Post-Fire Burned Area Detection Using Machine Learning and Burn Severity Classification with Spectral Indices in İzmir: A SHAP-Driven XAI Approach DOI Creative Commons
Halil İbrahim Gündüz, Ahmet Tarık TORUN, Cemil Gezgin

et al.

Fire, Journal Year: 2025, Volume and Issue: 8(4), P. 121 - 121

Published: March 21, 2025

This study was conducted to precisely map burned areas in fire-prone forest regions of İzmir and analyze the spatial distribution wildfires. Using Sentinel-2 satellite imagery, burn severity first classified using dNBR dNDVI indices. Subsequently, machine learning (ML) algorithms—RF, XGBoost, LightGBM, AdaBoost—were employed classify unburned areas. To enhance model performance, hyperparameter optimization applied, results were evaluated multiple accuracy metrics. found that RF achieved highest with an overall 98.0% a Kappa coefficient 0.960. In comparison, classification based solely on spectral indices resulted accuracies 86.6% (dNBR) 81.7% (dNDVI). A key contribution this is integration Explainable Artificial Intelligence (XAI) through SHapley Additive exPlanations (SHAP) analysis, which used interpret influence environmental variables area classification. SHAP analysis made decision processes transparent identified dNBR, dNDVI, SWIR/NIR bands as most influential variables. Furthermore, analyses confirmed variations reflectance across fire-affected are critical for accurate delineation, particularly heterogeneous landscapes. provides scientific framework post-fire ecosystem restoration, fire management, disaster strategies, offering decision-makers data-driven effective intervention strategies.

Language: Английский

Post-Fire Burned Area Detection Using Machine Learning and Burn Severity Classification with Spectral Indices in İzmir: A SHAP-Driven XAI Approach DOI Creative Commons
Halil İbrahim Gündüz, Ahmet Tarık TORUN, Cemil Gezgin

et al.

Fire, Journal Year: 2025, Volume and Issue: 8(4), P. 121 - 121

Published: March 21, 2025

This study was conducted to precisely map burned areas in fire-prone forest regions of İzmir and analyze the spatial distribution wildfires. Using Sentinel-2 satellite imagery, burn severity first classified using dNBR dNDVI indices. Subsequently, machine learning (ML) algorithms—RF, XGBoost, LightGBM, AdaBoost—were employed classify unburned areas. To enhance model performance, hyperparameter optimization applied, results were evaluated multiple accuracy metrics. found that RF achieved highest with an overall 98.0% a Kappa coefficient 0.960. In comparison, classification based solely on spectral indices resulted accuracies 86.6% (dNBR) 81.7% (dNDVI). A key contribution this is integration Explainable Artificial Intelligence (XAI) through SHapley Additive exPlanations (SHAP) analysis, which used interpret influence environmental variables area classification. SHAP analysis made decision processes transparent identified dNBR, dNDVI, SWIR/NIR bands as most influential variables. Furthermore, analyses confirmed variations reflectance across fire-affected are critical for accurate delineation, particularly heterogeneous landscapes. provides scientific framework post-fire ecosystem restoration, fire management, disaster strategies, offering decision-makers data-driven effective intervention strategies.

Language: Английский

Citations

1