Introduction

Forest fire, grassland fire, crop-residue burning (CRB) are prevalent worldwide since long1,2,3. Fires in Savannah, forests, shifting cultivation, wood fuel combustion and CRB contributed about 50, 24, 10, 11, and 5%, respectively, in biomass fires occurred in tropic during late 1970’s itself4. Cunningham et al.5 reports that frequency of extreme events increased by about 2.2-fold during 2003–2023, in which the last 7 years included the 6 most extreme ones. In spite of the fact that the total burnt area on earth may be declining, their study reports that fire behaviour is on a worsening trend in several regions, particularly the boreal and temperate conifer biomes, with serious implications on carbon storage and human exposure to wildfire disasters. In the recent 2023–2024 fire season, an area of 3.9 × 106 sq. km. was under fire and fire carbon (C) emission was to the tune of 2.4 Pg C. Noteworthy wildfire episodes were recorded in Canada, Greece, Western Amazonia, Hawaii and Chile6. UNEP7 has expressed concerns on escalating forest and peatland fires that are threatening the extensive ecosystem services received from these regions. Increasing trend in the proportion of areas under high-intensity fires is reported from boreal forests of Siberia, including the Arctic zone during the past decade8. Further, Australia is frequently devastated by wildfires, with the most recent episodes occurring during 2019–2020, costing in excess $4.4 billion in losses9.The evidences on the integral link of human health to wildfires is continuously emerging and has been reviewed recently10.Approximately 730 Tg biomass was estimated to have combusted annually in Asia during early parts of the decade of 2000 in which India’s CRB had a share of about 18%11. In recent times, CRB alone accounted for about 25% of total biomass burnt worldwide12,13. The short interval of about 2‒3 weeks in between rice harvesting and sowing of wheat have instigated farmers to burn paddy residues during Oct–Nov in India to clear fields for sowing14. Indian Agricultural Research Institute reports combustion of about 14 million tons (Mt) (about 63.6%) out of about 22 Mt paddy stubble generated in India each year15. But, in spite of a longer interval of about 46–48 days in between wheat harvesting and sowing of rice during April–May16, wheat stubbles are also found being put on fire in recent times17. Two important agricultural states of India, Haryana and Punjab, contributed about 48% to CRB during the first decade of twenty-first century18. Apart from CRB, severe incidences of fire are also reported from dry deciduous forests in India while other types of forests were found to be comparatively less prone to heavy fires19. More than 36% of India’s forest cover is prone to frequent forest fires while about 4% and 6% of forest cover are extremely and highly fire prone, respectively20. On the other hand, forest fires are regularly spotted on Indian landmass, mostly during summer. According to Forest Survey of India (FSI), 52,785 forest fires were detected by Moderate Resolution Imaging Spectroradiometer (MODIS) while Visible Infrared Imaging Radiometer Suite (VIIRS) was able to detect 345,989 forest fires during November 2020 to June 202121. The escalation in forest and biomass fires in specific parts of India has been flagged earlier but fire data had largely been extracted from Along Track Scanning Radiometer (ATSR) or MODIS22,23,24,25 that did not have comparable ability to acquire data on small fires like the Suomi-VIIRS has26.

Documented incidences of human influence in causing deliberate and sometimes inadvertent wildfires are concerning27. In mountain regions of North–eastern India, slash and burn agriculture is practiced for long to clear forest land for agriculture, leading to arson in forest patches28. Banerjee29 found that wind speed, temperature, closeness to roads and tree cover are the crucial limiting factors in wildfires incidences in Sikkim Himalaya of India. Other factors like altitude, forests composition, precipitation and population density of trees also influence wildfires30. Fire Radiative Power (FRP), a measure of instant radiative energy emanating from fires31, is depicted as pixel-integrated FRP in Mega Watts (MW). It is an important variable to characterize the extent of fire32.

