ISRO BAH 2026 - PS-3

StarTrek

Seeing India's air from space. AI-powered satellite intelligence that maps surface Air Quality and HCHO hotspots across the nation - even where no monitor exists.

0
CPCB Stations
0
Monitoring Gap
0
Pollutants Mapped
0
Satellite Sources
Scroll to explore
The Gap

Most of India is flying blind on air quality

CPCB's monitoring network covers only ~500 stations - mostly in urban centres. The vast majority of India's 1.4 billion people live beyond the reach of any ground monitor.

0 km+

Average distance to the nearest air quality monitor for rural India

Our solution: Use INSAT-3D, TROPOMI, and a CNN-LSTM model to convert satellite column readings into daily surface-level AQI maps - covering every square kilometre of India.

End-to-End Workflow

From satellite to surface truth

Our pipeline transforms raw satellite columns into actionable, daily surface AQI maps and HCHO hotspot intelligence.

01

Data Acquisition

INSAT-3D AOD, TROPOMI NO₂/SO₂/CO/O₃/HCHO, CPCB ground truth, IMDAA/ERA5 meteorology, FIRMS fire data

02

Preprocessing

Google Earth Engine harmonization, regridding, quality control, cloud masking, temporal alignment

03

Feature Engineering

Columns + PBLH + humidity + ventilation coefficient + wind + solar zenith + aerosol layer height + temporal features

04

CNN-LSTM Model

Hybrid deep learning: CNN captures spatial patterns, LSTM captures temporal dependencies. Predicts surface concentrations from satellite features.

05

Validation

Compare predictions against CPCB ground truth. Evaluate with RMSE, R², and MAE. Station-level and spatial cross-validation.

06

AQI & HCHO Maps

Daily surface AQI maps, HCHO hotspot detection via DBSCAN, source-region identification, fire transport analysis with HYSPLIT.

Innovation

The X-Factor features

Most teams map AOD straight to PM2.5. We add the physics that turns a column reading into a real ground-level value.

Boundary Layer Height

PBLH controls how deep pollutants mix. Shallow winter boundary layers trap pollution near the surface - this is the #1 predictor most models miss.

Relative Humidity

Humidity swells aerosol particles, inflating AOD relative to dry PM mass. Without RH correction, the column→surface mapping breaks down.

Ventilation Coefficient

PBLH × wind speed - a confirmed strong PM2.5 predictor over the Indo-Gangetic Plain. Combines vertical mixing with horizontal dispersion.

Photochemistry & HCHO/NO₂

Solar radiation drives O₃ and HCHO formation. The HCHO/NO₂ ratio indicates VOC- vs NOx-limited ozone regimes - key for Objective 2.

Dual Objectives

Two missions, one intelligence platform

Objective 1

Surface AQI Maps

Generate daily, gap-free spatial maps of surface Air Quality Index covering all of India. Every CPCB pollutant folds into a single AQI value.

  • CNN-LSTM trained on CPCB + satellite + meteorology
  • Multi-pollutant: PM2.5, PM10, NO₂, SO₂, CO, O₃
  • X-factors: PBLH, RH, ventilation, aerosol layer height
  • Validated vs ground truth: RMSE, R², MAE
  • High-resolution daily maps at national scale
Objective 2

HCHO Hotspot Detection

Map spatio-temporal HCHO hotspots during biomass-burning seasons, identify source regions, and track fire-driven transport.

  • TROPOMI HCHO column + FIRMS fire data
  • DBSCAN clustering for hotspot detection
  • Punjab/Haryana stubble, forest fires, sugarcane burning
  • HYSPLIT back-trajectories for transport analysis
  • Fire–HCHO correlation and seasonal evolution
Built With

Technology Stack

India-first, open-source, satellite-powered.

Python
Google Earth Engine
PyTorch
CNN-LSTM
XGBoost
xarray / rasterio
GeoPandas
INSAT-3D / TROPOMI
Google Colab
Plotly / Folium
HYSPLIT
DBSCAN
System Design

System Architecture

Data Sources
  • INSAT-3D (MOSDAC)
  • Sentinel-5P / TROPOMI
  • CPCB Ground Stations
  • IMDAA / ERA5
  • MODIS/VIIRS FIRMS
Ingestion
  • Google Earth Engine
  • Regridding
  • QC & cloud-mask
  • xarray · rasterio · GDAL
Feature Store
  • Satellite columns
  • Meteorology + PBLH
  • Humidity + ventilation
  • Emission proxies
  • Temporal features
Model Layer
  • CNN-LSTM (primary)
  • Random Forest (baseline)
  • XGBoost (baseline)
  • Validation vs CPCB
Output
  • Surface AQI maps
  • HCHO hotspot maps
  • Source-region analysis
  • Interactive dashboard
Team Star Trek

The crew

AD

Arka Dash

Team Leader
YM

Yatharth Maheshwari

Team Member
NS

Nimish Shinde

Team Member
AB

Aditya Bansal

Team Member