Clearing the air: Do India’s crop burning bans really work?

In December 2015, the National Green Tribunal instituted a ban on crop residue burning across five states. Utilising satellite data on crop fires and administrative data on fines levied, this article examines the efficacy of the ban. It finds that the ban did have a sizeable downward impact on fire counts – but with a lag of a year, and effects lasting for at most two years.

Across the world, and especially in developing countries, farmers burn inordinate amounts of crop residue, leading to nearly 3.5% of all greenhouse gas emissions (Ritchie 2020). The resulting pollution has been known to cause higher mortality, child stunting, and incidence of disease, and lower economic growth (Lin and Begho 2022). In India, crop residue burning (hereafter, CRB) is among the leading sources of air pollution, particularly in the northern states, and is estimated to cause one million deaths annually (Pandey et al. 2021).

Nonetheless, studies that analyse policies designed to mitigate CRB are limited. Previous studies have found notable success from policies such as unconditional cash transfers to farmers in India (Jack et al. 2025) and building of biomass power plants (Cao and Ma 2023, Nian 2023). The bans on CRB in China are shown to have limited effects (Cao and Ma 2023, Nian 2023). In our new research (Khanna, Mahajan and Sudarshan 2024), we study the efficacy of India’s largest ban on CRB using satellite data on crop fires and administrative data on fines levied for CRB after the ban. The ban was instituted by the National Green Tribunal (NGT) on 10 December 2015 across five states: Punjab, Haryana, Uttar Pradesh, Delhi, and Rajasthan.

Crop burning in India

We use daily data on crop fires from 2011 to 2020 for 12 states – every state the ban was instituted in (the ‘ban’ states) and all those that neighboured them¹ – from NASA’s MODIS Active Fires Product. The product furnishes the centroid of every 500 pixel where a biomass fire has been detected², its date of detection and a ‘confidence’ value or likelihood for the detected fire actually being a fire. We weight each fire by its confidence value and aggregate it to the 10 km x 10 km grid level as a measure of our main outcome variable.

Additionally, to explore the mechanism through which the ban may have operated, we compiled data for annual ‘environmental compensation’ (fines) collected between 2015 to 2019 by states that instituted the ban. We collated this information from state governments and district magistrate offices across the ban states.

To quantify the impact of the ban, we use daily data from 2011-2019 and employ a ‘difference-in-differences’³ strategy using 11 December 2015 as the beginning of the ‘treatment’ period and all neighbouring states to our ‘ban’ states as ‘control’ (not subjected to intervention regions. We observe that prior to the institution of the ban, the ‘treated’ (subjected to intervention) states and control states had similar trends in fires. Hence, any change in trends that we observe in treated states following the ban may be attributed to the intervention (Figure 1).

Figure 1. Impact of ban on fires: Event study estimates (Grid level)

Notes: (i) The plots show the event study estimates for each year relative to the year before the ban (2014). (ii) The dependent or outcome variable is the confidence-weighted sum of fires in a given grid – observed daily between 1 January 2011 and 31 December 2020 in Panel (a). (iii) This specification includes grid and date fixed effects and errors clustered by state in which the observed grid is placed in. (iv) The dotted line represents 2015, that is, year zero for the implementation for the ban and 95% confidence intervals (indicating that if we repeated the study many times and constructed a 95 % CI each time, about 95 % of those intervals would capture the true effect.) are plotted.

Short-lived success

In contrast to conventional wisdom, we find that the ban did, in fact, have a sizeable downward impact on fire counts. Specifically, we find that the years after the ban saw a reduction in fires equivalent to 30% of the pre-ban period average. Event study estimates show that almost all of this reduction took place in 2017 and 2018, that is, with a lag of a year, and that the effect lasted two years at most and waned immediately after that. Additionally, decomposing the effects by state yields that (i) the strongest (and statistically significant) reductions came from Punjab and Haryana and (ii) Uttar Pradesh saw no statistically significant decline. Specifically, relative to control states, Punjab and Haryana saw a fall in fire counts equivalent to approximately 9.6% and 27% of their pre-period grid averages respectively, after the ban.

