Climate change strongly affects future fire weather danger in Indian forests

The study investigates how climate change will alter fire weather danger across India’s diverse and fragmented forest ecosystems. Forest fires in India occur most of the year (except peak monsoon) and the country spans a wide range of forest types and climates, making it a biodiversity hotspot. Climate projections indicate substantial warming (about 4.4–4.8 °C by end-century vs. 1976–2005), which, along with changes in humidity, precipitation, and wind, is expected to reshape fire regimes. Existing global-scale analyses are too coarse (0.5–2°) to resolve India’s fragmented forests, show divergent results over India, and often lack appropriate forest masking, leading to unrealistic signals (e.g., high fire danger over deserts). Regional studies largely focus on dry subtropical systems and are not directly translatable to India’s range of ecoclimates. The research aims to produce a high-resolution, India-focused assessment of future fire weather using a calibrated Fire Weather Index (FWI) framework, quantify changes in mean fire weather, severe fire weather days, seasonal severity, and the timing/length of the fire weather danger season, and diagnose the dominant meteorological drivers (temperature, precipitation, relative humidity, wind) behind projected changes.

Prior work shows climate change is linked to increased severity, frequency, and season length of forest fires globally, with IPCC AR6 noting more severe fire weather in some regions and a possible ~50-day increase in mean fire season length under strong warming. Global studies have produced inconsistent projections over India (ranging from homogeneous increases to mixed responses and even decreases), attributable to differences in input resolution/quality, land cover, and fire module structures. Liu et al. highlighted the absence of forest masks in global assessments, which can yield unrealistic danger signals in non-forest areas. Observational and modeling literature identifies temperature, humidity, wind, and precipitation as key meteorological controls on fuel moisture, ignition likelihood, and fire spread, with precipitation suppressing fire probability beyond thresholds. Regional studies in India note high fire activity, especially in dry deciduous and evergreen forests, and anthropogenic ignitions (e.g., shifting cultivation) as important in the Northeast. Overall, the literature underscores the need for high-resolution, region-specific fire weather analysis and calibration of FWI to local conditions.

Study domain: 3680 × 3680 km centered on India at 10 km grid spacing. Time slices: baseline (2006–2015) and end-century (2091–2100) under RCP8.5. Forcing data: high-resolution (10 km) dynamically downscaled CESM (DSCESM) via WRF; variables include 12:00 IST 2-m temperature, relative humidity, accumulated precipitation (12:00–12:00), and wind speed. Wind speed bias in DSCESM was corrected by linear scaling using GSOD station observations (35 stations with complete 2006–2015 data), with performance verified against reanalysis. FWI model: Canadian Forest Fire Danger Rating System (CFFDRS) computing FFMC, DMC, DC, ISI, BUI, and FWI. MATLAB implementation follows Van Wagner’s formulations. Latitude/daylength adjustments applied to DMC and DC; PET for DC computed with assumptions appropriate to low latitudes (<20°N). Initialization: 5-year spin-up per scenario starting with default FFMC/DMC/DC (85/6/15) at each forest grid; indices stabilized after ~1 year. Forest zones: Combined Köppen climate classification and satellite land use/land cover to define five FWI zones: Himalayan mountains (HIM; cold, dry; alpine/subtropical pine), Northeast (NE; warm, humid; wet evergreen/deciduous), Central India (CEN; hot, dry; mixed deciduous/thorn), Deccan (DEC; hot, dry; thorn), Western Ghats (WG; warm, humid; semi-evergreen). Forest mask used to exclude non-forest fires. Evaluation against fires: MODIS daily active fire detections (1 km) filtered by forest pixels (2006–2015), aggregated to 10 km daily fire counts. Association tests between FWI and fire counts: Chi-square test of association/independence, Yule’s correlation, Fisher’s exact test with odds ratio. Spatial correspondence between baseline mean FWI percentiles and observed fire counts; fire-count vs. FWI-percentile scatter and best-fit curves by zone. Danger class thresholds: Five classes (Low, Medium, High, Very High, Extreme). Default approach (99th percentile-based with geometric progression) was insufficient in some zones (probabilities saturating early). An ensemble of five methods was used to define zone-specific thresholds: (1) logistic regression fire probability (thresholds at 0.2/0.5/0.7/0.9 probabilities), (2) percentile-based (22nd/45th/90th/97th), (3) geometric progression limiting extreme days (~3%), (4) fire-occurrence percentages (beyond which 40%/65%/87%/96% of fires occur), and (5) K-means clustering (five clusters on presence/absence-labeled fire-FWI pairs). The maximum value per class from each method was averaged to obtain final class limits; conditional probability distributions were checked for monotonicity. Future attribution and distributions: Sensitivity experiments to isolate each meteorological variable’s contribution (swap in end-century values for one variable at a time) and identify the dominant driver per pixel. Aridity classification via modified de Martonne index (arid <20, intermediate 20–35, humid >35). Epanechnikov KDE of daily FWI per zone for both periods; bootstrapped mean differences (95% confidence). Severity and season metrics: Daily Severity Rating (DSR = 0.0272 × FWI^1.77); Seasonal Severity Rating (SSR) computed seasonally (DJF, MAM, JJA, SON). The 90th percentile baseline SSR per zone used as severe-threshold reference. Fire weather danger season defined as continuous period when daily FWI climatology exceeds the Medium threshold; start/end dates derived for baseline and end-century to quantify shifts in timing and duration.

