A novel multi-spectral index for burned area detection using high-resolution satellite imagery

The limitations of existing indices, heavily depend on the shortwave infrared (SWIR) band for identifying burned areas—a band not available in SuperDove satellite imagery. The study offers the novel SuperDove burned area index (SBAI) for burned area identification using high-resolution SuperDove satellite imagery as existing indices fail to use available SuperDove spectral bands. SuperDove satellite imagery with 3.7-m resolution was used to perform the analysis and the study area was the burned region in Goizha Mountain in Sulaymaniyah, Iraq. Nine established indices and K-nearest neighbor (KNN), support vector machine (SVM), random forest (RF), and maximum likelihood (ML) machine learning techniques for burned area detection were compared to the performance of the SBAI. The investigations showed that SBAI underscored high accuracy with 80.7% in detecting areas burned by fire, which outperformed 8 out of 9 traditional indices and all used machine learning algorithms. SBAI showed lower noise and false detections compared to the established techniques. The difference normalized burn ratio dNBR was the only index among all techniques which slightly surpassed the performance of SBAI, with 83.4% accuracy, while the majority of indices showed poor performance when applied to SuperDove images. The research also underscored that well-designed indices could outperform advanced machine learning techniques in specific conditions of earth observation applications. The developed novel index significantly contributes to monitoring fire events and post-fire assessment by using high spatial and temporal resolution multispectral imagery, which offers valuable insights to geospatial analysts and land managers for accurate fire event mapping. This novel index fills the gap in using SuperDove satellite imagery accurately to detect burned areas, which previously was challenged by existing indices.