Applicability of the PROSPECT model for estimating protein and cellulose plus lignin in fresh leaves

Hyperspectral remote sensing of leaf biochemicals is critical for understanding many biochemical processes. Leaf biochemical contents (e.g., protein, cellulose and lignin) in fresh and dry leaves have been quantified from hyperspectral data using empirical models. However, they cannot be retrieved for fresh leaves by inverting radiative transfer models. We demonstrated the applicability of PROSPECT leaf optical properties model in the separation of specific absorption coefficients for protein and cellulose + lignin following a newly proposed algorithm, and evaluated the feasibility in estimating leaf protein and cellulose + lignin content through model inversion. Assessment was performed across a large variety of plant species benefiting from the Leaf Optical Properties Experiment (LOPEX) dataset. To alleviate ill-posed problems, inversion was performed over different spectral subsets. The PROSPECT model with newly calibrated specific absorption coefficients was able to accurately reconstruct leaf reflectance and transmittance. Leaf protein and cellulose + lignin were estimated at moderate to good accuracies for both fresh and dry leaves. The spectral subset of 2100-2300 nm yielded the most accurate estimation of leaf cellulose + lignin (R2 = 0.70, RMSE = 5.21E-04 g/cm2 ) and protein (R2 = 0.47, RMSE = 2.75E-04 g/cm2 ) in fresh leaves, which were comparable with those obtained from stepwise multiple linear regressions (protein: R2 = 0.83, RMSE = 3.91E-04 g/cm2 ; cellulose + lignin: R2 = 0.66, RMSE = 2.02E-04 g/cm2 ). Our results confirm the importance of selecting a proper spectral subset that contains sufficient information for a successful inversion. For the first time, we provide promising estimations of leaf protein in fresh leaves through inversion of a radiative transfer model, which can be applied at canopy level for regional mapping if coupled with a canopy reflectance model and air- or spaceborne hyperspectral imaging.