Modelling reservoir turbidity from medium resolution Sentinel-2A/MSI and Landsat-8/OLI satellite imagery

Abstract

This study investigates the use of Sentinel-2A (S2A) and Landsat-8 (L8) OLI for monitoring of turbidity in reservoir waters. Using observed in situ data from 18 sampling stations for Chebara Reservoir in Kenya, the study developed an empirical multivariate regression model for turbidity estimation from atmospherically corrected, band adjusted and spectral resolution standardized S2A and L8 bands. Best results for turbidity estimation were obtained from the regression of in situ data with B2 (blue) and B3 (green) bands as [Rrs(B2/B3)^2+Rrs((B2/B3)] for S2A and [Rrs((B3/B2)] for L8. Both S2A and L8 retrieved turbidity with high and nearly equal accuracy of R^2 > 0.75 from the visible and NIR bands, with nearly similar RMSE of 0.5 NTU and NMAE% being higher for S2A by more than 30% as compared to L8’s average NMAE% of 15%. The study shows that for both S2A and L8 sensors, and the proposed empirical regression algorithm suffices in the rapid and cost-effective quantification of turbidity inland reservoir waters. Using spatial interpolation for the visualization of the correlation between the predicted and observed turbidity, the L8 results were found to be more significant than the turbidity estimations using S2A bands.