A framework for natural resource management with geospatial machine learning: a case study of the 2021 Almora forest fires

Background Wildfires have a substantial impact on air quality and ecosystems by releasing greenhouse gases (GHGs), trace gases, and aerosols into the atmosphere. These wildfires produce both light-absorbing and merely scattering aerosols that can act as cloud condensation nuclei, altering cloud reflectivity, cloud lifetime, and precipitation frequency. Uttarakhand province in India experiences frequent wildfires that affect its protected ecosystems. Thus, a natural resource management system is needed in this region to assess the impact of wildfire hazards on land and atmosphere. We conducted an analysis of a severe fire event that occurred between January and April 2021 in the Kumaun region of Uttarakhand, by utilizing open-source geospatial data. Near-real-time satellite observations of pre-and post-fire conditions within the study area were used to detect changes in land and atmosphere. Super-vised machine learning algorithm was also implemented to estimate burned above ground biomass (AGB) to monitor biomass stock. Results The study found that 21.75% of the total burned area burned with moderate to high severity, resulting in a decreased Soil Adjusted Vegetation Index value (> 0.3), a reduced Normalized Differential Moisture Index value (> 0.4), and a lowered Normalized Differential Vegetation Index (> 0.5). The AGB estimate demonstrated a significant simple determination ( r 2 = 0.001702) and probability ( P < 2.2 10 −16 ), along with a positive correlation ( r ≤ 0.24) with vegetation and soil indices. The algorithm predicted that 17.56 tonnes of biomass per hectare burned in the Kumaun forests. This fire incident resulted in increased emissions of carbon dioxide (CO 2 ; ~ 0.8 10 −4 kg carbon h −1 ), methane (CH 4 ; ~ 200 10 −9 mol fraction in dry air), carbon monoxide (CO; 2000 10 15 molecules cm −2 total column), and formaldehyde (HCHO; 3500 10 13 molecules cm −2 total column), along with increased aerosol optical thickness (varying from 0.2 to 0.5). Conclusions We believe that our proposed operational framework for managing natural resources and assessing the impact of natural hazards can be used to efficiently monitor near-real-time forest-fire-caused changes in land and atmosphere. This method makes use of openly accessible geospatial data that can be employed for several objectives, including monitoring carbon stocks, greenhouse gas emissions, criterion air pollution, and radiative forcing of the climate, among many others. Our proposed framework will assist policymakers and the scientific community in mitigating climate change problems and in developing adaptation policies.