- Strong FWI–fire association: Across forested grid cells, baseline FWI is significantly associated with MODIS fire counts (Chi-square p < 0.05; Yule’s correlation = 0.88, p < 0.05; Fisher’s exact test p < 0.05). Fisher’s odds ratio = 16.74 (99% significance). Spatial maps and percentile-bin analyses show more fires at higher FWI percentiles in all zones, with zone-specific curve shapes (exponential in NE; cubic in HIM, CEN, DEC, WG). - Zone-specific thresholds: The NE zone requires substantially lower FWI thresholds for a given danger class (e.g., Medium ≈ 2.5; Extreme ≈ 3.5) than other zones (Medium ≈ 3–3.3; Extreme ≈ 4.3–4.8), reflecting higher fuel loading and anthropogenic ignitions (e.g., shifting cultivation). Ensemble-based thresholds yield monotonic increases in conditional fire probabilities by class and avoid saturation seen with the default method. - Mean FWI change by end-century: Nationally, mean FWI increases by roughly 5–10% (bootstrapped). CEN shows the largest mean difference (~0.54 FWI units), NE the smallest (~0.09). Spatially, increases in northern CEN, southern WG, NE, and most of HIM; decreases in western CEN and northern WG. Attribution shows temperature dominates changes in arid/intermediate regions (CEN, DEC, southern WG), while precipitation and/or relative humidity dominate in humid regions (NE, north WG), often reducing future FWI despite warming. - Severe fire weather days: By end-century, days exceeding Medium/High/Very High thresholds increase over large forested areas, with 20–60% increases in many dry regions (eastern CEN, DEC, HIM foothills, northern WG; hotspots include Tripura/Mizoram semi-evergreen and moist deciduous forests, Chir pine and Sal forests in Himalayan foothills, northern moist evergreen forests in WG, and mixed deciduous forests in Odisha/Andhra Pradesh). Some areas experience decreases, including western CEN and central NE, driven by increases in rainy days and/or RH that elevate fuel moisture. Opposite signals can occur between mean FWI change and changes in severe-day counts due to shifts in daily precipitation distributions. - Seasonal severity (SSR): Baseline SSR peaks in pre-monsoon (MAM), matching observed fire seasonality. By end-century, MAM SSR intensifies most: forest area exceeding the 90th percentile SSR increases by ~34.5%; regional increases include HIM (>5 °C warmer, ~8% drier), western CEN (3–5 °C warmer; ~30% SSR rise), and WG (~3 °C warmer). NE sees decreased MAM SSR despite ~3 °C warming due to ~15% RH and increased pre-monsoon rainfall. In JJA, SSR increases by ~50% in HIM/eastern CEN and ~30% in DEC; high-SSR regions of western CEN decrease. In SON, southern CEN increases (hotter and drier); moderate increases in parts of HIM/NE. In DJF, SSR mostly decreases by 20–50% across zones, yet the forest area exceeding the 90th percentile SSR still increases by ~26.5%. - Fire weather danger season: Daily FWI climatology shows pronounced annual cycles with pre-monsoon peaks. By end-century, the fire weather danger season lengthens across all zones by 3–61 days. Changes are due to earlier starts (CEN, HIM, DEC), later ends (NE, by ~15 days), or slight shifts (WG ~5 days). The maximum increase (~61 days) occurs in CEN. - National-scale impacts: Although mean FWI increases are modest (~5%), changes in the frequency of days exceeding operational danger thresholds can be large (up to +60% in dry regions; up to −40% in humid regions).

