Topical Keynote Sessions
Blending physical modelling and machine learning: new frontiers in spectroscopy data processing
|Conveners:||Jochem Verrelst (Universitat de Valencia)|
|Gustau Camps-Valls (Universitat de Valencia)|
|Jose Gomez Dans (University College London)|
Physically-based radiative transfer models (RTMs) help understand the interactions of radiation with vegetation and atmosphere. However, advanced RTMs can be computationally burdensome and too rigid for daily use. Alternatively, recent machine learning models can cope with large datasets with high accuracy but deploy too flexible models that may not respect the underlying physics. This special session will discuss the emerging field of synergistic use of physical models with machine learning techniques for efficient spectroscopy data processing. Advances in emulation approaches will be presented, as well new machine learning methods that encode RTM differential equations and learn the latent forces from empirical data, as well practical applications in the fields of sensitivity analysis, efficient inversion (or retrieval) and synthetic scene generation approaches. The session will close with a discussion on how advanced physical models and machine learning can live together and contribute to the analysis and processing of forthcoming imaging spectroscopy data streams.
Keywords: emulation, radiative transfer models, metamodels, machine learning
Imaging spectroscopy from unmanned aerial systems (UAS): Recent advances in technology and applications
|Conveners:||Pablo J. Zarco-Tejada (European Commission, Joint Research Centre, Ispra, Italy)|
|Helge Aasen (ETH Zurich, Switzerland)|
|Lammert Kooistra (Wageningen University, The Netherlands)|
|Arko Lucieer (University of Tasmania, Australia)|
The acquisition of spatially high resolution imaging spectroscopy data from unmanned aerial systems (UAS) has become a reality and enables addressing several challenging scientific questions related to environmental change, food security, or the assessment of hazardous events. Nevertheless, critical technical advances are still required before such data can be operationally acquired. Particularly limitations and restrictions imposed by unmanned aircraft vehicles (UAV) require future attention, e.g. optimizing the size and weight of sensors, or the miniaturization of auxiliary devices needed during flight. This session will discuss current technology available to facilitate imaging spectroscopy from UAS and will outline emerging applications in which the use of UAS based imaging spectroscopy is important. Current technological limitations that complicate the application of miniature-based imaging spectrometers will be discussed, e.g. issues related to the radiometric and spectral calibration of sensors, or the atmospheric correction of acquired images. Further, critical next steps to eventually facilitate a successful application of imaging spectroscopy for quantitative remote sensing will be defined.