Temporal variation in discriminating Acacia species using optical and radar data

This study investigated the efficacy of Sentinel-1 radar and Sentinel-2 optical data to classify Acacia cyclops and A. mearnsii species for each month of a year using the Extreme Gradient Boosting classifier. Sentinel-2 bands and their derivative indices yielded modest overall accuracies (69–77%) in detecting Acacia species across months. Comparable accuracies were obtained (75–79%) by adding Sentinel-1 radar data. The producer’s accuracies varied with month for both Acacia species (60–65%) when using Sentinel-2 products. Adding radar data improved these accuracies for nearly all months by 4–32%. In all cases, better accuracies were obtained in summer and autumn months. Variable importance comparison revealed that the Sentinel-2 visible bands, RedEdge bands and their derivative indices to be the highest contributors in the identification of the Acacia species. Although the radar data showed more importance in distinguishing Acacia from the other land cover groups than from each other, they did not improve the classification accuracies significantly. The inability of radar to differentiate the Acacia species was attributed to the lack of distinctive structural variation between the two species. The findings of this study show the importance of data acquisition time in the classification effort while more research is needed to exploit radar data.