Retrieving Canopy Variables by Radiative Transfer Model Inversion – an Automated Regional Approach for Imaging Spectrometer Data

A new, automated, regional approach is presented for the estimation of leaf area index, leaf chlorophyll, dry matter, and water content, based on the inversion of the combined leaf and canopy radiative transfer model PROSPECT+SAILh. The approach, named CRASh, is open to different types of imaging spectrometers, although it has been originally designed for airborne hyperspectral sensors with a wide field of view. Central concern is the exploitation of the complete spectral signature while minimizing the interdependency between the large number of spectral bands. Moreover, the model offers a new way of regularizing the ill-posed nature of radiative transfer model inversion in cases where no information on land cover or phenology is available in advance. For finding the solution of the inverse problem, the distance between modelled and measured spectrum, and between a priori estimate and model simulation is exploited using a lookup table (LUT) approach. In order to regularize the ill-posedness of the inverse problem, an automated spectral classifier, called SPECL, is integrated. This has the advantage that the solution can be optimized for specific vegetation cover types and plant physiological conditions, and allows the characterization of covariance between the different wavebands and variables. A priori estimates of the solution are predicted using regression equations based on the radiative transfer model simulations present in each LUT. The model can run in a completely automated mode or in a mode in which the user controls one or more of the inputs (e.g. land cover, prior knowledge on canopy characteristics, soil reflectance). The results discussed in this study rely on the completely automatic mode. Two case studies are presented to test the performance of the model. The first study involved the estimation of leaf area index (LAI), leaf dry matter, and water content from intensively managed grasslands in southern Germany using HyMap data from 2003. The CRASh algorithm shows significant improvement for LAI and leaf dry matter estimations compared to simpler algorithms minimizing only for the spectral distance, whereas prediction of leaf water content remains nearly unaffected. Moreover, the inclusion of covariance between variables has a positive effect on the stability of the solution, reducing ambiguity between several variables. Validation with ground measurements shows an average accuracy of the estimates of 71.7, 73.2, and 66.3% for leaf water content, leaf dry matter content and LAI, respectively. A case study on the estimation of chlorophyll content in cotton fields in Uzbekistan, based on Proba-1/CHRIS mode-5 imagery taken in 2006, shows some improvement with respect to a simpler minimization function, although improvement is not as striking as for the first case study. Average accuracy is good (80.4%) when estimations are compared to SPAD measurements, although it is significantly lower (41.4%) with respect to laboratory based chlorophyll measurements. Even if additional validation for different vegetation species, phenological conditions, view/sun constellations, and sensor configurations is Proceedings 5 EARSeL Workshop on Imaging Spectroscopy. Bruges, Belgium, April 23-25 2007 2 required to validate the model, the results underline the potential of incorporating the retrieval of land surface products in upcoming automatic processing chains for high resolution imaging spectrometers.