Fire and the Vulnerability of the Caatinga Biome to Droughts and Heatwaves

Abstract

This study analyzes the relationship between fires and climate extremes in the Caatinga biome from 2012 to 2023 by integrating Fire Radiative Power (FRP) from VIIRS (S-NPP and NOAA-20), Vapor Pressure Deficit (VPD) and air temperature from ERA5, drought indices (SPI-1 and SPI-6), and heatwave events from the Xavier database. Daily percentiles of maximum (CTX90pct) and minimum (CTN90pct) temperatures were used to characterize heatwaves. Spatial and temporal dynamics of fire patterns were identified using the HDBSCAN algorithm, an unsupervised Machine Learning clustering method applied in three-dimensional space (latitude, longitude, and time). A marked seasonality was observed, with fire activity peaking from August to November, especially in October, when FRP reached ~1000 MW/h. The years 2015, 2019, 2021, and 2023 exhibited the highest fire intensities. A statistically significant upward trend in cluster frequency was detected (+1094.96 events/year; p < 0.001). Cross-correlations revealed that precipitation deficits (SPI) preceded FRP peaks by about four months, while VPD and air temperature exerted immediate positive effects. FRP correlated positively with heatwave frequency (r = 0.62) and negatively with SPI (r = −0.69). These findings highlight the high vulnerability of the Caatinga to compound drought and heat events, indicating that fire management strategies should account for both antecedent drought conditions, monitored through SPI, and real-time atmospheric dryness, measured by VPD, to effectively mitigate fire risks.

1. Introduction

Extreme climatic events have become more intense, prolonged, and frequent as a result of global warming caused by human activities [1]. These phenomena may occur in isolation or simultaneously, characterizing compound events, or sequentially as cascading risks, amplifying their impacts across multiple sectors [2,3]. In the context of wildfires, droughts and heatwaves play distinct yet complementary roles: drought conditions act as a precursor by reducing soil and vegetation moisture and increasing fuel availability, whereas heatwaves exert a direct and immediate influence by intensifying ignition probability and fire spread through increased air temperature and atmospheric dryness [4]. The interaction between these processes significantly increases wildfire risk and contributes to environmental degradation and atmospheric pollutant emissions, with implications for regional and global climate systems [5,6,7,8].

The increase in air temperature and the intensification of aridity conditions associated with climate change have significantly altered wildfire dynamics in different regions of the world. Global- and continental-scale studies demonstrate that anthropogenic warming has extended the fire season, increased vapor pressure deficit, and reduced vegetation fuel moisture, creating more favorable conditions for fire ignition and spread [4,9,10]. In addition, the joint or sequential occurrence of droughts and heatwaves, often described as compound climate events, has been associated with more extensive and intense wildfires, as drought acts as a precursor by reducing vegetation moisture, while extreme heat intensifies combustion and flame spread [11,12]. These processes make semiarid ecosystems particularly vulnerable, reinforcing the need for integrated analyses that consider multiple climate extremes simultaneously [11,13,14,15].

In this context, the use of remote sensing data and climate reanalyses has become established as a fundamental approach to investigate the spatiotemporal patterns of wildfires and their relationship with climate extremes. Products derived from orbital sensors, such as VIIRS, allow the monitoring of fire location and intensity through the estimation of Fire Radiative Power (FRP), providing consistent information at regional and interannual scales [16,17]. In parallel, atmospheric reanalyses and interpolated climate databases have been widely employed to characterize droughts, heatwaves, and thermodynamic conditions associated with fire activity [15,18]. Although these datasets present uncertainties inherent to detection, assimilation, and spatial resolution processes, international studies demonstrate that their integration provides a robust basis for understanding the climatic mechanisms that control fire activity, especially in regions with scarce in situ observations, such as semiarid environments [11,14].

Fire constitutes an essential ecological agent in the dynamics of several terrestrial ecosystems around the world, playing a key role in maintaining the structure and functioning of plant and animal communities. In environments ranging from boreal forests to tropical savannas, it acts by removing accumulated biomass, controlling invasive species, and renewing vegetation adapted to this type of disturbance [19,20]. However, when fire frequency or intensity exceeds natural thresholds, it ceases to play a regulatory role and instead compromises ecological resilience, resulting in biodiversity loss, long-lasting changes in land cover, and substantial greenhouse gas emissions [21,22].

