Cerrado Fire Regimes Atlas
Introduction
This product is an integration of different fire metrics derived from the monthly MapBiomas Fire product in a landscape scale (~9 km).
Goal of the Product
The Cerrado Fire Regimes Atlas is designed as a strategic resource for fire ecologists, pyrogeographers, environmental managers, and policymakers.
While traditional monitoring systems focus on where and when fire occurs (burned area), this project aims to decipher the complex pyrogeography of the biome. By analyzing 40 years of historical data (MapBiomas Fire, 1985-2024) through the lens of Landscape Ecology and Machine Learning, this atlas provides a structural classification of fire activity.
Primary Objectives:
Advance Pyrogeographical Understanding: To support the scientific community in moving beyond binary “burned/unburned” analyses by quantifying distinct Fire Regimes—recurring patterns of frequency, seasonality, size, and shape—that drive ecosystem dynamics.
Distinguish Natural vs. Anthropogenic Fire: To differentiate between the natural, ecological fire cycles of the open savannas (Regime 3) and the anthropogenic, degradation-driven fires at the agricultural frontier (Regime 5).
Track Regime Shifts: To visualize where fire regimes are changing over time (e.g., areas shifting from natural cycles to suppression or intensification) due to land use change and climate variability.
Support Conservation Planning: To identify priority areas where the current fire regime diverges from the historical baseline, guiding prescribed burning or suppression efforts.
What this Product is NOT
It is equally important to understand the limitations of this tool to ensure its appropriate application.
🚫 It is NOT a Fire Intensity or Severity Map: This atlas classifies regimes based on the spatiotemporal structure of fire scars (e.g., frequency, seasonality, patch shape, fragmentation). It does not directly measure Fire Radiative Power (FRP) or spectral indices of burn severity (e.g., dNBR).
- Note: While structural metrics like Mean Core Area are often strongly correlated with intensity (larger, more solid fires tend to burn hotter), this tool infers regime characteristics from pattern, not from direct energy measurements.
🚫 It is NOT a Real-Time Alert System: This tool does not detect fires happening today. For active fire monitoring and emergency response, please refer to the operational systems provided by INPE (Queimadas) or NASA FIRMS.
🚫 It is NOT a Fire Spread Model: It classifies historical patterns. It does not simulate the physics of fire spread for specific ignition events based on current weather conditions.
🚫 It is NOT a Future Forecast: While it visualizes historical trajectories (1985–2024), it describes the state of the fire regime, not the probability of next year’s fire.
Data Sources and Processing
Fire Occurrence and Metrics
Processing
Fire Scars: Individual fire events were identified from the monthly data in a 9 km- cells grid.
Landscape Metrics: For each fire scar, a suite of landscape metrics was calculated to characterize its spatial pattern. To ensure statistical robustness, variable selection was performed to remove highly collinear metrics (using VIF or correlation thresholds).
Aggregation: The metrics were averaged over the entire study period (1985-2024) and also aggregated by decade to capture temporal trends.
Final Metrics
The selected metrics include:
| Metric Type | Metric Name | Ecological Interpretation (What it Measures) |
| Fire Abundance | Fire Scar Density | Measures the number of individual fire scars per unit of area. It reflects ignition patterns and landscape constraints on fire spread. High values suggest frequent ignitions or a fragmented landscape that prevents scars from becoming large. Low values indicate fewer, potentially larger fire events. |
| Fire Configuration | Fire Scar Fragmentation | Measures how the total burned area is broken up into separate, smaller scars. It reflects the continuity of fuel and burn conditions. A high value indicates a highly fragmented burned landscape, suggesting discontinuous fuels or varied fire behavior. A low value points to a few large, contiguous fire scars. |
| Fire Scar Shape | Fire Scar Shape Regularity | Describes how compact or convoluted fire scar shapes are on average. It’s an indicator of how uniformly fire spreads. Values closer to 1 indicate simple, compact scars, typical of fires in uniform fuel with no wind. Values closer to 0 indicate highly irregular shapes, presumably driven by wind, topography, or fuel heterogeneity. |
| Fire Configuration | Fire Scar Isolation Variation | Measures the variability in the distance between a fire scar and its nearest neighboring scar. It describes the spatial clustering of fire activity. A high value suggests an uneven spatial pattern where some scars are tightly clustered while others are very isolated. A low value indicates that fire scars are more evenly spaced. |
| Fire Scar Interior | Mean Fire Scar Core Size | Measures the average size of the “core” or interior area of fire scars, buffered from the unburned edge. It represents the average extent of uninterrupted, high-intensity burning. Larger values indicate larger, more completely burned scars, while smaller values suggest smaller or narrower scars with more edge influence. |
| Fire Scar Interior | Fire Scar Integrity Variation | Measures the variability in the proportion of each fire scar that is considered “core” area. It captures the consistency of burn completeness across different scars. A high value indicates a mix of scar types: some are solid with large cores, while others are “hollow” or linear with very little core area. |
| Fire Scar Interior | Fire Scar Cohesion Variation | Measures the variability in the internal fragmentation of individual fire scars. Specifically, it quantifies the variation in the number of disjunct core zones within fire patches. A high value indicates a landscape with heterogeneous fire behaviors, where some scars are solid and cohesive (single core) while others are structurally complex and broken into multiple disjoint interiors (multiple cores). |
| Fire frequency | Fire occurrence (monthly) | Measures if the cell was burned or not (binary, 0 or 1). When averaged across years, represents the proportion of years a pixel burned in each month (Jan-Dec). |
Classification Method of Fire Regimes
Algorithm: Model-based Clustering (mclust) was used to classify pixels into distinct fire regimes based on their fire metrics and monthly frequencies.