Mechanisms

What explains the above effects? Given the history of ban implementations in the past, farmers, at least at first, did not expect any fines to be levied leading to no effect in the first year after the ban. When some were in fact penalised in 2016, it is plausible they reacted cautiously during the next cropping cycle. This is because of the uncertainty about the degree and extent to which fines would be imposed.

Unsurprisingly, this is consistent with the magnitude of and trends in fines imposed. Our data show that fines were levied only starting April 2016-March 2017 (Figure 2), with Punjab and Haryana levying the maximum number of fines. Additionally, our back-of-the-envelope calculations suggest that the imposition did not scale well with time. We assume the proportion of farmers engaging in CRB in a state to be equal to the lower bound of the same across survey estimates⁴, and combine this with data on agricultural landholdings from the Agricultural Census of 2015. We find the ‘per offender’ incidence of fines to be extremely low and scaling slowly with time. Emblematically, the highest average incidence of fines for ‘burning’ landholdings during the years we have records for (2015-2019) was Rs. 37 in Haryana during 2018-2019, up from Rs. 32 in the previous year – starkly lower than the lowest prescribed fine under the ban of Rs. 2,500 (Figure 2).

Figure 2. Environmental compensation collected by states with ban on residue crop burning

Notes: (i) Panel A shows the total value of fines levied (in Rs. lakhs). (ii) Panel B shows the estimates for fine per burnt landholding (Rs.), obtained by dividing the total value of fines (obtained by authors from various state governments) by estimated count of landholdings that engage in CRB. These are estimated based on data from the Agricultural Census of 2015-16 and an estimate of percentage landholdings burnt in each state in the literature.

Furthermore, to check whether the attention to air quality in Delhi drives these results, we test whether the effect of the ban differs by distance to Delhi. We find that this is indeed, true and the effect of the ban dissipates beyond 750 km from Delhi. This result suggests that heightened public scrutiny and media attention near Delhi likely shaped enforcement actions in the adjoining areas.

Other checks

The use of satellite data often comes with concerns of measurement error – from sources such as the classification algorithm for the fires. In order to ensure that our results are robust to the same, we aggregate our fires data to the district-month level and thereafter conduct the difference-in-differences analysis. Alternatively, we also utilise data for a longer time period, that is, from 2002 instead of 2011. Our results remain qualitatively similar: we continue to find a similar quantitative effect on fires and similar decomposition in terms of the states the effect arises from.

We also test whether our results are robust to (a) including only neighbouring districts between our treated and control states and (b) controlling trends in our outcome variable arising from changes over time in each agro-ecological zone. Once again, we find that our results hold with the concentration of the decline in Punjab.

Policy implications

Our findings suggest that even weakly enforced environmental bans can generate temporary compliance if there is sufficient uncertainty about enforcement. This implies that early-stage deterrence may hinge more on perceived risk than on actual penalties. However, for such bans to achieve sustained success, governments must demonstrate consistent commitment to enforcement through either scaling up monitoring or levying meaningful fines. Without credible follow-through, behavioural change is likely to be short-lived, as seen in the return to pre-ban burning levels after 2018. Importantly, policy design should consider complementarities: combining regulatory bans with targeted subsidies for residue management technologies or enhancing public awareness, particularly during high-salience pollution episodes, could bolster compliance.

Notes:

In particular, these states included Uttarakhand, Himachal, Madhya Pradesh, Gujarat, Bihar, Jharkhand and Chhattisgarh.
Arrived at by a recognition algorithm that takes the relevant satellite’s images (underlying pixels) as inputs.
Difference in differences is an empirical strategy which estimates the causal effect of a ‘treatment’ (the ban in our case) by comparing differences in how treatment and control group outcomes change over time. It is a valid strategy under the assumption that both groups would have trended similarly in the outcome over time in the absence of the treatment.
The average rates across states noted by these studies vary – Punjab (50-90%), Haryana (10-50%), Uttar Pradesh (approximately 10%).