Biomass fires, remotely detected by sensors mounted on satellites, is an extremely useful method to study wildfires3,33,34. The data on thermal hotspots and fire activity reported by VIIRS and MODIS are known to conform in hotspot detection but a better spatial resolution of 375 m in VIIRS ensures better response for small fires35. The active fire data (375 m) reported by VIIRS that embarked on its mission on the polar orbiting Suomi National Polar-orbiting Partnership (Suomi-NPP) in 2012, complement the 1 km MODIS data acquired from Terra and Aqua satellites.

Wildfires and CRB lead to emissions of PM10, PM2.5, black carbon (BC) and several gases35,36,37 Episodic biomass burning are also linked to substantial regional air pollution38. Further, biomass fires are known to alter landscapes39, cause hydrologic and geomorphic changes40, perturb local and regional meteorology41, change surface albedo in post-fire periods42,reduce organic soil carbon and nitrogen43, destruct forest biomass41,44 and flora and fauna45, enforce long-term climate change46, making biomass and wildfires extremely hostile to environment. Wildfires have strong influence on forest succession and play an active role in the replacement of existing tree species by new ones with different adaptation behaviour47. Microbial succession in soil in the aftermath of forest fires is also widely known48,49. Wildfires are also damaging to the delivery of ecosystem services like timber supply, animal hunting, honey, mushroom, resin and other forest product supply by inflicting losses in forest canopy, forest products, habitat and biodiversity50,51,52. Fires also limit human access to forests, affecting livelihoods. Forests repeatedly destroyed by fires reportedly had lesser woody surface fuels and tree seedling densities as compared to forests experiencing only a sole and recent wildfire. Also, repeatedly burnt areas may recover differently than once-burnt areas, leading to alternative ecosystem structure53. It is therefore crucial to monitor historical and also, near real time (NRT) fire incidences on regular basis to counter recurrent fires and protect environment and natural resources.

This study was envisaged to analyze and report trends in fire counts and FRPs during 2012–2023, right from the beginning of data acquisition by Suomi-VIIRS up to 2023, considering the presence of a highly ‘fire sensitive’ I-Band with 375 m spatial resolution (at nadir) that can detect ‘small’ active fires with < 1 MW FRP in VIIRS’s34. This study gives insight on (i) temporal variability and trends of biomass fire spots on Indian landmass(ii) seasonality in the fire incidents (iii) fires hot spots in 20 km × 20 km grids vis a vis the populated urban centres. The spatial spread and clustering of biomass fires in ‘hot spots’ over Indian landmass vis a vis seasons and populated cities, will help understand potential impacts of emanating smoke and heat on urban centres.

Methods

Study area

Indian landmass (in between 8° 4′ and 37° 6′  north latitudes and 68° 7′ and 97° 25′  east longitudes), consisting of 28 states and 8 Union Territories54, covering an area of about 32,87,263 sq. km55 ,was chosen as the study area. Indian landmass measures about 3214 km (north–south) and about 2933 km (east–west). India had 275 cities with > 1 lakh (> 0.1 million) population as per last population census conducted in 2011.

