This paper introduces an efficient Bayesian method for inverting complex physical models, like the Hapke photometric model, on massive planetary datasets. By using a "learning approach" with Gaussian Locally-Linear Mapping (GLLIM), they significantly speed up the inversion process, making it feasible to analyze large hyperspectral image sequences from Mars.
What Problem Does This Paper Address?
Studying planetary surfaces, like Mars, often involves analyzing hyperspectral images to understand material composition and microtexture. This is typically done by "inverting" a physical model, such as the Hapke model, which relates how light reflects off a surface (reflectance) to its physical properties (like grain size, roughness, etc.). However, planetary datasets, especially from instruments like CRISM on Mars, are massive, containing millions of spectral measurements across many spatial points and viewing angles. Traditional inversion methods, like Markov Chain Monte Carlo (MCMC), become computationally prohibitive for such large datasets, making it difficult to extract detailed information efficiently.
Methods
The authors propose a two-step approach:
Surrogate Model Learning: They first build a statistical surrogate model of the complex forward physical model (e.g., Hapke model). This surrogate model is trained using a technique called Gaussian Locally-Linear Mapping (GLLIM). GLLIMs are chosen for their ability to approximate highly non-linear functions while remaining computationally tractable. The training involves generating a dataset of input parameters and corresponding model outputs, then maximizing the likelihood of this data using an Expectation-Maximization (EM) algorithm. This "learning phase" is computationally intensive but is performed only once.
Efficient Inversion: Once the surrogate model is trained, it can be used to invert observations. The inverse of the GLLIM surrogate model has an explicit formulation as a Gaussian Mixture, allowing for straightforward computation of the posterior distribution of the physical parameters. This enables efficient propagation of uncertainties and assessment of multi-modal solutions (where multiple parameter sets could explain the data).
Datasets
CRISM Multi-Angular Data: The primary dataset used for demonstration is a collection of eleven hyperspectral images from the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) instrument, specifically targeting the South Pole of Mars. This dataset was pre-processed with atmospheric correction.
Laboratory Spectro-photo-goniometer Data: The authors mention that their method has also been successfully applied to laboratory measurements, though specific details are not provided in this excerpt.
Key Insights and Ideas
Bridging the Gap: The core innovation is developing a method that can handle the massive scale of modern planetary remote sensing data, which was a bottleneck for traditional inversion techniques.
Bayesian Framework for Uncertainty: Leveraging the Bayesian framework naturally allows for the propagation of measurement uncertainties and provides a robust assessment of the uncertainty in the inverted physical parameters.
Handling Ill-Posed Problems: The complexity of radiative transfer models often makes inversion ill-posed. The proposed method's ability to provide a full posterior density, including assessing multi-modality, is crucial for understanding potential ambiguities in the solutions.
Scalability: The computational cost of the learning phase does not scale with the number of observations to be inverted, making it highly efficient for large datasets.
Key Results
The method successfully inverted the Hapke model on a challenging multi-angular CRISM dataset from the Martian South Pole.
The inverted Hapke parameters (specifically, $\omega$ and $\theta$) showed spatial regularity and meaningful correlations with known surface compositions (CO2 ice, H2O frost, mineral dust) and geological structures.
The approach significantly speeds up the inversion process compared to methods like MCMC, making it practical for large-scale analysis.
Limitations and Future Work
Computational Cost of Learning: While the inversion itself is efficient, the initial "learning phase" for building the surrogate model is computationally intensive.
Multi-Modality Assessment: While the method can assess multi-modality, further exploitation of this feature for a more nuanced interpretation of solutions is suggested.
Generalizability: The paper demonstrates the method on planetary data, but its applicability to other complex physical models and datasets (e.g., laboratory measurements) is highlighted as a direction for future work.