Model Selection: The VVV (ellipsoidal, varying volume, shape, and orientation) model with 5 components (clusters) was selected based on the Bayesian Information Criterion (BIC), which showed a plateau in performance improvement after 5 clusters, indicating an optimal balance between model complexity and fit.
Characterization of Fire Regimes
We crossed the fire regime classification with land use information and the fire metrics that predicted them. Here is a summary of the fire regimes characteristics:
| Fire Regime | Name | Description | Fire Metrics | Landscape context | Socioecological interpretation |
| 1 | Anthropogenic Suppression | This regime is characterized by very rare and suppressed fire activity. | All fire metrics (frequency, density, size) are below average (blue in the heatmap), except for fire shape regularity and fragmentation. | Dominated by pasture, and agricultural areas with high population density and anthropic use. | This represents the consolidated agricultural matrix where fire is actively suppressed to protect assets (crops, infrastructure) and due to the lack of continuous flammable vegetation. |
| 2 | Grassland Wildfire | This regime features infrequent but large and intense wildfires. | Characterized by high mean core area (lsm_c_core_mn) and core variability (lsm_c_dcore_sd), indicating large, continuous burn scars when fires do occur. Monthly frequency is generally low or average, but relatively higher in September (late dry season). | Associated with native open (grasslands, savannas, wet herbaceous) and native vegetation areas with low population density. However, these areas commonly share the use for temporary crops. | These are likely large natural remnants or protected areas where fuel accumulates over longer periods. When fires occur, the continuous grassy fuel allows them to spread extensively, creating large, solid scars. |
| 3 | Pyrophilic Savanna | This is the most active fire regime, characterized by frequent, fragmented burning throughout the year. | High values for fire density (lsm_c_pd), isolation variation (lsm_c_enn_sd), cohesion variation, and fire frequency across almost all months (Jan-Dec). | Strongly dominated by savanna and grassland formations, with lowest population density and antropic use. | This represents the core “fire-prone” savanna biome. The high frequency and fragmentation reflect the natural flammability of the vegetation and potentially the influence of human ignitions that are frequent but spatially limited by the mosaic of fuel types. |
| 4 | High-Biomass Ecotone | A regime defined by seasonal constraints and high biomass. | Shows a distinct seasonal avoidance pattern (lower frequency in peak dry months), high shape regularity (lsm_c_circle_mn) and patch density, and low isolation variation. | Defined by high NDVI (biomass) and low native vegetation and moderate population density. Located in the edges of the Cerrado. | These are ecotonal systems, with low dominace of open formations (grasslands and savannas) and with significant antropic pressure and degradation. Fire scars are abundant, large in extent and with regular shapes. |
| 5 | Frontier Interface | A transitional regime with variable and scattered fire activity, dominated by forest formation, mosaic uses and non vegetated areas. | Metrics are generally average or “messy,” with high variability in isolation (lsm_c_enn_sd) and integrity variation. Mean core areas of fires are lower than average (lsm_c_core_mn). | Characterized by a mix of forest, mosaic uses, and non vegetated areas. Relatively high biomass (NDVI). | This corresponds to the agricultural frontier (e.g., Matopiba) or the Amazon-Cerrado ecotone. Fire activity here is driven by the interaction between deforestation/clearing fires and the remaining forest edges, resulting in a spatially heterogeneous pattern. |
The Atlas
Interact with the Atlas! To download the original rasters. Access the GitHub or Zenodo pages (available soon). For more customizable downloads, see the Shiny App I created (available soon).
Map Legends
Metric Legends