Fire and FRP data retrieval

Active fire spots captured by VIIRS on NASA/NOAA Suomi-National Polar-orbiting Partnership (Suomi NPP) satellite and processed by NASA’s Fire Information Resource Management System (FIRMS) were extracted and analyzed for every month and year during 2012 to 2023. All the active fire spot data points recorded by VIIRS from the very first day of transmission from the S-NPP satellite in 2012 and archived in NASA-FIRMS Near Real-Time (NRT) active fire data product (375 m spatial resolution), were collected (https://firms.modaps.eosdis.nasa.gov). Subsequently, information on select parameters viz. geographical coordinates of fire spots, data acquisition date and time, levels of confidence in fire detection (low, nominal, high) and FRP of each fire was extracted from the collected database for detailed analysis and interpretation. In VIIRS, the acquired fire data is screened by a confidence algorithm in which detection confidence levels are set to ‘low’ (l), ‘nominal’ (n) and ‘high’ (h) for designating the quality of individual fire pixels that helps exclude false positive fire occurrences (https://data-nifc.opendata.arcgis.com/datasets/esri2::satellite-viirs-thermal-hotspots-and-fire-activity/about). To ensure incorporation of fewer false alarms in analysis, use of only nominal- and high-confidence fire pixels are prescribed, treating the low-confidence fire pixels as ‘no fire’. On the other hand, maximum fire detectibility is ensured when all three classes of fire pixels are considered56. In this study, all fires including l, n and h confidence levels were analyzed and reported to ensure maximum fire detectability. A certain fire may not get detected by a satellite, if (i) the fire started and ended outside the time frame of satellite overpass (ii) the fire is masked by cloud cover or tree canopy or heavy smoke (iii) when instruments develop glitches (iv) when the fire is too small or too cool to be detected. The onboard bow-tie deletion algorithm in VIIRS helps minimize detection of duplicate fire spots between adjacent scans at off-nadir position, resulting in fewer repeat fire detections than MODIS57. Apart from these general limitations of satellite remote sensing, VIIRS has the following specific limitations in fire detection: (i) Early afternoon orbit of VIIRS leads to large temporal separation between same-day data acquisitions58 (ii) During daytime, potential false alarms are witnessed due to reflected solar radiation from solar panels, hot and bright surfaces, sun glint from small lakes/ponds and clouds. In night, false alarms include hot smoke plumes and reflected solar energy from cloud tops near the terminator59. However, Data verification and validation for VIIRS performed at selected sites across the globe, including specific field campaigns in small-to-medium size (< 500 ha) fires60, showed a good agreement with VIIRS daytime and night-time fire data34, showcasing the reliability of VIIRS as an efficient wildfire sensor.

The MODIS (Aqua and Terra combined) fire spots were also extracted from FIRMS product parallelly with VIIRS, for the purpose of presenting an inter-comparison of fire spots retrieved by the two instruments.

Data analysis

The fire counts recorded by VIIRS over 2012‒2023 were extracted and the numbers of active fires in the most active months were analyzed. The Indian ‘fire-year’ cycle was constructed from the monthly variation in fire incidences in every year, following the concept proposed by Boschetti and Roy61 that inculcates the logic of calculating a 12 month ‘fire-year’ cycle from the month of significant escalation in number of wildfires. The concept of fire-year was invoked in this study after preliminary analysis of historical fire data (2012–2023) over India. A ‘fire year’ incorporates a 12 month period, starting from the month of remarkable escalation in fire activity up to the completion of 12 months which may or may not coincide with a calendar year. Therefore, based on wildfire activities, a fire year may be different in different parts of the world. The authors proposed the concept specifically for studying inter-annual variability of fire activity and found that a March–February ‘Fire Year’ was suitable for studying global fire activity. For the sake of a wider understanding, the ‘Fire Year’ may be likened to a ‘Fire Period’ or a ‘Fire Season’. The respective FRPs (MW) of fires were also extracted and analyzed. Further, inter-annual trends of yearly total fire count, median FRPs and annual sum FRPs of active fires were analyzed and assessed for inter-relationships by Ordinary Least Square (OLS) Method and coefficient of determination (R2) was tested for statistical significance. The year-wise trends of yearly total fire counts, yearly median FRPS and total yearly sum of FRPs over 2012 to 2023 were also analyzed by Sen’s nonparametric slope estimation methods. The COVID years of 2020 and 2021 were ignored for studying the inter-annual trends, considering potential impacts of perturbation in anthropogenic movements and activities during these two years on fires. The true slope of trends were estimated by Sen’s nonparametric method in which the slope is estimated as the median of all pair-wise slopes between each pair of points in the data set62,63. This method has advantages like allowing missing values and non-inclusion of data of any particular distribution. Also, Sen’s method is known to be not remarkably impacted by single data errors or outliers64. Frequency analysis of fire counts in select FRP ranges over the entire study period was conducted to understand the distribution pattern of fire counts under various FRP classes. Cluster analysis by Ward’s Method was performed on the year-wise datasets (n = 12 months under each year) on total fire counts and sum FRPs to identify statistically similar or dissimilar years in terms of total fire counts and sum FRPs, respectively. Ward’s Method, also known as Ward’s minimum variance method, is one of the agglomerative hierarchial approaches used for cluster analysis and is known to be the most suitable for quantitative variables. This approach analyzes the variance of clusters and minimizes within-cluster variance65.

