Ítem
Acceso Abierto
Modelado predictivo de la ocurrencia de focos de incendios forestales en Colombia mediante técnicas de machine learning
Título de la revista
Autores
Novoa Cardozo, Mateo
Fecha
2026-06-19
Directores
Sánchez Salazar, Fabián
ISSN de la revista
Título del volumen
Editor
Universidad del Rosario
Buscar en:
Métricas alternativas
Resumen
Este trabajo desarrolla un modelo de machine learning para predecir la ocurrencia semanal de focos de incendios forestales en Colombia, a partir de variables climáticas, geográficas e históricas. Los resultados muestran que la información climática rezagada permite anticipar el riesgo de incendio con una semana de antelación, aportando una herramienta útil para la prevención y la gestión del riesgo.
Abstract
Wildfires have become an increasing risk to ecosystems and communities in Colombia, particularly under conditions of climate change and intensive land use. In response to this challenge, this study develops a model for the weekly prediction of wildfire hotspot occurrence, formulated as a binary classification problem using machine learning techniques.
The model is built using historical municipal-level data from Colombia covering the 2010–2023 period, integrating climatic variables such as relative humidity, precipitation, and temperature, together with historical wildfire records and geographic features. Among the evaluated models, LightGBM achieved the best performance, reaching an AUC of 0.85 and an F1 score of 76% under temporal validation. In addition, the variable importance analysis showed that maximum temperature, minimum relative humidity, and maximum precipitation, with lags ranging from one to three weeks, were the most relevant predictors. Overall, these results show that climatic information can help anticipate wildfires one week in advance, providing a useful tool for prevention and risk management in Colombia.
Palabras clave
Incendios forestales , Machine learning , Predicción de incendios , Variables climáticas , Colombia , LGBM
Keywords
Forest fires , Machine learning , Spatiotemporal prediction , Climatic variables , Colombia
