Trophic status estimation of case-2 water bodies of the Godavari River basin using satellite imagery and artificial neural network (ANN)

Abstract

The dynamics of trophic status estimation of case-2 water bodies on a synoptic mode for frequent intervals is essential for water quality management. The present study attempts to develop trophic status estimation approaches utilizing Landsat-8 and Sentinel-2 images as inputs. The chlorophyll-a concentration, a proxy parameter for trophic status, was estimated using the empirical method, fluorescence line height (FLH) method, and artificial neural network (ANN) approaches using spectral reflectance values as inputs. The outcomes following the empirical approaches revealed the scope of kernel normalized difference vegetation index (kNDVI) (R2 = 0.85; RMSE = 2 μg/l) for estimating the chlorophyll-a concentration using Sentinel-2 images of the Godavari River basin. Though the performance of the FLH method (R2 = 0.91; RMSE = 1.6 μg/l) was superior to kNDVI-based estimation, it lacks the capability to estimate chlorophyll-a concentration above 20 μg/l. Due to the existence of eutrophic regions within the Godavari basin (28%), adopting better approaches like ANN for trophic status estimation is essential. To accomplish the same, the Levenberg–Marquardt algorithm-based ANN was developed using non-redundant bands of Sentinel-2 as inputs, and Sentinel-3 derived chlorophyll-a values as output. The developed architecture was successful in estimating trophic status estimations at all levels.