Criteria for the use of regression analysis for remote sensing of sediment and pollutants

Abstract An examination of limitations, requirements, and precision of the linear multiple-regression technique for quantification of marine environmental parameters is conducted. Both environmental and optical physics conditions have been defined for which an exact solution to the signal response equations is of the same form as the multiple regression equation. Various statistical parameters are examined to define a criteria for selection of an unbiased fit when upwelled radiance values contain error and are correlated with each other. Field experimental data are examined to define data smoothing requirements in order to satisfy the criteria of Daniel and Wood. Recommendations are made concerning improved selection of ground-truth locations to maximize variance and to minimize physical errors associated with the remote sensing experiment.