Space to Policy: Scalable Brick Kiln Detection and Automatic Compliance Monitoring with Geospatial Data

Brick Kilns

A brick kiln is a facility where bricks are produced through shaping, drying, firing, and cooling processes. The brick-making process involves four main steps: shaping, drying, firing, and exporting. Brick kilns are often located near rivers due to the availability of suitable soil.

Continuous Fire Brick Kilns

Brick kilns are classified into batch production kilns and continuous fire kilns. Continuous fire kilns are more energy-efficient as they reuse warm air from the combustion zone to dry unfired bricks. In India, approximately 75% of bricks are produced by continuous fire kilns, with FCBK and Zigzag kilns being the most common types.

Brick Kilns and Sustainability

Air Pollution

In 2017, India's total brick production was estimated at 233±15 billion bricks, consuming 990±125 PJ yr⁻¹ of energy. Brick kilns contribute an estimated 8-14% of India's air pollution. Advanced technologies, such as Zigzag kilns, reduce PM and CO emissions by approximately 60-70%.

United Nations Sustainable Development Goals

Brick production relies heavily on labor, employing approximately 15 million workers in India. This raises concerns related to forced labor, modern slavery, and human trafficking, which are addressed by UN Sustainable Development Goal (SDG) 8.7. Our research intersects with several SDGs, including addressing premature deaths from air pollution (SDG 3.4 & 3.9), mitigating CO₂ emissions (SDG 9.4), monitoring PM2.5 and PM10 emissions (SDG 11.6), reducing fossil-fuel subsidies (SDG 12.c), and tracking greenhouse gas emissions (SDG 13.2).

Government Policies

The National Clean Air Program (NCAP), launched in 2019, aims to reduce air pollution levels to comply with National Ambient Air Quality Standards (NAAQS). The Environment (Protection) Amendment Rules, 2022, mandated the conversion of all brick kilns within 10 km of non-attainment cities to cleaner technologies by 2024. Brick kilns must be located at least 800 meters from residential areas and fruit orchards, and a minimum distance of 1 km must be maintained between two brick kilns.

Object Detection Models

Object detection involves identifying objects in an image by predicting their bounding boxes and classifying them into predefined categories. Oriented Bounding Boxes (OBBs) offer a more precise representation of object areas, especially for irregularly oriented objects. In the context of brick kilns, OBBs facilitate accurate spatial extent estimation, which is essential for assessing brick production capacity and calculating emission factors for air quality modeling.

Satellite Imagery

Selecting suitable satellite imagery requires balancing cost and computational efficiency. High-resolution commercial imagery, such as Maxar (30-50 cm) and Airbus products, offers detailed views but is often too expensive for research. Brick kilns, on average, cover 150 meters in latitude and longitude and thus span approximately 15 × 15 pixels in Sentinel-2 imagery. Hence we utilized freely available moderate-resolution imagery (4.77 meters) under a research license

1. 3.1.1 Planet Labs

We utilize satellite imagery from Planet Labs in this work . Our study area spans five Indian states, covering over 520,000 km² (15.8% of India). The large 4096 × 4096 sized images are divided into overlapping 640 × 640 sized crops with a 64-pixel overlap (≈300 meters), ensuring that brick kilns do not get cut at the image edges. We selected imagery from the first quarter of 2024. We annotated high-resolution imagery and transferred the labels to Planet imagery, as detailed in the next section.

2. 3.1.2 Esri Wayback Imagery

Esri, a leading provider of GIS technology delivers high-resolution basemaps (up to 30 cm) through a Web Map Service (WMS). We utilized Esri imagery from February 8, 2024, to align with the first-quarter Planet mosaic of 2024.