Background
Forest fires are a widespread phenomenon globally, encompassing both immediate and long-term ramifications for the ecosystem.The consequences of forest fires exhibit a wide range of effects, including changes in above ground biomass (AGB) (Amiro et al. 2000(Amiro et al. , 2001;;Pérez-Cabello et al. 2012), as well as reductions in subterranean physical, chemical, and microbial processes (Cerdà 1998;Aaltonen et al. 2019).Forest fires also generate significant quantities of trace gases and particulate matter.These emissions contribute to the deterioration of both local and regional air quality, while also affecting the global climate (Bae et al. 2019;Yin et al. 2019;Datta 2021).
Forest fires can lead to extreme climatic events, such as more frequent and intense droughts, heat waves, rising sea levels, and melting glaciers.The occurrence of these events poses substantial threats to global livelihoods (Versini et al. 2013;Sharma and Pant 2017;Heyer et al. 2018;Berenguer et al. 2021;Nojarov and Nikolova 2022;Sharifi 2022;Turco et al. 2023).The impacts of forest fires can endure for extended periods, resulting in a decline in net ecosystem productivity over the long run (Taylor et al. 2014;Pellegrini et al. 2018).
In recent times, researchers from around the globe have acquired an understanding of the pivotal role played by forests in mitigating climate change (dos Reis et al. 2021;Kolanek et al. 2021).According to the India State of Forest Report 2021 (https:// fsi.nic.in/ forest-report-2021-detai ls), forests account for 23.58% of India's total geographical area; understanding carbon stock is crucial for comprehending the dynamics of forest stands, their productive capacity, and sustainable management (Zheng et al. 2016;Kirchmeier-Young et al. 2019;Wang and Zhang 2020).
large areas, including inaccessible locations, with high temporal precision (Agus et al. 2020).The utilization of remotely sensed data in fire management encompasses mapping burn severity, detecting hot spots in forest fires, identifying smoke, and managing pre-fire conditions.Various satellite sensors, such as polar-orbiting sensors (Moderate Resolution Imaging Spectroradiometer, MODIS; Advanced Very High Resolution Radiometer, AVHRR; the Landsat series; the Sentinels satellites; Visible Infrared Imaging Radiometer Suite, VIIRS) and geostationary sensor systems (Geostationary Operational Environmental Satellite, GOES; Spinning Enhanced Visible and Infrared Imager, SEVIRI; Communication, Ocean and Meteorological Satellite, COMS; Himawari-8), have been employed for forest fire detection.On-board satellite sensors capture vertical views of the Earth's surface and provide multi-band images in spectral regions such as short-wave infrared (SWIR), near-infrared (NIR), midinfrared (MIR), and long-wave infrared (LWIR), facilitating in-depth monitoring of pre-and post-fire dynamics.
The utilization of machine learning algorithms in conjunction with satellite-derived data has emerged as a prominent methodology for estimating the impacts of forest fires.Machine learning, which falls within the realm of artificial intelligence in computer science, entails the utilization of machines or computers to think and respond in a manner akin to the human mind.This enables software applications to enhance their accuracy in predicting outcomes without explicit programming.Machine learning algorithms employ historical data as input to forecast new output values.The learning process involves training the system through supervised or unsupervised learning.The application of these algorithms varies depending on the purpose and type of data being utilized.Linear regression, logistic regression, random forest, k-mean clustering, and decision trees are among the algorithms that fall within the framework of machine learning.Jang et al. (2019) proposed a threshold-based random forest machine learning system for identifying fires in South Korea using geostationary satellites.Geostationary satellite data may exhibit a high false alarm rate when identifying fire pixels, but the use of machine learning models has significantly mitigated this issue.Van Hoang et al. (2020) employed multi-criteria analysis of individual forest-fire-related characteristics to determine the causes of frequent forest fires in Vietnam, utilizing Landsat-7 data.However, the temporal coverage of Landsat-7 satellite is deficient in terms of revisiting the area of interest, thereby limiting its effectiveness for continuous monitoring.Rabiei et al. (2022) conducted a study to determine the effect of forest fires on droughts by examining fire risk zones in Golestan, Iran, utilizing ecological indicators and precipitation data from 2005 to 2020.Ghorbanzadesh et al. (2019) utilized machine learning and geographic information system multi-criteria decision-making techniques, incorporating environmental and land use data, in order to propose a social vulnerability index and a forest fire susceptibility index.
Post-fire dynamics involve the impacts of fire emissions on the atmosphere, which play a critical role in the degradation of local air quality.Emissions of pollutants during a forest fire event affect the radiative forcing of the climate (Haywood et al. 1999).Forest fires directly emit particulate matter (PM), as well as numerous gaseous compounds.Notable gaseous compounds generated during wildfires include carbon dioxide (CO 2 ), nitrogen oxides (NO x ), carbon monoxide (CO), methane (CH 4 ), and hundreds of volatile organic compounds (VOCs), including a significant number of oxygenated VOCs (Lazaridis et al. 2008).This chemical complexity sets wildfire smoke apart from typical industrial pollution.Noyes et al. (2020) employed multi-angle imaging spectroradiometer data to analyze the evolution of smoke plumes, including size-selective deposition, new-particle formation, and the locations within the plume where particulate pollutants dominate.A significant challenge in understanding the impact of fires on air quality arises from the substantial variability observed from one fire to another, in terms of both the quantity and composition of emissions.According to the CORINAIR 1990(Grösslinger et al. 1996) inventory, forest fires in 1990 contributed approximately 0.2% of NO x , 0.5% of non-methane VOCs, 0.2% of CH 4 , 1.9% of CO, 1.2% of nitrous oxide (N 2 O), and 0.1% of ammonia (NH 3 ) in Europe (Wang et al. 2020).The adverse effects of these brief emissions within a confined region can have a more profound impact on public health, specifically impacts leading to respiratory symptoms and illnesses such as bronchitis, asthma, pneumonia, upper respiratory infection, and impaired lung function, as well as cardiac diseases.
In contrast to other sources of human-induced emissions, the measurement and evaluation of emissions from forest and agricultural biomass fires are lacking in scholarly literature due to the difficulties associated with estimating their temporal and spatial distribution.Dispersion modeling of emitted substances is complicated by the burning processes involved.The act of burning consists of multiple stages, each producing distinct compounds, while the burned material itself is heterogeneous and resistant to mathematical description.Consequently, significant disparities between predicted and observed levels of air pollution may arise.Numerous researchers have employed machine learning algorithms to evaluate the impact of fires on air quality (Ma et al. 2020;Mohajane et al. 2021;Bar et al. 2022).
The goal of our investigation was twofold: (1) to examine the consequences of a severe forest fire event that occurred between January and April 2021 in the Almora district of Uttarakhand, India, through the utilization of multiple satellite observations; and (2) to develop a machine learning based method for tracking the forestfire caused alterations in ecological assets using satellite images.

Study site
Uttarakhand is geographically located between latitudes 28.716667 to 31.45 and longitudes 77.566667 to 81.03333334 in northern India.It shares borders with Tibet and China to the north, Nepal to the east, the Indian state of Himachal Pradesh to the west, and the Indian state of Uttar Pradesh to the south.The state of Uttarakhand is primarily characterized by lofty mountains and rugged terrain, which account for 93% of its total area.The elevation within Uttarakhand ranges from 300 to 7816 m above sea level, this significant variation in topography giving rise to diverse ecosystems within the state.The recorded forest area in Uttarakhand measures 24,240 km 2 , representing 45.32% of its overall geographical expanse (Forest Survey of India 2015).
Broadly, Uttarakhand can be classified into six major vegetation types, corresponding to elevation.Above an elevation of 4500 m, the state is predominantly covered by ice, glaciers, and rocky landscapes.The western Himalayan alpine shrub and meadows can be found between elevations of 3000 m and 4500 m.The temperate western Himalayan subalpine conifer forests exist within the range of 2600 m to 3000 m, forming a tree line below the upper, treeless meadows.Between 1500 to 2600 m, one can find the temperate western Himalayan broadleaf forests.The Himalayan sub-tropical pine forests extend from 900 to 1500 m.The lower Himalayas or Upper Gangetic plains are characterized by dry and moist deciduous forests.The lowlands of the state, adjacent to Uttar Pradesh, are covered by dry savannas and grasslands.
Forest fires in Uttarakhand occur during fall (October through December, post-monsoon) and summer (March through May, pre-monsoon) seasons.Steep terrain, high summer temperatures, strong winds, and the presence of flammable materials contribute to significant impact from, and extensive spread of, forest fires in Uttarakhand.In mountainous areas, forest fires tend to spread rapidly uphill due to the upward flow of hot air, resulting in moisture loss and elevated temperatures in the upper regions.Additionally, isolated trees found at higher elevations and shrubs located at lower elevations exhibit a significant vulnerability to ignition as a result of their diminished levels of moisture, increasing their likelihood of combustion.Furthermore, rolling and burning forest material promote and re-ignite fire at new locations downslope.
Longleaf Indian pine (Pinus roxburghii Sarg.) forests having high inflammable resin content are also highly susceptible to forest fires.Most of the forest fires are caused by human activities.Some fires are accidental, but most of the fires are initiated deliberately for some purpose such as to collect sal (Shorea robusta Roth) seeds left after the forest is burned, to conceal illegal timber extraction, to improve grass growth, to scare away wild animals, etc.The number of fire events in Uttarakhand is reported to have increased due to increased anthropogenic disturbances as well as changes in climate.These fires cause significant impacts to natural resources, which can be mapped and monitored using satellite images (Fig. 1).
The commencement of this fire outbreak in our study dates back to October 2020 and persisted until April 2021.Initially confined to specific locations within the region in from October through December 2020, the fire gradually expanded to a considerable portion of Almora forests in January 2021.Subsequent to mid-January, the intensity of the fire escalated, and the fire continued to grow until April 2021.This incident resulted in the scorching of a total area measuring 478,449.1 ha in Almora forests, exhibiting varying degrees of severity, and significantly influencing the climatic equilibrium of the region.