Summary updated
Abstract
IntroductionWhen studying the Martian surface, the composition of the materials is established on the basis of spectral detection, unmixing, and physical modelling using images produced by hyperspectral cameras. Information on the microtexture of surface materials such as grain size, shape, roughness and internal structure can also be used as tracers of geological processes (Fernando et al. PSS 2016). This information is accessible under certain conditions thanks to hyperspectral image sequences acquired over a site of interest from eleven different angles by the Compact Reconnaissance Imaging Spectrometer for Mars (CRISM@MRO, Murchie et al. JGR-Planets 2009) . Similar multi-angular observations can also be realized in the laboratory with spectro-photo-goniometers, on planetary analog materials or real extraterrestrial matter (Potin et al. Icarus 2019) . In both cases, the interpretation of the surface Bidirectional Reflectance Distribution Factor extracted from these observations, in terms of composition and microtexture, is based on the inversion of the Hapke model, a semi-empirical photometric model that relates physically meaningful parameters x to the reflectivity y of a granular material for a given geometry of illumination and viewing (Schmidt and Fernando Icarus 2015).This work presents an efficient method, based on a learning approach in a Bayesian framework, able to invert the Hapke model on large experimental or remote sensing datasets, as illustrated on a challenging CRISM multi-angular sequence of images. 1 Inversion characteristicsIn the CRISM case, the dataset provides both spatial and spectral dimensions. More specifically, each of the Nspatial spatial point of the scene is observed at Nspectral wavelengths, yielding Nspatial× Nspectral vectors of reflectance values. This high number of observations to be inverted usually makes approaches like Markov Chain Monte-Carlo simulations unacceptably slow. The approach we propose starts with a computer intensive learning phase that builds a statistical model for the Hapke model; it is then applied for all observables yobs.Each observable is a vector of D reflectances, one for each geometry of measure. In the case of CRISM data, D = 11 is moderate, but in laboratory, D may be one order of magnitude higher. The statistical model we propose has the advantage to be able to handle high dimensional settings while remaining computationally tractable.The Bayesian framework offers a natural solution to propagate uncertainties on measures, which are formalized as variance on the prior distribution, while the variance of the posterior distribution assesses the uncertainty induced by the inversion.The complexity of the forward radiative transfer model often makes our inverse problem ill-posed. In practice, multiples solutions might be acceptable. More than a point estimator, our statistical model provides a full posterior density, whose multi-modality can be assessed and interpreted.2 Bayesian inversion via a learning approachFrom a statistical point of view, we formalize our model as couple of random variables (X,Y), where Y = F(X) + ε. X has a prior distribution on the physical parameters space, F is the forward model and ε is a centered Gaussian noise accounting for measure uncertainties. The main idea is to use a two steps approach: first, approximate this model by a parametric surrogate model, second use the surrogate model to invert the observations.We chose the family of the so-called Gaussian Locally-Linear Mapping (GLLiM) (Deleforge et al StatComp 2015) . Such models are expressive enough to approximate highly non-linear, complex, forward models, while remaining tractable.The learning step is performed by generating a training dictionary composed of samples from (X,Y), on which the likelihood is maximized by a standard EM algorithm. Regarding the dimensionality of the problem, the number of parameters is proportional to D, making this model suitable even for high dimensional observables. Moreover, the computational cost of the learning phase does not scale with the number of observations to be inverted.The second step is performed by computing the posterior distribution of the random variable X given yobs. The inverse of the surrogate GLLiM has an explicit formulation in the form of a Gaussian Mixture. From this posterior density, the mean and variance are straightforward to compute. To handle a multi-solution scenario, the mixture can be further exploited.3 Massive inversion of spectro-imagesThe dataset comes from a multi-angular observation of the South Pole of Mars by CRISM. The targeted scene presents spatially segregated C02 ice, H2O frost, and mineral dust (Douté and Pilorget EPSC 2017). After fusion and atmospheric correction of eleven hyperspectral images (Ceamanos et al. JGR-Planets 2013), the dataset provides both spatial and spectral dimensions, totaling Nobs = 154650 = 3093 × 50 measurements vectors yobs, which makes MCMC approach unacceptably slow. Maps of Hapke’s parameter values are generated from the results of our inversion and superposed with transparency onto the full resolution CRISM nadir image, which serves as a geological control background image.The results are satisfying from the application point of view. The colour composition of Figure 1 reflects the variation of ω at three wavelengths and corresponds well with the spatial distribution of the three previous materials and their known spectral optical properties. The map of the roughness parameter θ averaged over the spectral dimension (Figure 2) is color coded by intervals of values whose spatial variations are correlated with the composition and the structures of the terrains. In general, we find that all predicted parameters preserve some spatial regularity and show meaningful correlations with the composition and the geology.   ConclusionWe propose an efficient statistical and learning method for inverting physical models on massive remote sensing or experimental data. In the case of the Hapke model, the method has proved to produce satisfactory results for CRISM multi-angular acquisitions as illustrated in this paper and for laboratory spectrophotometric measurements as illustrated in Potin et al., this conference.
BibTeX
@misc{kugler2020efficient,
author = {B. Kugler and Florence Forbes and S. Douté},
title = {{An efficient Bayesian method for inverting physical models on massive planetary data}},
year = {2020},
doi = {10.5194/epsc2020-335},
}