Spatial distribution of active fire spots in months of highest occurrence in summer (March or April, as found through fire data analysis) and winter (November, as found through fire data analysis) were mapped on Indian landmass by generating shape files from fire coordinates extracted from FIRMS fire product and integrating the layers onto the shape file of India having first and second levels of administrative divisions (country and state borders). Fire density maps were prepared for the months of highest occurrences in summer and winter i.e. March and November of 2022 and March or April and November of 2023 in 20 km × 20 km grids, to delineate gridded fire hot spots over Indian landmass. The density maps were prepared on a GIS platform by dividing the entire study area into 20 km × 20 km grids followed by geometric processing of number of fire spots appearing in each grid and selective colour marking of grids by creating fire-spot number bins. The shape file of Indian cities having more than 1 lakh population (> 0.1 million) was prepared by extracting the city coordinates reported in India’s last population census database conducted in 2011 and overlaid on fire density maps. These maps facilitated visual assessment of proximity of these busy urban centres to the fire hot spots of various intensities (i.e. with different classes of fire counts), developed in 20 × 20 sq. km. grids. The major cities and towns within the seasonal hot spots were labeled for easy identification.

Results and discussions

Suomi-VIIRS detected far higher number of fires than what Terra and Aqua MODIS together did (Table 1). The numbers of fires detected with nominal and high confidence levels by VIIRS were in the range of 82.1–84.2% of all fires in the calendar years of 2012 to 2023 while in MODIS, the fires detected with > 50% confidence level were 79.1–81% of total fire counts. The temporal variation in monthly total fire counts during 2012‒2023 is presented in Fig. 1, showing wide temporal variation in monthly total fire counts over the study period. The summer fire count started to increase remarkably from March every year, peaking either in March itself (2012, 2013, 2015, 2016, 2018, 2021, 2022) or in April (2014, 2017, 2019, 2020, 2023), subsequently declining to reach low numbers by July with ensuing monsoon. At the onset of winter, fire count started to increase again from October onwards, peaking in November and declining thereafter. The fire counts at winter peaks were much lower than summer peaks always, underlining the dominance of summer as the most intense season of fires. Yearly total fire count in 2019 and 2020 were found to have declined to near 2014 and 2015 numbers, possibly reflecting the effect of COVID related slowdown on anthropogenic activities linked to fires.

Table 1 Inter-comparison of yearly (calendar-year) fire counts retrieved by SUOMI-VIIRS and MODIS (Aqua and Terra combined) over Indian landmass.
Full size table
Fig. 1
figure 1

Variation in monthly total fire count over Indian landmass during 2012 to 2023.

Full size image

As per the observed historical fire activity in India during 2012–2023 (Fig. 1), noticeable increase in fires was observed from February onwards consistently every year, corroborating the findings of Global Forest Watch for India as a whole66 and Indian Institute of Remote Sensing (IIRS) for one of the most wildfire-affected states in India, Uttarakhand. Global Forest Watch reported that peak fire season typically began in mid-February in 2024 based on wildfire data extracted from VIIRS. As per IIRS, who used wildfire data extracted from one of the sensors used by NASA-FIRMS to come to a conclusion, forest fire season in Uttarakhand starts from mid-February, characterized by onset of dry weather and increasing presence of drying vegetation as fuel for fire due to rising temperature which persists up to mid-June67. Considering the remarkable escalation in number of fires from March onwards every year in India, as found in this study (Fig. 1), the 12-month cycle of March–February was chosen as the Indian ‘Fire Year’ or ‘Fire Season’ for presenting inter-annual variability in fire activity.