Sentinel-2
Sentinel-2 satellites have been specifically engineered to provide land remote sensing data for the Copernicus program of the European Commission (https:// www.esa.int/ Appli catio ns/ Obser ving_ the_ Earth/ Coper nicus/ Intro ducing_ Coper nicus).The objective of the Sentinel-2 mission entails the utilization of two satellites in order to facilitate vegetation, land cover, and environmental monitoring.The initial satellite, Sentinel-2A, was successfully launched by the European Space Agency on 23 June 2015 and is currently operating in a sun-synchronous orbit with a repetition cycle of 10 days.The second satellite, Sentinel-2B, identical to its predecessor, was launched on 7 March 2017.Together, these satellites effectively cover all land surfaces, large islands, and inland and coastal waters of the Earth every 5 days.
The Multi Spectral Instrument of the Sentinel-2 satellites captures data from 13 spectral bands spanning from the visible and near-infrared (NIR) to the shortwave infrared (SWIR) wavelengths (Table 1).These spectral bands have a coverage width of 290 km along their orbital path.The spectral band classification of the Sentinel-2 satellites is composed of four bands with a resolution of 10 m: blue (band 2, B2, 490 nm), green (band 3, B3, 560 nm), red (band 4, B4, 665 nm), and near-infrared (NIR, band 8, B8, 842 nm).Additionally, there are six bands with a resolution of 20 m, which consist of four narrow bands utilized for vegetation characterization (red-edge, band 5, B5, 705 nm; NIR, band 6, B6, 740 nm; NIR, band 7, B7, 783 nm; and NIR, band 8a, B8a, 865 nm) and two larger SWIR bands (band 11, B11, 1610 nm; and band 12, B12, 2190 nm) employed for applications such as snow or ice or cloud detection and vegetation moisture stress assessment.Lastly, three bands with a resolution of 60 m are primarily used for cloud screening and atmospheric corrections: 443 nm for aerosol detection (ultra blue, band 1, B1), 945 nm for water vapor analysis (SWIR, band 9, B9), and 1375 nm for cirrus cloud detection (SWIR, band 10, B10).The four bands with a resolution of 10 m have been specially designed to delineate the fundamental classes of land cover.Similarly, the six bands with a resolution of 20 m fulfill the criteria for enhanced land-cover classification and the retrieval of geophysical parameters.On the other hand, the bands with a resolution of 60 m are primarily dedicated to atmospheric corrections and cirrus-cloud screening.For a concise overview of the band information of the Sentinel-2 satellites, please refer to Table 1.In our study, we utilized two Sentinel-2 images from within the designated study area: a pre-fire image and a post-fire image.The pre-fire image was acquired by Sentinel-2A on 16 January 2021, while the post-fire image was retrieved on 1 April 2021.Both of these Sentinel-2A-acquired images underwent processing in SNAP (Sentinel Application Platform; https:// earth.esa.int/ eogat eway/ tools/ snap), including cloud correction, resampling, and subsetting to band numbers 3, 8, and 12.
It is known that healthy vegetation exhibits high reflectance in the visible spectrum but low reflectance in the shortwave infrared (SWIR) portion.Conversely, burned areas tend to display relatively low reflectance in the near-infrared (NIR) band and heightened reflectance in the SWIR bands.To assess changes in forest fire parameters, we employed various indices.One such index was the Normalized Difference Vegetation Index (NDVI), which measures the greenness and density of vegetation (Rouse et al. 1974).Chlorophyll has a strong tendency to absorb visible light, while the cell structure of leaves strongly reflects NIR light.When a plant experiences dehydration, it absorbs more NIR light rather than reflecting it.In the case of Sentinel-2, the NDVI is calculated by combining the NIR and red bands.Specifically, the NDVI from Sentinel-2 can be computed using the B8 band (NIR) and the B4 band (Red) with Eq. 1: Gao (1996) introduced the Normalized Difference Moisture Index, known as NDMI, which defines moisture levels using the NIR and shortwave infrared SWIR bands.While NIR reflectance is affected by leaf internal structure and dry-matter content, rather than water content, the SWIR band reflects changes in both vegetation water content and the spongy mesophyll structure in vegetation canopies.The combination of NIR and SWIR enhances the accuracy of determining vegetation water content by eliminating variations caused by leaf internal structure and dry-matter content.The SWIR region of the electromagnetic spectrum primarily depends on the water content present in the internal leaf structure, resulting in a negative correlation between leaf water content and SWIR reflectance.NDMI can be computed using Sentinel-2 B8 (NIR) and B11 (SWIR) (Eq.2): Delegido et al. (2011)  In contrast to empirically derived NDVI products, the Soil Adjusted Vegetation Index (SAVI) provides a more stable parameter for studying vegetation as it accounts for variations in soil color, soil moisture, and saturation effects caused by dense vegetation.In an effort to improve NDVI, Huete (1988) developed a vegetation index that considers the differential extinction of red and near-infrared wavelengths through the vegetation canopy.This index is a transformation technique that reduces the influence of spectral vegetation caused by red and near-infrared (NIR) wavelength soil brightness indices.The following formula (Eq.4) can be used to calculate SAVI using Sentinel-2 B8 (NIR) and B4 (Red) bands: where L is an index value dependent upon the amount of green vegetation cover.
Several other parameters, such as aerosol optical thickness (AOT), Normalized Burn Ratio (NBR), and above ground biomass (AGB), have been used by numerous researchers to quantify the impact of forest fires in different application domains.

