A computationally efficient inverse modelling approach of inherent optical properties for a remote sensing model


Inverse modelling of inherent optical properties (IOP) is an alternative to the in situ measurements of IOP requiring specialized instruments. However, inverse modelling using Monte Carlo models may require very large computational time due to a large number of dynamic model runs needed to search the optimum parameter values. We present a new approach to reduce this computational time. Mathematical relationships were developed for wavelength and concentration dependence of IOP values of suspended mineral based on four parameters. Optimal values of these four parameters were calculated by minimizing the mean sum of error between the physical hyperspectral optical-Monte Carlo (PHO-MC) model predicted reflectance to measured reflectance values for selected 33 reflectance measurements for a set of 11 wavelengths and three suspended sediment concentrations. The computation time was significantly reduced by several orders of magnitude by: (1) replacing the PHO-MC model with 11 wavelengths specific pseudo-simulator models developed by applying artificial neural network approach; and (2) using a nondominated sorted genetic algorithm –II (NSGA II) to search the global optimal solution of four parameters of IOP equations. Determined IOP values of suspended minerals were then successfully validated by using them as input to PHO-MC model to predict reflectance values for an independent set of 287 combinations of 41 wavelengths and seven suspended sediment concentrations.

Date of this Version










Link Out to Full Text