Multi-season RapidEye imagery improves the classification of wetland and dryland communities in a subtropical coastal region

Abstract

Remote sensing is considered a valuable tool for monitoring the impacts of global change on vegetation species composition, condition and distribution. Multi-season imagery has been shown to improve the classification of vegetation communities though the contribution of the winter season in multi-seasonal classifications remains to be assessed. The capability of multi-season images of RapidEye, a new generation high spatial resolution (5 m) space-borne sensor containing an additional red-edge band, was evaluated for the classification of several wetland and dryland communities. RapidEye images were obtained for four seasons (winter, spring, summer and autumn) between 2011 and 2012 for a subtropical coastal region of South Africa. The separability of nine wetland and dryland communities was assessed for each season using the Partial Least Square Random Forest (PLS-RF) algorithm. The four-seasons approach yielded a higher overall classification accuracy (OA = 86 ± 2.8%) when compared to using any single-season classification. The highest single-season accuracies were obtained in spring (80 ± 2.9%), summer (80 ± 3.1%), and autumn (79 ± 3.4%) compared to the winter (66 ± 3.1%). A three-season combination of autumn, winter and spring yielded the highest average OA (86 ± 3.1%), maximised the user’s accuracies and minimised the number of comparable pairs confused. The inclusion of indices in the classification scenarios showed a minor (±1 percentage points) difference in the average overall and user’s accuracies compared to the classification results where only bands were used. The red-edge band of RapidEye increased the overall and average user’s accuracy for most of the scenarios by 2–6 percentage points and thus contributed to the separability of communities which are dominated by evergreen tree species.