Sentinel-5P
The Sentinel-5P satellite provides precise and detailed measurements of various atmospheric parameters such as air quality, ultraviolet (UV) radiation, ozone, and climate forcing.The TROPOspheric Monitoring Instrument (TROPOMI) imaging spectrometer, located on board the Sentinel-5P satellite, covers specific wavelength bands ranging from UV to SWIR channels.These bands include UV-1 (270 to 300 nm), UV-2 (300 to 370 nm), VIS (370 to 500 nm), NIR-1 (685 to 710 nm), NIR-2 (755 to 773 nm), SWIR-1 (1590 to 1675 nm), and SWIR-3 (2305 to2385 nm).The primary aim of this sensor is to monitor the presence of trace gases in the atmosphere, with the intention of managing air quality.To accomplish this, the sensor employs passive remote sensing techniques to measure the solar radiation reflected by and radiated from the Earth's atmosphere at the top of the atmosphere.The obtained data is used to monitor concentrations of carbon monoxide (CO), nitrogen dioxide (NO 2 ), and ozone (O 3 ); the UV aerosol index; and various geophysical parameters of clouds.The spatial resolution of the measurements is below 8 km for wavelengths above 300 nm, and below 50 km for wavelengths below 300 nm.The impact of forest fire events has been investigated using measurements of columnar HCHO molecules, columnar CO, and mole fraction of CH 4 in dry air obtained from the TROPOMI on the Sentinel-5P satellite.For our analysis, we utilized the Sentinel-5P products both before and after the fire event, specifically focusing on the designated region of interest to study the emissive effects.The data products were processed in the R programming language (R Core Team 2020) using the S5P_ process library, which is specifically designed for post processing of Sentinel-5P data.The processing workflow includes resampling, masking the area of interest, and applying a scale factor to the measured values according to the specified syntax: S5P_process(input, product, my_res, my_aoi, extent_only, ref_raster, inter-pol_method, apply_scale_factor)

Orbiting carbon observatory
The compound CO 2 , which is a significant constituent of GHGs (Greenhouse gases), exhibits only minimal fluctuations over a span of 2 to 3 months.Nevertheless, in the context of its correlation with forest fires, the Orbiting Carbon Observatory (OCO) data on CO 2 concentrations was utilized to examine its presence pre and post-fire in the region of Uttarakhand.The OCO-2 dataset provides bias-corrected measurements of CO 2 with a temporal resolution of 1 day.The Orbiting Carbon Observatory represents the inaugural National Aeronautics and Space Administration (NASA) mission that was been specifically designed to gather space-based assessments of atmospheric carbon dioxide with utmost precision, resolution, and coverage.It encompasses three spectrometers of high resolution that simultaneously take measurements of sunlight reflected in the near-infrared CO 2 region close to 1.61 and 2.06 μm, in addition to atmospheric molecular oxygen (O 2 ) at 0.76 μm (A-Band).

Above ground biomass
Above ground biomass (AGB) is a crucial metric for evaluating the impact of forest fires.The European Centre for Medium-Range Weather Forecasts consistently monitors climate change parameters to meet the requirements of the global scientific community (Santoro and Cartus 2021).We processed the global datasets of forest above ground biomass provided by the Climate Change Initiative (https:// clima te.esa.int/ en/ about-us-new/ clima techange-initi ative/) for the year 2020.These datasets were subsetted to focus on our study area, and we conducted pixel-wise multivariate regression analysis to estimate the loss of above ground biomass caused by forest fires in Uttarakhand.The entire dataset was processed using a train-test algorithm, with a split of 70% for training and 30% for testing.The independent variables NDVI, SAVI, and NDI45 were modeled with the AGB data for 2020.The coefficients of these independent variables pre-fire were utilized to obtain the dependent variable (AGB) from the post-fire independent variables.

