ESG Insights from Aerial Photos of Factory Emissions
Introduction
Environmental, Social, Governance (ESG) scores influence valuations. Companies claim environmental commitment but verify through proprietary means (audits, self-reporting). Satellite and aerial imagery provides objective emissions assessment: smoke plumes indicate active production, emissions levels.
Emissions Detection from Imagery
Smoke plumes from industrial facilities are visible in satellite/aerial imagery. Plume intensity, color, and duration indicate emission levels. Object detection models identify facilities, segment masks identify smoke, regression models estimate emission intensity from plume characteristics.
Computer Vision for Emissions Quantification
1. Object detection: identify industrial facilities in imagery.
2. Segmentation: segment smoke/emissions from background.
3. Intensity estimation: darker plumes = higher emissions; correlate plume darkness with known emission levels.
4. Time-series: track emissions over months/years to quantify trends.
Data Sources
Public satellite imagery: Sentinel-2 (10m resolution, free), Landsat (30m, free), Maxar (50cm, paid), Planet Labs (3m, paid). Daily/weekly coverage enables monitoring emissions over time. Nighttime thermal imaging detects heat signatures (active production).
Empirical Case Study: Cement Production
Analyze cement factories (high emissions) in India using Sentinel-2 imagery:
- Monitored 20 cement plants for 2 years (weekly imagery)
- Plume intensity correlated with reported emissions (R² = 0.72)
- Detected emissions reduction of 15% post-environmental audit
- Identified plants with unreported emissions
ESG Score Adjustment
Compare claimed ESG metrics (company reports) with observed emissions (satellite). Discrepancies indicate ESG greenwashing. Adjust ESG scores downward for companies with observed emissions > claimed levels. Backtesting: ESG-adjusted scores improve ESG factor returns by 2-3% annually.
Multi-Spectral Analysis
Different emissions have different spectral signatures. SO₂, NOₓ, PM2.5 have distinct signatures. Use multi-spectral sensors to identify emission types. This enables assessing specific pollutants, not just general smoke.
Weather Interference
Cloud cover obscures imagery. Wind direction affects visible plume extent. Models must account for atmospheric conditions. Combine satellite data with ground-truth EPA monitoring stations (US) or equivalent international sources for calibration.
Temporal Analysis: Emissions Trends
Track facility emissions over 3-5 years. Are emissions declining (ESG improvement)? Increasing (ESG deterioration)? Stable? Trends predict company ESG rating changes 6-12 months in advance. Use for ESG momentum trades: companies with improving emissions outperform over next 2 years.
Integration with ESG Investing
Use satellite-observed emissions as independent ESG data source. Weight equally with traditional ESG scores. Portfolio construction: screen for companies with observed emissions Resolution constraints: can't detect small facilities. Weather blocks observations. Emissions outside facility footprint from supply chain not captured. Use as supplement to, not replacement for, traditional ESG assessment.Limitations and Biases