Comparative Analysis of dNBR, dNDVI, SVM Kernels, and ISODATA for Wildfire-Burned Area Mapping Using Sentinel-2 Imagery

Comparative Analysis of dNBR, dNDVI, SVM Kernels, and ISODATA for Wildfire-Burned Area Mapping Using Sentinel-2 Imagery

Accurate and rapid delineation of wildfire-affected areas is essential in the era of climate-driven increases in fire frequency. This study compares and analyzes four techniques for identifying wildfire-affected areas using Sentinel-2 satellite imagery: (1) calibrated differenced Normalized Burn Rat...

Full description

Saved in:
Bibliographic Details
Main Authors: Sang-Hoon Lee, Myeong-Hwan Lee, Tae-Hoon Kang, Hyung-Rai Cho, Hong-Sik Yun, Seung-Jun Lee
Format: Article
Language:English
Published: MDPI AG 2025-06-01
Series:Remote Sensing
Subjects:
burn area mapping
classification
machine learning
remote sensing
Sentinel-2
dNBR
Online Access:https://www.mdpi.com/2072-4292/17/13/2196
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Accurate and rapid delineation of wildfire-affected areas is essential in the era of climate-driven increases in fire frequency. This study compares and analyzes four techniques for identifying wildfire-affected areas using Sentinel-2 satellite imagery: (1) calibrated differenced Normalized Burn Ratio (dNBR); (2) differenced NDVI (dNDVI) with empirically defined thresholds (0.04–0.18); (3) supervised SVM classifiers applying linear, polynomial, and RBF kernels; and (4) unsupervised ISODATA clustering. In particular, this study proposes an SVM-based classification method that goes beyond conventional index- and threshold-based approaches by directly using the SWIR, NIR, and RED band values of Sentinel-2 as input variables. It also examines the potential of the ISODATA method, which can rapidly classify affected areas without a training process and further assess burn severity through a two-step clustering procedure. The experimental results showed that SVM was able to effectively identify affected areas using only post-fire imagery, and that ISODATA enabled fast classification and severity analysis without training data. This study performed a wildfire damage analysis through a comparison of various techniques and presents a data-driven framework that can be utilized in future wildfire response and policy-oriented recovery support.
ISSN:2072-4292

Similar Items