OpenStreetMap

Table 1 Datasets extracted from OpenStreetMap and filters applied to extract them

Shapefile	Data	Filters
gis_osm_landuse_a_free_1.shp	Habitation	fclass="residential"
gis_osm_landuse_a_free_1.shp	Orchards	fclass="orchard"
gis_osm_landuse_a_free_1.shp	Nature reserves	fclass="nature_reserve"
gis_osm_buildings_a_free_1.shp	Schools	type="school"
gis_osm_buildings_a_free_1.shp	Hospitals	type="hospital"
gis_osm_buildings_a_free_1.shp	Religious places	type="temple" or "mosque" or "church"
gis_osm_roads_free_1.shp	National & Express highways	ref starts with "NH" or "NE"
gis_osm_roads_free_1.shp	State highways	ref starts with "SH"
gis_osm_roads_free_1.shp	District highways	ref starts with "MDR"
gis_osm_water_a_free_1.shp	Wetland	fclass="wetland"
gis_osm_waterways_free_1.shp	Rivers	fclass="river"
gis_osm_railways_free_1.shp	Railway tracks	No filter

Government Data

Population Data

Initial Data Preparation

An initial labeled dataset is required to train the object detection models. We select the regions for initial labeling with the following criteria:

• i) highly polluted;
• ii) highly populated (highly impacted by brick kilns);
• iii) high likelihood of kilns (as suggested by an air quality expert); and
• iv) non-attainment cities.

(1) Delhi Airshed, India

Delhi, India’s capital, is highly populated and consistently ranks among the world’s most polluted cities. While no brick kilns are located within Delhi due to strict regulations, the surrounding region has a dense concentration of kilns. We selected an 80×80 grid over Delhi with a 0.01-degree resolution, as defined by Guttikunda.

(2) Lucknow Airshed, Uttar Pradesh, India

Lucknow, the capital of Uttar Pradesh (India’s most populous state), is classified as a non-attainment city under NCAP. A 60×60 airshed grid with 0.01-degree resolution was chosen for annotation, following the definition by Guttikunda.

(3) West Bengal Small Airshed, West Bengal

A small airshed near the Haldi River in West Bengal was selected due to the high density of brick kilns in the area. A 639 km² region surrounding the identified kilns was annotated.

(4) Ahmedabad Buffer Region

Ahmedabad, a non-attainment city, was studied in collaboration with the Gujarat Pollution Control Board (GPCB). Following the Environment (Protection) Amendment Rules, 2022, we geo-located brick kilns within a 10 km buffer of Ahmedabad city.

Across these four regions, we manually scanned over 15,018 km² area and identified 1,621 brick kilns, completing the annotations in 125 hours.

Table 2 Statistics of initial dataset

Airshed	Area (km2)	CFCBK	FCBK	Zigzag	Total Kilns	Annotation time (hours)
Delhi Airshed	6937	2	35	746	783	58
Lucknow Airshed	3962	26	225	241	492	33
West Bengal Small	639	0	89	110	199	5
Ahmedabad 10 km buffer	3480	38	108	1	147	29
Total	15018	66	457	1098	1621	125

Annotation Process

The annotation process involves labeling brick kilns with oriented bounding boxes (OBBs) and assigning a category to represent the kiln's technology. A custom labeling interface using Leafmap was developed, utilizing the Esri Wayback Satellite Imagery WMS layer. The region of interest was divided into a 1 km² grid, and each grid cell was manually inspected. OBBs were drawn around the identified kilns, and the labeled data were overlaid onto cropped, geo-referenced Planet images. This process ensured accurate and undistorted annotations during reprojection from Esri to Planet imagery.

Results

yolo11m-obb model is yielding the best ‘Weighted mAP’ among all models, and thus, we use it for brick kiln detection.

Table 3 Performance of various models on the initial dataset.

Model	CFCBK	FCBK	Zigzag	Weighted mAP
yolov8l-obb	0.61	0.58	0.83	0.62
yolov8x-obb	0.63	0.55	0.82	0.63
yolov11x-obb	0.66	0.57	0.80	0.66
yolov11l-obb	0.68	0.51	0.76	0.66
yolov8m-obb	0.68	0.54	0.79	0.66
yolov11m-obb	0.73	0.61	0.83	0.71