IgniWise Dataset — Prescribed Burn Window Prediction for Spain (1983–2015)

Read full text Summarise with VIRAT
Authors: Sergio Romera Martínez
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Machine learning dataset for predicting safe prescribed burn windows across 48 Spanish provinces, supporting the IgniWise open-source system. Dataset contents:- training_data.csv: 11,996 records derived from historical fire occurrence data (MITECO / IEPNB, 1983–2015) with 20 features including meteorological variables (statistically approximated from Spanish climate distributions), FWI indices (Van Wagner, 1987), and real geographic features per province- random_forest_v1.pkl: Trained Random Forest model (200 estimators, 5-fold CV accuracy: 99.5% ± 0.002)- provincias_geo.geojson: Spanish province geometries enriched with real topographic (Copernicus DEM GLO-30), vegetation (Sentinel-2 NDVI via GEE), and land cover features (CORINE Land Cover 2018) Note on meteorological features: weather variables in training_data.csv are statistical approximations calibrated to provincial climate. Operational predictions use real-time data from OpenWeatherMap. The 99.5% CV accuracy reflects model fit to the classification rules rather than independent real-world validation, which requires field data from executed prescribed burns. Sources: MITECO/IEPNB (fire occurrences), Copernicus DEM GLO-30 (topography),Sentinel-2/GEE (NDVI), CORINE Land Cover 2018 (vegetation cover).FWI methodology: Van Wagner, C.E. (1987), Canadian Forest Service. License: CC BY 4.0 | Web: igniwise.com | Code: github.com/TrueRomanZe/igniwise

Fields of Study & Subjects

Fire effects on ecosystems Rangeland and Wildlife Management Forest Management and Policy

Publication details

Publication date
21 March 2026
Language
ES
Document type
Dataset