Burn severity
In order to track the regrowth of vegetation over burned areas, López Garcia and Caselles (1991) examined LANDSAT-5 thematic mapper data.The thermal band (10.4 to 12.5 μm) and the normalized difference in reflectance between NIR (0.76 to 0.90 μm) and MIR (2.08 to 2.35 μm) were found to be the most appropriate parameters for mapping burned areas.The Normalized Burn Ratio (NBR) is used to determine the location of the burn by measuring the difference in light reflected by burned versus healthy green vegetation.Burned vegetation has a low NBR value, while healthy green vegetation has a high value.NBR values are also lower in areas with bare soil or dry, brown vegetation compared to areas with green vegetation.Cansler and McKenzie (2012) discovered inconsistent NBR values across ecosystems and proposed two metrics for image-based burn severity: the delta Normalized Burn Ratio (dNBR) and the relative differenced Normalized Burn Ratio (RdNBR), with dNBR being slightly superior.However, areas with low pre-fire vegetation cover may face challenges with dNBR due to little absolute change in NBR between pre and post-fire.In such situations, the relativized burn severity is useful.Subtracting a post-fire NBR image from the pre-fire NBR image yields dNBR.
Several studies have been conducted using LAND-SAT-8, Sentinel-2, VIIRS, and MODIS datasets to examine the long-term effects (3 to 4 years) of forest fires on the burned area in Uttarakhand (Sannigrahi et al. 2020;Bar et al. 2022).The long-term regeneration of vegetation following a forest fire can affect the accuracy of burn severity measurements.In our study, we examined the effects of a forest fire that started 2 to 3 h before the satellite acquired the image.The NBR was computed using the SWIR Band-12 and NIR Band-8 of Sentinel-2.By using the delta Normalized Burn Ratio (dNBR), which measures the absolute difference between pre-and postfire NBR, we analyzed and classified the burned area of the Almora forest based on the severity of the burn.Equation 5 defines the calculation for NBR, which allows for the categorization of burn severity based on dNBR (Table 2).
According to our analysis, the forests of Almora exhibit burns of varying severity, ranging from 0.27 (moderatelow) to 1.1.51(High).The presence of numerous longleaf Indian pine in the area, characterized by their combustible needles rich in resin, makes the area susceptible to forest fires.As temperatures increase during the summer season, the forest floor becomes more prone to combustion due to the abundance of pine needles.Detailed analysis of the fire incident on 1 April 2021 revealed that the Almora region in Uttarakhand experienced a severe forest fire, affecting a total area of 478,449.1 ha.The extent and proportion of burn severity in the affected area is presented in Table 3.
(5) NBR = (NIR − SWIR) The assessment of burn severity in the study area is facilitated by Fig. 2. The bright yellow points indicate areas with high burn severity, characterized by dNBR values > 0.66.

Environmental changes
The forest fire in Almora had a significant impact on the environment and ecology of the Kumaun region in Uttarakhand.These changes were effectively examined through the evaluation of aerosol optical thickness, NDVI, NDMI, and SAVI.Environmental changes within the designated study area were assessed by examining the pixel-wise difference between post-fire and pre-fire images.
Forest fires release substantial amounts of particulate aerosols, which affect air quality and contribute to the radiative forcing of the climate.The characterization of emitted aerosols plays a crucial role in air quality modeling.Soflev et al. ( 2009) conducted a study utilizing a series of fire events in Europe to evaluate the emissive properties of particulate aerosols by using brightness temperature anomalies and fire radiative power.Groundbased measurements of aerosol concentrations and optical thickness were employed to characterize these events in regions where smoke had a significant influence on fine particle matter concentrations compared to other sources of pollution.Saarnio et al. (2010) analyzed individual smoke plumes from forest fires in Finland and identified submicrometer particles, primarily composed of particulate organic matter, as the main contributor to the increase in particulate matter concentration.Various studies have been conducted on forest fires in Uttarakhand to assess the impact of meteorological factors, carbon stock, forest cover, and climate variability.Due to the limited availability of literary resources in the area, it is crucial to understand the physical and chemical properties of aerosols emitted from smoke plumes generated by fires in the forests of Uttarakhand.
The aerosol optical thickness (AOT) values ranged from 0 to 0.2 pre-fire (16 January 2021) and increased to 0.6 post-fire (1 April 2021).The absolute difference between the pre-and post-fire values indicated an  increase in fire-emitted aerosols from 0 to 0.4, raising concerns about local air quality.The effects of land changes caused by fires have been extensively observed in terms of plant moisture content, soil composition, and vegetation cover.The fire incident that occurred on 1 April 2021 had a significant impact on various forest types found in the state.These include subalpine conifer forests, temperate western Himalayan broadleaf forests, sub-tropical pine forests, and lower Himalayan dry and moist deciduous forests.Among these, the subalpine conifer forests and the dry and moist deciduous forests were the most affected.These forests are categorized as protected areas, highlighting the importance of their preservation.
The subalpine conifer forests exhibited a notable decrease in Normalized Difference Vegetation Index (NDVI) by 0.4 to 0.9, while the dry and moist deciduous forests showed a decrease in NDVI difference of 0.2 to 0.6.Moreover, the plant moisture content in these forests changed from 0 to 0.3 and 0 to 0.5, in subalpine conifer forest and in dry and moist deciduous forest, respectively.Conversely, the Soil Adjusted Vegetation Index (SAVI) indicated a differential index of soil brightness ranging from 0.1 to 0.5, which reduced uncertainty in NDVI measurements over soil.The impact of forest fires on the area is clearly demonstrated by the decline in green vegetation, plant moisture content, and soil brightness.Figure 3 illustrates the effects of a forest fire on the environment.

