Abstrak Artikel
The Mid-Infrared Burn Index (MIRBI) was designed for savanna ecosystems, yet its threshold sensitivity and spectral separability in Indonesian tropical montane savannas remain poorly documented. This study evaluates MIRBI for the Bukit Anak Dara fire (Sembalun, East Lombok, 31 August – 2 September 2024) using Landsat 8 OLI–TIRS Level-2 SR data acquired on 29 August and 14 September 2024. The difference image dMIRBI was tested at nine thresholds (7,000–25,000) and evaluated against a 319.61 ha control polygon visually interpreted from Sentinel-2A imagery (12 September 2024) using recall, precision, F1, IoU, kappa, and overall accuracy. Spectral separability statistics, ROC analysis, automatic Otsu thresholding, and bootstrap confidence intervals supported the evaluation. The dMIRBI distributions of burned and unburned pixels are highly separable (JM = 1.759; M-statistic = 2.176; AUC = 0.9960). The empirical threshold dMIRBI ≥ 15,000 yields 318.87 ha (0.74 ha or 0.2% from control) with F1 = 0.937 (95% CI [0.930; 0.942]), IoU = 0.881, kappa = 0.927, and OA = 98.4%. The Otsu threshold (21,906) gives 281.61 ha (only 4.4 ha less than the official BKPH Rinjani ). Only 52.2% of SiPongi+/FIRMS hotspots fall inside the control polygon, confirming the complementarity of hotspots and spectral indices. Keywords: dMIRBI, Landsat, Otsu, Burned Area Mapping, Spectral Separability
Pratinjau PDF
Pastikan PyMuPDF terpasang lalu coba muat ulang.