An Emulator Toolbox to Approximate Radiative Transfer Models with Statistical Learning

Juan Rivera,Jose Gómez-Dans,Gustau Camps-Valls,Jochem Verrelst,José Moreno,Jordi Muñoz-Marí

doi:10.3390/rs70709347

Juan Rivera, Jose Gómez-Dans + Show 4 more

Open Access

https://doi.org/10.3390/rs70709347

Copy DOI

Abstract

Physically-based radiative transfer models (RTMs) help in understanding the processes occurring on the Earth’s surface and their interactions with vegetation and atmosphere. When it comes to studying vegetation properties, RTMs allows us to study light interception by plant canopies and are used in the retrieval of biophysical variables through model inversion. However, advanced RTMs can take a long computational time, which makes them unfeasible in many real applications. To overcome this problem, it has been proposed to substitute RTMs through so-called emulators. Emulators are statistical models that approximate the functioning of RTMs. Emulators are advantageous in real practice because of the computational efficiency and excellent accuracy and flexibility for extrapolation. We hereby present an “Emulator toolbox” that enables analysing multi-output machine learning regression algorithms (MO-MLRAs) on their ability to approximate an RTM. The toolbox is included in the free-access ARTMO’s MATLAB suite for parameter retrieval and model inversion and currently contains both linear and non-linear MO-MLRAs, namely partial least squares regression (PLSR), kernel ridge regression (KRR) and neural networks (NN). These MO-MLRAs have been evaluated on their precision and speed to approximate the soil vegetation atmosphere transfer model SCOPE (Soil Canopy Observation, Photochemistry and Energy balance). SCOPE generates, amongst others, sun-induced chlorophyll fluorescence as the output signal. KRR and NN were evaluated as capable of reconstructing fluorescence spectra with great precision. Relative errors fell below 0.5% when trained with 500 or more samples using cross-validation and principal component analysis to alleviate the underdetermination problem. Moreover, NN reconstructed fluorescence spectra about 50-times faster and KRR about 800-times faster than SCOPE. The Emulator toolbox is foreseen to open new opportunities in the use of advanced RTMs, in which both consistent physical assumptions and data-driven machine learning algorithms live together.

Highlights

Since the advent of optical remote sensing, physically-based radiative transfer models (RTMs) have deeply helped in understanding the processes occurring on the Earth’s surface and their interactions with vegetation and atmosphere
The normalized RMSE (NRMSECV ) indicates that relative errors fall below 3%, but significant differences across the three MO-Machine Learning Regression Algorithm (MLRA) and sampling size can be observed
neural networks (NN) was validated as best performing for the datasets of 500, 1000 and 5000 samples, closely followed by kernel ridge regression (KRR)

Summary

Introduction

Since the advent of optical remote sensing, physically-based radiative transfer models (RTMs) have deeply helped in understanding the processes occurring on the Earth’s surface and their interactions with vegetation and atmosphere. RTMs describe absorption and scattering, and some of them even describe sun-induced chlorophyll fluorescence, the microwave region and thermal emission They are useful in a wide range of applications, including designing vegetation indices, performing sensitivity analyses, developing inversion models to accurately retrieve vegetation properties from remotely sensed data (see Verrelst et al [4] for a review) and to generate artificial scenes as would be observed by an optical sensor, e.g., [5]. Canopy RTMs can be categorized as “economically” and “non-economically”

Methods

Results

Discussion

Conclusion