The analysis demonstrates that future fire weather in India will be governed by the balance among multiple meteorological drivers—temperature, precipitation, and relative humidity—rather than temperature alone. In arid and intermediate regions, strong warming with relatively limited moistening raises FWI and increases the number of severe fire weather days. Conversely, in humid regions, projected increases in precipitation and/or RH can offset warming, leading to reduced FWI and fewer severe days. This nuanced, region-specific response explains previously divergent signals in global assessments and highlights that mean changes in FWI may not align with changes in the number of severe days due to shifts in rainfall occurrence and intensity. The study confirms that FWI is a robust predictor of fire occurrence in India (significant association with MODIS fire counts), but also that thresholds for fire danger classes must be tailored to each forest zone—especially NE, where dense fuels and prevalent anthropogenic ignitions drive high fire activity at lower FWI. High-resolution modeling (10 km) and a forest mask are essential to capture India’s fragmented forests and mesoscale precipitation variability, improving fidelity over coarse global studies. Practically, the results support regionalized fire management strategies, with enhanced preparedness in hotspots during MAM and in zones showing large increases in severe-day counts and SSR. The extension and timing shifts of the fire weather danger season will require adjustments to staffing, surveillance, and resource deployment calendars.

This study provides the first high-resolution, India-wide assessment of how climate change will affect fire weather across distinct forest zones using a calibrated CFFDRS-FWI framework driven by downscaled projections. Main contributions include: (1) Empirical confirmation that FWI robustly represents fire occurrence in Indian forests and must use zone-specific danger thresholds; NE exhibits substantially lower thresholds due to fuel and ignition characteristics. (2) Projected mean FWI increases of roughly 5–10% by end-century translate into much larger shifts in the frequency of severe fire weather days (up to +60% in arid/intermediate regions and up to −40% in humid regions), with temperature dominating changes in dry regions and precipitation/RH in humid regions. (3) The pre-monsoon season (MAM) will intensify over broad areas (about 55% of India’s forests), and the fire weather danger season will lengthen by 3–61 days (maximum in CEN). (4) High-resolution methods and a forest mask reveal fine-scale heterogeneity, enabling more actionable insights than coarse global studies. Future work should integrate ignition processes (natural and anthropogenic) and ecosystem vulnerability to produce comprehensive risk assessments, and employ multi-model ensembles to better characterize projection uncertainties.

- Fire weather is only one component of overall fire risk; ignition sources (natural/anthropogenic) and ecosystem vulnerability are not explicitly modeled here. - Projections rely on a single downscaled ESM scenario (CESMv1, RCP8.5); a broader multi-model, multi-scenario ensemble would better constrain uncertainties but is computationally demanding. - While wind speed biases were corrected using linear scaling against GSOD stations and other variables were within acceptable biases, residual biases and observational limitations may persist. - MODIS active fire detections may miss small/understory fires or be affected by cloud/smoke, and presence/absence sampling in threshold methods may introduce uncertainties despite ensemble design. - Zone-level thresholds and attributions, while robust at 10 km, may still smooth finer-scale microclimatic and fuel heterogeneity relevant to site-level management.