W. Wang, J. Hu, and H.-X. Hu (PRC)
Decision tree, land cover classification, MODIS, vegetation index time series, remote sensing
Four land cover classification methods, i.e., maximum likelihood, neural network, spectral angle mapper and decision tree, are applied to investigate the explanatory power of multi-temporal MODIS Enhanced Vegetation Index (EVI) data to provide land-cover estimates at approximately 250 m resolution for sub-tropical humid southern China. The results show that, the decision tree method has the best performance among the four methods applied in the present study, including the method of maximum likelihood, neural network, spectral angle mapper and the decision tree, and the decision tree manually constructed based on knowledge has a better performance than the automatically generated tree based on classification and regression trees algorithm. The reason that the method of maximum likelihood, neural network, and spectral angle mapper perform badly is probably because those methods are more data-dependent than the decision tree method, especially compared with the knowledge-based decision tree method, whereas high quality continuous MODIS EVI data are difficult to retrieve because of complicated landscape and warm humid south sub-tropical climate.
Important Links:
Go Back
