The Accuracy of Gililawa Darat Wildfire Spread Estimation Using Burn Severity and WRF-Sfire Model

Abstract

In this research, the origin of the wildfire area is assessed by using the potential of burn severity and the WRF-SFIRE model. This research focuses on the mountainous savanna region, by taking the case of the Gililawa Darat wildfire event. The most accurate index among Sentinel-2B optical burn severity indices, i.e. dNBR, MIRBI, dMIRBI, CSI, dCSI, NDVI, dNDVI, EVI, and dEVI and among Landsat-8 OLI/TIRS thermal and mixed burn severity indices, i.e. LST, dLST, LST/EVI, and d(LST/EVI) was used to map the areas with low burn severity, an indication generally found at origin area. A series of fire spread simulation from these areas was conducted using WRF-SFIRE to assess the accuracy of each simulation in reproducing the burned area. The burn severity accuracy assessment showed that dNBR and dCSI indices had the highest value of Overall Accuracy and Kappa Hat Coefficient, i.e. 91.67% and 0.889 (almost perfect agreement). However, dNBR was the most suitable index for mapping burn severity in the region due to its goodness-of-fit measure for linear regression model with the R-squared value of 0.7856. The assessment of thermal and mixed burn severity indices based on Landsat-8 OLI/TIRS resulted in LST, LST/EVI, and d(LST/EVI) gained the same overall accuracy of 58.33% and Kappa Hat Coefficient of 0.444 indicating moderate agreement, whereas dLST performed poorer than these indices. However, it is not recommended to use these burn severity indices in the region due to the nonlinearity of severity level with the index value. According to WRF-SFIRE simulations result, it was found that fire ignition started from low burn severity area coordinates which have a distance of 0 to 334 metres from the origin area resulting in fire area witan h overall accuracy value range from 77.04% to 81.90% and Kappa Hat Coefficient value range from 0.536 to 0.626. The simulation from the origin area resulted in an overall accuracy of 81.57%, a Kappa hat coefficient of 0.613, underestimated burned area ratio of 0.08, overestimated burned area ratio of 0.23, and a backing fire perimeter difference ratio of 0.4 to the reference.
 Keywords: wildfire; Gililawa Darat; wildfire spread; burn severity; WRF-SFIRE, fire origin area