In Brazil, the fifth-largest country in territorial extent, the situation is particularly concerning due to the high incidence of wildfires, often associated with severe droughts and increasingly intense heatwaves [23,24,25,26,27]. These extreme events not only favor fire propagation but also increase greenhouse gas emissions, creating a degrading feedback loop.

In the case of the Caatinga, a semi-arid biome marked by strong seasonality and limited water availability, extreme climatic conditions (e.g., droughts) amplify the risk of large wildfires [16,19]. The biome also faces significant environmental challenges resulting from human activities, such as forest clear cut and soil degradation, which have severely impacted its ecosystems and biodiversity [28,29]. With endemic biodiversity and essential ecosystems, the Caatinga is increasingly threatened by fires, often originating from uncontrolled agricultural burnings [30]. These fires are exacerbated both by anthropogenic factors and by the projected intensification of extreme climatic conditions, including prolonged droughts and rising temperatures, reflecting trends observed in other fire-prone biomes [11,19,28]. Thus, understanding the interaction between fire and extreme climate events becomes fundamental for anticipating impacts, developing adaptation policies, and promoting effective management strategies in vulnerable ecosystems such as the Caatinga.

Natural fires in the Caatinga are rare due to limited fuel availability, strong rainfall seasonality, and the low frequency of lightning strikes, which classifies the biome as ecologically fire-independent. Unlike fire-adapted ecosystems, such as the Cerrado, Caatinga vegetation exhibits limited evolutionary adaptation to recurrent burning, making it particularly vulnerable to wildfire impacts [31,32].

Despite the low incidence of natural ignitions, fire is widely used for anthropogenic purposes in the Caatinga, mainly for land clearing, pasture management, and agricultural practices. When combined with prolonged droughts and heat extremes, these human-induced ignitions substantially increase the likelihood of uncontrolled wildfires, often occurring in ecosystems with low resilience to fire [33,34,35].

Wildfires can drastically alter ecological landscapes within a short period, while vegetation recovery may take several years and does not necessarily lead to the reestablishment of pre-fire conditions. Recent remote sensing–based studies have demonstrated that post-fire vegetation recovery is strongly modulated by burn severity and climatic conditions, with long-term shifts in vegetation structure and hydrological functioning frequently observed after extreme fire events [36]. In fire-sensitive environments, such as the Caatinga, recurrent fires under increasingly extreme climatic conditions may therefore result in persistent degradation, reduced ecosystem services, and delayed or incomplete vegetation recovery.

This study investigates the relationship between wildfires and climate extremes in the Caatinga biome. We analyze how droughts and heatwaves influence fire dynamics by applying machine learning algorithms to identify spatiotemporal patterns of wildfires. Our approach aims to contribute to the understanding of the climatic mechanisms that modulate fire activity in semiarid regions in order to provide insights on the development of more effective mitigation management strategies under extreme climate conditions.

2. Materials and Methods

2.1. Data and Study Area

Hourly data on active fire hotspot detections were obtained from the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor aboard the Suomi National Polar-orbiting Partnership (S-NPP) and NOAA-20 satellites, covering the period 2012–2024. Operated by the National Oceanic and Atmospheric Administration (NOAA), these satellites provide consistent monitoring of fire activity, enabling the assessment of fire dynamics in the Caatinga biome in northeastern Brazil (Figure 1). This biome represents the main semi-arid ecosystem in South America, characterized by low humidity, irregular rainfall, and prolonged droughts [26]. For analyses correlating fire activity with climate extremes, the period 2012–2023 was considered, as complete climate data are only available until 2023.

The VIIRS sensor has been in operation since 2012 and covers the Earth’s surface twice a day, acquiring data in five spectral bands with a spatial resolution of 375 m [37]. These data were obtained through the Fire Information for Resource Management System (FIRMS), maintained by the National Aeronautics and Space Administration (NASA): (https://firms.modaps.eosdis.nasa.gov/, accessed on 9 September 2025).

Vegetation cover dynamics were evaluated using the Normalized Difference Vegetation Index (NDVI) derived from the VIIRS “VNP13A1” product (version 002), which provides data at a spatial resolution of 500 m. This product is based on temporally composited surface reflectance observations, which reduces the influence of short-term noise and provides stable estimates of vegetation conditions.

The dataset spans the period from 2012 to 2024 and was obtained from NASA through the Google Earth Engine platform (https://earthengine.google.com/, accessed on 18 September 2025). The NDVI data were used to assess the spatiotemporal variability of vegetation conditions and their relationship with fire activity throughout the Caatinga biome.

