ARTMO: Machine Learning Regression Algorithms Toolbox for mapping

Author: Jochem Verrelst (Jochem.verrelst @ uv.es). Presented: 12 May 2021

The focus of today’s presentation will be on ARTMO’s (Automated Radiative Transfer Models Operator). Machine Learning Regression Algorithm (MLRA) toolbox and some sub-tools.
For the generation and interpretation of hyperspectral data, ARTMO brings together a huge collection of diverse leaf, canopy and atmosphere radiative transfer models into a synchronized user-friendly GUI environment. Mapping of biophysical and biochemical vegetation traits from optical remote sensing images always requires an intermediate modeling step to transform acquired spectral observations into useful estimates. This modeling step can be approached with either parametric regression, nonparametric regression, physically-based or hybrid methods. Here, emphasis is put on the nonparametric approaches: The MLRA toolbox provides a suite of these nonparametric algorithms to enable semiautomatic mapping of terrestrial vegetation properties. For this, machine learning regression algorithms are used to build models for the prediction of traits (or variables) of interest by using training datasets of input-output data pairs, which usually come from concurrent measurements of the variables and the corresponding radiometric observations. MLRAs have the potential to generate adaptive and robust relationships and, once trained, they are very fast and efficient for processing large (satellite) scenes. Typically, MLRAs are able to cope with the strong nonlinearity of the functional dependence between the biophysical variable and the observed reflected radiance. Therefore, they may be most powerful candidates for mapping applications (agriculture or other ecosystems).
The presentation will be composed of a tutorial (brief theoretical session) and a practical demonstration part with the following ARTMO tools:
(1) machine learning regression algorithms (MLRA) as the main tool,
(2) dimensionality reduction (DR) methods,
(3) band ranking (GPR-BAT), and
(4) active learning (AL) methods.

The full software framework can be freely downloaded at https://artmotoolbox.com/.