Above ground biomass (AGB)
Biomass combustion is a significant consequence of forest fire, as it plays a crucial role in assessing the carbon stock.The examination of burned biomass was conducted using a supervised machine learning approach, specifically by constructing a multivariate linear regression model of above ground biomass (AGB) with the variables of Normalized Difference Vegetation Index (NDVI), Normalized Difference Index 45 (NDI45), and Soil-Adjusted Vegetation Index (SAVI).Resampled AGB data, which possessed the same extent and resolution as the NDVI, NDI45, and SAVI data, was processed in the R programming language using the gstat and forecast libraries.The AGB data, along with NDVI, NDI45, and SAVI, was extracted at 25,411,681 points and organized according to a test-train methodology with a split ratio of 70% for testing and 30% for training.In this case, AGB served as the dependent variable while NDVI, NDI45, and SAVI served as the independent variables.The statistical analysis of the model revealed the significance of the relationships between NDVI, NDI45, and SAVI, with r 2 = 0.001702 and P < 2.2 10 −16 .SAVI and NDI45 exhibited negative t-values with respect to AGB, which may be attributed to the random sampling of the population.Nevertheless, the significance of the model and the lower P-value indicated a strong individual relationship with AGB.To further validate the model statistics, an analysis of variance (ANOVA; Fisher 1919) test was conducted.ANOVA partitions the observed aggregate variability within a dataset into two components: systematic factors and random factors.The systematic factors exert a statistical influence on the dataset, whereas the random factors do not.ANOVA (also known as Fisher's analysis of variance) represents an extension of the t-and Z-tests.Test results demonstrated the significance of the fitted model, as evidenced by the higher F-value for each estimator.The model and ANOVA statistics are presented in Table 4.
After confirming the statistical significance of the model, we applied it to estimate AGB both pre and postfire.The predicted AGB exhibited a strong positive correlation with NDVI, NDI45, and SAVI both pre and post-fire.Specifically for pre-fire, AGB demonstrated a positive correlation of 0.38 and 0.41, with NDVI and SAVI, respectively.This observation underscores the relationship between green vegetation, moisture in plants, and healthy soil.Conversely, the post-fire correlogram reveals the impact of fire on vegetation, soil, and plant moisture, as evidenced by a decreased positive correlation of 0.24 and 0.20 with NDVI and SAVI, respectively, and a small negative correlation of − 0.19 with NDI45 (Fig. 4).
Significant loss of AGB in the Almora forests is reported during the post-fire event.A pixel-wise comparison of estimated AGB revealed a considerable decrease in biomass across a large area of the Almora Forest Range.Prior to the fire event, most pixels in the region concerned denoted a maximum biomass of 21.03 t ha −1 , whereas post-fire, this decreased to 17.38 t ha −1 (Fig. 5).

