Executive summary

Globally, crop residue burning (CRB) is a major environmental and public health concern. It releases greenhouse gases and particulate matter and also degrades soil quality and health (Vadrevu et al. 2019; Tawfeeq et al. 2025). Despite its harmful impacts, it remains a widely adopted post-harvest field-clearing practice in both developed and developing countries, with 450 million tonnes of crop residue burnt every year (CCAC 2023). In India, CRB is a national-scale problem and a major source of air pollution in northern Indian cities. For instance, it contributes as much as 35–40 per cent to Delhi’s particulate matter 2.5 (PM_2.5) on peak burning days in late October and early November (Govardhan et al. 2023).

India has committed substantial public resources to the effort – roughly INR 4,200 crore during 2018–26 – through schemes such as the Central Sector Scheme for Promotion of Agricultural Mechanisation for In-Situ Management of Crop Residue (MoAFW 2026) and the National Policy on Biofuels. Evaluating whether such interventions are working and determining where to direct efforts next require accurate monitoring of the situation on the ground. Such monitoring underpins enforcement, targeted action, and course correction.

There are two satellite-based methods for monitoring these fires, but both have structural limitations. The first method is based on active fire counts detected by polar-orbiting satellites. Currently, the Commission for Air Quality Management in National Capital Region and Adjoining Areas (CAQM) follows a fire count-based protocol developed by the Indian Space Research Organisation (ISRO) to monitor CRB. These satellites have a coarser spatial resolution of 375 m and 1 km and revisit the same region once or twice a day, potentially missing small, short-lived fires that occur outside their narrow overpass windows (Schroeder et al. 2014). The underestimation has increased as peak burning has shifted to the evening, outside the satellite overpass windows (Singh et al. 2025a, 2025b).

The second monitoring method estimates burnt area using remote sensing (Giglio et al. 2006). In Indian conditions, the overall accuracy of the burnt area method using Sentinel-2 imagery is about 75 per cent, with a true positive rate (TPR) of 59 per cent (Ignatious et al. 2025). However, the Sentinel-2 constellation generally revisits the same area only once every five days, missing the fields burnt and subsequently managed (ploughing / tilling) in the interim (Ignatious et al. 2025). Moreover, cloud cover and other atmospheric conditions, along with smoke from fires, could obscure detections.

This study asks a practical question: can this gap be narrowed by combining the imagery India already relies on with complementary commercial data? We propose a method that combines openly available Sentinel-2 imagery with commercial multispectral imagery from Planet Labs to improve CRB estimates.

Data and methodology

We used openly available Sentinel-2 (Level-2A) and commercially available PlanetScope (Level-3B, Surface Reflectance) data. As mentioned above, the Sentinel-2 constellation generally revisits the same region every five days. However, certain strips on Earth’s surface are revisited every other day or every third day. A portion of eastern Punjab spanning 9,700 sq km lies in one such strip. The region of interest (ROI) of this study lies within this region. We obtained Sentinel-2 images (using Google Earth Engine) between 23 October and 7 November 2024 for this ROI. We then procured PlanetScope images for the region to fill the observation gap between Sentinel-2 passes. Our objective was to estimate the extent of daily burning over two to three consecutive days during the peak CRB period.

We then followed the threshold-based approach discussed in Ignatious et al. (2025) to compute burnt area on Sentinel-2 images. Extending the approach for PlanetScope imagery, we extracted pixels of burnt and unburnt fields and analysed spectral indices – Simple Ratio (SR), Normalised Difference Vegetation Index (NDVI), Char Index (CI), Burn Area Index (BAI) and Bare Soil Index (BSoI). BAI and CI could potentially differentiate between the two categories, and we defined four test thresholds for them. We tested these indices and thresholds across 14 randomly selected villages and computed performance metrics such as accuracy, TPR, false positive rate (FPR), and false negative rate (FNR), among others, to identify the best-performing threshold.

Key findings

• The BAI has an overall accuracy of 82 per cent and correctly classifies 54 per cent of burnt pixels in PlanetScope images. Setting BAI between the 1st and 95th percentiles outperformed the other test thresholds. It misclassifies only 0.23 per cent of actual unburnt pixels as burnt. Table ES1 shows the performance metrics of the thresholds tested in this study.
• The algorithm developed using PlanetScope images in tandem with Sentinel-2 increased the number of burnt land parcels detected by two to five times compared with Sentinel-2 alone. Figure ES1 shows the number of fields detected using the algorithm across five randomly selected villages. This implies that combining commercial and openly available satellite imagery can improve CRB detection.

Figure ES1. The inclusion of PlanetScope images in tandem with Sentinel-2 images improved the detection of burnt land parcels by two to five times compared to using Sentinel-2 alone

Source: Authors’ analysis
Abbreviations: S2: scenario 1; S2PS: scenario 2.
Note: Axis uses a square-root scale to keep small and large counts both readable

Recommendations

Our analysis shows that combining commercial and openly available satellite imagery improves CRB detection. Therefore, to improve CRB monitoring in India, we recommend the following.

• The CAQM should revise the existing CRB monitoring protocol to include burnt area estimates alongside fire counts, adopting a hybrid approach that combines openly available Sentinel-2 imagery with commercial imagery to better detect CRB.
• India should leverage its existing space ecosystem for continuous CRB monitoring. In the long run, India should have its own constellation for environmental monitoring and reduce reliance on international satellites. Public–private partnerships, such as the recent one between the Indian National Space Promotion and Authorization Centre (IN-SPACe) and Pixxel (The Hindu 2025), can enable Indian authorities to use indigenous data to improve CRB monitoring.