CHETNA-Brick Sector: Estimating GHG and Pollutant Emissions from Brick Kilns in India Using Sentinel-2 Imagery and Deep Learning

India, the world’s second-largest brick producer, operates over 100,000 kilns. These kilns emit 170 kt of PM2.5 (15% of the national total) and 120 Mt of CO₂ (6% of the national total) annually, along with substantial SO_x and NO_x emissions. Transitioning from traditional Fixed Chimney Bull’s Trench Kilns (FCBTKs) to cleaner technologies, such as Zigzag Kilns (ZZKs), has the potential to reduce coal consumption by 20% and particulate matter emissions by 70%. However, comprehensive datasets for kiln locations across India remain scarce. This study contributes to the CHETNA project (City-wise High-resolution carbon Emissions Tracking and Nationwide Analysis), which leverages artificial intelligence and advanced datasets to deliver high-resolution, near real-time daily CO₂ and air pollutant emissions data for over 100 Indian cities. 

To address this gap, we leveraged Sentinel-2 imagery, with a spatial resolution of 10–20 m, to develop a cost-effective and scalable approach. Most existing studies focus on specific geographic areas, such as northern India, and rely on expensive, high-resolution satellite imagery that is often not readily available, limiting their broader applicability. In contrast, our study represents the first nationwide mapping of brick kilns in India, using openly accessible satellite data and advanced machine learning models.

Using a curated dataset of 9,600 geo-tagged labels covering 18,000 km², we developed a method combining Sentinel-2 imagery with convolutional neural networks (CNN) to detect brick kilns and classify their operational technologies (e.g., FCBTK, Zigzag). Labels were annotated using Google Earth layers on QGIS and validated based on distinct visual features, such as oval or rectangular ochre-colored shapes. The model leverages RGB bands to detect active kilns, while the addition of NIR, SWIR, and NDVI metrics enhances its ability to identify abandoned kilns, often concealed by vegetation, and reduces false positives.

The model achieved a precision of 0.90, a recall of 0.89, and an accuracy of 0.91 on the test set. Detected kiln centroids were highly accurate, with precise GPS coordinates matching their actual locations. Nationwide, the model identified 44,000 brick kilns in India for 2022. We benchmarked multiple models to optimize false positive reduction and improve technology classification. Building on these results, we applied the model to neighboring countries in the Indo-Gangetic Plain (IGP), spanning Pakistan, Bangladesh, and parts of Nepal, which also contribute significantly to the brick kiln industry, identifying approximately 20,000 kilns in 2022.

Beyond location mapping, we are generating annual gridded emission maps for CO₂ and pollutants such as PM2.5, black carbon, and NO_x. These maps provide time-series insights into emission trends, reduce uncertainties in carbon and pollutant emissions, quantify reductions achieved through cleaner technologies, and identify regional hotspots. By focusing on underregulated, high-emission sectors like brick kilns, this study offers critical insights for targeted mitigation strategies and sustainable urban planning. It equips policymakers with tools to evaluate regulations and demonstrates the feasibility of using Sentinel-2 imagery for cost-efficient, large-scale monitoring.