To assess heatwave and drought events over the analyzed period, data from the dataset developed by Xavier et al. [38] were used. This dataset provides standardized indices of temperature and precipitation extremes for Brazil from 2012 to 2023 and is derived from gridded daily meteorological fields constructed through the spatial interpolation of observations from a large network of weather stations across the country [38]. As with any interpolated climatological dataset, uncertainties are inherently associated with the spatial distribution and density of meteorological stations, as well as with the interpolation procedures [34,35,38]. The gridded data are publicly available at https://sites.google.com/site/alexandrecandidoxavierufes/brazilian-daily-weather-gridded-data, accessed on 15 September 2025). Meteorological variables from the ERA5 reanalysis [28], made available by the Copernicus Climate Data Store (CDS) (https://cds.climate.copernicus.eu/, accessed on 12 September 2025), were also used to calculate the Vapor Pressure Deficit (VPD), an indicator of atmospheric dryness closely related to fire risk [9,10,39]. ERA5 provides data at a horizontal spatial resolution of approximately 0.25° (~31 km) [18].

It is recognized that remote sensing data are subject to uncertainties inherent to the detection process, spatial and temporal resolution, atmospheric conditions, and instrumental limitations, especially in applications related to active fire detection and FRP estimation [16,17,40]. In the case of hotspot and FRP data derived from the VIIRS sensor, previous studies indicate high reliability in the detection of medium- and high-intensity fires, particularly due to the 375 m spatial resolution and the greater radiometric sensitivity of the sensor compared to earlier products, such as MODIS [16,17].

Similarly, reanalysis products (ERA5) and climate indices derived from interpolated data [38].present uncertainties associated with the density of the observational network and data assimilation processes, particularly in semi-arid regions. Despite these limitations, these datasets are widely used and validated in the literature for studies of climate extremes and fire dynamics, and their combination helps reduce individual uncertainties, providing a robust and consistent basis for integrated analyses of fires and climate extremes [11,18].

2.2. Clustering Methodology for Fire Dynamics

The analysis of wildfire dynamics was carried out using Fire Radiative Power (FRP), a variable widely used as an indicator of the intensity and energy emitted by burning events [16,17]. The date and time information of each active fire hotspot were combined into a single temporal variable, allowing the establishment of a chronological order of the records and the application of spatiotemporal analysis techniques.

The Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) algorithm was adopted to identify patterns in the FRP data, as described and methodologically validated by Bezerra et al. [41] for the state of Alagoas. HDBSCAN is an evolution of DBSCAN [42] that incorporates a hierarchical and adaptive approach, enabling the detection of clusters with varying densities [43,44,45,46].

Then, the geographic coordinates (latitude and longitude) were converted into approximate linear distances (km), considering the Earth’s curvature, while the temporal dimension was expressed in days. In this way, each active fire hotspot was represented in a three-dimensional space (latitude–km, longitude–km, time–days), allowing simultaneous capture of the spatial and temporal dynamics of the fires.

The algorithm estimates the core distance for each point x, defined as the distance to its k-th nearest neighbor, as shown in Equation (1):

$$C_k\left(x\right) = d(x, N_k(x\left)\right)$$

(1)

where $N_k\left(x\right)$ is the k-th nearest neighbor of point (x) and d ( , ) represents the Euclidean distance. Based on this, the mutual reachability distance between two points (a) and (b) is defined as in Equation (2):

$$d_{mreach,k}(a,b) = \max \left\{C_k\left(a\right), C_k\left(b\right), d(a,b)\right\}$$

(2)

Based on these distances, HDBSCAN constructs a Minimum Spanning Tree (MST), from which a hierarchy of clusters is derived. This hierarchy is subsequently segmented to identify the most stable clusters, whose stability ($S_{cluster}$) is calculated using Equation (3):

$${ S}_{cluster= \sum p\in C_{cluster}}{\lambda }_p-{\lambda }_{death}$$

(3)

where $C_{cluster}$ is the set of points in the cluster, ${\lambda }_p$ represents the inverse distance value at which point p joins the cluster, and ${\lambda }_{death}$ is the value at which the cluster splits into subgroups.

