Optimization and verification of a new analytical radiative transfer model

Kurzfassung

The analysis of satellite or airborne remote sensing data relies on reference measurements on the ground. For water covered areas, spectral reference measurements need to be carried out in and above the water column. Due to the wave-focusing effect a high variability is induced to the measurements. A new analytical model allows to account for this effect by treating the direct and diffuse parts of the light field separately. Parameterizations of the downwelling direct and diffuse attenuation coefficients, Kdd and Kds, were developed in this study. For this purpose numerical simulations with the well established radiative transfer model Hydrolight 5.0 were carried out. The program was modified to calculate the direct and diffuse components separately. The developed analytical parameterization was verified against these calculations and compared to a recently published semi analytical model by Pan and Zimmerman, which also treats direct and diffuse radiation separately. The diffuse part of this model could be improved with the developed parameterization concerning the sun zenith angle dependency. The physically meaningful, new analytical model performs better, on a wide range of simulated water compositions than the original, and even the improved Pan-Zimmerman model. The new parameterization is not restricted to absorption dominated water bodies. Field measurements were performed to collect a dataset which is suitable to independently validate the model. The model was fitted to these field measurements and the results of one fit parameter, the sensor depth, were compared to independent measurements. The model yields equally small mean errors and standard deviations for clear and overcast sky conditions, even at extremely high sun zenith angles. In comparison to a widely used model of Gordon, the mean standard deviation is decreased by a factor of five, which is a result of the successful correction of the wave focusing effect. The absolute differences are on average by a factor of three lower due to the improved parameterizations of Kdd and Kds.