VIIRS 375 m active-fire detections and fire radiative power in the four biomes of Northeast Brazil (2020–2025): processed data and R scripts
Authors/Creators
Description
# VIIRS 375 m active-fire detections and fire radiative power in the four biomes of Northeast Brazil (2020–2025): processed data and R scripts
Version 1.0
This dataset contains the processed data and R scripts used in:
> Araújo, L.B., Olinda, R.A., Barbosa, D.S., Angeoletto, F. Spatial clustering and radiative power of active fires across the four biomes of Northeast Brazil (2020–2025). Submitted to *Remote Sensing Applications: Society and Environment*.
The citation and DOI of the article will be added here after publication.
## Authors
- Lucas Barboza de Araújo, Universidade Estadual da Paraíba (UEPB), Campina Grande, Brazil
- Ricardo Alves de Olinda, Universidade Estadual da Paraíba (UEPB), Campina Grande, Brazil
- Domingos Sávio Barbosa, Universidade Federal de Rondonópolis (UFR), Laboratório de Tecnologia e Gestão Ambiental (LTGA), Rondonópolis, Brazil. Contact: domingos@ufr.edu.br
- Fabio Angeoletto, Universidade Federal de Rondonópolis (UFR), Rondonópolis, Brazil; PUMAS – Pessoas Unidas pelo Meio Ambiente, Agricultura & Sociedade, Chapada, Brazil
## Contents
```
README.md
LICENSE_data.txt CC BY 4.0 (data)
LICENSE_code.txt MIT (scripts)
data/
viirs_active_fire_NE_Brazil_2020-2025.csv 307,129 detections
biomes_NE_Brazil_IBGE2019.gpkg 4 biome polygons (layer "biomes")
data_dictionary.csv column definitions
scripts/
00_load_data.R shared data loading (sourced by the others)
01_descriptive_figures.R monthly and annual summaries (Figs. 3–5)
02_temporal_tests.R Friedman, sign, Mann–Kendall and Sen tests (Table 2, Section 3)
03_maps_kernel_density.R kernel density by half-year (Fig. 6) and Table 1
04_point_pattern_functions.R Ripley's K, J-function and mark correlation (Figs. 7, 8, S1)
```
## Data sources
**Active-fire detections.** Data were downloaded from the INPE BDQueimadas database (https://terrabrasilis.dpi.inpe.br/queimadas/bdqueimadas/). The dataset uses the VIIRS 375 m active-fire product (Schroeder et al., 2014) and covers 1 January 2020 to 31 December 2025. It includes every detection in the nine states of Northeast Brazil.
The original export was split into 12 half-year files. The authors kept only the month, state, biome, coordinates and fire radiative power (FRP) of each detection. In this deposit, the 12 files are merged into one table without removing or changing any record. The only changes are these:
- columns renamed and reordered;
- year, half-year and month number added from the original file names;
- state names converted to their two-letter codes;
- coordinates rounded to 5 decimal places, about 1 m, which removes floating-point artefacts;
- the column `biome_ibge2019_overlay` added (see below).
**Biome polygons.** The polygons come from IBGE (2019), *Biomas e Sistema Costeiro-Marinho do Brasil, 1:250 000*, clipped to Northeast Brazil. They are distributed in the projected coordinate system used in the analyses: Albers equal-area conic on the GRS80 ellipsoid, with standard parallels at 2°S and 15°S, latitude of origin at 7°S and central meridian at 43°W (`+proj=aea +lat_1=-2 +lat_2=-15 +lat_0=-7 +lon_0=-43 +ellps=GRS80 +units=m`). The original IBGE data are available at https://www.ibge.gov.br/geociencias/informacoes-ambientais/.
## Main table: `viirs_active_fire_NE_Brazil_2020-2025.csv`
UTF-8, comma-separated, decimal point, one row per detection. See `data/data_dictionary.csv`.
| Column | Description |
|---|---|
| `year` | Year of detection (2020–2025) |
| `half` | 1 = January–June, 2 = July–December |
| `month` | Month of detection (1–12) |
| `state` | State abbreviation (AL, BA, CE, MA, PB, PE, PI, RN, SE) |
| `biome` | Biome as assigned by BDQueimadas: Amazon, Atlantic Forest, Caatinga, Cerrado. **All analyses use this column.** |
| `biome_ibge2019_overlay` | Biome from point-in-polygon overlay with the unsimplified IBGE (2019) polygons in this deposit. "outside" = point outside all four polygons (33 points). Added for transparency and not used in the analyses. |
| `longitude`, `latitude` | Geographic coordinates of the pixel centre, in decimal degrees, WGS84 (EPSG:4326) |
| `frp_mw` | Fire radiative power (MW) |
**Known limitations of the data:**
- Detections from **March of 2021, 2022, 2023 and 2024 were not included** in the analysis due to data unavailability. March is included in 2020 (198 detections) and 2025 (591 detections).
- The export kept the month but not the day, time, satellite (S-NPP, NOAA-20, NOAA-21) or confidence level. So the data cannot be aggregated into fire events or restricted to a single satellite.
- Five detections have FRP = 0 MW. They were kept.
- 35 pairs of detections share year, month and coordinates. They may be detections on different days and were kept.
- The BDQueimadas biome attribute (`biome`) and the overlay with the IBGE (2019) polygons (`biome_ibge2019_overlay`) disagree for 24,405 detections (7.9%). Most of these lie along the Cerrado–Caatinga boundary.
## How to reproduce
Requirements:
- R 4.3 or later;
- packages: `sf`, `dplyr`, `tidyr`, `ggplot2`, `ggspatial`, `scales`, `raster`, `rmapshaper`, `spatstat` (including `spatstat.geom` and `spatstat.explore`).