The yearly (fire-year) number of fires ranged from 492,282 (2013–14) to 731,154 (2021–22). The next two most populated years in terms of yearly fire counts were 2016–17 (675,740) and 2018–19 (650,765). The yearly total fire counts, found to be fluctuating during 2012‒2023, had a growth rate of 6022 per year (excluding COVID years of 2020–21 and 2021–22) as per Ordinary Least Square (OLS) method with a R2 value of 0.076 which was found to be statistically non-significant at 5% level (p value = 0.4381) (Fig. 2). When COVID years were included, fire count growth rate increased to 7217 per year with a R2 value of 0.12, which was also statistically non-significant at 5% level (p value = 0.2685) (Fig. 3). Sen’s slope, a non-parametric method used to study linear relationships which is not affected by outliers in the datasets62, is shown as an alternative trend along with OLS trend. Cluster analysis of all the years under study, each represented by monthly total fire count in respective year (n = 12), revealed that each member year in the pairs of 2017 and 2023, 2016 and 2022, 2013 and 2014 and 2012 and 2018 were statistically similar to each other (Fig. 4).

Fig. 2
figure 2

Yearly trends in annual fire count over 2012–13 to 2023–24 fire-years (without COVID years of ss 2020 and 2021) estimated by OLS and Sen’s Method.

Full size image
Fig. 3
figure 3

Yearly trends in annual fire count over 2012–13 to 2023–24 fire-years estimated by OLS and Sen’s Method.

Full size image
Fig. 4
figure 4

Dendrogram depicting statistically similar and dissimilar years (fire-years) in terms of monthly total fire counts.

Full size image

FRP of individual fires fluctuated within a range of near-zero to 1378.6 MW. The histogram of fire counts falling under various FRP ranges during 2021‒2023 represents a positively skewed distribution, reflecting low number of fires beyond the FRP of 500 MW (Fig. 5). The observed difference in individual FRPs is expected due to the variability in amount of biomass burnt, combustion efficiency, biomass composition and moisture content in biomass68. The OLS trend of median FRPs per fire (MW) over the study period (excluding COVID years of 2020 and 2021) was slightly negative (− 0.027 MW per year) with a statistically significant R2 value of 0.608 at 5% level (p value = 0.0078) (Fig. 6), indicating negligible but statistically significant trend of reduction in median FRP per fire during study period. When COVID years were included, the decreasing trend of median FRP per fire was observed to be − 0.020 per year with a R2 value of 0.411, which was also statistically significant at 5% level (p value = 0.0247) (Fig. 7). Sen’s slope is also depicted with the OLS trends.

Fig. 5
figure 5

Histogram of fire counts under various FRP ranges recorded during 2012–2023.

Full size image
Fig. 6
figure 6

Yearly trends in median FRP over 2012–13 to 2023–24 fire-years (without COVID years of 2020 and 2021) estimated by OLS and Sen’s Method.

Full size image
Fig. 7
figure 7

Yearly trends in median FRP over 2012–13 to 2023–24 fire-years estimated by OLS and Sen’s Method.

Full size image

The annual (fire-year) sum of FRPs ranged from 3,140,642 in 2023–24 to 4,672,833 MW in 2021–22. The OLS trend of annual sum FRPs (MW) over the study period (excluding COVID years of 2020 and 2021) was 1039 MW per year with a statistically non-significant R2 value of 0.00017 at 5% level (p value = 0.9717) (Fig. 8). The yearly sum FRPs registered an increasing trend from 2012-13 fire year onwards with a rate of 25,534 MW per year when all years under study were considered (Fig. 9). This linear increasing OLS trend was statistically non-significant with a R2 value of 0.042 (p value = 0.5204). Sen’s slope is also depicted with the OLS trends.

Fig. 8
figure 8

Yearly trends in annual sum FRP over 2012–13 to 2023–24 fire-years (without COVID years of 2020 and 2021) estimated by OLS and Sen’s Method.

Full size image
Fig. 9
figure 9

Yearly trends in annual sum FRP over 2012–13 to 2023–24 fire-years estimated by OLS and Sen’s Method.

