Vegetation canopy fluorescence and reflectance retrieval by model inversion using optimization : abstract

The simultaneous retrieval of a vegetation canopy's fluorescence (F) and reflectance (R) from high spectral resolution top-of-atmosphere (TOA) radiance spectra is possible thanks to the smoothness of the spectra of reflectance and fluorescent radiance. Spectral fitting methods and the Fraunhofer line detection (FLD) method exploit this spectral smoothness to retrieve F and R by modelling these as smooth mathematical functions such as low degree polynomials and spline functions. The retrieval problem becomes one to optimize the fit between modelled and measured spectra of both F and R by adjusting the coefficients describing the F and R spectral functions. Since vegetation fluorescence takes place in the wide spectral range from 650 to 850 nm, the fitting of both F and R in this range with mathematical functions requires a large number of degrees of freedom, with an increased risk of illposedness of the retrieval problem. Therefore, in this contribution an alternative approach was tried to model R using a light version of the SAIL model, which has only 10 variables, but yet allows modelling R over the 400 to 2400 nm spectral range. In addition, two parameters were used to describe the fluorescence spectrum with two end member spectra, bringing the total number of degrees of freedom equal to 12. Because of this wide spectral range supported by SAIL_light, the spectral data used as input for the retrieval of F and R are not limited to those provided by the ESA candidate Earth Explorer 8 FLuorescence Explorer (FLEX) mission alone. In addition, data from the Sentinel-3 mission, expected to fly in tandem with FLEX, can be used to further constrain the reflectance retrieval. In a numerical experiment a database of 31 simulated TOA spectral radiance observations by the FLORIS instrument on board FLEX and by the OLCI and SLSTR sensors on board Sentinel-3 was generated with the coupled models SCOPE and MODTRAN5. By means of SAIL_light and MODTRAN5, TOA radiance spectra were generated, and the instrument characteristics (Spectral Response Functions and noise) were applied to obtain realistically simulated observations by the three sensors FLORIS, OLCI and SLSTR. A box-constrained Levenberg-Marquardt optimization routine was applied to retrieve F and R. The retrieved F levels were compared to those in the database. Statistical analysis indicates that retrievals are possible after about 11 model iterations, but that systematic errors of about 10% in F are still found. Currently, most of the computational burden is related to the propagation of ground signals through the atmosphere and simulation of the spectral sampling by the three sensors. Special techniques are required to reduce computation time in this respect.