New Delhi, Nov. 11 -- Farm fires in Punjab and Haryana hit their highest count of the season on Diwali even as new estimates put the scale of the problem vastly higher than what's being counted. Moreover, while the north-westerly wind that carries their smoke to Delhi slackened in the festival's aftermath, easing the impact on the Capital, it is forecast to return on Wednesday and Thursday, an HT analysis of satellite and weather data shows. Till October 28, INSAT-3DS, a weather satellite run by the Indian Space Research Organisation (ISRO), detected almost no farm fire in the two states. From Oct 29 the count rose every day, from 18 to 447 on November 8, Diwali, and the daily average in the week to November 9 was 12.5 times that of the week to November 1. The count fell sharply on November 9; the reason is not known. (See Chart 1) INSAT-3DS stays over the same spot and scans the region every half hour. Each fire it detects is a patch of ground several kilometres across that is hot enough to be burning. A different pattern emerges if the method of detection is changed. Hiren Jethva, an atmospheric scientist at NASA's Goddard Space Flight Center, wrote on X on October 3 that a statistical model, which predicts the number of fires from how green the crop was as measured by NASA's MODIS satellite instrument, "ranks 2026 Punjab-Haryana post-monsoon fire season as one of the intense burning periods". The model expects about 14,000 fires, as MODIS would count them. On November 9, Jethva posted on X that the current farm fire figures were likely an undercount, with an empirical calculation using a combination of satellites and considering data till 4 pm: both Punjab and Haryana had likely logged closer to 34,000 farm fires till November 9. Referring to data from the NOAA-20 VIIRS satellite (375m resolution) for both Punjab and Haryana from October 1-November 9, he said 4,089 fires were logged around 1:30pm. Similarly, data from the Aquad MODIS satellite (1km resolution) showed 4,196 fires had been logged around 4 pm local time. "After empirical adjustment to VIIRS 375 m fire-count baseline, the total reaches to around 34,000 fires -- a conservative estimate," he said on Monday evening. HT cannot check that ranking against the fire counts of earlier years. NASA's satellites have missed more of the stubble burning season after season as the activity moved to later in the day, so a comparison with past counts would mostly measure how much they missed. ISRO's counts are not comparable from one season to the next. The closest like-for-like measure is the area burned, which the research group iFOREST maps from European Sentinel-2 satellite images; burnt fields show up in them whatever time of day they were set alight. A CAQM official, when contacted, said the fact that fields are being set on fire later in the day is well-known now. "The Indian Agricultural Research Institute (IARI) has also begun utilising the AQUA satellite to capture instances later in the day, but counts beyond 4 pm are still being missed," he said, adding they are putting to use drones and even low-cost sensors to gather more data on-ground. Smoke from the fields travels to Delhi on a north-westerly wind 1 km to 3 km above the ground. The India Meteorological Department's weather balloon, launched from Safdarjung in Delhi, measured that wind at 29 km/h on November 7. On Diwali afternoon it had eased to 15 km/h, and by the early hours of Tuesday it was down to 6 km/h and blowing from the north-north-east, no longer from the direction of the fields. On Diwali afternoon, the mixing layer, the band of air above the ground in which pollution spreads and thins, was 364 metres deep, against a usual depth of about 1,300 metres in early November, so pollution had far less room to thin out. (See Chart 2) Using winds from two weather models, the US Global Forecast System (GFS) and the European Centre for Medium-Range Weather Forecasts (ECMWF), HT traced where the air reaching Delhi at 2pm each day had been over the previous three days. On Diwali, that air had come through the fire belt, the farmland of Punjab and northern Haryana where most of the fires burn. On November 9 and November 10 it came from the east and north-east, passing no INSAT-3DS detections and no more than nine NASA-detected fires. The fires typically burn around 260 km away from Delhi. The air reaching the city on Diwali afternoon had passed over most of the fires roughly 30 to 35 hours earlier, the trace shows. At the speeds that have been forecast for Wednesday and Thursday, the air would cover that distance in about 10 hours. The models forecast the north-westerly will return at 27 km/h on Wednesday and 23 km/h on Thursday. Both trace the air reaching Delhi on Thursday afternoon back through the fire belt. On Wednesday, the ECMWF traces it through the belt; the GFS, at about 800 metres above the ground, has it skirting the belt. Near the ground, the models forecast light winds, and no rain through November16. Rain would wash some of the smoke out of the air. (See Chart 2) NASA's fire count, the one most often quoted, shows a far smaller rise. It too hit its season high on Diwali, at 385, but its daily average in the week to November 9 was only 2.4 times that of the week to November 1. HT compared only how fast the two counts grew, because NASA counts individual fires and INSAT-3DS counts larger hot patches of ground. (See Chart 1) Most of the burning now happens after NASA's satellites have passed. They cross Punjab in daylight once, in the early afternoon; this season, 98% of the fires INSAT-3DS detected came between 4pm and 6pm. Jethva wrote that actual fire counts for 2022 to 2025 were "multifold lower" than the model expected because peak burning moved to the late afternoon, which NASA's fire-counting instruments, MODIS and VIIRS, do not see. (See Chart 3) Meteosat, a European weather satellite, also stays over one spot; it scans every 15 minutes and measures fire heat, the energy that fires give off. In 2021 it saw 53% of the season's fire heat in Punjab and Haryana after 3pm; by 2025 the share was 98%. Over the same years NASA's season count fell to 6% of its 2021 level, while the area burned, by iFOREST's measure, fell only to 81%. (See Chart 4)...