Effects of forest fire emissions
Forest fires serve as a primary origin for aerosols and greenhouse gases, both of which have significant implications.Biomass burning leads to the generation of black carbon, a particulate pollutant of considerable importance.Within the atmosphere, black carbon absorbs radiation from the sun across all wavelengths, resulting in an increase in near-surface temperature.Conversely, GHGs are responsible for the escalation of near-surface temperature and the phenomenon of global warming.Greenhouse gasses obstruct the passage of reflected long-wave radiation from the Earth's surface into space, thereby diminishing the rate of cooling and disrupting the planet's heat budget.In the context of our investigation, the emissions of CO 2 , CO, CH 4 , and HCHO during the Almora forest fires are significant parameters.A detailed analysis of the emissions before and after the fire demonstrates a notable decline in air quality.
Forest fires are a prominent contributor to greenhouse gases, particularly through the emission of carbon dioxide.Carbon dioxide constitutes the majority of total wildfire smoke emissions, accounting for 90% of carbon emissions, and consequently exerts significant radiative forcing (Urbanski 2014).The global anthropogenic Table 4 Statistical analysis results of a multivariate linear regression model of above ground biomass (AGB) with the variables Normalized Difference Vegetation Index (NDVI), Normalized Difference Index 45 (NDI45), and Soil-Adjusted Vegetation Index (SAVI) (r 2 = 0.001702 and P < 2.2 10 −16 ), and their coefficients of regression (β), along with an analysis of variance.Results include standard error (SE), t-value (t), sum of squares (SS), mean of squares (MS), and F-value (F).Data used were collected from our study of the consequences of severe fire in the Almora district of Uttarakhand, India, between 16 January and 1 April 2021  The burned biomass image is the absolute difference between the pre-and post-fire images.The color gradient represents low (blue), moderate (green), and high (red) above ground biomass greenhouse gas emissions in 2019 were equivalent to 59 billion tonnes of CO 2 , accounting for 75% of total 2019 GHG emissions, with HCHO, NO 2 , and fluorinated gases contributing methane 18%, 4%, and 2%, respectively (IPCC 2022).Although plants assimilate CO 2 in subsequent growing seasons following fire, frequent fires and changing climate conditions may impede the ability of ecosystems to fully recover, resulting in a net increase in CO 2 emissions (Bowman et al. 2021).
Numerous researchers have conducted studies on the impact of forest fires on CO 2 emissions worldwide.For instance, Rajab et al. (2009) analyzed Atmospheric Infrared Sounder-(AIRS-) retrieved CO 2 data over Malaysia and observed heightened CO 2 levels during the dry season (February to April), when biomass burning occurs in the region.Mannan et al. (2019) estimated CO 2 emissions based on average dry matter g m −2 , burned area, combustion factor, and burning efficiency.The severe impact of CO 2 on global climate change has prompted the scientific community to develop methodologies for estimating CO 2 emissions.Setiani et al. (2021) employed the Google Earth Engine platform to analyze Landsat-8 and Sentinel-2 images in Indonesia from 2016 to 2019.Their findings indicated that the events primarily affected grassland and tropical forest areas, as well as a fraction of agricultural areas, with total estimated carbon emissions of 2.5 × 10 3 t km −2 burned area, with CO 2 emissions being the highest, followed by CO emissions.Ndalila et al. ( 2022) developed a fine-scale emission inventory with spatial patterns of Australian tropical savannas, utilizing fire severity and vegetation mapping alongside favorable emission factors for Australian vegetation types.A comparison of the results with the Global Fire Emission Database for the 2013 Forcett-Dunalley Fire demonstrated the greater reliability of CO 2 estimation for Australian fires as compared to fine particulate matter.Lv and Shi (2023) quantified the loss of forest carbon sequestration capacity due to forest fires in China during 2020, estimating the release of 35,017.42 to 98,486.5 t of CO 2 .
All of these investigations have documented the significance of CO 2 emissions, which serve as the primary source for the escalation of temperature and global climate alterations.There have been very few reports that highlight the immediate CO 2 emissions from the diverse Uttarakhand forests in terms of ecological classifications.The majority of studies have relied on historical patterns of CO 2 to evaluate the repercussions of climate change, with only a small fraction concentrating on event-based analysis.In our study, the emissions of greenhouse gases (GHGs) and carbon monoxide (CO) were observed in two homogeneous clusters of forest zones, namely temperate broadleaf forests and moist and dry deciduous forests.It was observed that there was an increase in CO 2 by 0.8 × 10 4 kg C h −1 post-fire in temperate broadleaf forests, which is a cause for concern in terms of climatic instability if the frequency of fire events continues to rise.
Carbon monoxide (CO) is another crucial element of trace gases that indirectly contributes to climate forcing.CO has a substantial impact on the hydroxyl radical concentrations in the troposphere, which in turn affects the duration of greenhouse gases like methane and halocarbons in the atmosphere (Ossola et al. 2022).CO can be generated through the photodegradation of dissolved organic matter, among other sources.Numerous effects of atmospheric CO have been documented, including the formation of ozone and other atmospheric particles.It has been determined that CO has detrimental effects on human health by reducing blood oxygen levels, thus impeding the organs' ability to receive oxygen.Consequently, the most common side effects of CO exposure are fatigue, headaches, disorientation, and dizziness, all of which are caused by insufficient oxygen delivery to the brain.Olivier et al. (2005) conducted a study on the global trends of GHGs and their precursors through source characterization and revealed that CO resulting from forest fires accounts for half of the global annual CO emissions (~ 300 million tonnes).The primary source of CO emissions from forest fires is the incomplete combustion of biomass.Bela et al. (2022) employed the University of Colorado Airborne Solar Occultation Flux (CU AirSOF) method in conjunction with high-resolution vegetation and fuel datasets from the Biomass Burning Flux Measurements of Trace Gases and Aerosols (BB-FLUX) campaign to measure direct CO emissions from fire plumes during California forest fires.They reported the CO emissions factors based on various vegetation and forest types.In our study, a comparison of temperate broadleaf forests to moist and dry deciduous forests following fire revealed a greater fraction of CO and an increased columnar CO concentration (~ 2000 × 10 15 molecules cm −2 ; Fig. 6).Similar discrepancies in CO emissions have also been reported in terms of emissions factors and fuel consumption for different ecosystems (Campbell et al. 2007;Akagi et al. 2011;van Leeuwen et al. 2014).Methane (CH 4 ), a critical greenhouse gas, is responsible for approximately 20% of the warming caused by long-lived gases.According to the IPCC 2013 report, escalating methane emissions are a significant contributor to the rising concentration of greenhouse gases in the Earth's atmosphere, and account for up to one-third of near-term global heating.Methane causes direct radiative forcing, which is second only to that of carbon dioxide (Butler et al. 2020).
Due to photochemically induced reactions with oxygen compounds, methane (CH 4 ) can lead to the escalation of Fig. 6 Visual representation of concentrations of greenhouse and trace gases that were injected into the atmosphere in our study area where we conducted research to determine the consequences of severe fire in the Almora district in southeast Uttarakhand, India, on (A) 16 January 2021, pre-fire; on (B) 1 April 2021, post-fire; and (C) their absolute differences shorter-lived ozone and water vapor in the atmosphere.This amplification of methane's warming influence in the near-term is referred to as indirect radiative forcing.Consequently, the interactions also generate longer-lived and less potent carbon dioxide (CO 2 ).Considering both direct and indirect radiative forcing, the rise in atmospheric methane accounts for approximately one-third of the global heating in the short term.Numerous inquiries have been conducted to incorporate the impact of forest fires on greenhouse gas emissions.Through the utilization of remote sensing and observational techniques, the majority of studies (van der Werf et al. 2010;Shan et al. 2020;Kelly et al. 2021) have examined burned areas, revealing that methane hydrates and soil carbon fluxes are the primary sources of CH 4 emissions in wildfireprone regions.
Our study identifies two clusters of heightened (dry air mole fraction ~ 200 × 10 −9 ppbv) methane concentrations within temperate broadleaf forests in the western Himalaya, with a scarcity of emission points in the moist deciduous forests (Fig. 6C).The temperate broadleaf forests are located in the high-altitude zones of the region, where the elevated CH 4 levels may be attributed to the combined effects of fire emissions and the indirect impact of CO.In contrast, the lower regions of the study area, characterized by moist deciduous forests, house isolated wetlands clusters, and dispersed population zones where both natural and human activities, such as agriculture and waste disposal, serve as the primary sources of CH 4 emissions.
One of the most prevalent volatile organic compounds (VOCs) emitted by fires is formaldehyde (HCHO), which plays a significant role as a precursor to oxidants.The emissions of HCHO are influenced by factors such as the type of fuel used, modified combustion efficiency, and overall carbon emissions (Liu et al. 2017).Emissions of HCHO can vary by more than a factor of two across different biomes, including tropical forests, savannas, boreal forests, and temperate forests.HCHO is not only generated through direct emissions but also through the oxidation of VOCs in fire plumes.Alvarado et al. (2020) demonstrated that HCHO enhancements in wildfire plumes persist for several days downwind, as observed using data from the TROPOspheric Monitoring Instrument (TROPOMI).HCHO also significantly contributes to the production of peroxy radicals (HO 2 ), which in turn affects the formation of ozone and other secondary pollutants.In our study, it was observed that the production of columnar HCHO increased (~ 3500 × 10 13 molecules cm −2 ) in the temperate broadleaf forest zones compared to the moist deciduous forest zones during post-fire events (Fig. 6D).This difference in HCHO production is a consequence of differing combustion efficiencies between the two ecosystems.Uttarakhand has documented a rising occurrence of forest fire events, which have adverse effects on the ecology, wildlife habitats, carbon stock, and regional air quality.
Emissions have the capacity to affect radiation budgets on both regional and global scales, as well as affect the properties of clouds and the water cycle.Aerosols produced by the combustion of forests escape from the boundary layer of the atmosphere and have the potential to remain suspended in the air for many days.These aerosols can modify the regional radiation budget and persist beyond the duration of the fire, leading to a degradation in air quality that can extend for hundreds of kilometers in the downwind direction.

