Prediction of Burned Areas Using the Random Forest Classifier in the Minas Gerais State
Tipo de material:
ArtigoAssunto(s): Recursos online:
Em: Floresta e Ambiente (Brazil) v. 27(3) p. 1-7; (2020)Sumário:
Abstract
Fire behavior prediction models can assist environmental agencies with fire prevention and control. This study aimed
to adjust a fire prediction model for the state of Minas Gerais, Brazil. Using the R program and hotspots provided
by the National Institute for Space Research (INPE) for 2010, prediction of the probability of fires through the
Random Forest algorithm was conducted using the Bootstrapping method. The model generated a prediction map
with global kappa value of 0.65. External validation was performed with hotspots in 2015. Results showed that 58%
of the hotspots are in areas with ignition probability > 50%, being 24% of them in areas with 25-50% probability,
and 17% in areas with < 25% probability. These results were considered satisfactory, demonstrating that the model
is suitable for predicting fires.
Keywords: fires; modeling; environmental monitoring.
| Tipo de material | Biblioteca atual | Coleção | Número de chamada | Informaçaõ do volume | Situação | Devolução em | Código de barras |
|---|---|---|---|---|---|---|---|
Periódicos
|
Biblioteca Nacional de Agricultura - Binagri Agrobase - Periódicos | Periódicos agrícolas | 2020 27(3) | Online | 2025-0457 |
Publicação on-line; 19 ref.; 1 table; 5 illus.; Summary (En)
Abstract
Fire behavior prediction models can assist environmental agencies with fire prevention and control. This study aimed
to adjust a fire prediction model for the state of Minas Gerais, Brazil. Using the R program and hotspots provided
by the National Institute for Space Research (INPE) for 2010, prediction of the probability of fires through the
Random Forest algorithm was conducted using the Bootstrapping method. The model generated a prediction map
with global kappa value of 0.65. External validation was performed with hotspots in 2015. Results showed that 58%
of the hotspots are in areas with ignition probability > 50%, being 24% of them in areas with 25-50% probability,
and 17% in areas with < 25% probability. These results were considered satisfactory, demonstrating that the model
is suitable for predicting fires.
Keywords: fires; modeling; environmental monitoring.
BINAGRI