Set the working directory to the root folder of this dataset and run:
```r
source("scripts/01_descriptive_figures.R")
source("scripts/02_temporal_tests.R") # base R; ~10 s
source("scripts/03_maps_kernel_density.R") # ~1 min
source("scripts/04_point_pattern_functions.R") # several hours
```
All results are written to `outputs/`. `02_temporal_tests.R` reproduces exactly the values reported in the article, including:
- annual totals of 40,446 detections in 2020 and 60,441 in 2025;
- Friedman χ² = 18.0 (W = 1.00) for counts and 16.0 (W = 0.89) for FRP;
- for the Amazon trend, τ = 1.00, p = 0.003, Sen's slope = 646 detections per year;
- a seasonality sign test with p = 0.031 in every biome.
### Analysis settings
- The spatial analyses use the projected coordinate system described above. The biome polygons are simplified with `rmapshaper::ms_simplify(keep = 0.05)`.
- **Table 1** counts detections by the BDQueimadas biome attribute, restricted to points inside the simplified polygons. Because of this, it sums to 307,076 detections, compared with 307,129 in the full table.
- **Kernel density (Fig. 6):**
- Gaussian kernel with edge correction;
- bandwidth = mean of Scott's-rule bandwidths over the 12 half-years (≈ 71 km);
- result expressed in detections per km².
- **K-function, J-function and mark correlation (Figs. 7, 8 and S1):**
- detections pooled over 2020–2025 by biome, with the biome polygon as the window;
- distances r = 0–3,000 m in steps of 50 m;
- pointwise envelopes from 40 simulations of complete spatial randomness inside each polygon;
- edge corrections: border for K, Kaplan–Meier for J.
- For K and J, points are selected by the BDQueimadas attribute, and points outside the polygon of that biome are discarded by `spatstat`. For the mark correlation, points are selected by spatial intersection with the polygon.
- Simulation envelopes depend on the random seed (`set.seed(2026)`), so repeated runs give slightly different envelopes.
## Licences
- Data (`data/`): Creative Commons Attribution 4.0 International (CC BY 4.0).
- Code (`scripts/`): MIT License.
- The original data belong to INPE (BDQueimadas) and IBGE. Please cite them as well.
## How to cite
Cite the article and this dataset:
> Araújo, L.B., Olinda, R.A., Barbosa, D.S., Angeoletto, F., 2026. VIIRS 375 m active-fire detections and fire radiative power in the four biomes of Northeast Brazil (2020–2025): processed data and R scripts (v1.0). Zenodo. https://doi.org/10.5281/zenodo.XXXXXXX
## References
IBGE, 2019. Biomas e Sistema Costeiro-Marinho do Brasil: Compatível com a Escala 1:250 000. Instituto Brasileiro de Geografia e Estatística, Rio de Janeiro.
Schroeder, W., Oliva, P., Giglio, L., Csiszar, I.A., 2014. The New VIIRS 375 m active fire detection data product: algorithm description and initial assessment. Remote Sens. Environ. 143, 85–96. https://doi.org/10.1016/j.rse.2013.12.008
## Funding
This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001.
## Added in v1.1: distance from biome boundaries
`scripts/05_boundary_distance_bands.py` (Python 3; pandas, geopandas, shapely ≥ 2, pyogrio, scipy) computes the distance of each detection to the Cerrado–Caatinga boundary and, as a contrast, to the Caatinga–Atlantic Forest boundary (shared outlines of the unsimplified IBGE 2019 polygons, Albers equal-area projection). The side of the boundary is the IBGE polygon containing each detection (`biome_ibge2019_overlay`). Detection density (detections 100 km⁻² yr⁻¹) is computed per year in distance bands (0–10, 10–25, 25–50, 50–100, 100–200, >200 km), and the ratio between density within 25 km and beyond 100 km is tested across years with an exact sign test (n = 6).
Run from the dataset root: `python scripts/05_boundary_distance_bands.py` (about 5 min). Outputs in `outputs/`: `boundary_distance_bands_by_year.csv`, `boundary_distance_bands_summary.csv` (Table 3 of the article) and `boundary_near_vs_interior_ratio.csv`.
## Added in v1.1: random labelling and inhomogeneous K-function
`scripts/06_markcorr_rlabel_kinhom.py` (Python 3; numpy, pandas, geopandas, shapely ≥ 2, scipy) re-implements two spatstat procedures, per biome, on three independent random subsamples of detections inside the unsimplified IBGE (2019) polygons (random thinning leaves both functions unchanged in expectation):
1. **Mark correlation of FRP, k_mm(r), under random labelling** (FRP permuted among the observed locations; 199 permutations; subsamples of 8,000 detections; r = 0.25–3 km in 125 m bins), with pointwise 95% envelopes and a global maximum-deviation test.
2. **Inhomogeneous K-function, K_inhom(r)**, with a leave-one-out Gaussian kernel intensity (bandwidth 10 km, 2 km grid) and border edge correction, compared with the minimum and maximum of 39 simulations of an inhomogeneous Poisson process with the fitted intensity (subsamples of up to 15,000 detections).
Run from the dataset root: `python scripts/06_markcorr_rlabel_kinhom.py` (about 45 min). Outputs: `outputs/markcorr_random_labelling.csv` and `outputs/kinhom_inhomogeneous_poisson.csv` (Supplementary Fig. S2 of the article). Random seed fixed (20261106).
Files
NE_Brazil_fire_VIIRS_2020-2025_v1.1.zip
Additional details
Funding
- Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES)
Dates
- Collected
-
2026Data
Software
- Programming language
- R