.venv\Scripts\Activate.ps1

2. Install dependencies

pip install -r requirements.txt

3. Configure Sentinel Hub (optional, for live data)

$env:SENTINELHUB_CLIENT_ID="your-client-id"
$env:SENTINELHUB_CLIENT_SECRET="your-client-secret"
$env:SENTINELHUB_ENABLED="true"
$env:SENTINELHUB_MAX_CLOUD="0.2"

If you want to run without live Sentinel-2 data, omit the credentials or disable Sentinel mode:

$env:SENTINELHUB_ENABLED="false"

4. Run the app

python -m uvicorn main:app --reload

Open the UI at:

http://127.0.0.1:8000

Architecture

• FastAPI backend for analysis endpoints and static asset serving
• Sentinel Hub integration in sentinel_client.py for live Sentinel-2 imagery
• Satellite processing in satellite_processing.py for NDVI / NDWI / NDBI and land cover classification
• Analysis orchestration in ai_module.py for change detection, time series, timeline logic, fallback behavior, and forecasting

Caching

• SQLite analysis cache: data/cache.db
• Sentinel imagery cache: data/sentinel_cache/

Frontend

• React (CDN)
• Leaflet map layers
• Plotly charts

Satellite Analysis Workflow

1. Resolve a location from city name or coordinates.
2. Fetch Sentinel-2 bands (B02, B03, B04, B08, B11) for multiple years when available.
3. Compute:
   • NDVI for vegetation
   • NDWI for water
   • NDBI for built-up / urban areas
4. Classify land cover into vegetation, water, urban, and ice.
5. Compare years (2016, 2018, 2020, 2022, 2024) for change detection.
6. Generate heatmaps, previews, and time-series charts.
7. Forecast the next 5 years using a linear trend model.
8. Produce AI-generated explanations based on current analysis.

Source Modes

ClimateLens can operate in different source modes depending on data availability:

• sentinel - live Sentinel-2 imagery was used
• mixed - some years used Sentinel-2 and some years used fallback data
• demo - location-aware fallback imagery was used
• config_missing - Sentinel config is missing
• auth_failed - Sentinel authentication failed
• no_scenes - no suitable Sentinel scenes were found
• download_failed - imagery request/download failed
• processing_failed - imagery processing failed

Sentinel Hub Setup

Set environment variables locally and never commit secrets.

Required:

• SENTINELHUB_CLIENT_ID
• SENTINELHUB_CLIENT_SECRET

Optional:

• SENTINELHUB_MAX_CLOUD (default: 0.2)
• SENTINELHUB_CACHE_DIR (default: data/sentinel_cache)
• SENTINELHUB_ENABLED (true / false)

Notes

• Sample scenes are generated on first run under data/
• Analysis caching is stored in data/cache.db
• Sentinel imagery cache is stored in data/sentinel_cache/
• The API logs whether Sentinel credentials are present, whether authentication succeeds, whether scenes are found, and the exact fallback reason when live data is not used
• The map uses OpenStreetMap and optional Esri satellite tiles when online

Known Limitations

• Some locations or years may fall back to demo mode because no suitable Sentinel scenes are available
• Cloud filtering and scene selection are simplified
• The AI assistant is context-aware and constrained, but not a general-purpose LLM analyst
• Demo fallback mode is synthetic, although it is location-aware

Tests

python -m pytest