Full size image

There existed a highly significant linear relationship between annual total fire count and annual sum FRPs with a statistically significant R2, both when COVID years were excluded (R2 = 0.839, p value = 0.002227) or included (R2 = 0.89, p value = 0.001119) in analysis (Fig. 10). This relationship was also reiterated by the concomitant variations in fire count with respective FRPs over the study period (Fig. 11). Cluster analysis of study years based on respective monthly total FRPs (n = 12 each year) indicated that the individual years within the pairs of 2017 and 2023, 2014 and 2020, 2016 and 2022, 2012 and 2015 were statistically similar to each other in terms of monthly sum FRPs (Fig. 12). It has been reported by Vadrevu and Lasko69 that sum of FRPs from VIIRS data was 2.5 times higher than MODIS sum FRP, as number of fires detected by VIIRS was more in number in an agricultural landscape. Li et al.26 found that MODIS missed an appreciable number of low-intensity fires in Africa, leading to underestimation of daily MODIS FRP by at least 42.8% than VIIRS FRP.

Fig. 10
figure 10

Linear variation in sum annual FRPs with total annual fire counts over 2012–13 to 2023–24 (with and without inclusion of COVID years).

Full size image
Fig. 11
figure 11

Concomitant fluctuations in monthly sum FRPs with monthly fire counts over 2012–2023.

Full size image
Fig. 12
figure 12

Dendrogram depicting pairs of statistically similar and dissimilar years (fire-years) in terms of sum FRPs.

Full size image

In summers of 2012–2023, fire spots were mainly populated in Central–South–Eastern part of India. The spots covered parts of Odisha, Madhya Pradesh, Chhattisgarh, Telengana, Maharashtra, Andhra Pradesh and Jharkhand, also covering parts of North–Eastern Indian states like Meghalaya, Tripura, Mizoram, lower Assam and Nagaland (Fig. 13). In winter, parts of Northern India in the states of Punjab, Haryana and Uttar Pradesh were mostly affected, followed by central and southern parts of India that hosted scattered fire spots (Fig. 14). The conspicuous clustering of fire spots in major rice and wheat cultivating regions of India in summer and winter builds up a strong case in favour of residue burning in the respective crop harvesting seasons that coincide with periods of March–May and Oct–Nov in India, respectively. Punjab, Haryana, Bihar, Jharkhand, Andhra Pradesh, Madhya Pradesh, Uttar Pradesh, Tamil Nadu, Rajasthan, Maharashtra and Gujarat are the major winter-wheat growing states in India70. For winter wheat, stubbles are removed in summer (March–April–May) and put on fire, resulting in a good number of summer fires. The states of Odisha, Andhra and Karnataka, Kerala, West Bengal, Uttar Pradesh, Bihar are known to cultivate summer rice that is harvested around March–April–May71, indicating to the possibility of paddy stubble burning around this time. Apart from CRB, the prominent numbers of summer forest fire events reported by FSI in India also are expected to substantially contribute to the number of fire spots detected by VIIRS in March–April–May. Some fire clusters overlapped with forested regions of India, predominantly dry/moist deciduous and tropical semi-evergreen types72 in Madhya Pradesh, Chhattisgarh, Jharkhand, Odisha, Maharashtra, Karnataka, Kerala, Manipur, Meghalaya, Tripura, Mizoram, Nagaland73,74, indicating occurrence of forest and wildfires. Bahuguna and Upadhyay75 reported that about 50% of forest fires of India occurred in NE region and95% of them originated from shifting cultivation. During January 2001 to April 2014, around 143,761 forest fire incidents were detected in NER76. Also, month-wise fire incidences in NER was maximum in April in Arunachal Pradesh, Assam and Tripura and the minimum in February in all other states, except in Tripura with no fire. States of Nagaland, Manipur, Meghalaya and Mizoram experienced highest number of fires in March. Overall, the decreasing order of fire incidences was: March (54%) > April (43%) > February (3%) in 2014. In NER, evergreen/semi-evergreen forest had highest susceptibility to fire followed by deciduous forests while alpine forests had least vulnerability. Manmade forest fire caused by shifting cultivation is reported to be one of the primary causes of fire incidences in NER. The fire-prone forest areas in India can be traced from the Van Agni Geoportal of FSI77 which shows extensive coverage of high, very high and extreme fire prone zones in NE India especially in states of Tripura, Meghalaya, Manipur, Nagaland and Mizoram. Further, high to extreme fire-prone zones are also located in several patches in states of Odisha, Telengana and Chhattisgarh which are all observed to be affected by fires during 2012–2021.

