Mapping bamboo with regional phenological characteristics derived from dense Landsat time series using Google Earth Engine

Abstract

ABSTRACTBamboo is an important plant not only because of its vital role in supporting biodiversity and land restoration, but also due to its contribution to poverty eradication. Although remote sensing has an advantage for monitoring vegetation, bamboo mapping is challenging due to the spectral similarity between bamboo and forest types. To overcome difficulties in bamboo mapping, we experimented with a phenology-based approach using dense Landsat time series data in Hainan Island, China. We constructed temporal profiles of Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Green Chlorophyll Vegetation Index (GCVI), and Land Surface Water Index (LSWI) for estimating the phenological variation among bamboo and adjacent forest types. We then compared the classification results under an all-season feature set and single-season feature sets. Our results indicated that the phenological variations of bamboo differed greatly from adjacent forest types, implying a good potential for bamboo identification. Compared to the conventional spectra-based approach, these results also emphasized the importance of phenological features. Validation using the k-fold (k = 10) approach showed this experiment achieved reasonable levels of accuracy for bamboo mapping (PA = 88.8%; UA = 74.6%). A bamboo distribution map in Hainan Island is useful to resource inventory in the second largest island in China. The success of the method in this tropical region suggests that it might be applicable to other parts of the tropical world.