The HDBSCAN implementation used min_cluster_size = 5 and min_samples = 2, with the Euclidean metric. These parameters were selected after sensitivity tests conducted on the dataset, in which min_cluster_size ranged from 3 to 10 and min_samples from 1 to 5. The choice of ranges and selection criteria followed the calibration of the algorithm proposed by Bezerra et al. [41] for the state of Alagoas. The quality of the clusters was evaluated considering the spatial and temporal coherence of the clusters, the persistence of the groupings, and the distinction between clusters and noise, ensuring greater robustness and reliability in identifying the spatiotemporal patterns of fires. Thus, each hotspot was classified as belonging to a cluster or identified as noise, representing isolated occurrences. This distinction allowed for an objective characterization of recurrence cores and the spatiotemporal propagation of fires.

Fire duration was analyzed using the Kaplan–Meier [47] survival estimator, a non-parametric method commonly applied to time-to-event analysis. In this study, the “event” corresponds to fire extinction, and the survival function represents the probability of a fire remaining active over time. This method enables the characterization and comparison of fire persistence among the states within the Caatinga biome.

Based on the annual aggregation of the fire detection clusters, an additional analysis was conducted to assess long-term temporal trends in fire activity. To evaluate temporal trends in fire activity, a linear regression analysis was applied to the annual number of fire detection clusters derived from the HDBSCAN results. The statistical significance of the trend was assessed using the slope coefficient and the associated p-value, adopting a 95% confidence level (p < 0.05). This approach allows identifying long-term increasing trends in fire occurrence over the study period.

2.3. NDVI

To assess the vegetation variability in the Caatinga biome between 2012 and 2024, the Normalized Difference Vegetation Index (NDVI) from VIIRS sensor images, version 003, was used, obtained from the “NASA/VIIRS/003/VNP13A1” collection available on the Google Earth Engine (GEE) platform. NDVI is widely recognized as a robust indicator for vegetation monitoring and is essential for analyzing vegetation cover, environmental degradation, and ecosystem resilience. It is also commonly applied to monitor agricultural crops, forest cover, and drought conditions [48,49].

As defined by Tucker [50], the NDVI is calculated by Equation (4):

$$NDVI= {\displaystyle \frac{NIR-RED}{NIR+RED}}$$

(4)

where NIR represents the surface reflectance in the near-infrared spectral band and RED corresponds to the surface reflectance in the red spectral band.

In GEE, the VNP13A1 collection provides preprocessed NDVI data with a spatial resolution of 500 m. To investigate the temporal dynamics of vegetation in the Caatinga, annual mean NDVI values were calculated for each year from 2012 to 2024, allowing the identification of areas with distinct vegetation dynamics related to degradation, recovery, or seasonal cycles.

2.4. Heatwaves and Drought

To identify heatwaves, daily maximum and minimum temperature data from Xavier et al. [38] for the period 2012 to 2023 were used, corresponding to the temporal range available in the dataset. A heatwave is defined as a period in which the maximum and/or minimum temperatures exceed a certain threshold, usually based on percentiles [51,52,53,54]. In this study, the CTX90pct and CTN90pct indices were applied, meaning that a heatwave is a period of three or more consecutive days with daily maximum (minimum) temperatures above the 90th percentile. Once identified, these heatwaves were characterized in terms of frequency, total number during the analyzed period, and duration (days). Frequency was determined as the number of observed heatwave events, and duration as the average number of days per event [54].

To assess drought conditions in the study region, the Standardized Precipitation Index (SPI) [55] was used, due to its wide application in monitoring and forecasting precipitation deficits [56,57]. The SPI allows for drought detection at different temporal scales (1, 3, 6, 12, and 24 months), enabling monitoring of meteorological, agricultural, and hydrological droughts. The SPI is calculated as:

$$\mathrm{SPI} = {\displaystyle \frac{X-\mu }{\sigma }}$$

(5)

where X is the accumulated precipitation over the considered period, μ is the historical mean precipitation for the same period, and σ is the standard deviation of precipitation.

The Vapor Pressure Deficit (VPD) was used as an indicator of atmospheric dryness, closely related to wildfire risk. High VPD values indicate drier air, increasing vegetation transpiration and, consequently, susceptibility to ignition and fire spread [50,51].

For the calculation of VPD, air temperature at 2 m (T, $T_{\mathbf{2}m}$) and dew point temperature at 2 m ($T_d$, $d_{2m}$) from the ERA5 reanalysis were used for the period 2012 to 2023. Saturation vapor pressure ($e_{sat}$) and actual vapor pressure ($e_{td}$) were calculated from T and $T_d$, respectively, using Tetens’ equation [58,59]:

$$e_{sat}=6.112 \times exp\left({\displaystyle \frac{17.67\times T}{T+243.5}}\right)$$

(6)

$$e_{td}=6.112\times exp\left({\displaystyle \frac{17.67\times T_d}{T_d+243.5}}\right)$$

(7)

where $e_{sat}$ represents the saturation vapor pressure (hPa), $e_{td}$ is the actual vapor pressure (hPa), T denotes the air temperature (°C), and $T_d$ corresponds to the dew point temperature (°C). The Vapor Pressure Deficit (VPD) was then calculated as the difference between the saturation vapor pressure and the actual vapor pressure:

$$\mathrm{VPD} = e_{sat}-e_{td}$$

(8)

where VPD represents the vapor pressure deficit in hPa, reflecting atmospheric dryness.