Fig. 13
figure 13

Indian landmass dotted with fire spots in the months of maximum occurrence (March or April) in summer in each year during 2012 to 2023 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image
Fig. 14
figure 14

Indian landmass dotted with fire spots in months of maximum occurrence (November) in winter in each year during 2012 to 2023 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image

Fire spot density in the worst affected summer and winter months of 2022 (March and November) and 2023 (April and November) were analyzed and mapped in 20 km × 20 km grids over Indian landmass vis a vis the number of fire spots detected in the grids. The grids can also be designated as fire hot spots of various degrees in terms of fire counts. The locations of cities with > 1 lakh (0.1 million) population were overlaid on Indian landmass to visualize the proximity or overlapping of cities with various classes of fire hot spots hosting specific numbers of fires (Figs. 15, 16, 17 and 18). It is observed that many of summer and winter fire hot spots (400 km2 area) indeed hosted some well-known and populated cities. In the winter (November) of 2022 and 2023, well-known cities like Amritsar, Abohar, Jalandhar, Ludhiana, Patiala and Pathankot in state of Punjab were found to be within 400 km2 hot-spot grids (100–750). Further away, Chandigarh, NCR of Delhi, and cities like Hisar, Sonepath, Karnal, Faridabad and Gurgaon in state of Haryana, Nizamabad and Kakinada in Telengana etc. were also located in hot spot grids in either 2022–23 or 2023–24 and wildfire impacts may also be expected at some of these urban centres. Notably, the said impacts were not measured or assessed but only perceived based on geographical proximity. Several other cities in states of Uttar Pradesh (Ghaziabad, Muzaffarnagar, Saharanpur, Bareilly, Hapur, Bulandshahar, Gorakhpur, Varanasi, Rampur); Madhya Pradesh (Jabalpur, Bhopal, Gwalior); West Bengal (Kulti, Durgapur, Asansol); Dibrugarh and Dispur in Assam; Imphal in Manipur etc. are also located within some prominent fire hot spots either in 2022–23 or in 2023–24 winter season (Fig. 15-16). There are more number of cities and small towns positioned in an around the wintertime wildfire hot spots that are not separately discussed here.

Fig.  15
figure 15

Fire hot spots relative to locations of cities with > 0.1 million population in 20 km × 20 km grids over India in the month of maximum detected fire number (Nov.) in winter 2022 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image
Fig.  16
figure 16

Fire hot spots relative to locations of cities with > 0.1 million population in 20 km × 20 km grids over India in the month of maximum detected fire number (Nov.) in winter 2023 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image

On the other hand, some of the well-known and populated cities located near summer fire hot spots were Delhi; Karnal, Sonepat and Panipat in Haryana; Gorakhpur, Muzaffarnagar, Meerut in Uttar Pradesh; Mumbai, Thane, Pune, Nasik and Nagpur in Maharashtra; Jabalpur, Ratlam, Gwalior, Sagar, Sannai, Bhopal, Sagar, Vidisha etc. in Madhya Pradesh; Hyderabad, Secunderabad, Nellore, Cuddapah etc. in Andhra Pradesh; Chennai and Coimbatore in Tamil Nadu; Trichur in Kerala; Asansol and Durgapur in West Bengal; Dhanbad in Jharkhand; Cuttack and Bhubaneswar in Odisha; Shillong in Meghalaya; Dibrugarh, Guwahati in Assam; Aizwal in Mizoram and Imphal in Manipur etc. There are many other towns and small cities positioned near summertime wildfire hot spots that are not separately discussed here (Fig. 17-18). The proximity of many of these populated towns and cities to the fire hot spots either in 2022 or in 2023 summer season underlines the potential scope of having wildfire-related impacts on environment. FSI’s Van Agni portal provides interactive viewing of forest fires and fire clusters vis a vis forest administrative boundaries, forest type and cover, fire prone forest area etc. in which a user can visualise NRT forest fires detected by MODIS & VIIRS during last three days. The portal also thematically depicts forest fire risk zones based on last several years’ data. The wildfire hot-spot maps provided in this paper demonstrate seasonally classified grids based on number of wildfires that would assist in easy visualization and comparison of highly affected, moderately affected and non-affected grids in terms of various ranges of fire numbers. Importantly, position of populated cities or groups of cities falling within specific grids can be visualized that could possibly indicate the extent of intervention needed for addressing potential short-term local impacts of wildfires in these towns and cities.

