Removal of Surface Reflection from Above-Water Visible—Near Infrared Spectroscopic Measurements

Abstract

Water quality estimation in fresh and marine water systems with in situ above-water spectroscopy requires measurement of the volume reflectance (ρ_v) of water bodies. However, the above-water radiometric measurements include surface reflection (L_r) as a significant component along with volume reflection. The L_r carries no information on water quality, and hence it is considered as a major source of error in in situ above-water spectroscopy. Currently, there are no methods to directly measure L_r. The common method to estimate L_r assumes a constant water surface reflectance (ρ_s) of 2%, and then subtracts the L_r thus calculated from the above-water radiance measurements to obtain the volume reflection (L_v). The problem with this method is that the amount of ρ_s varies with environmental conditions. Therefore, a methodology was developed in this study for direct measurement of water volume reflectance above water at nadir view geometry. Other objectives of this study were to analyze the contribution of L_r to the total water reflectance under various environmental conditions in a controlled setup and to develop an artificial neural network (ANN) model to estimate ρ_s from environmental conditions. The results showed that L_r contributed 20–54% of total upwelling radiance from water at nadir. The ρ_s was highly variable with environmental conditions. Using sun altitude, wind speed, diffuse lighting, and wavelength as inputs, the ANN model was able to accurately simulate ρ_s, with a low root mean square error of 0.003. A sensitivity analysis with the ANN model indicated that sun altitude and diffuse light had the highest influence on ρ_s, contributing to over 82% of predictability of the ANN model. Therefore, the ANN modeling framework can be an accurate tool for estimating surface reflectance in applications that require volume reflectance of water.