To investigate the temporal relationships between fire activity and climate extremes, a cross-correlation analysis was performed between monthly anomalies of FRP and climate variables, including SPI-1, SPI-6, vapor pressure deficit (VPD), air temperature at 2 m (T2m), and heatwave frequency. Pearson correlation coefficients were calculated for temporal lags ranging from −6 to +6 months, allowing the identification of both antecedent and simultaneous relationships.

Figure 2 illustrates the overall framework of fire analysis in the Caatinga biome from 2012 to 2024. This schematic highlights the sequential steps, from data acquisition and preprocessing to clustering, trend analysis, and cross-correlation with climate extremes, providing a comprehensive overview of the analytical approach.

3. Results

3.1. Analysis of Fire Dynamics in the Caatinga Biome

Figure 3 shows the monthly and annual distribution of Fire Radiative Power (FRP) in the Caatinga biome for 2012 to 2024. A clear seasonality is observed (Figure 3a), peaking between August and November, with the highest recorded in October (~1000 MW/h). During these peak months, the FRP distributions exhibit higher medians (represented by the white boxes) and greater concentration of data at higher FRP levels (indicated by the wider parts of the violins). This pattern coincides with the dry season in the Caatinga, when the combination of low humidity, high temperatures, and scarce precipitation favors the occurrence and spread of more intense fires [41,60]. Similar results were observed by Ardila et al. [61].

In the annual behavior (Figure 3b), considerable variability in FRP intensity is observed across the years. Between 2012 and 2024, the years with the highest fire intensity in the Caatinga were 2015, 2019, 2021, and 2023, showing higher medians and wider, elongated violins, indicating that fires reflected a consistent trend of increasing intensity and frequency. These results agree with Bezerra et al. [19], who also identified FRP peaks in 2021 and a higher frequency of events in 2023, particularly during the months of September to November.

Figure 4 complements the analysis of Figure 3 by showing the cumulative spatial distribution of fire detection clusters and their annual evolution between 2012 and 2024. The spatial map (Figure 4a) highlights areas of recurrent fire concentration, mainly in the central-northern and western Caatinga, regions with more dense vegetation and higher accumulation of dry biomass, as well as intensive land use for extensive livestock farming and fire-based management practices [62]. We note that these areas may function as pyrogenic hotspots, where recurrent fires can lead to soil degradation and biodiversity loss [63].

The temporal analysis (Figure 4b) of the number of clusters shows a significant increasing trend over the series, as indicated by a positive slope of +1094.96 clusters/year and a p-value of 0.000, confirming that the rise in fire frequency is not random but statistically robust. These results corroborate the findings of Rocha et al. [63], who reported that 2021 had the largest burned area in the Caatinga Biome, with the western and central-northern regions exhibiting the highest cumulative burned area between 1985 and 2023.

The survival curves of wildfires by state within the Caatinga biome are presented in Figure 5. These curves depict the probability of a fire remaining active over time, providing an empirical representation of the duration and persistence of fire events. This analysis enables the comparison of the temporal behavior of fires among the states, highlighting differences in fire persistence and extinction dynamics, which are essential for understanding the characteristic fire regime of the Caatinga.

In general, it is observed that in all Caatinga states the probability of fire continuation decreases rapidly in the first few days, indicating that most events are short-lived. The state of Alagoas (Figure 5a), although presenting a maximum duration of slightly more than 80 days, showed the highest probability of a fire remaining active compared to the other states. This higher persistence is consistent with periods characterized by precipitation deficits and elevated atmospheric dryness, which favor sustained combustion, and may be further influenced by local meteorological conditions, vegetation characteristics, or land use patterns that support prolonged fire activity [24,41].

