Genetic algorithm-based method for forest type classification using multi-temporal NDVI from Landsat TM imagery

Abstract

Remote-sensing technology has been a  useful tool for mapping and characterizing forest cover types and species composition, providing valuable information for effective forest management. This study investigates the application of a genetic algorithm (GA)-based approach on Normalized Difference Vegetation Index (NDVI) to separate local forest communities at Huntington Wildlife Forest (HWF), located in New York State of the United States, into deciduous, mixed/coniferous and nonforests using Landsat TM imagery. Overall accuracy, producer’s accuracy, user’s accuracy and kappa coefficient of agreement are employed to assess the performance of the proposed method. Its overall effectiveness is supported by the accuracy of 80.41% and kappa coefficient of 0.56, and its capability of separating the forest cover types is endorsed by the class-wise accuracy measures. This method shows advantages in its limited demands for input features, that only multi-temporal NDVI indices are required; and in its simple and efficient mechanism, which refers to threshold optimization and feature selection.