Conclusion
We propose a comprehensive operational framework for the management of natural resources in India through the current work.As far as our knowledge goes, the operational services in India, provided by various agencies, are currently limited to the reporting of active fire points identified in MODIS and VIIRS data.MODIS provides the total count of fire points observed within the last 24 h.In our study, we have evaluated the ecological and climatic impacts of forest fires and have ultimately determined that anthropogenic factors exert the strongest influence on the local and regional climate of Uttarakhand.Utilizing a supervised machine learning algorithm, we quantified the loss of above ground biomass resulting from severe fire events in Uttarakhand and have reported a biomass stock loss of up to 17.55 tonnes per hectare within the study area.We also investigated the immediate emissions resulting from these fire events in order to assess their contribution to greenhouse gases (GHGs), trace gases, and volatile organic compounds (VOCs).Our findings reveal that fire events have significantly contributed to increased emissions of CO 2 , CH 4 , CO, and HCHO.These emitted species are highly concerning pollutants on a global scale, with both direct and indirect effects on global warming and climate change.Our results demonstrated that the temperate broadleaf forests and moist deciduous forests have emitted GHGs and trace gases that contributed to local climatic imbalances.These forests are home to a diverse range of endangered species, many of which are highly endemic, and fire incidents such as these have a detrimental impact on the ecological stability of the region.
Furthermore, our analysis also examined the gaps in biomass estimation and emissions in the Kumaun forests of Uttarakhand.We have shown that the near real-time estimation of biomass and carbon loss resulting from forest fire events can be efficiently achieved by applying a machine learning framework to the most current of satellite images.However, more accurate fine-scale analyses require improved data on fuel types and their corresponding emission factors in the Kumaun forests.Numerous research endeavors are currently underway to detect forest fire incidents at an early stage, and remote sensing algorithms have proven to be particularly effective in monitoring forests.The methodology presented in our paper will contribute to ongoing research efforts related to active fire monitoring and impact assessment, both in India and globally.

Fig. 1
Fig. 1 Location of the study area in which we conducted research to determine the consequences of severe fire in the Almora district in southeast Uttarakhand, India, between 16 January and 1 April 2021.The black outline represents India's international boundary, red outline represents Uttarakhand state boundary, and blue bounding box represents our study area

Fig. 2
Fig. 2 Burn severity of the study area in which we conducted research to determine the consequences of severe fire in the Almora district in southeast Uttarakhand, India, between 16 January and 1 April 2021.Burn severity, defined by dNBR values, is indicated on a color gradient from red (low severity) through orange (moderate severity), and yellow (high severity)

Fig. 4
Fig. 4 Correlation matrix of Above Ground Biomass (AGB) with Normalized Difference Moisture Index (NDMI), Soil-Adjusted Vegetation Index (SAVI), Normalized Difference Index 45 (NDI45), and Normalized Difference Vegetation Index (NDVI) on training data (A) 16 January 2021, and on test data (B) 1 April 2021, taken from our research to determine the consequences of severe fire in the Almora district in southeast Uttarakhand, India, between these two dates.Negative correlation of AGB with NDI45 and reduced positive correlation with NDMI, SAVI, and NDVI on 1 April 2021 (B) shows the consequences of fire in terms of reduced carbon stock, vegetation, and soil moisture.The numbers inside the ovals represent the correlation coefficient between each pair.The more narrow the oval is, the stronger the positive correlation.Likewise, oval colors from light green through dark red represent negative to strong positive correlation, respectively

Table 1
Specifications of the spectral bands utilized by Sentinel-2 satellites to create the images used in our study of the consequences of severe fire in the Almora district of Uttarakhand, India, between 16 January and 1

April 2021 Sentinel-2 spectral band specifications Number, descriptions Central wavelength (nm) Bandwidth (nm)
introduced an optimal index known as the Normalized Difference Index 45 (NDI45).

Table 2
Burn severity classification as determined by delta Normalized Burn Ratio (dNBR), as used in our study of the consequences of severe fire in the Almora district of Uttarakhand, India, between 16 January and 1 April 2021

Table 3
Proportion of burn severity detected in the burned area in our study of the fire in the Almora district of Uttarakhand, India, between 16 January and 1 April 2021.The total study area was 1,418,791 ha, of which 478,449.1 ha (33.72%) burned