Some states within the Caatinga biome, such as Rio Grande do Norte (Figure 5h), Paraíba (Figure 5e), and Bahia (Figure 5b), exhibited longer survival tails of approximately 140, 130, and 100 days, respectively. These long-duration episodes, although relatively rare, tend to occur under extreme climatic conditions, particularly during persistent droughts and heatwaves, when high VPD and elevated air temperatures reduce fuel moisture and suppress fire extinction processes [12,64]. In addition, the high availability and continuity of combustible material resulting from vegetation degradation and the recurrent use of fire in agricultural practices may act as amplifying factors, enhancing fire persistence once favorable climatic conditions are established [63].

3.2. Spatial and Temporal Analysis of Extreme Events in the Caatinga Biome

Figure 6 shows the frequency and duration (in days) of heatwaves in the Caatinga Biome calculated from the maximum (CTX90) and minimum (CTN90) temperature indices, highlighting differences between the two indices. Heatwaves based on CTX90 are observed to be more frequent and prolonged, indicating a greater persistence of elevated maximum temperatures over time.

According to Figure 6a, based on the CTX90pct index, the years 2015, 2019, and 2023 exhibited the highest frequencies of heatwaves, with 809, 860, and 665 events, respectively. In terms of duration, these years also recorded the highest total number of days of occurrence, with 1645, 1145, and 910 days, respectively. These results are consistent with the findings of Bezerra et al. [28] for the Cerrado Biome and northeastern Brazil, which reported elevated frequencies and persistence of extreme heat events during specific years.

Similarly, Figure 6b, based on the CTN90pct index, indicates that the years 2013, 2016, and 2019 had the highest frequency peaks, with 750, 658, and 597 events, respectively, while the longest durations were recorded in 2013 (940 days), 2016 (902 days), and 2018 (849 days), indicating that both maximum and minimum temperatures contribute significantly to the occurrence of extended heatwaves in the region.

Figure 7 illustrates the spatial distribution of heatwave frequency and duration for both thresholds (CTX90 and CTN90), allowing the identification of regional occurrence patterns and highlighting the biome’s vulnerability to these climatic extremes. The analysis of heatwave frequency and duration for both indices shows that the areas of highest occurrence are mainly concentrated in the central region and, notably, in areas near the eastern coast. When comparing frequency, the CTX90 index (Figure 7a) shows a larger number of areas affected by high heatwave occurrences, whereas CTN90 (Figure 7b) reveals a more homogeneous distribution. Regarding duration, a critical aspect emerges: both indices, CTX90 (Figure 7c) and CTN90 (Figure 7d), demonstrate strong event persistence, with maximum durations of up to 54 days in several regions of the Caatinga.

In summary, the Caatinga faces not only recurrent heatwave events but also prolonged persistence of these extremes. These findings are consistent with Bezerra et al. [45], who reported similar patterns of high frequency and long duration along the northeastern coast. Moreover, climate projections suggest that global warming above 4 °C tends to intensify the occurrence and magnitude of extreme events, such as heatwaves, promoting the expansion of arid and semi-arid areas [65]. Thus, the results of Figure 6 underscore the need for climate adaptation strategies, such as integrated monitoring of extreme events, proactive public drought policies, and sustainable management of water resources, to reduce the Caatinga’s vulnerability to future climate change.

Figure 8 presents the climatology of heatwave frequency and duration over the period 2012 to 2023. The highest heatwave frequencies, according to the CTX90 index (Figure 8a), occurred in October (1372), November (1239), and December (1135). This coincides with the main fire season in the Caatinga Biome, with October being the peak month, as shown in Figure 3a. In terms of duration, the highest records were observed in November (1545), December (1446), and March (1097), likely due to the combination of persistent high temperatures and low humidity levels, typical of the dry season in the northeastern semi-arid region [54,61,65]. In Figure 8b, the climatology for the CTN90 index shows that the highest frequency records occurred in December (938), March (866), and November (820). Regarding duration, peaks were observed in March (1127), November (1060), and January (1052), demonstrating the persistence of heatwaves even outside the peak of the dry season [65,66].

Figure 9 presents the conditions of water deficit in the Caatinga Biome, represented by SPI-1 (Figure 9a) and SPI-6 (Figure 9b), where negative values indicate precipitation scarcity, while positive values reflect pluviometric surplus. A clear seasonal variation is observed, which is characteristic of the high irregularity of rainfall in the Caatinga [20,26,65]. This variability highlights the alternation between periods of intense drought and episodes of above-average precipitation, typical of the semi-arid regime of the region [67].

