Mapping with different classification methods of burnt forest areas by using medium resolution satellite images

Every year tens of thousands of hectares of forest are being destroy due to fires. The regular mapping of burnt forest areas is essential for damage assessment and interventions. Nowadays, the free availability of medium-resolution satellite imagery offers significant advantages in mapping burnt areas. Firstly, the change detection of burnt forest areas with the different band indices were successfully mapped by using the bands of medium resolution Sentinel 2A imagery before and after the fires. Difference Vegetation Index (DVI), Normalized Difference Vegetation Index (NDVI), Normalized Burn Area Index (NBR), and Raw Normalized Burn Area Index (NBR-Raw) were used in this study. Change detection has been performed by using both pixel-based and object-based classification approaches. Secondly, using the test fields, the object-based and pixel-based classification methods are applied to map the burnt forest areas. In this study, the same algorithm, training data set and parameters were used for the comparison of the two classification approaches. To evaluate all the results obtained, 2400 randomly generated control points in the study area were used to calculate the user accuracy, the producer accuracy, overall accuracy and kappa values. According to the accuracy results analysis, the best change detection result was the object-based DVI index with 92% accuracy. The lowest accuracy value of 66.98%. was obtained from the object-based classification without indices. Furthermore, when pixel-based and object-based classification approaches are compared, object-based change detection yields higher accuracy values than the pixel-based and all indexes.