Spectral Unmixing based Approach to Quantify Structural Vegetation Diversity in the Riparian Zones of River Yamuna: A Study for the Delhi Stretch

Riparian zones are essential for the vitality of river systems and their adjacent environment while offering an abundance of ecosystem services. However, these ecosystems are increasingly exposed to various stressors with human activities like habitat fragmentation instigated by land use intensification, hydroclimatic changes, and environmental impacts, being the primary contributors. The growing availability of satellite datasets has enhanced the capacity and efficiency of monitoring these ecosystems. Based on these viewpoints, the objective of this work is to present a systematic methodology to examine the structure of vegetation in riparian forests using the spectral heterogeneity in their reflectance for the Delhi stretch of river Yamuna. The spectral variation hypothesis suggests that a higher spectral variability is positively related to plant species richness. Taking that into account, the primary objectives of the present study are: (1) determination of differentiable spectral clusters for riparian vegetation and deriving endmember spectra for each cluster from the scene using Vertex Component Analysis (VCA), (2) unmixing the vegetation spectra and determination of Shannon’s Index (SHDI) as an indicator of structural diversity in the study area. Landsat 7 ETM and Landsat 8 and 9 OLI scaled surface reflectance products have been used for the analysis. Pixels corresponding to the forested region are identified using land use landcover maps. The principal component analysis was first carried out to reduce the high correlation among the image bands. The clusters for deriving the endmembers were determined from the output principal components using K- means clustering. The optimal number of clusters (k) were obtained using a tolerance-based plateau detection for iterative k - value against its average mean centroid distance (AMCD) at each step prior to retracing the cluster identities in the reflectance space. The endmember spectra are identified using VCA for each cluster and the method of Non-Negative Least Squares (NNLS) is employed to optimize the endmember reflectance function. Since the endmembers are identified for multiple years, we have used Spectral Angle Mapping (SAM) to identify the classes of similar endmember types within each year and across multiple years before determining the SHDI values based on the proportion of the total area covered by the endmembers The results show a decreasing trend in SHDI which implies that there is a decline in structural diversity at 30 meter scale within the riparian zones and thereby the area is becoming dominated by fewer vegetation types over time. The study reveals that spectral unmixing-based SHDI serves as a remote, repeatable metric for assessing riparian vegetation structure. Monitoring alterations in diversity facilitates the identification of homogenization and habitat complexity loss, thereby aiding in early warning, restoration targeting, and assessment of management or land-use policy effects.

Keywords: Riparian zones, spectral endmembers, spectral unmixing, VCA, Shannon diversity index.