Sub-Pixel Classification of Multi-date Satellite Images for Accurate Change Detection in Pichavaram Mangroves, Tamilnadu, India

Abstract

The Pichavaram mangrove wetland ecosystem contains species such as Rhizophora apiculata and Avicennia marina which are interspersed with the creeks and marshland landforms. Various studies have indicated that over the years the areal extent of this mangrove ecosystem has decreased due to both human impact and natural causes (erosion). However, field visits have indicated an increase in the areal extent due to re-forestation by the officials of the forestry department. To verify the fact, mapping of the changes in the Pichavaram mangrove ecosystem has to be attempted using multi-date satellite images. However, accurate mapping of these mangroves using multispectral satellite image data is difficult due to the presence of mixed pixels, which in turn is due to rich diversity, presence of brackish water bodies, and a complex network of tidal creeks and canals. While using remote sensing technology, conventional image classification (per-pixel) approaches result in misclassification and inaccurate estimates of the areal extent of mangroves. Such inaccuracy, in turn, leads to an indication that there has been an alarming rate of degradation of these mangroves. This paper examines the issues in the approaches to estimate the areal extent of mangroves and suggests that the sub-pixel approach (spectral unmixing) can be the appropriate technique to carry out change detection in mangrove ecosystems, instead of visual interpretation or per-pixel classification approaches. Satellite images of Pichavaram mangroves, acquired over a period of three decades were subjected to per-pixel and sub-pixel classification to determine the area covered by the mangroves during different dates. While per-pixel approach over-estimated the areal extent (due to the inclusion of water-spread area under the mangrove class), the sub-pixel approach gave estimates that agreed well with the actual areal extent of the mangroves. Thus, the need for an appropriate approach such as the sub-pixel classification to map mangroves is demonstrated in this paper.

Keywords: Mangroves, hard classification, sub-pixel classification, estimation of accurate area