Fuzzy neural network classification of global land cover from a 1 degrees AVHRR data set

Phenological differences among broadly defined vegetation types can be a ba sis for global scale landcover classification ata very coarse spatial scale . Using an annual sequence of composited normalized difference vegetation i ndex (NDVI) values from AVHRR data set composited to 1 degrees DeFries and Townshend (1994) classified eleven global land-cover types with a maximum l ikelihood classifier. Classification of these same data using a neural netw ork architecture called fuzzy ARTMAP indicate the following: i) When fuzzy ARTMAP is trained using 80% of the data and tested on the remaining (unseen ) 20% of the data, classification accuracy is more than 85% compared with 7 8% using the maximum likelihood classifier; ii) classification accuracies f or various splits of training/testing data show that an increase in the siz e of training data does not result in improved accuracies; iii) classificat ion results vary depending on the use of latitude as an input variable simi lar to the results of DeFries and Townshed; and iv) fuzzy ARTMAP dynamics i ncluding a voting procedure and the numbrer of internal nodes can be used t o describe uncertainty in classification. This study shows that artificial neural networks are a viable alternative for global scale landcover classif ication due to increased accuracy and the ability to provide additional inf ormation on uncertainty. (C)Elsevier Science Inc., 1999.