The most pronounced water deficits were observed mainly between August and October, coinciding with the end of the dry season, the peak fire season (Figure 3a), and the occurrence of heatwaves (Figure 9a), when soil moisture reaches its lowest levels and vegetation becomes more vulnerable to water stress [60,68,69,70].

These results corroborate previous studies that highlight the strong seasonality of the water balance in the Caatinga, with a short rainy period concentrated between February and May, followed by a prolonged dry period extending until the beginning of spring [20,71].

3.3. Relationship Between Fire and Extreme Events

The cross-correlation analysis (Figure 10) shows different temporal responses of FRP in relation to the main climate indices evaluated for the Caatinga biome. It is observed that the vapor pressure deficit (VPD) and the mean air temperature (T2m) presented positive and significant correlations at lag zero, with maximum coefficients of approximately r = 0.56 and r = 0.44, respectively. This behavior indicates that hotter and drier conditions are directly associated with increased fire intensity, reflecting the immediate role of thermodynamic variables in the ignition and propagation of fires [11,14,28,72].

On the other hand, the standardized precipitation indices (SPI1 and SPI6) exhibited maximum negative correlations with a lag of about four months, suggesting that periods of water deficit are preceded by peaks in fire activity. This result reinforces the influence of antecedent drought conditions, which promote the accumulation of dry biomass and reduce fuel moisture, creating an environment conducive to the occurrence of fires [72,73,74].

The frequency of heatwaves (HWfreq) showed a positive correlation, but of lower magnitude compared to the other variables, indicating a simultaneous association between the occurrence of extreme temperature events and the intensification of FRP. Although the isolated effect of heatwaves on the monthly variability of fire is not statistically dominant, their temporal coincidence with peaks in FRP activity suggests an amplifying role, enhancing the thermodynamic conditions already favorable to the ignition and propagation of fires [14,28].

Overall, the results demonstrate that fire activity in the Caatinga is mainly modulated by the interaction between antecedent drought conditions (SPI) and current thermodynamic conditions (VPD and Temperature). The negative lags observed for precipitation indices highlight the precursor role of drought in the accumulation of fuel material, while the simultaneous positive correlations with VPD and temperature emphasize the direct influence of the atmospheric conditions of the event month on the intensification of fires.

Figure 11 complements the analysis presented in Figure 10 by correlating FRP and the main climate indices over the period from 2012 to 2023. While Figure 10 summarizes the average response of the variables at different temporal lags, highlighting the antecedent role of drought and the immediate influence of thermodynamic conditions, Figure 11 allows observing the seasonal variability of these relationships, showing how the patterns of association between fire and climate change over time.

It is possible to notice in Figure 11a a significant positive correlation between FRP and the frequency of monthly heatwaves (r = 0.62, R² = 0.38, p = 0.0324), indicating that months with a higher occurrence of heat events present greater fire intensity. This result highlights the direct influence of extreme temperature conditions on the spread of fires, possibly due to the reduction in water content in vegetation and increased biomass flammability [14,75,76].

The correlation between FRP and the SPI indices shows a significant negative correlation (r = −0.69, R² = 0.48, p ≈ 0.012) for both SPI-1 (Figure 11b) and SPI-6 (Figure 11c). This trend suggests that periods with water deficit, represented by negative SPI values, are associated with higher FRP values.

These results also agree with Bezerra et al. [28], who found positive correlations between FRP and heatwaves, as well as negative correlations between FRP and drought indices in the Cerrado Biome. Comparatively, while Figure 10 provides an overview of the average correlations and their lags, Figure 11 reveals how these correlations change over time, showing that the intensity and the sign of the relationship vary according to the strength and duration of extreme climate events.

Figure 12 presents the annual NDVI values in the Caatinga Biome during the period from 2012 to 2024. In general, it is observed that areas with sparser or more degraded vegetation (low and negative NDVI values) predominate in the central and eastern portions of the biome.

Throughout the period from 2012 to 2019, a pronounced stress or degradation of vegetation is observed in the central and northern portions of the Caatinga Biome, evidenced by low NDVI values, possibly associated with a period marked by severe drought [77]. Between 2020 and 2023, a more expressive improvement in NDVI is noted, especially in the central zone, indicating more favorable conditions for vegetation development, despite the persistence of stress areas in the north. In 2024, recovery seems to become more evident, with a significant increase in NDVI across much of the biome, although some isolated areas continue to show low indices. This is likely reflecting local vulnerabilities and possible persistent climatic or anthropogenic impacts [41,78].

