Estimating wetland vegetation abundance based on spectral mixture analysis: a comparison between LSMA and FCM classification methods

Abstract

Vegetation abundance is a critical indicator for measuring the status of vegetation. It is also important for evaluating the eco-environment of wetland. In this article, linear spectral mixture analysis (LSMA) and fuzzy c-means (FCM) classification methods were applied to estimate vegetation abundance in Wild Duck Lake Wetland, one of the typical freshwater wetlands in North China, based on Landsat Thematic Mapper (TM) data acquired on 27 June 2011. Due to its effectiveness in characterizing vegetation activity and greenness, the normalized difference vegetation index (NDVI) was incorporated into the six reflective bands of the Landsat TM image to provide enough dimensionality to support the use of the a five-endmember LSMA model, which includes terrestrial plants, aquatic plants, high albedo, low albedo, and bare soil. Then, a fully constrained LSMA algorithm was performed to obtain vegetation abundance in our study area. An FCM classification algorithm was also used to generate vegetation abundance. Finally, both results were modified using the extracted water area of Wild Duck Lake Wetland, which was obtained with the combination of NDVI and normalized difference water index. The root mean square error (RMSE) and the coefficient of determination (R2) were calculated to assess the accuracy of vegetation abundance by using a WorldView-2 multispectral image. Validation showed that although there were slight differences between the vegetation abundance images, they shared similar spatial patterns of vegetation distribution: high vegetation abundance values in agricultural areas and riparian areas, moderate in grassland areas, and low in residential areas. The FCM classification generated an R2 of 0.791, while the LSMA yielded a result with an R2 of 0.672. Additionally, the RMSE also indicated that the FCM classification can obtain a much better result than LSMA: the former’s RMSE is 0.091 and the latter is 0.172. The result suggests that the FCM classification based on the nonlinear assumption can handle mixed pixels more effectively than LSMA.