Use of TOC2TOA and GSA toolbox (ARTMO) to analyze the impact of water and air quality parameters on the TOA

Abstract

Machine learning (ML) models are developed to solve complex problems through recursive and iterative analysis, so that they can find relationships between sets of variables without being explicitly programmed to perform the task. Because of this versatility, ML methods can be used in a wide variety of fields, e.g. remote sensing. Knowledge of key variables that drive Top Of the Atmosphere (TOA) radiance on a surface is of importance for obtaining biophysical variables. Coupled water-atmosphere Radiative Transfer Models (RTMs) allow linking water variables directly to TOA radiance. Global Sensitivity Analysis (GSA) of RTMs enables the computation of the contribution of each input variable to the output variance. To do so, we developed a statistical learning model (emulator) that allows approximating RTM outputs through a machine learning algorithm with low computation time. A Random Forest emulator was used to reproduce lookup tables of TOA radiance. GSA total sensitivity results quantified the driving variables of emulated TOA radiance. In general, atmospheric variables play a more dominant role than water variables, probably as a consequence of the low reflectance of water. Only the presence of a chlorophyll spike in the spectral range of the green colour is found.