4. Discussion and Conclusions

This study advances the understanding of wildfire dynamics in the Caatinga biome by framing fire activity as the outcome of interacting climate extremes operating across distinct temporal scales. Rather than being driven by isolated meteorological anomalies, wildfires in the Caatinga emerge from the coupling between antecedent hydroclimatic stress and contemporaneous thermodynamic conditions. This perspective aligns with recent advances in climate–fire research, which emphasize compound and cascading processes as dominant drivers of extreme fire behavior, particularly in semiarid regions [14,15,28].

Based on the patterns identified in this study, drought conditions emerge as a fundamental preconditioning factor in the Caatinga fire regime. The temporal relationships observed between precipitation deficits and subsequent fire activity indicate that prolonged drought progressively reduces soil and vegetation moisture, increases fuel continuity, and promotes the accumulation of dry biomass prior to the fire season. These slow-onset processes establish a background state of enhanced flammability that precedes periods of intensified fire activity. Similar drought-driven preconditioning mechanisms have been reported in other fire-prone regions, where antecedent moisture deficits regulate fuel availability and spatial coherence, thereby increasing landscape susceptibility to ignition and fire spread [4,9,10].

Superimposed on this drought-driven conditioning, short-term atmospheric processes exert an immediate control on fire behavior. Elevated air temperature and increased vapor pressure deficit intensify vegetation water stress, accelerate fuel desiccation, and enhance combustion efficiency at the time of fire occurrence. This interaction between slow and fast climate drivers reflects a hierarchical structure in wildfire controls, in which drought sets the stage and heat-related variables regulate fire intensity and persistence [39,79,80]. Such a mechanism is consistent with conceptual models developed for semiarid and Mediterranean-type ecosystems, where atmospheric dryness acts as a critical proximate trigger once fuels are available [11,14].

Heatwaves represent an additional amplifying factor within this framework. Although heatwaves alone do not necessarily determine fire occurrence, their concurrence with drought and elevated atmospheric dryness enhances fire intensity and reduces extinction probability by intensifying fuel desiccation and thermal stress. This synergistic behavior is characteristic of compound extreme events, in which the co-occurrence of heatwaves and droughts produces disproportionate impacts on wildfire behavior, particularly in semiarid and Mediterranean-type environments [9,11,14,15].

The increasing recurrence of fire clusters suggests a gradual modification of the fire regime in the Caatinga biome, consistent with changes reported across different regions of Brazil under increasing climate variability and environmental stress [11,12,34]. Although fire is not an intrinsic ecological driver in this system, as the Caatinga is classified as a fire-independent biome, ignition sources are predominantly associated with human activities, particularly the use of fire for agricultural management and land preparation, which is widely documented in Brazilian landscapes, including the semiarid Northeast [20,33].

From an ecosystem perspective, recurrent fires under extreme climatic conditions may exceed the adaptive capacity of Caatinga vegetation. Unlike fire-adapted biomes, such as the Cerrado, repeated burning in the Caatinga can lead to long-term degradation, biodiversity loss, and reduced ecosystem services, including carbon storage and soil protection [21,31]. Evidence from remote sensing studies suggests that vegetation recovery in semiarid environments is strongly constrained by post-fire climatic conditions, with drought prolonging recovery times and, in some cases, promoting persistent shifts in vegetation structure [29,36].

The findings of this study have direct implications for fire management and climate adaptation strategies in the Caatinga biome. Effective mitigation requires moving beyond reactive approaches focused solely on fire suppression and toward anticipatory strategies that explicitly consider the climatic processes conditioning fire activity in fire-sensitive ecosystems, as discussed in the fire management and climate adaptation literature [11,14,21]. Based on the results presented here, it is suggested that integrated early-warning systems prioritize periods following prolonged droughts and incorporate real-time indicators of atmospheric dryness and extreme heat, thereby improving the identification of high-risk conditions and supporting preventive decision-making within the Caatinga biome [18].

In conclusion, wildfire activity in the Caatinga biome is best understood as the result of interacting climate extremes acting across multiple temporal scales. Antecedent drought conditions govern fuel availability, while concurrent heat and atmospheric dryness modulate fire intensity and persistence. Under projected climate change scenarios for northeastern Brazil, which indicate an intensification of both droughts and heatwaves, the frequency of compound extreme events and associated fire impacts is expected to increase [12,54]. These trends reinforce the urgency of developing adaptive fire management strategies tailored to semiarid, fire-sensitive ecosystems, in which climatic pressures interact with human ignition sources to shape emerging fire regimes [14,72].