Fig.  17
figure 17

Fire hot spots relative to locations of cities with > 0.1 million population in 20 km × 20 km grids over India in the month of maximum detected fire number (March) in summer 2022 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image
Fig.  18
figure 18

Fire hot spots relative to locations of cities with > 0.1 million population in 20 km × 20 km grids over India in the month of maximum detected fire number (April) in summer 2023 (Maps created by QGIS version 3.34.4-Prizren, https://qgis.org/download/).

Full size image

Studies indicate that wildfires can have measurable and statistically significant effect on urban centres78. In US, a large number of houses were found to have been built since 1990 within or near natural vegetation, in areas designated as the Wildland-Urban Interface (WUI), and by 2017, about one in three houses and one in ten hectares of land were located in the WUI. This observed growth in WUI in US is expected to aggravate wildfire-related problems on humans and their livelihood in future79. In California during 1985–2013, interface WUI areas hosted about 50% of buildings that were destroyed in wildfire, whereas intermix WUI had about 32% of the destroyed buildings80. Chen et al.81 have reported that expanding wildfires and human settlements have put higher number of people and infrastructure within the realm of WUI areas under risk. They estimated that 7.07% (12.54%) of WUI areas, hosting about 4.47 million (10.11 million) people, are coming within a 2400 m (4800 m) buffer zone of wildfire threat.   Countries like United States, Brazil, China, India and Australia have highest share in WUI areas. In case of large wildfires, air parcels can carry smoke to large distances in short time and air quality in urban centers near fire clusters may get affected substantially, as have been reported from India82,83, Australia84, Southeast China85, apart from perturbations in local and regional climate and radiation budgets in a few aggravated cases86. Fire hot spot identification by density mapping technique, as used in this paper, has been deployed by Tonini et al.87 in Portugal also that facilitated easy visualization of the fire hot spots.

Conclusion

Analysis of fire spots and their FRPs during 2012–2023 could answer some of the questions posed at the onset of the study. There are yearly fluctuations in wildfire number and total annual FRP and evidences on slight increase in annual fire counts and sum FRPs over 2012 to 2023. But, these trends were found statistically insignificant. But nevertheless, the hint of increase should be viewed with caution due to potentially grave impacts of repeated wildfires on environmental sustainability. There is a clear seasonality in fire incidents, summer being the worst affected season. Winter fires are much lesser in number and mostly centred around specific areas in a few North Indian states viz. Punjab and Haryana followed by Uttar Pradesh. Existence of some populated cities (> 0.1 million population) within fire hot spot grids is apparent, implying the scope of potential impacts of fires on urban life, and air environment. It may be noted that these perceived impacts on nearby urban centres are inferred from proximity of cities to wildfire clusters and not from direct measurement of exposure or effects.

Studies on positional changes in fire-prone areas over the last decade and likely future scenarios, likelihood of having high-intensity fires in the changing climate, estimation of burnt-areas, mapping of vulnerable WUIs and innovations in fire-resilient living under local conditions are potentially important to know in India through future studies, which were not covered under this study. Other potential additions to this work in future could be to identify and clearly differentiate between fire types and assess the extent of burnt areas, keeping in view the challenge of doing so in a large study domain i.e. the entire Indian landmass. Proactive participation of local inhabitants along with interventions by state administrations are urgently needed to streamline and promote sustainable eco-practices to minimize incidence of natural fires, inadvertently caused fires and arson in India.