Deciphering Optical Water Types of Wetlands using multispectral Earth observation datasets

Wetlands are important and highly productive ecosystems in a variety of geomorphic settings ranging from inland to coastal environments. Wetlands are very dynamic in nature and are driven by the water and sediment fluxes carried by the streamlets throughout the year. Wetlands are under tremendous pressure all over the world due to various natural and anthropogenic factors, and therefore, require an immediate attention for their conservation. The available studies on wetland have given much less importance to the internal dynamics of the wetlands, which is primarily driven by hydrology and Land Use Land Cover (LULC) changes. Here, we propose to use the Optical Water Types (OWTs) concept to understand the hydrodynamics within the wetland.

The OWTs are the aquatic counterpart of terrestrial LULC classification and can be created by clustering of optically sensitive parameters like chlorophyll content, turbidity, suspended organic and inorganic matter using remote sensing reflectance, absorption, and scattering parameters. The Forel Ule (FU) color index, a visual color comparison scale of water bodies ranging from blue to cola brown (1-21), used a similar idea but is fairly limited in scope. The hyperspectral datasets have distinct absorption and reflection spectrum for various optically sensitive parameters, and therefore, they are particularly suited for this work. However, the availability of the high-resolution hyperspectral data is very limited and hence this research explores the possibility of deciphering the OWTs using multispectral datasets.

A possible approach to create OWTs is using the spectral indices of the multispectral datasets which are sensitive to the optical parameters instead of using the FU color index as a single parameter. In this work, various spectral indices which are independent and highly sensitivity to chlorophyll content, turbidity, suspended organic and inorganic matter are identified using the principal component analysis. The OWT clusters are created using the iso-cluster unsupervised classification similar to the LULC classification but the spectral indices are taken into account instead of directly using the spectral bands of satellite datasets. In this work, the Sentinel – 2A and 2B datasets are used to create independent OWT clusters of the Chilika (a Coastal wetland, along the east coast of India covering an area of 1,165 km²) and Kaabar Tal (an inland wetland in north Bihar plains, India covering an area of 51 km²) using the supervised classification method. The developed framework is very simple and robust in nature but the only disadvantage is that the clusters are variable in the temporal context. However, the temporal variations can be integrated with the spatial analysis to understand the wetland dynamics in the context of both space and time.