Spatial spectral contextual image analysis approach to map water quality of inland waterbody using AVIRIS-NG datasets

Abstract

Remote sensing has already proven its capability in water quality studies. The technology enables the mapping of water quality parameters that affect its optical properties. Generally, multi-spectral remote sensing data has been utilized for such analysis. Mostly, the band rationing or simple regression approaches have been adapted to map the water quality, however, they result in qualitative estimates rather than quantitative. On the other hand, sensor technology has made steep advancements from multi-spectral to hyperspectral. The hyperspectral data provides contiguous spectral information over several hundred narrow spectral bands. This emerging technique provides a more in-depth means of investigating spatial, spectral, and temporal variations, leading to more quantifiable/accurate estimates of the process under consideration. At present, very few satellite-based hyperspectral data are available. Recently, the Indian Space Research Organisation has initiated an airborne hyperspectral remote sensing campaign to cover the important earth features in India using Airborne Visible/Infrared Imaging Spectrometer – Next Generation (AVIRIS-NG) sensor in collaboration with the National Aeronautics and Space Administration (NASA). As the AVIRIS-NG hyperspectral data measures the spectral response in the wavelength range from 380 to 2510 nm at very narrow spectral bandwidth of 5 nm. These advanagtes of AVIRIS-NG datasets enable researchers to analyse the spectra of each and every pixel of the image spatially to match its spectral characteristics with the spectra collected in the field. In the present study, this spectral similarity approach has been expoilted within the context of turbidity, the optical property of water for Chilika Lake, Odisha, India; using the AVIRIS-NG data and field reference spectra. The lake was surveyed synchronous to AVIRIS-NG flights over it during the month of December 2015. A field-based spectral library with respect to various concentrations of turbidity in the lake has been generated using sophisticated instruments like field spectro-radiometer, handheld Global Navigation Satellite System (GNSS) receiver, and turbidimeter. The spectral angle mapper (SAM) spatial-spectral contextual image analysis has been carried out to map the turbidity of the lake. It was observed that almost at each and every pixel the image spectra were matching with the field-collected spectra. Further, the match between two spectra (field and image) measured in terms of SAM score was as high as 0.9 for most of the instances. Moreover, the estimated turbidity was in the agreement (R2 = 0.96) with the turbidity measured on the field. It was realized that the spectral similarity approach using hyperspectral remote sensing data provides more quantitative estimates as compared to multi-spectral data analysis.